Cancer Immunotherapy for Hepatocellular Carcinoma: Current Drug Strategies, Resistance, Biomarkers, and Future Directions.

Introduction

Hepatocellular carcinoma (HCC) accounts for most primary liver cancers and remains a leading cause of cancer-related death worldwide. Its incidence is driven by several major etiologic factors, including chronic hepatitis B virus (HBV) or hepatitis C virus (HCV) infection, alcohol-related liver disease, and metabolic dysfunction-associated steatotic liver disease.1,2 Most HCCs arise in the setting of chronic liver disease, frequently with underlying cirrhosis. Treatment must therefore control the malignancy without precipitating hepatic decompensation.

This dual burden of cancer and liver dysfunction distinguishes HCC from many other solid tumors. It also complicates the interpretation of survival, treatment-related toxicity, and radiologic response.3–5 These differences in outcome are not random. Hepatic reserve determines treatment tolerance, the risk of competing liver-related mortality, and eligibility for subsequent therapy.

Disease etiology may influence immune activation, antigenicity, and the tolerogenic liver microenvironment. Geographic treatment patterns further affect access to combination regimens, surveillance, locoregional therapy, and post-progression options, thereby shaping both the populations enrolled in trials and the results observed in routine practice.

For more than a decade, sorafenib was the reference systemic treatment for advanced HCC.6,7 Lenvatinib later emerged as a non-inferior first-line alternative.8,9 However, both agents produced modest survival benefits, and deep or durable tumor responses were uncommon.10 The introduction of immune-checkpoint inhibitors changed the therapeutic landscape.11 Early studies of nivolumab and pembrolizumab showed that a minority of patients could achieve substantial and durable tumor regression.12,13 Nevertheless, subsequent randomized trials also revealed the limitations of checkpoint-inhibitor monotherapy. Nivolumab and Pembrolizumab did not meet the prespecified threshold for a statistically significant overall survival (OS).14,15 These findings redirected development from single-agent checkpoint inhibition toward combination strategies designed to overcome multiple barriers to antitumor immunity. This transition produced several effective first-line treatment platforms.

Atezolizumab plus bevacizumab combines PD-L1 blockade with inhibition of vascular endothelial growth factor (VEGF). The STRIDE regimen combines durvalumab with a single priming dose of the cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) inhibitor tremelimumab. Camrelizumab plus rivoceranib pairs PD-1 blockade with potent VEGFR2 inhibition.

More recently, nivolumab plus ipilimumab demonstrated an OS benefit over lenvatinib or sorafenib in the phase III CheckMate 9DW trial (Figure 1).16–19 Recent guidelines from the European Association for the Study of the Liver (EASL), the American Association for the Study of Liver Diseases (AASLD), and the Asian Pacific Association for the Study of the Liver (APASL) now recommend these ICI-based combinations as first-line standards for eligible patients.20,21 Contemporary reviews have further consolidated the evidence supporting these regimens, while emphasizing that treatment selection must integrate liver function, bleeding risk, and patient-specific contraindications. However, not all combination strategies have succeeded. The phase III LEAP-002 trial of lenvatinib plus pembrolizumab and the COSMIC-312 trial of cabozantinib plus atezolizumab failed to meet their primary endpoints, highlighting that biological rationale and higher response rates do not always translate into survival benefits.22,23 Recent single-cell and translational studies have begun to define molecular subgroups—such as “immune-competent” versus “angiogenesis-driven”—that may explain differential responses to these regimens and guide future biomarker development.24 These divergent results will be discussed in Immunologic Context–Immunotherapy in Earlier Disease.

Figure 1Evolution of systemic immunotherapy for unresectable HCC. The timeline summarizes major therapeutic milestones in HCC, beginning with the pre-immunotherapy era (sorafenib, 2007) and the first wave of immune checkpoint inhibitor (ICI) monotherapy (nivolumab, CheckMate 040; pembrolizumab, KEYNOTE-224). The combination breakthrough was marked by atezolizumab plus bevacizumab (IMbrave150), followed by dual-checkpoint strategies (STRIDE in HIMALAYA; nivolumab plus ipilimumab in CheckMate 9DW) and PD-1/VEGFR2 inhibition (camrelizumab plus rivoceranib, CARES-310).

The most recent phase reflects ongoing refinement and expansion, including strategies after first-line therapy, earlier-stage disease, and novel agents and combinations, with a focus on subsequent therapy, biomarkers, and patient selection.A horizontal timeline with six dated sections. 2007-2016, Pre-Immunotherapy Era: Sorafenib (2007); Modest survival benefit; Limited treatment options. 2017-2019, First Wave (ICI Monotherapy): Nivolumab (CheckMate 040); Pembrolizumab (KEYNOTE-224); Proof of concept for immune checkpoint inhibition in HCC.

2020, Combination Breakthrough: Atezolizumab plus Bevacizumab (IMbrave 150); Significant OS improvement; New standard of care established. 2022, Dual Checkpoint Strategies: Durvalumab plus Tremelimumab (HIMALAYA); Nivolumab plus Ipilimumab (CheckMate 9DW); Validated dual checkpoint therapy options. 2023, PD-1/VEGFR2 Inhibition: Camrelizumab plus Rivoceranib (CARES-310); Confirmed activity of PD-1 plus VEGFR2 combination in phase III trial.

2024- and Beyond, Refinement & Expansion: Strategies after first-line therapy; Earlier stage disease; Novel agents and combinations; Focus on subsequent therapy, biomarkers and, patient selection.A timeline infographic of systemic therapy milestones for unresectable hepatocellular carcinoma.Notes: Cross-trial comparisons in this Figure are hypothesis-generating only and do not establish superiority of one regimen over another.

Despite these advances, most patients eventually experience disease progression. Emerging evidence has identified tumor-associated macrophages, B cells, and metabolic reprogramming as key contributors to primary and acquired resistance, offering potential targets for next-generation combination therapies.25–27 The expanding treatment landscape has also introduced clinical questions that were less prominent during the sorafenib era. The optimal regimen may differ according to portal hypertension and variceal bleeding risk, autoimmune disease, previous liver transplantation, metabolic liver disease, or impaired hepatic reserve.

Additional uncertainties concern whether major tumor regression can enable conversion to resection or locoregional therapy, how long treatment should continue in responding patients, and which therapy should be selected after progression on a first-line immunotherapy-based combination. Recent phase Ib and Phase II studies of conversion therapy in intermediate-stage HCC suggest that systemic ICI-based treatment followed by resection may be feasible in selected patients, although prospective validation is ongoing.28 The present review addresses these questions systematically. Literature Approach outlines our literature approach.

Immunologic Context provides the immunologic context, covering immune tolerance and heterogeneity, angiogenesis as an immune barrier, and checkpoint biology. Clinical Development of Immune-Checkpoint Therapy appraises the clinical development of immune-checkpoint therapy, from early monotherapy through combination regimens, including regimen selection and response assessment. Immunotherapy in Earlier Disease examines immunotherapy in earlier disease, including adjuvant, neoadjuvant/perioperative, and locoregional strategies.

Resistance to Immunotherapy discusses resistance mechanisms, encompassing tumor-intrinsic, microenvironmental, acquired, and host- and liver-related factors. Biomarkers and Precision Immunotherapy focuses on biomarkers and precision immunotherapy, including tissue, blood-based, microbiome, imaging, and artificial intelligence approaches. Emerging Immune Therapies explores emerging immune therapies such as adoptive cell therapy, vaccines, oncolytic approaches, and bispecific antibodies.

Underrepresented Populations and Hepatic Reserve addresses underrepresented populations and hepatic reserve. Subsequent Therapy and Real-World Evidence reviews subsequent therapy and real-world evidence. Future Directions outlines future directions.

The central premise of this review is that treatment failure in HCC comprises not only tumor-driven resistance but also liver-related competing mortality and exhaustion of hepatic reserve. We therefore use hepatic reserve as an organizing thread and evaluate efficacy alongside liver-related competing death. This narrative review examines cancer immunotherapy for HCC, with particular emphasis on current drug strategies, pharmacologic and immunologic rationale, pivotal clinical evidence, mechanisms of resistance, and biomarker development.

It also discusses emerging immune therapies. Cholangiocarcinoma and other primary hepatic malignancies fall outside the main scope of this review.

Literature Approach

This narrative review was developed through targeted searches of PubMed for English-language publications available through May 31, 2026. One updated IMbrave050 analysis published in June 2026 was added during revision because of its direct relevance to adjuvant therapy and its potential impact on the interpretation of the IMbrave050 findings. To ensure completeness, we also hand-searched the reference lists of pivotal trials and key reviews, and checked ClinicalTrials.gov for ongoing or completed trials of interest.

We did not systematically search Embase, Web of Science, the Cochrane Library, conference abstracts, or regulatory documents, as this is a narrative review rather than a systematic evidence synthesis or pooled analysis.

Search terms included “hepatocellular carcinoma”, “immune checkpoint inhibitor”, “PD-1”, “PD-L1”, “CTLA-4”, “VEGF”, “phase III trial”, “neoadjuvant therapy”, “adjuvant therapy”, “transarterial chemoembolization”, “biomarker”, “liquid biopsy”, “treatment resistance”, “CAR-T cell”, “cancer vaccine”, and related terms and combinations when additional evidence was required for specific topics.

Main inclusion criteria were as follows: (1) English-language, peer-reviewed publications; (2) clinically interpretable outcomes relevant to HCC immunotherapy; and (3) sufficient methodological detail to assess the strength of evidence. Studies were excluded if they were conference abstracts, editorials, commentaries, case reports, or not published in English.

Randomized phase III trials were prioritized when comparative efficacy and safety data were available, because they provide the most reliable estimates of relative treatment effects and are least susceptible to selection and confounding biases. Phase I and II studies were included when they established an important therapeutic concept or evaluated an emerging strategy for which phase III evidence was unavailable. Translational studies were selected when they connected a defined immune mechanism with human HCC samples, treatment response, resistance, or clinical outcomes.

When trials addressing similar questions produced conflicting results, we did not pool them or declare a single winner. Instead, we compared their eligibility criteria, control-arm design, endpoint definitions (including dual primary endpoints and alpha allocation), baseline liver function, patient and etiologic composition, subsequent therapies, and maturity of follow-up to explain the divergence. This approach was intended to preserve the complexity of the evidence rather than oversimplify it.

To clarify the strength of the evidence underlying each statement, we indicate the primary source of evidence for key conclusions throughout the review. Statements derived from phase III randomized controlled trials are identified as the highest level of comparative evidence. Findings from phase I/II single-arm studies are labeled as preliminary or proof-of-concept evidence.

Conclusions drawn from retrospective or real-world cohorts are identified as hypothesis-generating and subject to selection bias. Preclinical or translational studies are cited only to provide mechanistic context and are not presented as clinical evidence. Where evidence is conflicting or immature, this is stated explicitly.

Immunologic Context

Immune Tolerance and Heterogeneity

The liver is continuously exposed to dietary antigens, microbial products, and other molecules delivered from the gastrointestinal tract through the portal circulation. To limit inappropriate inflammation against these recurring signals, the hepatic immune environment maintains a context-dependent bias toward tolerance. Liver sinusoidal endothelial cells, Kupffer cells, dendritic cells, hepatocytes, and hepatic stellate cells contribute through unconventional antigen presentation, inhibitory signaling, cytokine production, deletion or dysfunction of activated lymphocytes, and the induction of regulatory T-cell states.29–34

Chronic liver injury can reinforce this tolerogenic environment. Persistent viral infection, steatohepatitis, alcohol-related injury, fibrosis, and cirrhosis generate continuous antigen exposure and inflammatory signaling. Over time, this process can produce immune dysfunction rather than effective immune clearance.35–37 Regulatory T cells, myeloid-derived suppressor cells, immunosuppressive tumor-associated macrophages, and dysfunctional natural killer cells may accumulate during HCC development.4,38 Effector CD8⁺ T cells may also be present, but persistent antigen stimulation and inhibitory receptor signaling can drive progressive functional exhaustion (Figure 2A).37,39Figure 2Immunologic of HCC: tolerance, angiogenesis and checkpoint biology.

(A) Immune tolerance and heterogeneity. The tolerogenic liver microenvironment, shaped by chronic injury and portal circulation, promotes immunosuppression and Treg induction. HCC exhibits three immune phenotypes: active, exhausted, and poor (cold).

(B) Angiogenesis as an immune barrier and anti-angiogenic therapy. Tumor-derived VEGF drives immune exclusion by impairing DC maturation, suppressing endothelial activation, and limiting T-cell trafficking. Appropriate VEGF inhibition/vascular normalization restores immune infiltration and reduces hypoxia, whereas excessive pruning may worsen it.

(C) CTLA-4 and PD-1/PD-L1 blockade are not interchangeable. PD-1/PD-L1 blockade acts on primed/exhausted T cells in tumor, blood, and lymph nodes, restoring effector function. CTLA-4 blockade acts during early T-cell priming in lymph nodes, enhancing CD28 costimulation and broadening the TCR repertoire.

These distinct mechanisms make the two strategies non-interchangeable. Upward arrows (↑) indicate an increase/upregulation; and downward arrows (↓) indicate a decrease/downregulation. Image created with BioRender.com, with permission.(A) Immune tolerance and heterogeneity.

Gut-to-liver portal circulation delivers dietary antigens and microbial products to the liver; chronic injury drivers listed (HBV, HCV, steatohepatitis, alcohol-related injury, fibrosis, cirrhosis). A boxed tolerogenic liver microenvironment contains Kupffer cells, dendritic cells, hepatocytes, hepatic stellate cells and liver sinusoidal endothelial cells, with arrows to unconventional antigen presentation, inhibitory signaling and cytokine production, leading to lymphocyte dysfunction and Treg induction. Side icons show Treg, MDSC, immunosuppressive TAM accumulation and dysfunctional NK cells.

Persistent antigen exposure drives PD-1, TIM-3 and LAG-3 signaling and CD8 superscrip plus T-cell exhaustion. Three HCC immune phenotypes depicted: active, exhausted, poor (cold). (B) Angiogenesis as immune barrier.

Tumor-derived VEGF impairs dendritic cell maturation, suppresses endothelial activation and limits T-cell trafficking; appropriate VEGF inhibition/vascular normalization restores infiltration and reduces hypoxia, whereas excessive pruning worsens both. (C) CTLA-4 versus PD-1/PD-L1 blockade, not interchangeable. PD-1/PD-L1 blockade acts on primed or exhausted T cells in tumor, blood and lymph nodes, restoring effector function; CTLA-4 blockade acts during early priming in lymph nodes, enhancing CD28 costimulation and broadening TCR repertoire. up arrow, increase/upregulation; down arrow, decrease/downregulation.Three-part schematic of HCC immune tolerance, vascular endothelial growth factor effects and checkpoint blockade.Abbreviations: HCC, hepatocellular carcinoma; HBV, hepatitis B virus; HCV, hepatitis C virus; LSEC, liver sinusoidal endothelial cell; Treg, regulatory T cell; MDSC, myeloid-derived suppressor cell; TAM, tumor-associated macrophage; NK, natural killer; PD-1, programmed cell death protein 1; PD-L1, programmed death-ligand 1; TIM-3, T-cell immunoglobulin and mucin domain-containing protein 3; LAG-3, lymphocyte activation gene 3; TIGIT, T-cell immunoreceptor with Ig and ITIM domains; CD8+, cluster of differentiation 8 positive; VEGF, vascular endothelial growth factor; VEGFR, vascular endothelial growth factor receptor; DC, dendritic cell; CTLA-4, cytotoxic T-lymphocyte-associated antigen 4; APC, antigen-presenting cell; CD80/86, cluster of differentiation 80/86; CD28, cluster of differentiation 28; IFN-γ, interferon-gamma; TNF-α, tumor necrosis factor-alpha.

HCC is not immunologically uniform. Single-cell and spatial studies have identified heterogeneous exhausted and regulatory T-cell states, together with distinct macrophage, dendritic-cell, endothelial, fibroblast, and tumor-cell neighborhoods.4,40 Transcriptomic analyses have identified an immune class that accounts for approximately one quarter of HCCs and contains biologically distinct active and exhausted subtypes (Figure 2A).41,42 Multi-region sequencing further shows that immune classification can vary across different areas of the same tumor.43–45

Angiogenesis as an Immune Barrier

HCC is typically a highly vascular tumor, and VEGF signaling is a central component of its biology. VEGF has important immunologic effects in addition to promoting angiogenesis. Tumor-derived VEGF can inhibit dendritic-cell differentiation and maturation.46 It can also suppress endothelial activation, reduce the expression of adhesion molecules and T-cell-recruiting chemokines, and thereby limit lymphocyte adhesion and entry into tumor tissue.

VEGF signaling may further support suppressive myeloid and regulatory immune programs.47–50

These effects provide a mechanistic rationale for combining VEGF inhibition with immune-checkpoint blockade. Appropriately dosed inhibition of the VEGF pathway may transiently normalize tumor vasculature, improve perfusion, reduce hypoxia, and facilitate lymphocyte trafficking. It may also reprogram tumor-associated macrophages and other suppressive myeloid populations.

Through these mechanisms, anti-angiogenic treatment can restore several steps of the cancer-immunity cycle rather than acting solely by restricting tumor blood supply (Figure 2B).49,51–53

The effects of anti-angiogenic therapy should not be interpreted as a uniform class effect. The biological consequences of vascular inhibition depend on the molecular target, dose, schedule, and duration of treatment. Vascular-normalizing exposure may improve perfusion and immune-cell access, whereas excessive vascular pruning can worsen hypoxia and restrict immune infiltration.54 Antibodies that neutralize VEGF and small-molecule inhibitors that target VEGFR together with additional kinases may therefore produce different vascular and immunologic effects.

CTLA-4 and PD-1/PD-L1 Blockade are Not Interchangeable

PD-1 and PD-L1 blockade predominantly restores or sustains the function of antigen-experienced T cells that have been restrained by chronic PD-1 signaling.55 Responsive populations include progenitor-like exhausted CD8⁺ T cells that retain proliferative capacity, as well as related tumor-reactive clones in tumor tissue and tumor-draining lymph nodes. PD-1 pathway inhibition can therefore act within the tumor, peripheral blood, and lymphoid compartments. Its effects are not restricted to the terminal effector phase of the immune response.

CTLA-4 has a prominent role during T-cell priming and early clonal expansion.55 It limits CD28-dependent costimulation by binding the shared ligands CD80 and CD86 with high avidity and by reducing their availability on antigen-presenting cells. CTLA-4 blockade can consequently enhance T-cell activation, promote clonal expansion, and broaden the peripheral T-cell receptor repertoire.56–60

This mechanistic distinction is clinically relevant but should not be interpreted as an absolute separation of function. Conceptually, PD-1 or PD-L1 blockade may be most effective when tumor-reactive T cells have already been generated but are functionally constrained. CTLA-4 blockade may provide additional benefit when insufficient priming, limited clonal diversity, or regulatory T-cell activity remains an important barrier.

However, both pathways act across overlapping anatomical and developmental compartments, and their relative contribution is likely to vary among patients (Figure 2C).

Accordingly, efficacy and safety should be interpreted at the regimen level. Indirect comparisons must account for differences in eligibility criteria, control-arm composition, CTLA-4 exposure, follow-up duration, subsequent treatment, geographic distribution, and underlying liver-disease etiology (Figure 2). The absence of a head-to-head trial means that HIMALAYA and CheckMate 9DW cannot establish the superiority of one dual-checkpoint strategy over the other.

Clinical Development of Immune-Checkpoint Therapy

Key efficacy and safety data for the pivotal trials discussed in this section are summarized in Table 1; trial-level eligibility, geographic distribution, and etiologic subgroups are compared in Table 2. The mechanistic rationale for these distinct combination strategies—particularly the non-interchangeable biology of CTLA-4 and PD-1/PD-L1 blockade—is illustrated in Figure 2C and discussed in CTLA-4 and PD-1/PD-L1 Blockade are Not Interchangeable.Table 1Pivotal Randomized Trials Shaping Systemic Immunotherapy for Unresectable HCCRegimenAtezolizumab + BevacizumabTremelimumab + Durvalumab (STRIDE)Camrelizumab + RivoceranibNivolumab + IpilimumabTrial(Year, Region)IMbrave150(2020, Global)HIMALAYA(2022, Global)CARES-310(2023, Global)CheckMate 9DW(2025, Global)Distinctive featuresPD-L1+VEGF blockade Established standardPD-L1+CTLA-4 blockade, single priming dose of TremelimumabPD-1+VEGFR2 inhibitionPotent anti-angiogenicPD-1+ CTLA-4 blockadeN (Exp/Ctrl)336/165393/389272/271335/333ControlSorafenibSorafenibSorafenibSorafenib or LenvatinibMedian follow-up, mo17.6 vs10.433.18 vs 32.2322.1 vs 14.952.5Median OS, mo (95% CI)19.2 vs 13.416.43 vs 13.7723.8 vs 15.223.7 vs 20.6OS HR (95% CI)0.66 (0.52–0.85)0.78 (0.65–0.93) †0.64 (0.52–0.79)0.79 (0.65–0.96)OS p value<0.0010.0035<0.00010.018Median PFS, mo6.9 vs 4.33.78 vs 4.075.6 vs 3.79.1 vs 9.2PFS HR (95% CI)0.65 (0.53–0.81)0.90 (0.77–1.05)0.54 (0.44–0.67)0.87 (0.72–1.06)ORR, %27.3 vs 11.920.1 vs 5.126.8 vs 5.936 vs 14CR, %5.5 vs <03.1 vs 01.8 vs 0.74 vs 1DCR, %73.6 vs 55.360.1 vs 60.777.9 vs 53.5NRGrade 3/4 TRAE, %43 vs 4650.5 vs 52.427.4 vs 18.441 vs 42Discontinuation due to TRAE, %15.5 vs 10.38.2 vs 11.027.2 vs 4.818 vs 10PMID32402160; 349025303831989237499670; 4130867640349714Notes: Data are presented as experimental arm versus control arm. †: HIMALAYA OS HR is reported with a 96.02% CI; other HRs are reported with 95% CIs. DCR was not reported for CheckMate 9DW (NR).

Event numbers and subsequent treatment proportions were not uniformly reported in primary publications. Cross-trial comparisons in this table are hypothesis-generating only and do not establish superiority of one regimen over another.Abbreviations: HCC, hepatocellular carcinoma; OS, overall survival; PFS, progression-free survival; ORR, objective response rate; CR, complete response; DCR, disease control rate; TRAE, treatment-related adverse event; HR, hazard ratio; CI, confidence interval; NR, not reported; STRIDE, Single Tremelimumab Regular Interval Durvalumab; Exp/Ctrl, experimental/control; vs, versus.Table 2Comparison of Eligibility, Exclusion, Geographic Distribution, and Etiologic Subgroups in Pivotal First-Line Immunotherapy Trials for Unresectable HCCTrialIMbrave150 (Atezo + Bev)HIMALAYA (STRIDE)CARES-310 (Camre + Rivo)CheckMate 9DW (Nivo + Ipi)Key inclusion criteriaAge≥18 years≥18 years≥18 years≥18 yearsHCC diagnosisHistologically or cytologically confirmed, or clinically diagnosed according to AASLD criteria in patients with cirrhosisHistologically confirmed, BCLC B (not eligible for locoregional) or CHistologically/cytologically confirmed, BCLC B or C, not amenable to or progressing after surgery/locoregional therapyHistologically confirmed, unresectable or metastatic, not eligible for curative surgical or locoregional therapy, or progressed after such therapyPrior systemic therapyNoneNoneNoneNoneLiver functionChild-Pugh AChild-Pugh AChild-Pugh AChild-Pugh AECOG PS0–10–10–10–1Measurable lesion≥1 untreated lesion per RECIST v1.1≥1 measurable lesion per RECIST v1.1≥1 measurable lesion per RECIST v1.1≥1 measurable untreated lesion per RECIST v1.1Key exclusion criteriaMain portal vein invasionExcluded (Vp4)Excluded (main trunk thrombosis)Not excluded per se; Vp4 permittedExcluded (Vp4)Varices/bleedingUntreated/high-risk varices excluded; EGD requiredActive/prior GI bleeding within 12 months excludedEGD required per protocolPortal hypertension with history of bleeding due to esophageal or gastricvarices within 6 monthsAutoimmune diseaseExcludedExcludedExcludedExcludedHBV/HCV coinfectionHBV/HCV coinfection excludedHBV/HCV coinfection excludedNot specified as exclusionActive HBV/HCV or HBV/HDV coinfection excludedHepatic encephalopathyExcludedExcludedNot specifiedHistory within 12 months excludedGeographic distributionRegionsGlobal (Asia excluding Japan vs ROW)Global (Asia excluding Japan vs ROW)Global (83% Asian; 17% non-Asian)Global (Asia 40%, America/Europe 43%, ROW 17%)No. of countries/sites17 countries, 111 sites16 countries, 181 sites13 countries, 95 sites25 countries, 163 sitesEtiologic subgroupsHBV~49% vs 46%~31% vs 30.6%76% vs 73%~34% vs 35%HCV~21% vs 22%~28% vs 27.5%8% vs 11%~28% vs 29%Non-viral~30% vs 32%~41% vs 41.9%15% vs 17%~37% vs 36%Etiology as stratification factorNoYes (HBV vs HCV vs non-viral)NoYes (HBV vs HCV vs uninfected)Notes: Interpretive notes: Generalizability: IMbrave150 and CheckMate 9DW enrolled globally with substantial non-Asian representation. HIMALAYA enrolled ~41% from Asia (excluding Japan).

CARES-310 was predominantly Asian (83%), which may limit generalizability to non-Asian populations with different etiologic profiles. Etiology: CARES-310 enrolled predominantly HBV-related HCC, reflecting the Asian epidemiology. In contrast, IMbrave150, HIMALAYA, and CheckMate 9DW enrolled more balanced proportions of HBV, HCV, and non-viral etiologies.

Exclusion criteria: IMbrave150 and CheckMate 9DW excluded Vp4 portal vein invasion, whereas CARES-310 permitted it. This difference may affect the applicability of results to patients with major vascular invasion. Cross-trial comparisons in this table are hypothesis-generating only and do not establish superiority of one regimen over another.Abbreviations: HCC, hepatocellular carcinoma; Atezo, atezolizumab; Bev, bevacizumab; Camre, camrelizumab; Rivo, rivoceranib; Nivo, nivolumab; Ipi, ipilimumab; STRIDE, Single Tremelimumab Regular Interval Durvalumab; AASLD, American Association for the Study of Liver Diseases; BCLC, Barcelona Clinic Liver Cancer; ECOG PS, Eastern Cooperative Oncology Group performance status; RECIST, Response Evaluation Criteria in Solid Tumors; EGD, esophagogastroduodenoscopy; GI, gastrointestinal; Vp4, tumor thrombosis involving the main portal vein trunk; ROW, rest of world; NAFLD, non-alcoholic fatty liver disease; vs, versus.

Early Monotherapy: Proof of Principle and Its Limits

CheckMate 040 (single-arm phase I/II) provided the first strong clinical signal for nivolumab in advanced HCC. The objective response rate (ORR) was 15% in the dose-escalation cohort and 20% in the dose-expansion cohort. Responses were observed in patients previously treated with sorafenib and across viral and non-viral etiologic subgroups, and several responses were durable.12 KEYNOTE-224 (single-arm phase I/II) subsequently reported a confirmed ORR of approximately 17% with pembrolizumab in patients previously treated with sorafenib.13 These single-arm studies established proof of principle that HCC can respond to PD-1 blockade.

They also supported accelerated regulatory approvals in selected jurisdictions, although the regulatory status of individual indications subsequently evolved.

RATIONALE-301 (Phase III RCT) evaluated tislelizumab versus sorafenib as first-line monotherapy in 674 patients with unresectable HCC. The trial met its primary endpoint of OS non-inferiority (median OS, 15.9 vs 14.1 months; HR 0.85; 95% CI 0.71–1.02), although superiority was not demonstrated. ORR was 14.3% versus 5.4%, and median duration of response was 36.1 versus 11.0 months.55 These findings support PD-1 monotherapy as a reasonable option when combination therapy is contraindicated or unavailable, while confirming that monotherapy does not achieve the response rates of effective combinations.

In the second-line setting, KEYNOTE-394 (Phase III RCT) demonstrated that pembrolizumab significantly improved OS versus placebo in an Asian population previously treated with sorafenib or oxaliplatin-based chemotherapy (median OS, 14.6 vs 13.0 months; HR 0.79; 95% CI 0.63–0.99).61 A meta-analysis of KEYNOTE-240 and KEYNOTE-394 confirmed the pooled OS benefit (HR 0.79; 95% CI 0.67–0.93), with an ORR of 15.4% versus 2.8%.62

These trials support two conclusions. First, PD-1 monotherapy has clinically meaningful and sometimes durable activity (Figure 1). It may remain an option in selected settings when combination therapy is contraindicated or unavailable, subject to regional approval and treatment context.

Second, durable responses in a minority of patients do not necessarily translate into population-level survival superiority in an unselected trial population. Current development has therefore shifted toward rational combinations and biomarker-informed patient selection.

Anti-PD-(L)1 Plus VEGF Inhibition

IMbrave150 randomly assigned patients with previously untreated, unresectable HCC to atezolizumab plus bevacizumab or sorafenib.16 The combination significantly improved OS, progression-free survival (PFS), and ORR compared with sorafenib, and delayed deterioration in patient-reported quality of life and functioning (Table 1).63,64 It subsequently became a widely adopted first-line standard for eligible patients.

ORIENT-32, a randomized phase II/III study conducted in China, evaluated sintilimab plus a bevacizumab biosimilar (IBI305) versus sorafenib in 571 patients with unresectable HBV-related HCC. The combination significantly improved both OS and PFS versus sorafenib (median PFS, 4.6 vs 2.8 months; HR 0.56; 95% CI 0·46-0·70).65 Updated analyses reported an ORR of 21.0% versus 4.7% by RECIST 1.1 and 25.1% versus 7.7% by mRECIST, with a median duration of response of 20.3 months. These results extend the evidence for PD-1 plus VEGF inhibition to a predominantly HBV-related, Chinese population, although cross-trial comparisons with global trials remain hypothesis-generating only.

Before bevacizumab-containing therapy, patients with cirrhosis should undergo guideline-concordant assessment for portal hypertension and gastroesophageal varices. High-risk varices should be treated before therapy is started. Recent clinically significant bleeding, uncontrolled hypertension, severe proteinuria, major thromboembolic risk, impaired wound healing, or other contraindications to VEGF inhibition may favor a non-VEGF regimen (Table 2).20,66

Dual-Checkpoint Blockade

The phase III HIMALAYA trial compared the STRIDE regimen—a single 300-mg priming dose of tremelimumab combined with durvalumab, followed by durvalumab every 4 weeks—with durvalumab monotherapy or sorafenib.17 STRIDE significantly improved OS compared with sorafenib (Table 1), and durvalumab monotherapy met the prespecified criterion for non-inferiority to sorafenib. The trial excluded patients with tumor thrombosis involving the main portal-vein trunk (Table 2), which limits direct extrapolation to patients with Vp4 portal-vein tumor thrombosis.

CheckMate 9DW (Phase III RCT) compared nivolumab plus ipilimumab with investigator-selected lenvatinib or sorafenib in patients with previously untreated, unresectable HCC.19 An early crossing of the OS curves was observed, but the survival benefit favoring nivolumab plus ipilimumab became evident with longer follow-up (Table 1). The safety profile was consistent with the known profile of dual-checkpoint blockade; grade 3/4 treatment-related adverse events were balanced between arms, and immune-mediated adverse events were manageable with established algorithms (Table 1). Eligibility criteria, exclusion of Vp4 portal-vein tumor thrombosis, and global enrollment are summarized in Table 2.

Nivolumab plus ipilimumab improved OS and produced an ORR of approximately 36%, including complete responses in about 7% of patients. Responses were substantially more durable than those achieved with lenvatinib or sorafenib. CheckMate 9DW established nivolumab plus ipilimumab as an effective first-line platform, with significant OS improvement and durable responses.

These trials were not designed for head-to-head comparison; indirect comparisons are hypothesis-generating only and cannot establish superiority of one dual-checkpoint strategy over another.

PD-1 Blockade Plus VEGFR Inhibition

The phase III CARES-310 trial compared camrelizumab plus rivoceranib with sorafenib in patients with previously untreated, unresectable HCC. The combination significantly improved PFS, OS and the OS benefit was maintained with longer follow-up (Table 1).18,67 CARES-310 validated camrelizumab plus selective VEGFR2 inhibition as an effective therapeutic platform. However, its efficacy must be balanced against toxicity, monitoring requirements, dose modification, and regional availability.

Not every checkpoint-inhibitor–kinase-inhibitor combination has demonstrated survival superiority. LEAP-002 compared pembrolizumab plus lenvatinib with lenvatinib plus placebo. The combination produced numerically favorable OS and PFS results, but neither primary endpoint crossed its prespecified statistical boundary (Table 1).22 The strong performance of the lenvatinib control arm reduced the incremental separation between the treatment groups.

COSMIC-312 (phase III RCT) compared atezolizumab plus cabozantinib with sorafenib. The trial used dual primary endpoints (PFS and OS) with prespecified alpha allocation and separate statistical boundaries. It met its PFS endpoint at the primary analysis but did not improve OS at the final analysis.23,68 This discordance may reflect the different statistical rules governing the two endpoints, as well as differences in kinase selectivity, dose intensity, control-arm activity, patient and etiologic composition, treatment discontinuation, subsequent therapy, and statistical assumptions.

The negative or mixed results of LEAP-002, COSMIC-312, CheckMate 459, and KEYNOTE-240 illustrate how statistical design and trial conduct can influence whether a true treatment effect is detected. LEAP-002 and KEYNOTE-240 allocated most alpha to OS and minimal alpha to PFS, and both narrowly missed their prespecified boundaries despite numerically favorable results. CheckMate 459 and COSMIC-312 were confounded by higher subsequent therapy use, including immune checkpoint inhibitors (ICIs), in the control arm, which likely diluted the OS benefit.

Control-arm strength (lenvatinib in LEAP-002) and patient selection further contributed. These trials should not be interpreted as evidence of biological inactivity; rather, they highlight the need to interpret negative results in the context of alpha allocation, control-arm activity, subsequent therapy, and follow-up maturity. Cross-trial comparisons remain hypothesis-generating only (Table 1 and Table 2).

Regimen Selection and Response Assessment

No randomized head-to-head trial has compared the major first-line immunotherapy combinations (Table 1 and Table 2). Treatment selection must therefore integrate the populations studied in the pivotal trials with the expected toxicity profile, tumor tempo, hepatic reserve, comorbidities, drug availability, and patient preference.20

Atezolizumab plus bevacizumab has mature global phase III evidence but it requires careful assessment of portal hypertension, gastroesophageal varices, recent bleeding, hypertension, proteinuria, thromboembolic risk, and wound healing (Table 2). STRIDE avoids VEGF-pathway inhibition and limits CTLA-4 exposure to a single priming dose, making it an important option when bevacizumab is unsuitable. Nivolumab plus ipilimumab produces a high ORR, including complete and durable responses, but requires intensive monitoring and timely management of immune-mediated toxicity.

Camrelizumab plus rivoceranib has demonstrated significant PFS and OS benefits, although VEGFR2-related adverse events and dose-management requirements are substantial. PD-1 or PD-L1 monotherapy may remain reasonable in selected settings when combination therapy is contraindicated or unavailable, subject to regional approval and treatment context. Response rates have generally been lower than those reported with effective combination regimens.

Underlying liver-disease etiology should not, by itself, determine treatment selection. Translational studies and exploratory meta-analyses have raised concern that non-viral HCC, particularly steatohepatitis-associated HCC, may be less sensitive to checkpoint inhibition. In preclinical NASH-HCC models, PD-1 blockade expanded activated or dysfunctional CD8⁺ T-cell populations that promoted liver injury without restoring effective antitumor surveillance.69 However, the accompanying clinical analyses were exploratory, and non-viral HCC represents a heterogeneous group that includes metabolic, alcohol-related, cryptogenic, and other etiologies.

Clinical evidence has remained inconsistent. Some observational studies have reported poorer outcomes with immune-checkpoint inhibition in non-viral HCC, whereas real-world cohorts of patients treated with atezolizumab plus bevacizumab have not consistently identified etiology as an independent determinant of survival or treatment duration.70 Responses and durable disease control have been documented in patients with non-viral and steatohepatitis-associated HCC. Etiology is therefore biologically relevant and should be incorporated into prospective stratification and translational analyses, but it is not currently a validated predictive biomarker for excluding patients from checkpoint-inhibitor therapy or selecting among first-line regimens.

Tumor response in HCC is commonly assessed using RECIST 1.1 and, in selected settings, modified RECIST (mRECIST). RECIST 1.1 measures changes in total lesion diameter, whereas mRECIST measures the enhancing viable component of target HCC lesions.71,72 Immunotherapy-based regimens, particularly those combined with anti-angiogenic therapy, may reduce arterial enhancement without producing an equivalent early reduction in total lesion size. Necrosis, hemorrhage, perfusion changes, and prior embolization can further complicate radiologic interpretation.

Available HCC cohorts suggest that true pseudoprogression is uncommon. In a retrospective multicenter study of 158 nivolumab-treated patients, no pseudoprogression was identified among patients who underwent repeat imaging after initial unconfirmed progression.73 However, evidence remains limited for contemporary immunotherapy combinations, and rare cases have been reported.

Treatment beyond radiologic progression may be considered in carefully selected patients who remain clinically stable, retain adequate hepatic reserve, and have equivocal, limited, or mixed progression that can be reassessed promptly. The rationale may include uncertainty about true progression, continued regression of other lesions, oligoprogression amenable to local treatment, or ongoing clinical benefit. A retrospective/real-world multicenter HCC cohort suggested that selected patients can achieve subsequent disease control after continuing anti-PD-1 therapy beyond initial progression, although treatment-selection bias limits causal interpretation.74

Continuation is generally inappropriate when progression is accompanied by clinically significant symptom worsening, deteriorating performance status, rising bilirubin, new or worsening ascites or encephalopathy, rapid vascular progression, or other evidence of declining hepatic reserve. In HCC, the opportunity to initiate an effective next-line therapy may be lost quickly once hepatic decompensation occurs.

Conversion decisions should consider the original extent and biology of disease, the depth and duration of response, residual vascular invasion, extrahepatic disease, future liver remnant, portal hypertension, hepatic reserve, and the likelihood that residual radiologic abnormalities contain viable tumor. Future prospective studies should define technical and oncologic conversion criteria before treatment begins, specify the required duration of disease control, and analyze surgery or local therapy as a time-dependent intervention rather than retrospectively classifying only successful cases.

Immunotherapy in Earlier Disease

Adjuvant Therapy

Recurrence remains common after curative-intent resection or ablation. After resection, large tumor size, multifocality, microvascular invasion, satellite nodules, and poor differentiation are consistently associated with increased recurrence risk.75–77 These findings provide a strong clinical rationale for adjuvant treatment in patients with high-risk disease.

IMbrave050 (Phase III RCT) compared 1 year of atezolizumab plus bevacizumab with active surveillance in patients at high risk of recurrence after curative-intent resection or ablation (Table 3). At the prespecified interim analysis, the combination significantly improved independently assessed recurrence-free survival.78 However, longer follow-up materially changed the interpretation. In the updated analysis, the RFS hazard ratio was 0.90, and the initial benefit was no longer maintained.

OS remained immature.79Table 3Immunotherapy in Earlier-Stage and Locoregional HCC: Key Studies Meeting Inclusion CriteriaStudy (Setting; Design; Reference)Population/Treatment/ComparatorKey Findings and LimitationsIMbrave050 (Adjuvant; Phase III RCT; PMID 37871608, 41580093)High-risk recurrence after curative resection/ablation (n = 668); Atezolizumab + bevacizumab × 1 year vs active surveillanceInterim: improved RFS (HR 0.72; 95% CI 0.53–0.98; p=0.012). Updated: RFS HR 0.90 (95% CI 0.72–1.12), benefit not maintained; OS immature. Limitations: early benefit not durable; mature follow-up required.Nivolumab ± ipilimumab (Neoadjuvant/perioperative; Randomized Phase II; PMID 35065057)Resectable HCC (n=27); Perioperative nivolumab alone or + ipilimumab20/27 resected; no surgery delayed due to grade ≥3 TRAE; MPR (≥70% necrosis) 3/9 vs 3/11.

Limitations: small, selected patients; pathologic response not validated as surrogate.Cabozantinib + nivolumab (Neoadjuvant/perioperative; Phase 1b; Translational; PMID 34796337)Resectable HCC (n = 15); Perioperative cabozantinib + nivolumabImmune-rich aggregates and TLS in pathologic responders; distinct response/resistance programs. Limitations: small, translational; hypothesis-generating.EMERALD-1 (Locoregional: TACE; Phase III RCT; PMID 39798579)Unresectable HCC eligible for embolization (n = 157); Durvalumab + bevacizumab + TACE vs TACE + placeboPFS 15.0 vs 8.2 mo; durva + TACE alone not significant; OS immature. Limitations: no bevacizumab-with-TACE control; OS immature.LEAP-012 (Locoregional: TACE; Phase III RCT; PMID 39798578)Unresectable, non-metastatic HCC (n = 847); Pembrolizumab + lenvatinib + TACE vs TACE + dual placeboPFS 14.6 vs 10.0 mo; increased toxicity; OS immature.

Limitations: OS immature; toxicity requires intensive management.PETAL (Locoregional: TACE; Phase Ib; PMID 38578610)HCC after TACE (n = 15); Pembrolizumab following TACEORR 53%; median OS 33.5 mo. Limitations: phase Ib; single-arm; hypothesis-generating.PEMRAD (Locoregional: SBRT; Phase II; PMID 41585952)Advanced HCC after sorafenib (n = 18); Pembrolizumab + SBRTPublished phase II experience; activity in advanced HCC post-sorafenib. Limitations: phase II; no phase III RCT; investigational.Notes: Evidence notes: Phase III RCT evidence is available for adjuvant IMbrave050 and for TACE combinations (EMERALD-1, LEAP-012).

Neoadjuvant/perioperative and SBRT evidence remains phase II or translational. For Y-90 and thermal ablation combined with ICIs, no peer-reviewed clinical trial meeting the inclusion criteria of this review was identified. Cross-trial comparisons are hypothesis-generating only and do not establish superiority of one strategy over another.Abbreviations: HCC, hepatocellular carcinoma; RCT, randomized controlled trial; RFS, recurrence-free survival; OS, overall survival; PFS, progression-free survival; ORR, objective response rate; TRAE, treatment-related adverse event; MPR, major pathologic response; TLS, tertiary lymphoid structures; TACE, transarterial chemoembolization; SBRT, stereotactic body radiotherapy; ICI, immune checkpoint inhibitor; Y-90, yttrium-90 radioembolization.

The trial also highlights fundamental challenges in adjuvant HCC development. Post-treatment recurrence can arise through at least two biological routes. Some recurrences are clonally related to the original tumor and represent residual intrahepatic or systemic dissemination.

Others are genetically independent tumors that arise de novo from the chronically diseased liver. Paired genomic and multi-omics studies have directly demonstrated these distinct evolutionary patterns.80,81 A transient delay in recurrence is therefore not equivalent to a durable reduction in recurrence risk.

Future adjuvant studies require mature follow-up and reproducible definitions of high recurrence risk. They should incorporate candidate measures of molecular residual disease, including circulating tumor DNA and other tumor-informed biomarkers, while recognizing that these tools are not yet validated surrogate endpoints in HCC. Analyses should also distinguish early and late recurrence, report recurrence patterns and salvage therapy, and use time-dependent methods that differentiate a temporary delay in recurrence from a durable reduction in recurrence risk.

The IMbrave050 experience—an early recurrence-free survival benefit that was not maintained with longer follow-up—demonstrates that adjuvant trials require mature follow-up and time-dependent analyses to distinguish transient delay from durable benefit.

Neoadjuvant and Perioperative Strategies

Neoadjuvant therapy offers a biological and translational opportunity that differs from postoperative treatment. The intact tumor may provide a continuing source of antigen during immune activation, while subsequent resection permits direct assessment of residual viable tumor and treatment-induced changes in the immune microenvironment. However, these proposed advantages remain biological hypotheses rather than established predictors of long-term clinical benefit.

A randomized phase II study evaluated perioperative nivolumab alone or nivolumab plus ipilimumab in 27 patients with resectable HCC (Table 3). The study established feasibility and proof of biological activity, but it was not large enough to determine whether adding ipilimumab improved pathologic response or long-term outcomes.82

Pathologic response definitions in neoadjuvant HCC trials remain inconsistent. MPR is variously defined as ≥70% or ≥90% tumor regression, and a cross-trial patient-level analysis identified ≥90% regression as the optimal threshold for predicting relapse-free survival (RFS).83,84 pCR, defined as no viable tumor, occurred in 18% of patients and was associated with longer RFS, but is less common than MPR. Necrosis percentage and viable tumor percentage are complementary, not interchangeable.

Importantly, pathologic response has not been validated as a trial-level surrogate for OS in prospective randomized trials.85 It should therefore be regarded as a promising intermediate endpoint for translational research, not as a validated surrogate for survival.

Perioperative studies are particularly valuable for translational research because they can integrate pretreatment biopsy, serial blood sampling, interval imaging, and the complete resection specimen. Analyses from the cabozantinib–nivolumab study identified immune-rich aggregates and tertiary lymphoid structures in pathologic responders, while subsequent spatial transcriptomic work identified distinct programs associated with response, resistance, and recurrence.86,87 Tissue and imaging analyses from perioperative PD-1 studies have similarly begun to define candidate markers of pathologic response.88

These early findings are encouraging but cannot define routine practice. The studies were small, included selected patients with adequate hepatic reserve, and were conducted at highly experienced centers. Pathologic response definitions also differed among trials, and pathologic response has not yet been validated as a surrogate for RFS or OS in HCC.

Cross-trial comparisons of pathologic response rates are further limited by differences in eligibility, treatment duration, surgical timing, and pathologic assessment, and should be considered hypothesis-generating.

Locoregional Treatment Combinations

Locoregional therapies—including transarterial chemoembolization (TACE), radioembolization, thermal ablation, and external-beam radiotherapy (including stereotactic body radiotherapy, SBRT)—can induce immunogenic cell death and modulate the tumor microenvironment. These effects provide a rationale for combining locoregional treatment with immune checkpoint blockade. However, the direction and magnitude of immune modulation differ substantially according to modality, treated volume, dose, and underlying liver environment.89,90 We therefore discuss each modality separately.

Transarterial Chemoembolization (TACE)

TACE induces ischemic necrosis and antigen release, and hypoxia upregulates PD-L1/PD-1, creating a therapeutic window for checkpoint blockade. Two phase III randomized trials met inclusion criteria. EMERALD-1 showed that durvalumab plus bevacizumab with TACE significantly improved PFS versus TACE plus placebo, whereas durvalumab with TACE alone did not show the same benefit.91 LEAP-012 showed that pembrolizumab plus lenvatinib with TACE significantly improved PFS versus TACE plus dual placebo, with increased toxicity.92 A phase Ib study (PETAL) reported promising activity for TACE followed by pembrolizumab.93 OS remains immature for the phase III trials.

These studies are summarized in Table 3.

SBRT and External-Beam Radiotherapy

SBRT induces immunogenic cell death, activates cGAS–STING, and upregulates ZBP1. A phase II trial (PEMRAD) evaluating pembrolizumab plus SBRT in advanced HCC after sorafenib met inclusion criteria and is summarized in Table 3.94 Other single-arm and early-phase studies suggest promising activity, but no phase III randomized trial has been published. These combinations remain investigational.

Y-90 Radioembolization and Thermal Ablation

Y-90 radioembolization and thermal ablation can induce immunogenic and immunomodulatory effects that provide a rationale for combining them with ICIs, but no peer-reviewed clinical trial meeting the inclusion criteria of this review was identified for either modality in HCC. Both combinations therefore remain investigational, and no efficacy or safety conclusions can be drawn from published peer-reviewed evidence.

Cross-Modality Considerations

Studies of immunotherapy plus locoregional therapy are heterogeneous in design, patient selection, technique, sequencing, and endpoints. Cross-trial comparisons are hypothesis-generating only and do not establish superiority of one strategy over another. Table 3 summarizes the studies meeting inclusion criteria.

Resistance to Immunotherapy

Tumor-Intrinsic and Microenvironmental Resistance

Primary resistance can arise at several stages of the antitumor immune response. A tumor may fail to generate or present sufficiently immunogenic antigens; tumor-reactive T cells may be generated but fail to enter the tumor; or infiltrating lymphocytes may remain functionally suppressed despite PD-1 or PD-L1 blockade. In HCC, these defects can be produced by ineffective antigen presentation, dendritic-cell insufficiency, immunosuppressive myeloid populations, spatial stromal barriers, abnormal vasculature, and inhibitory VEGF-, TGF-β-, adenosine-, hypoxia-, or metabolite-dependent programs.

These mechanisms frequently coexist and may vary among lesions within the same patient.95–97

HCC cells can also actively manipulate antigen presentation to escape immune surveillance. Beyond MHC-I loss and antigen-processing defects, tumor-intrinsic programs can alter antigen handling and display, reducing CD8⁺ T-cell recognition. In HCC preclinical models, targeting Annexin A2 was recently shown to reactivate tumor-associated antigen presentation and relieve immune tolerance in liver cancer.98 These findings suggest that ineffective antigen presentation is not only a static genetic defect but also a dynamically regulated, potentially targetable barrier.

Prospective validation is needed to determine whether such interventions restore sensitivity to checkpoint blockade.

Activation of WNT/β-catenin signaling is among the best-characterized tumor-intrinsic mechanisms of immune exclusion. In HCC preclinical models, tumor-intrinsic β-catenin activation impairs the recruitment of cross-presenting dendritic cells, thereby limiting tumor-specific T-cell priming and subsequent intratumoral T-cell accumulation.99 Pan-cancer analyses have also associated WNT/β-catenin pathway activation with a non-T-cell-inflamed phenotype.100 In HCC models, β-catenin activation promoted immune escape and resistance to anti-PD-1 therapy, providing direct disease-specific mechanistic evidence.101

An early clinical genomic study associated WNT/β-catenin pathway alterations with poor disease control and shorter survival during checkpoint blockade in HCC.102 However, these findings do not justify excluding an individual patient from immunotherapy solely on the basis of CTNNB1 status. CTNNB1 genotype does not invariably capture functional β-catenin activity, and its association with treatment outcome has differed according to assay, tissue source, sample timing, treatment regimen, and definition of resistance. Small clinical cohorts have also shown that some patients with CTNNB1-mutated HCC can derive benefit from regimens that combine checkpoint and VEGF-pathway inhibition.

WNT/β-catenin activation should therefore be regarded as a biologically credible candidate resistance biomarker rather than a validated treatment-exclusion marker.

An immune-cold phenotype can also reflect vascular or stromal exclusion rather than an absence of recognizable tumor antigens (Figure 3A). TGF-β-active fibroblast programs can retain T cells within the peritumoral stroma and prevent their penetration into tumor-cell regions, while abnormal vessels can impair perfusion, adhesion, and lymphocyte trafficking.103,104 In HCC clinical, immune and angiogenic phenotypes are partly interconnected, further supporting the concept that vascular biology can shape immunotherapy sensitivity.105Figure 3Primary resistance, acquired resistance, and host/liver-related factors contributing to immunotherapy failure in hepatocellular carcinoma (HCC). (A) Primary resistance (pre-existing): (i) Low antigen/MHCI presentation (poor T-cell recognition); (ii) Wnt/β-catenin activation (DC exclusion, reduced T-cell recruitment); (iii) VEGF-driven aberrant vessels/hypoxia (limited lymphocyte trafficking); (iv) Treg/MDSC/TAM-mediated immunosuppression; (v) Fibrotic stroma/physical barrier (immune-cold phenotype).

(B) Selection during therapy: PD-1/PD-L1 blockade releases T cells to attack tumors, but immune pressure selects resistant states. (C) Acquired resistance (initial response then progression): (i) Immune editing/neoantigen loss; (ii) MHC-I loss (β2M/HLA alterations); (iii) IFN-γ–JAK–STAT defects (JAK1/2, STAT1/2 mutations); (iv) Progressive T-cell exhaustion (upregulation of PD-1, LAG-3, TIM-3, TIGIT); (v) Alternative checkpoints and MDSC/TAM recruitment; (vi) Intratumoral heterogeneity/resistant subclone expansion (oligoprogression). (D) Host/liver factors: Declining liver function; portal hypertension/decompensation (ascites, bleeding, HE, inflammation); antibiotics (dysbiosis); corticosteroids (immunosuppression); malnutrition; sarcopenia; gut microbiome dysbiosis (associated with poor ICI outcomes).

Solid arrows indicate direct promotion/progression; red crosses/T-bars indicate inhibition or blockade; upward arrows (↑) indicate an increase/upregulation; and downward arrows (↓) indicate a decrease/downregulation. Image created with BioRender.com, with permission.(A) Primary resistance: (i) tumor cells with low or absent MHC class I antigen presentation; (ii) Wnt/β-catenin activation excluding dendritic cells and reducing T-cell recruitment; (iii) VEGF-driven aberrant vessels and hypoxia (↑hypoxia, ↓lymphocytes); (iv) Treg/MDSC/TAM-mediated suppression, shown as T-bars; (v) fibrotic stroma excluding T cells. (B) During PD-1/PD-L1 blockade (crossed-out interaction), released T cells attack tumor cells while immune pressure selects resistant states.

(C) Acquired resistance: (i) neoantigen loss, pre- vs post-treatment clones; (ii) MHC class I loss, pre- vs post-treatment; (iii) IFN-γ–JAK–STAT alterations (JAK1/2, STAT1/2); (iv) T-cell exhaustion (↑PD-1, LAG-3, TIM-3, TIGIT); (v) alternative checkpoints with MDSC/TAM recruitment; (vi) resistant subclone expansion. (D) Host/liver factors listed as text: declining liver function, decompensation (ascites, bleeding, hepatic encephalopathy), antibiotics/dysbiosis, corticosteroids, malnutrition, sarcopenia. Solid arrows, promotion; red crosses/T-bars, inhibition; up arrow/down arrow, increase/decrease.Multi-panel schematic of immunotherapy failure mechanisms in hepatocellular carcinoma.Abbreviations: HCC, hepatocellular carcinoma; MHC-I, major histocompatibility complex class I; VEGF, vascular endothelial growth factor; Treg, regulatory T cell; MDSC, myeloid-derived suppressor cell; TAM, tumor-associated macrophage; PD-1, programmed cell death protein 1; PD-L1, programmed death-ligand 1; IFN-γ, interferon-gamma; JAK1/2, Janus kinase 1/2; STAT1/2, signal transducer and activator of transcription 1/2; LAG-3, lymphocyte activation gene 3; TIM-3, T-cell immunoglobulin and mucin domain-containing protein 3; TIGIT, T-cell immunoreceptor with Ig and ITIM domains.

Metabolic dysregulation provides an additional barrier to antitumor immunity. Iron overload and altered sphingolipid/S1P signaling can reprogram myeloid and lymphoid cells and promote HCC progression. In HCC preclinical models, Spns2, an S1P transporter, is essential for iron-regulated HCC metastasis, linking metabolic stress to immune evasion.26 These constraints may act alongside vascular and stromal exclusion, although their contribution to checkpoint-inhibitor resistance requires prospective validation.

Acquired Resistance

Acquired resistance refers to disease progression after an initial objective response or a clinically meaningful period of disease control during immune-checkpoint blockade (Figure 3B and C). Direct HCC clinical evidence for these mechanisms remains limited; most evidence derives from HCC preclinical models and other tumor types. Resistance may arise through immune editing and loss of immunogenic neoantigens, impairment of HLA class I antigen processing or presentation, disruption of IFN-γ–JAK–STAT signaling, or selective expansion of resistant tumor clones.

Additional candidate mechanisms include progressive T-cell dysfunction, compensatory expression of alternative inhibitory receptors, and remodeling toward a more suppressive myeloid or stromal microenvironment.

The clinical pattern of acquired resistance may be systemic or spatially restricted. Discordant progression among lesions may reflect resistant subclones or organ-specific microenvironments rather than complete loss of systemic immune activity. Oligoprogression therefore raises the possibility of treating the resistant sites locally while continuing a systemic regimen that remains effective elsewhere.

In a multicenter retrospective cohort of HCC patients who had achieved durable benefit from immunotherapy, oligoprogression was common among those who subsequently progressed, and locoregional treatment was associated with longer survival.106 However, the nonrandomized design and strong treatment-selection bias prevent causal conclusions. Prospective studies are needed to define the number and location of lesions suitable for local treatment, the role of continued immunotherapy, and the appropriate comparator strategy.

Rechallenge with immune-checkpoint inhibitors after acquired resistance is an emerging clinical strategy and is discussed in Subsequent Therapy and Real-World Evidence.

Host- and Liver-Related Treatment Failure

Tumor progression is not the only reason that immunotherapy fails to produce durable clinical benefit in HCC. In HCC clinical, declining hepatic reserve can interrupt treatment, preclude subsequent therapy, or cause competing mortality before antitumor benefit becomes clinically meaningful. Hepatic decompensation should therefore be treated as a distinct and clinically relevant outcome rather than being subsumed entirely under tumor progression or OS.107

HCC-specific cohorts have associated concurrent antibiotic exposure with poorer survival during immunotherapy.108 Nevertheless, antibiotic use has also been associated with adverse outcomes among patients receiving non-ICI therapies, suggesting that it may partly reflect infection, hepatic decompensation, or general disease severity rather than an ICI-specific resistance mechanism.109 Necessary antimicrobial therapy should therefore not be withheld to preserve immunotherapy efficacy. The more appropriate clinical objective is to avoid unnecessary exposure while treating infections promptly and effectively.

Malnutrition and sarcopenia may reduce physiologic reserve and the ability to sustain treatment. In a retrospective cohort of HCC patients treated with a PD-1 inhibitor, sarcopenia was associated with shorter PFS, although its relationship with OS and objective response was less consistent after adjustment.110 A broader meta-analysis across solid tumors has also associated sarcopenia with poorer outcomes during checkpoint blockade.111 These findings do not establish sarcopenia as an ICI-specific resistance biomarker. In HCC clinical, low muscle mass frequently coexists with cirrhosis, hypoalbuminemia, systemic inflammation, reduced performance status, and limited access to subsequent treatment.

The gut microbiome represents another plausible host determinant. Landmark studies in melanoma and other cancers demonstrated prognostic associations between gut-microbial composition and response to PD-1 blockade and provided experimental evidence that microbiota transfer can alter antitumor immunity.112,113 Small HCC clinical cohorts have also identified dynamic microbial signatures associated with anti-PD-1 response. However, the reported taxa and composite indices have not been consistently reproduced or prospectively validated for clinical selection.

Microbiome composition is influenced by liver-disease etiology, cirrhosis severity, geography, diet, hospitalization, antibiotics, proton-pump inhibitors, lactulose, rifaximin, and other medications (Figure 3D).

These host factors cluster strongly with advanced cirrhosis and are difficult to separate from general prognosis. An observed association with survival does not establish that a factor specifically modifies the biological efficacy of immunotherapy. Predictive relevance requires evidence that the factor has a differential effect during checkpoint blockade compared with an appropriate non-ICI control, ideally supported by a treatment-by-factor interaction.

Host and liver characteristics should therefore not be treated as immutable exclusion biomarkers. Prospective studies should test whether modifiable interventions—such as optimized management of portal-hypertension complications, infection prevention, nutritional support, physical rehabilitation, and rational antimicrobial stewardship—preserve hepatic reserve, increase delivered treatment exposure, or improve quality of life and survival. Microbiome-directed interventions require randomized evaluation before routine use.

Such studies should distinguish improvement in immune sensitivity from improvement in general health, treatment tolerance, or competing liver-related outcomes.

None of these host factors has been validated as a predictive biomarker for selecting or excluding immunotherapy in HCC. Their reported associations are predominantly prognostic or reflect general disease severity, and predictive utility would require demonstration of a differential treatment effect compared with an appropriate non-ICI control.

Biomarkers and Precision Immunotherapy

Throughout this section, we distinguish three biomarker categories. A prognostic biomarker is associated with clinical outcome regardless of treatment. A pharmacodynamic biomarker changes during treatment and reflects biological effect but does not establish differential benefit.

A predictive biomarker identifies differential benefit from a specific treatment relative to an alternative, ideally through a prespecified biomarker-by-treatment interaction in a randomized trial. To date, no validated predictive biomarker is available to guide regimen selection for HCC immunotherapy.

Tissue Biomarkers

PD-L1 expression has shown enrichment value in some HCC cohorts but has insufficient sensitivity, reproducibility, and treatment-specific predictive validation for routine patient selection. In CheckMate 040, responses occurred in both PD-L1-positive and PD-L1-negative tumors, although higher baseline expression was associated with greater response probability in exploratory analyses.12,114 A negative result should therefore not be used to exclude a patient from checkpoint-inhibitor therapy.

Most HCCs have a relatively low or intermediate tumor mutational burden, and MSI-high or mismatch-repair-deficient disease is exceptionally uncommon. In a genomic series of 755 advanced HCCs, the median tumor mutational burden was approximately four mutations per megabase, fewer than 1% of tumors were classified as TMB-high, and only one of 542 evaluable tumors was MSI-high.115 Tumor mutational burden has not been prospectively validated as a regimen-specific selection biomarker in HCC (Table 4).Table 4Candidate Biomarkers for HCC ImmunotherapyDomainRepresentative CandidatesPrimary RoleMain LimitationClinical ReadinessTissue immune statePD-L1, inflammatory signatures, tertiary lymphoid structures, spatial immune architectureBiological stratification and hypothesis generationSampling heterogeneity, assay variation, and absent prospective treatment interactionsInvestigationalTumor genomicsWNT/β-catenin, TMB, MSI/dMMR, antigen-presentation pathwaysMechanistic classification; rare tumor-agnostic indication for MSI/dMMRLow prevalence or insufficient treatment-specific validation in HCCLimited/context dependentBloodAFP, NLR and related inflammatory scores, ctDNA, circulating immune phenotypesPrognosis, disease monitoring, and research pharmacodynamicsCirrhosis-related confounding, pre-analytical variation, and uncertain thresholdsPrognostic use only (non-predictive)MicrobiomeDiversity, taxa, and functional signaturesResearch on the gut-liver-immune axisSmall cohorts and strong dietary, drug, geographic, and cirrhosis effectsInvestigationalImaging and AIRadiomics, deep learning, multimodal modelsNon-invasive risk and response modelingRetrospective design, leakage, limited external validation, and weak calibration evidenceInvestigationalNotes: No biomarker listed in this table is currently validated as a predictive biomarker for selecting among approved immunotherapy regimens in HCC. Clinical readiness definitions: Investigational: Not validated for routine clinical use; evidence remains exploratory or early-phase.

Limited/context dependent: Evidence restricted to specific contexts (eg, rare tumor-agnostic indications) and not broadly applicable. Prognostic use only (non-predictive): Associated with clinical outcome regardless of treatment, but does not demonstrate differential benefit between treatments.Abbreviations: HCC, hepatocellular carcinoma; PD-L1, programmed death-ligand 1; TMB, tumor mutational burden; MSI, microsatellite instability; dMMR, deficient mismatch repair; AFP, alpha-fetoprotein; NLR, neutrophil-to-lymphocyte ratio; ctDNA, circulating tumor DNA; AI, artificial intelligence.

Rare dMMR/MSI-high HCCs may nevertheless be sensitive to PD-1 blockade under tumor-agnostic treatment principles. Across solid tumors, mismatch-repair deficiency has been associated with high neoantigen burden and clinically meaningful responses to checkpoint inhibition.116 However, HCC-specific efficacy estimates remain uncertain because this molecular subgroup is extremely small. Neither low TMB nor microsatellite stability should be interpreted as evidence that an HCC cannot respond, because viral antigens, antigen clonality, HLA presentation, and the immune microenvironment are not captured by TMB alone.

Spatial context is essential when interpreting tissue immunity. The same number of CD8⁺ T cells may have different implications when the cells directly contact malignant cells, remain confined to the invasive margin, accumulate around vessels, or are separated from tumor nests by fibrotic stroma. Single-cell and spatial analyses have demonstrated complex immune, stromal, and malignant-cell neighborhoods in primary liver cancers.117 Highly multiplexed imaging in HCC has further shown that tumor–immune cellular proximity can provide information beyond cell density alone.118

Clinical translation remains limited by tissue availability, sampling bias, platform-specific resolution, antibody-panel and computational variation, cost, turnaround time, and the absence of prospectively validated decision thresholds. Many spatial studies also use resection specimens from early-stage disease, whereas treatment decisions in advanced HCC are often based on small archival biopsies. Future biomarker trials should therefore prespecify assay platforms and thresholds, use independent validation cohorts, incorporate serial or multisite sampling where feasible, and formally test treatment-by-biomarker interactions rather than reporting outcome associations within a single treatment arm.

Blood-Based Biomarkers and Liquid Biopsy

Alpha-fetoprotein remains an established prognostic and disease-monitoring marker in HCC. In patients with an elevated and measurable baseline AFP concentration, an early decline during systemic therapy has been associated with radiologic response and improved survival, as a pharmacodynamic marker, but it does not predict differential benefit between treatments. An exploratory analysis of GO30140 and IMbrave150 showed that early AFP dynamics could help identify response or progression during atezolizumab plus bevacizumab treatment.119 However, AFP kinetics are not informative in AFP-nonproducing tumors, and the optimal threshold and assessment time vary among studies.

AFP has not been validated as a treatment-specific predictor of relative benefit from immune-checkpoint blockade over another systemic therapy.

Inexpensive peripheral inflammatory markers—including the neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, C-reactive protein (CRP), the CRP-to-albumin ratio, and the systemic immune-inflammation index—have repeatedly been associated with response and survival in HCC. The CRAFITY score, which combines baseline CRP and AFP, was developed and validated as a prognostic and risk-stratification tool in patients receiving PD-1 or PD-L1 immunotherapy and was associated with radiologic response, PFS, and OS.120 Nevertheless, it remains primarily a prognostic and risk-stratification tool rather than a validated predictor of treatment-specific benefit.

Circulating tumor DNA (ctDNA) may provide a more direct and dynamic estimate of tumor-derived molecular burden and clonal evolution. Baseline ctDNA burden is being studied primarily as a prognostic marker, whereas early decline or clearance of tracked tumor-derived variants may function as a pharmacodynamic measure of treatment response. In patients receiving atezolizumab plus bevacizumab, early changes in plasma cfDNA and detectable tumor-derived variants were associated with radiologic response and clinical outcomes.121 These findings suggest that molecular response may complement imaging, but ctDNA-guided treatment switching has not been shown to improve survival.

Clonal hematopoiesis can introduce plasma variants that are not tumor-derived. Matched leukocyte sequencing can reduce this source of error but does not eliminate all uncertainty. Tumor-informed assay design, adequate plasma input, standardized collection and processing, predefined sampling time points, error-suppressed sequencing, and prospectively specified analytical thresholds are required.

Current expert recommendations recognize the prognostic validity of molecular residual-disease detection in several cancers but do not support routine treatment escalation or de-escalation on the basis of ctDNA without evidence of clinical utility.122 In HCC, ctDNA should currently be used primarily in clinical trials and structured translational studies rather than as a stand-alone trigger for switching, intensifying, or discontinuing therapy.

Correlation with response or survival is insufficient to establish clinical utility. A useful predictive biomarker must identify differential benefit from one treatment relative to an appropriate alternative and must improve a defined clinical decision. This ideally requires a prespecified biomarker-by-treatment interaction within a randomized trial or a randomized biomarker-guided strategy.

Future liquid-biopsy studies should distinguish prognostic value, pharmacodynamic response monitoring, early detection of resistance, and true treatment-selection utility rather than grouping these functions under a single biomarker claim.

Microbiome, Imaging, and Artificial Intelligence

All microbiome- and imaging/AI-based findings discussed in this section remain investigational. They should not be used to guide clinical decisions outside of clinical trials or prospective registries.

The gut–liver axis provides a strong biological rationale for microbiome research in HCC. Microbial products, pathogen-associated molecular patterns, bile-acid derivatives, and other metabolites reach the liver through the portal circulation and can influence hepatic innate and adaptive immunity. Conversely, liver function, bile-acid metabolism, and portal hypertension can alter the intestinal environment and microbial ecosystem.123

Small HCC cohorts have produced inconsistent microbiome findings. An early longitudinal metagenomic study reported differences in microbial dynamics and candidate bacterial species between responders and nonresponders during anti-PD-1 therapy. In contrast, a prospective exploratory cohort of 36 patients did not identify a robust association between baseline microbial diversity or overall community composition and objective response.

Although several taxa were enriched among patients with prolonged disease control, their baseline abundance did not predict OS.124 These studies should therefore be considered hypothesis-generating rather than evidence for a clinically validated microbial signature.

Interpretation is complicated by the sensitivity of the microbiome to external and host-related factors. Diet, geography, household environment, medications, and health status account for substantial interindividual variation in microbial composition.125,126 Proton-pump inhibitors can independently alter the gut microbiome.127 Antibiotics, infection, hospitalization, and medications used to manage cirrhosis—including lactulose and rifaximin—may also modify microbial composition while simultaneously identifying patients with more advanced liver disease. Liver-disease etiology, portal-systemic shunting, ascites, hepatic encephalopathy, and decompensation should therefore be incorporated into microbiome analyses rather than treated as background noise.

Microbiome manipulation should not currently be recommended to improve immunotherapy outcomes in HCC. Fecal microbiota transplantation, defined microbial consortia, probiotics, prebiotics, and dietary interventions remain investigational in this setting.128 Prospective intervention studies must demonstrate not only changes in microbial composition but also improved treatment outcomes and acceptable liver-specific safety. Infection, bacterial translocation, hepatic encephalopathy, and multidrug-resistant organism transmission require particular attention in patients with cirrhosis.

Many published HCC radiomics models remain retrospective, single-center, and based on relatively small or highly selected datasets.129 Scanner manufacturer, acquisition protocol, reconstruction parameters, contrast dose and timing, respiratory phase, slice thickness, segmentation, image preprocessing, and feature-extraction software can materially alter the measured features. Systematic evaluations have shown that many radiomic features are not consistently repeatable or reproducible across technical settings.130,131

Beyond radiomics, AI and deep learning are increasingly applied to digital pathology. Models can extract spatial immune features from H&E whole-slide images—such as tumor-infiltrating lymphocyte density and tumor–immune proximity—and predict transcriptomic signatures of atezolizumab–bevacizumab response.132 Multimodal spatial omics with graph deep learning can also identify immune microenvironments linked to response or resistance.133 However, these models remain retrospective, platform-dependent, and unvalidated for prospective treatment selection; like radiomics, they should currently be considered investigational.

For precision immunotherapy, the decisive question is not whether imaging or AI can predict outcome within a treated cohort. It is whether a model identifies differential benefit from one regimen relative to another and improves a specific clinical decision. This requires randomized or carefully controlled datasets, prespecified model thresholds, comparison with established clinical variables, and prospective evaluation of biomarker-guided treatment selection (Table 4).

Likewise, no microbiome-based, imaging-based, or AI-derived biomarker is currently validated as predictive for regimen selection in HCC immunotherapy.

Emerging Immune Therapies

Adoptive Cell Therapy

Glypican-3 (GPC3) is expressed in a substantial subset of HCCs and is among the most clinically advanced targets for CAR-T-cell therapy in this disease. However, its expression varies across patients and may be heterogeneous within individual tumors. Early phase I studies of GPC3-directed CAR-T cells have demonstrated the feasibility of cell manufacturing, lymphodepletion, infusion, and post-treatment monitoring.

Objective responses have been observed, but only in a minority of heavily pretreated patients, and their durability and reproducibility remain uncertain.

In two early phase I studies involving 13 patients with advanced GPC3-positive HCC, conventional GPC3 CAR-T cells produced two partial responses and provided initial evidence of clinical feasibility.134 A subsequent phase I study evaluated GPC3 CAR-T cells engineered to coexpress RUNX3. Among six evaluable patients, one achieved a partial response and two achieved stable disease. All treated patients developed cytokine-release syndrome, including grade 3 events, although the events resolved with management and no immune-effector-cell-associated neurotoxicity syndrome was reported.135

Major barriers include heterogeneous or insufficient antigen expression, the potential for antigen-low or antigen-negative escape, inadequate trafficking into tumor tissue, limited cellular persistence, progressive T-cell dysfunction, and the immunosuppressive and metabolically hostile liver-tumor microenvironment. Cytokine-release syndrome is an established clinical toxicity, whereas the risk of on-target, off-tumor injury remains incompletely characterized for GPC3-directed products.

T-cell-receptor-engineered therapy expands the targetable antigen space by recognizing intracellular proteins processed and presented as peptide–HLA complexes. AFP-derived peptides are prominent candidate targets in HCC. Preclinical studies identified HLA-A2-restricted AFP-specific TCRs capable of redirecting human T cells against AFP-positive HCC cells.136 This approach subsequently progressed to first-in-human evaluation.

In a phase I trial of ADP-A2AFP, 21 participants, including 20 with advanced HCC, received lymphodepletion followed by affinity-enhanced AFP-specific TCR-T cells. The trial demonstrated a manageable safety profile and preliminary antitumor activity in appropriately selected HLA- and AFP-positive patients.137

HBV-derived antigens represent another important target in virus-associated HCC. Experimental studies established that HBV-specific TCR-engineered cells can recognize HBV-infected cells and HBV-related HCC.138 Early clinical reports subsequently demonstrated the feasibility of administering HBV-specific TCR-T cells in patients with HBV-related HCC, including patients with extrahepatic metastatic disease.139 A dose-escalation phase I study of short-term HBV-specific TCR expression reported generally manageable safety and preliminary clinical activity, including a prolonged partial response in one patient.140

Cancer Vaccines and Oncolytic Approaches

Therapeutic vaccines targeting AFP, GPC3, telomerase, shared tumor-associated antigens, or personalized neoantigens aim to generate or expand tumor-specific T-cell responses. HCC is an attractive setting because several relevant antigens are well characterized. However, established tumors contain abnormal vasculature, fibrotic and stromal barriers, suppressive myeloid populations, regulatory T cells, metabolic stress, and checkpoint-dependent inhibition that may prevent vaccine-induced lymphocytes from entering tumors or sustaining effective cytotoxic activity.

Personalized neoantigen vaccination may reduce some limitations of shared self-antigens by targeting mutations selected from an individual tumor. A single-arm phase I/II study evaluated a personalized DNA-plasmid vaccine encoding up to 40 neoantigens, together with plasmid-encoded interleukin-12 and pembrolizumab, in patients with advanced HCC previously treated with a multikinase inhibitor. The RECIST 1.1 ORR was 30.6%, including complete responses in three of 36 patients, and neoantigen-specific T-cell responses were detected in 19 of 22 evaluable patients.141

The principal developmental challenge is not simply the induction of circulating antigen-specific T cells. Effective vaccination must generate T cells that expand and persist, traffic into liver tumors, penetrate malignant-cell nests, recognize clonally and spatially heterogeneous cancer cells, and remain functional within an immunosuppressive and metabolically hostile cirrhotic liver. Peripheral ELISpot, tetramer, or cytokine responses do not by themselves establish that the relevant cells are present in sufficient numbers, in the correct spatial compartment, or in a functional state capable of tumor killing.

Oncolytic viruses provide a complementary approach. They can preferentially replicate within malignant cells, produce direct tumor-cell lysis, release tumor antigens, and activate innate and adaptive immune sensing. Pexa-Vec, an engineered vaccinia virus expressing granulocyte–macrophage colony-stimulating factor, has been the most extensively studied oncolytic platform in HCC.

The phase III PHOCUS trial compared Pexa-Vec followed by sorafenib with sorafenib alone in patients with advanced HCC who had not received previous systemic therapy. The study was terminated after an interim analysis failed to show an overall-survival benefit.142 This negative result demonstrates that early evidence of viral replication, immune activation, distant-lesion effects, and dose-related survival signals did not translate into confirmed survival improvement in a large randomized setting.

Oncolytic viruses therefore remain investigational in HCC. Future development should prioritize rational combinations supported by tumor and blood pharmacodynamic evidence, standardized measures of viral replication and immune conversion, and randomized comparisons against contemporary immunotherapy-based standards. Demonstration of local immune activation or tumor necrosis should not be treated as a surrogate for durable systemic disease control or survival.

Bispecific Antibodies and Emerging Checkpoints

Bispecific antibodies can redirect endogenous immune cells toward tumor-associated antigens or inhibit two immune-regulatory pathways within a single molecule. GPC3×CD3 T-cell engagers are among the most prominent antigen-directed bispecific strategies in HCC. By simultaneously binding GPC3 on malignant cells and CD3 on T cells, these agents can create an immune synapse and induce T-cell activation independently of the native T-cell-receptor specificity.

Emerging inhibitory pathways in HCC include LAG-3, TIGIT, TIM-3, and the CD39–CD73–adenosine axis. LAG-3 expression has been associated with dysfunctional tumor-infiltrating lymphocytes in HCC.143 Fibrinogen-like protein 1 was subsequently identified as an important inhibitory ligand of LAG-3, providing additional biological relevance to this pathway in liver cancer.144 Nevertheless, neither LAG-3 expression nor FGL1 abundance has been prospectively validated as a biomarker for LAG-3-directed therapy in HCC.

In the randomized phase II AdvanTIG-206 study, first-line ociperlimab plus tislelizumab and a bevacizumab biosimilar was compared with tislelizumab plus the bevacizumab biosimilar. Confirmed ORRs were 37.1% and 40.6%, respectively, indicating no improvement from adding TIGIT blockade.145 Grade 3 or higher treatment-related adverse events and treatment discontinuation were more frequent in the ociperlimab-containing group. This result provides direct evidence that adding another checkpoint inhibitor to an active PD-1–anti-VEGF backbone can increase treatment burden without improving efficacy.

Expression of a candidate checkpoint should not be equated with dependence on that pathway. A receptor may mark activation, exhaustion, or compensatory feedback without being the dominant constraint on antitumor immunity. Similarly, a response observed after adding a new antibody does not prove that the intended pathway was responsible unless target engagement and downstream immune modulation are demonstrated.

Early trials should therefore prespecify pharmacodynamic evidence appropriate to the target. TIGIT studies should evaluate receptor occupancy, CD226 signaling, and T-cell and NK-cell function. LAG-3 studies should assess LAG-3, FGL1, relevant cellular compartments, and downstream proliferative or effector recovery.

Adenosine-pathway studies should measure CD39 and CD73 activity, extracellular purine metabolites, adenosine-receptor signaling, and changes in suppressive myeloid and lymphoid states. GPC3×CD3 studies should link tumor-antigen density, drug exposure, immune-synapse formation, T-cell expansion, cytokine release, and tumor-cell killing.

Development should prioritize mechanistically defined and biomarker-enriched populations rather than adding another checkpoint inhibitor to an unselected combination. Biomarker enrichment should integrate target and ligand expression, cellular source, spatial localization, downstream pathway activity, and relevant counter-receptors rather than relying on a single immunohistochemical cutoff. Randomized trials should formally test whether the biomarker identifies incremental benefit from the matched intervention and should compare efficacy gains with the additional immune toxicity and loss of hepatic reserve (Table 5).Table 5Emerging Immune-Therapy Platforms in HCCPlatformRepresentative Target or StrategyKey Evidence ObservedPrincipal BarrierAdoptive cellsGPC3 CAR-T; AFP- or HBV-directed TCR-T; CIK cellsManufacturing feasibility confirmed; sporadic objective responses noted in early phaseAntigen heterogeneity, trafficking, persistence, manufacturing, HLA restriction, and liver-specific safetyVaccinesAFP, GPC3, telomerase, personalized neoantigensImmunogenicity (T-cell clonal expansion) verified; yet peripheral responses poorly correlate with tumor shrinkagePeripheral immune responses have not consistently translated into durable clinical benefitOncolytic virusesPexa-Vec/JX-594Local intratumoral bioactivity confirmed (Phase I/II); however, monotherapy survival benefit unproven (Phase III)Phase III failure, intratumoral delivery, neutralizing immunity, and cirrhosis-related safetyBispecifics and new checkpointsGPC3xCD3; TIGIT, LAG-3, TIM-3, CD39-CD73-adenosinePreclinical antitumor activity robust; early Phase I clinical signals are emergingCytokine release, target heterogeneity, pathway redundancy, and need for target-engagement evidenceNotes: All platforms listed are investigational and have not yet been approved for routine clinical use in HCC.

Evidence remains early-phase or preclinical, and cross-trial comparisons are hypothesis-generating only.Abbreviations: HCC, hepatocellular carcinoma; GPC3, glypican-3; CAR-T, chimeric antigen receptor T cell; AFP, alpha-fetoprotein; HBV, hepatitis B virus; TCR-T, T-cell receptor-engineered T cell; CIK, cytokine-induced killer cells; HLA, human leukocyte antigen; TIGIT, T-cell immunoreceptor with Ig and ITIM domains; LAG-3, lymphocyte activation gene 3; TIM-3, T-cell immunoglobulin and mucin domain-containing protein 3.

Drug Design and Pharmacokinetic Considerations

The clinical performance of emerging immunotherapies is shaped by molecular design. Bispecific antibodies require optimization of format, valency, affinity, Fc engineering, half-life, and CD3-binding affinity to balance antitumor activity against cytokine release. CAR-T and TCR-T constructs depend on the antigen-binding domain, costimulatory and signaling domains, safety switches, persistence, manufacturing, and HLA restriction.

For combination regimens, dosing sequence, exposure–response relationships, overlapping toxicity, drug-drug interactions, and anti-drug antibodies may determine whether a biologically plausible combination translates into clinical benefit. These design and pharmacokinetic features are therefore integral to the development of HCC immunotherapy.

Underrepresented Populations and Hepatic Reserve

Registrational trials of first-line immunotherapy-based regimens in advanced HCC predominantly enrolled patients with Child-Pugh A liver function, ECOG performance status 0–1, adequate hematologic and organ function, and protocol-defined control of chronic viral hepatitis. These eligibility criteria created relatively selected populations with preserved hepatic reserve. The atezolizumab–bevacizumab registration trial additionally required assessment of gastroesophageal varices and excluded patients with untreated or incompletely treated varices at high risk of bleeding because of the bevacizumab component.16 HIMALAYA and other dual-checkpoint trials similarly enrolled patients with preserved liver function and performance status, but did not reproduce every bevacizumab-specific bleeding-risk criterion.17

Routine practice includes older and frail adults, patients with Child–Pugh B cirrhosis, renal impairment, controlled HIV infection, pre-existing autoimmune disease, extensive portal-vein invasion, prior hepatic decompensation, and other comorbidities that were excluded or poorly represented in pivotal trials. Although pre-existing autoimmune disease has historically been an exclusion criterion, retrospective real-world data suggest that immune-checkpoint inhibitors can be cautiously used in selected patients whose autoimmune condition is quiescent and well controlled, with careful monitoring for disease flare and immune-related adverse events.146 These groups should not be treated as a single population. Their risks arise through different mechanisms, including limited hepatic reserve, treatment intolerance, competing mortality, infection, immune activation of pre-existing disease, vascular complications, and reduced access to subsequent therapy.

Real-world cohorts are essential for describing treatment delivery, effectiveness, and safety in these broader populations. They can identify actual treatment duration, dose modification, hepatic decompensation, uncommon toxicity, reasons for discontinuation, and access to subsequent therapy. However, comparative interpretation is limited by treatment-selection bias, heterogeneous imaging schedules, missing laboratory and toxicity data, differences in supportive care, and incomplete documentation of why a specific regimen was selected or stopped.

Real-world studies are therefore strongest for describing treatment exposure and outcomes in underrepresented populations; comparative-effectiveness claims require substantially stronger design and analysis.

The most consistent observation is that impaired hepatic reserve is associated with markedly shorter OS. Whether this reflects reduced antitumor activity, shorter treatment exposure, competing liver-related mortality, inability to receive subsequent therapy, or a combination of these factors varies among cohorts. Measurable tumor responses can occur in selected patients with Child-Pugh B liver function, but response rates differ substantially according to regimen, patient selection, and severity of hepatic dysfunction.

In a real-world cohort of 203 patients treated with nivolumab, the ORR was 2.8% in patients with Child-Pugh B liver function, compared with 15.9% among patients with Child-Pugh A, median OS was also substantially shorter in the Child-Pugh B group. Within Child-Pugh B, patients with scores of 8 or 9 had worse survival than those with a score of 7.147

A systematic review and meta-analysis of ICI-treated Child-Pugh B HCC similarly found that objective responses remained possible but that PFS and OS were substantially inferior to those observed in Child-Pugh A patients.148 The pooled evidence was dominated by observational studies with heterogeneous regimens, liver-function distributions, and patient-selection criteria. It therefore supports cautious feasibility in selected patients rather than establishing a standard treatment for unselected Child-Pugh B disease.

Child-Pugh B should not be treated as a homogeneous category. A patient with a score of 7 due primarily to modest hypoalbuminemia differs substantially from one with the same numerical score driven by active ascites or worsening bilirubin. Patients with recurrent encephalopathy, refractory ascites, recent variceal bleeding, or rapidly deteriorating synthetic function have a different risk profile from patients with stable Child-Pugh B7 disease.

The cause and reversibility of hepatic dysfunction may be as important as the numerical score: tumor-related liver impairment may improve with effective therapy, whereas advanced cirrhosis-related dysfunction may not. Child-Pugh B is a prognostic factor and should not be interpreted as a predictive biomarker for selecting among immunotherapy regimens.

In real-world cohorts, patients with Child-Pugh B liver function treated with immune-checkpoint inhibitors have consistently lower response rates and shorter OS than those with Child-Pugh A. Scores of 8–9 and active decompensation—ascites, encephalopathy, recent variceal bleeding, or worsening bilirubin—carry a particularly poor prognosis. Treatment decisions should consider individual Child-Pugh components, portal-hypertension complications, performance status, and the likelihood of receiving subsequent therapy.

No validated predictive biomarker currently guides regimen selection in this population.

Real-world studies should report baseline and longitudinal ALBI or modified ALBI grade, individual Child-Pugh components, ascites and encephalopathy status, portal-hypertension complications, platelet count, and the presumed cause of liver-function impairment. They should distinguish improvement, stability, and deterioration of hepatic reserve during treatment. Time to hepatic decompensation, cancer-specific and liver-related mortality, treatment exposure, relative dose intensity, reasons for discontinuation, discontinuation before first imaging, and receipt of subsequent therapy should be reported alongside response, PFS, and OS.

Liver transplant recipients require particular caution with ICIs, because the PD-1/PD-L1 pathway also maintains allograft tolerance and blockade may trigger rejection. In pre-transplant bridging, ICIs achieve downstaging in approximately 75–82% of patients, with 1-year and 3-year survival of approximately 95% and 70–80% (single-arm/retrospective data).149 However, rejection risk is 16–28% and is highest when the washout period is <90 days. An individual patient data meta-analysis confirmed a 26.4% rejection rate and identified washout duration as the key determinant; risk falls below 10% when washout exceeds 50 days.150,151 Post-transplant ICI use for recurrent HCC is limited to case reports and small series, with rejection in 20–25% and rejection-related mortality in 10%–12%.152 In a multicenter retrospective analysis of 52 patients, rejection occurred in 13% and was associated with lower survival; escalation of immunosuppression and calcineurin inhibitor use at ICI initiation reduced rejection risk (P = 0.04 and P = 0.013).153 No validated predictive biomarker identifies safe candidates, and prospective randomized data are lacking.

Decisions should therefore be individualized through multidisciplinary collaboration at experienced centers, and ICI use should remain confined to clinical trials or prospective registries where feasible.

Subsequent Therapy and Real-World Evidence

Treatment after progression on a first-line immunotherapy-based combination remains a major evidence gap in advanced HCC. The pivotal trials that established sorafenib, regorafenib, cabozantinib, and ramucirumab were conducted before contemporary first-line immune combinations became routine. The registration trials for regorafenib, cabozantinib, and ramucirumab required previous sorafenib exposure, whereas sorafenib itself was developed in the first-line setting.154–156 These studies demonstrate antitumor activity in HCC but do not directly establish the efficacy or optimal sequencing of these agents after atezolizumab plus bevacizumab, dual-checkpoint blockade, or an immune-checkpoint inhibitor combined with a kinase inhibitor.

Retrospective cohorts support the feasibility of sorafenib, lenvatinib, regorafenib, and cabozantinib after atezolizumab plus bevacizumab. A multicenter comparison suggested that lenvatinib produced longer PFS than sorafenib, with no clear overall-survival difference, but treatment allocation was not randomized.157 Additional real-world sequencing studies have reported clinical activity with several kinase inhibitors after atezolizumab plus bevacizumab, while also demonstrating substantial heterogeneity in liver function, treatment selection, and access to subsequent therapy.158 Ramucirumab remains a potential option for selected patients with α-fetoprotein concentrations of at least 400 ng/mL, but direct evidence after prior bevacizumab-containing immunotherapy is particularly limited.

These observational studies do not establish an optimal post-immunotherapy sequence. Most combine patients with primary resistance, acquired resistance, treatment intolerance, and different patterns of progression. Their results are also limited by small or uneven treatment groups, preserved-liver-function selection, immortal-time and treatment-selection bias, and the absence of randomized head-to-head comparisons.

Checkpoint-inhibitor rechallenge is also being used in clinical practice. Recent narrative and multicenter studies in HCC suggest potential benefit in selected patients, although evidence remains retrospective and heterogeneous.159,160 In particular, adding CTLA-4 blockade after prior PD-1 or PD-L1 therapy has produced objective responses in a subset of patients with HCC.161 Multicenter observational data have similarly shown that sequential immune-checkpoint treatment after atezolizumab plus bevacizumab can achieve renewed disease control in selected patients.162 However, no randomized trial has established the incremental benefit of rechallenge or its optimal timing relative to kinase-inhibitor therapy. These uncertainties require prospective randomized evaluation.

Post-progression treatment should therefore be individualized according to the pattern and tempo of resistance, prior treatment exposure, reason for discontinuation, residual hepatic reserve, α-fetoprotein concentration, prior immune toxicity, feasibility of local treatment, and availability of subsequent systemic options. Biomarker-rich trials using paired pretreatment and progression samples are needed to determine whether distinct acquired-resistance mechanisms should direct patients toward local therapy, kinase inhibition, CTLA-4-containing rechallenge, or another mechanism-based intervention.

The IMMUreal cohort has further described systemic-treatment sequences after atezolizumab plus bevacizumab in routine practice.163 Such registries provide valuable information about subsequent-treatment use and outcomes, but patients who remain sufficiently fit to receive another therapy represent a selected subgroup. Those who die, develop hepatic decompensation, or lose performance status before second-line treatment are excluded from treatment-specific analyses, potentially overstating the apparent benefit of subsequent therapy.

Sequence analyses are particularly vulnerable to immortal-time bias, informative treatment discontinuation, and time-varying confounding. Liver function changes during first-line treatment, influences second-line selection, and independently affects survival. Simple comparisons limited to patients who received a second-line drug cannot capture patients who lost that opportunity because of early progression or hepatic decompensation.

These methodological priorities for future registries are discussed in Future Directions.

Future Directions

The next phase of HCC immunotherapy should prioritize four unresolved questions.

1. Predictive biomarker validation. No validated predictive biomarker currently guides regimen selection.

Future trials should test prespecified biomarker-by-treatment interactions within randomized designs rather than reporting associations within a single arm. Candidate tissue, blood, microbiome, imaging, and AI-based markers should be evaluated for whether they improve a defined treatment decision.

2. Rational subsequent therapy after modern first-line therapy. Primary progression, acquired progression, progression after CTLA-4 exposure, discontinuation due to immune or vascular toxicity, and progression after treatment-free immune control are distinct failure states and should not be pooled.

Randomized trials are needed after atezolizumab plus bevacizumab, STRIDE, nivolumab plus ipilimumab, and other contemporary combinations, including rational evaluation of ICI rechallenge, CTLA-4-containing salvage, kinase inhibitors, and local therapy for oligoprogression (see Subsequent Therapy and Real-World Evidence).

3. Preservation of hepatic reserve. Liver function should be treated as a longitudinal clinical outcome, not merely as an entry criterion.

Antiviral therapy, variceal control, management of ascites and encephalopathy, infection prevention, nutritional support, muscle preservation, and careful integration of locoregional treatment should be prospectively incorporated into trials and real-world datasets.

4. Mechanism-defined development of emerging therapies. Adding agents empirically to active doublets may increase toxicity without incremental benefit.

Development should begin with a defined failure mechanism: inadequate priming (vaccines, dendritic-cell activation, oncolytics), vascular or stromal exclusion (vascular/microenvironmental modulation), myeloid dominance (macrophage, chemokine, or adenosine targeting), antigen-low or heterogeneous tumors (multi-antigen cell therapy or broader TCR recognition), and compensatory checkpoint signaling (LAG-3, TIGIT, TIM-3) (Figure 4). Figure 4Future directions for precision immunotherapy in hepatocellular carcinoma (HCC). The framework highlights five priority areas to advance precision immunotherapy in HCC: (A) Predictive biomarkers—validate marker-by-treatment interactions, integrate tumor, immune, ctDNA, microbiome, imaging, and liver function, and prioritize clinical utility over biological interest.

(B) Subsequent therapy—recognize distinct resistant states after modern first-line therapy, use serial profiling to identify actionable dependencies, and conduct randomized trials after current standards. (C) Preserve liver reserve—treat hepatic function as a longitudinal outcome, manage portal hypertension, antiviral therapy, nutrition, and sarcopenia, and ensure access to subsequent therapy. (D) Mechanism-driven innovation—use adaptive trials with biomarker guidance, define the failure mechanism first, and match the right intervention to the right population.A flow diagram of precision immunotherapy in hepatocellular carcinoma with four priority boxes.

Limitations

This is a narrative review based on targeted PubMed searches and hand-searching of reference lists, not a systematic review. We did not search all databases, conference abstracts, or regulatory documents, and no formal quality assessment or meta-analysis was performed. Evidence for Child-Pugh B and other underrepresented populations relies largely on observational real-world data subject to selection bias and confounding.

No head-to-head randomized trials have compared first-line immunotherapy combinations, so cross-trial comparisons are hypothesis-generating only. Biomarker evidence remains largely prognostic or pharmacodynamic, and emerging therapies are still early-phase. These limitations should be considered when interpreting our conclusions.

Conclusion

The therapeutic landscape of advanced HCC is now defined by several phase III platforms, but their evidence is not uniform. IMbrave150 established atezolizumab plus bevacizumab as a first-line standard, with mature global phase III evidence and a median OS of 19.2 months. HIMALAYA validated the STRIDE regimen, demonstrating improved OS versus sorafenib and non-inferiority of durvalumab monotherapy.

CheckMate 9DW showed that nivolumab plus ipilimumab improved OS and produced durable responses compared with lenvatinib or sorafenib. CARES-310 confirmed camrelizumab plus rivoceranib as an effective PD-1–VEGFR2 platform. In contrast, LEAP-002 and COSMIC-312 showed that biological rationale and higher response rates do not ensure survival superiority, and the updated IMbrave050 findings demonstrated that an early recurrence-free-survival signal in the adjuvant setting may not be maintained with longer follow-up.

Mechanistically, HCC arises in a tolerogenic liver microenvironment shaped by chronic injury, VEGF-driven immune exclusion, and non-interchangeable CTLA-4 and PD-1/PD-L1 pathways. Resistance may be primary, acquired, or host- and liver-related, and no validated predictive biomarker currently guides regimen selection. PD-L1, TMB, and MSI/dMMR have limited utility; AFP, inflammatory scores, and ctDNA are mainly prognostic or pharmacodynamic; microbiome, imaging, and AI remain investigational.

Emerging therapies—including adoptive cell therapy, vaccines, oncolytic approaches, and bispecific antibodies—require mechanism-driven development rather than empirical addition.

First-line immunotherapy combinations are not interchangeable. For patients without major contraindications to VEGF inhibition, atezolizumab plus bevacizumab has the most mature evidence. When bevacizumab is unsuitable—because of high-risk varices, recent bleeding, uncontrolled hypertension, severe proteinuria, thromboembolic risk, or impaired wound healing—STRIDE or nivolumab plus ipilimumab are important alternatives; STRIDE limits CTLA-4 exposure to a single priming dose, whereas nivolumab plus ipilimumab produces high response rates but requires intensive immune-toxicity monitoring.

Camrelizumab plus rivoceranib is an effective option where available and when VEGFR2 inhibition is appropriate. PD-1 or PD-L1 monotherapy may remain reasonable in selected patients when combination therapy is contraindicated or unavailable. Selection must integrate tumor tempo, hepatic reserve, individual Child–Pugh components, portal-hypertension complications, autoimmune disease, prior transplantation, metabolic liver disease, prior immune toxicity, and the probability of receiving subsequent therapy.

Child-Pugh B should not be treated as a homogeneous category; patients with active ascites, encephalopathy, recent variceal bleeding, or worsening bilirubin have a distinct and poorer risk profile. Preservation of hepatic reserve is a therapeutic objective in itself, because loss of liver function can preclude subsequent effective therapy.

The central unmet need is the absence of validated predictive biomarkers and randomized sequencing data after modern first-line combinations. Future progress will depend on matching defined failure mechanisms to appropriate interventions while preserving hepatic reserve and future treatment opportunities.