Predictive Biomarkers of Immune Checkpoint Inhibitor-Based Mono- and Combination Therapies for Hepatocellular Carcinoma

Introduction

As the predominant type of primary liver cancer, hepatocellular carcinoma (HCC) accounts for over 90% of primary liver malignancies and is the sixth most prevalent and the third most deadly human cancer worldwide, contributing to approximately 900,000 cases and 800,000 deaths per year ^1-3. Many therapeutic options have been well established for treating HCC, including surgical therapies (such as liver transplantation and resection) ^{4, 5}, locoregional therapies (such as radiotherapy, ablation, and embolization) ^{6, 7}, and systemic therapies (such as chemotherapy and molecular targeted therapy) ^{8, 9}. Moreover, immunotherapies such as immune checkpoint inhibitor (ICI) therapy have been developed as a promising treatment modality for HCC ^{10, 11}. However, the therapeutic efficacy of these treatment modalities varies among patients and remains to be improved. Therefore, the discovery of valuable biomarkers for predicting therapeutic responses and outcomes in HCC patients is an important goal to select the most suitable patient for the most suitable treatment (the so-called personalized therapy) to improve patient survival.

Immune checkpoint molecules include the co-inhibitory receptors expressed by effector T cells and the corresponding ligands expressed by tumor cells and stromal cells ^{12, 13}. Through the interaction between the receptors and ligands, immune checkpoint molecules play an important role in suppression of the activation and function of effector T cells, thereby facilitating tumor cell escape from T cell-mediated anti-tumor immune responses ^{14, 15}. As a result, many monoclonal antibodies that block the binding of immune checkpoint receptors to ligands (the so-called ICIs) have been generated as effective immunotherapeutic agents to restore T cell-mediated killing of tumor cells ^{16, 17}. The ICIs that have been licensed or are in clinical research for HCC treatment include the agents targeting the co-inhibitory receptors such as cytotoxic T lymphocyte-associated antigen 4 (CTLA-4) and programmed death 1 (PD-1) and the agents targeting the ligand of PD-1, programmed death ligand 1 (PD-L1) ^18-20. In addition, the combination of ICIs targeting CTLA-4 and PD-1 or PD-L1, the combination of ICIs targeting PD-1 or PD-L1 and tyrosine kinase inhibitors (TKIs), and the combination of ICIs targeting PD-L1 and vascular endothelial growth factor (VEGF) inhibitors have been also licensed or are under clinical validation for HCC therapy ^18-20. It has been shown that ICI-based mono- and combination therapies are active, tolerate, and clinically beneficial against HCC, although patients may concurrently receive various locoregional therapies. However, the overall response rates remain unsatisfied in HCC patients, with only about 15% and 30% for ICI-based mono- and combination therapies, respectively ^18-20. Therefore, the development of clinically useful predictive biomarkers for identifying HCC patients who are more likely to respond to ICIs is urgently needed to advance personalized therapy for better patient outcomes.

This review provides a comprehensive summary of the hitherto published literature, which unravel various promising biomarkers at pre-treatment, on-treatment, and post-treatment time points in tissue, blood, and stool samples for predicting the therapeutic responses and clinical benefits of different categories of ICI-based therapies in HCC patients, including ICI mono-therapy and combination therapy with other ICIs or TKIs or VEGF inhibitors.

Predictive biomarkers of ICI mono-therapy for HCC

Several biomarkers have been validated with predictive value in HCC patients receiving ICI mono-therapy in multiple lines of studies (Table 1). The study conducted by Sangro et al. ²¹ analyzed the pre-treatment expression of PD-L1 in tumor tissues of 195 HCC patients treated with PD-1 ICI (nivolumab) and showed that high PD-L1 expression level (≥ 1%) was associated with better median overall survival (OS) (28.1 versus (vs.) 16.6 months, P value = 0.03) than low PD-L1 expression level (< 1%). The study conducted by Feun et al. ²² evaluated the pre-treatment levels of a panel of cytokines and chemokines in blood of 28 HCC patients treated with PD-1 ICI (pembrolizumab) and identified that low transforming growth factor-beta (TGF-β) level (< 200 pg/mL) predicted longer median OS (not reached (NR) vs. 7 months, P value = 0.005) and progression-free survival (PFS) (NR vs. 2 months, P value = 0.008) than high TGF-β level (≥ 200 pg/mL). The study conducted by Zhang et al. ²³ combined the pre-treatment blood levels of C-reactive protein (CRP) and alpha-fetoprotein (AFP) to stratify 101 HCC patients treated with PD-1 ICI (such as nivolumab, toripalimab, sintilimab, and pembrolizumab) into low-risk (CRP ≤ 20.9 mg/L and AFP ≤ 400 ng/mL), medium-risk (CRP > 20.9 mg/L or AFP > 400 ng/mL), and high-risk (CRP > 20.9 mg/L and AFP > 400 ng/mL) subgroups and confirmed that patients with a low-risk score had the highest disease control rate (DCR) (82% vs. 65% vs. 35%, P value = 0.002) and the best OS (P value < 0.001) and PFS (P value < 0.001) compared with patients with a medium- or high-risk score. The study conducted by Dong et al. ²⁴ combined the pre-treatment albumin-bilirubin (ALBI) grade and age to stratify 38 HCC patients treated with PD-1 ICI (such as sintilimab and camrelizumab) combined with locoregional therapy (such as conventional radiotherapy (CRT), hypofractionated radiotherapy (HFRT), and stereotactic body radiotherapy (SBRT)) into low-risk (ALBI grade 1 and age ≥ 53 years) and high-risk (ALBI grade 2 and age < 53 years) subgroups and revealed that patients with a low-risk score had higher objective response rate (ORR) (50% vs. 14%, P value = 0.001) and longer median OS (NR vs. 10.1 months, P value = 0.003) and PFS (15.3 vs. 2.7 months, P value < 0.001) than patients with a high-risk score. In addition, the study conducted by Choi et al. ²⁵ measured the pre-treatment counts of neutrophils and lymphocytes in blood of 194 HCC patients treated with PD-1 ICI (nivolumab) and indicated that low neutrophil to lymphocyte ratio (NLR) (< 3) was correlated with longer median OS (61.3 vs. 21.0 weeks, P value < 0.001) and PFS (11.0 vs. 7.1 weeks, P value = 0.01) than high NLR (≥ 3). The study conducted by Hung et al. ²⁶ also calculated the NLR in blood of 45 HCC patients treated with PD-1 ICI (nivolumab) and verified that low pre-treatment NLR (≤ 2.5), on-treatment NLR (< 4.1), or post-treatment NLR (≤ 2.7) predicted DCR (sensitivity/specificity, 57%/97%; 86%/58%; 64%/87%, respectively) and longer PFS (P value = 0.004; P value = 0.006; P value = 0.001, respectively) than high pre-treatment NLR (> 2.5), on-treatment NLR (≥ 4.1), or post-treatment NLR (> 2.7). The study conducted by Dharmapuri et al. ²⁷ detected both NLR and platelet to lymphocyte ratio (PLR) in blood of 103 HCC patients treated with PD-1 ICI (nivolumab) combined with or without locoregional therapy (such as transarterial chemoembolization (TACE) and transarterial radioembolization (TARE)) and showed that low pre-treatment or post-treatment NLR (all < 5) predicted longer median OS (23 vs. 10 months, P value = 0.004; 35 vs. 9 months, P value < 0.001, respectively) and PFS (16 vs. 5 months, P value = 0.022; 35 vs. 5 months, P value < 0.001, respectively) than high NLR (≥ 5). When PLR was divided into low (≤ 118 for pre-treatment; ≤ 125 for post-treatment), medium (> 118 to < 224 for pre-treatment; > 125 to < 229 for post-treatment), and high (≥ 224 for pre-treatment; ≥ 229 for post-treatment) level groups, patients with a low pre-treatment or post-treatment PLR had the longest median OS (35 vs. 10 vs. 15 months, P value = 0.05; NR vs. 19 vs. 10 months, P value = 0.013, respectively) compared with patients with a medium or high PLR. The study conducted by Huang et al. ²⁸ determined the pre-treatment NLR, PLR, systemic immune-inflammation index (SII), and lymphocyte to monocyte ratio (LMR) in blood of 110 HBV-related HCC patients treated with PD-1 ICI and demonstrated that low NLR (< 5), PLR (< 140), or SII (< 970) or high LMR (≥ 1.8) was associated with longer median OS (7.3 vs. 6.0 months, P value = 0.0007; 7.3 vs. 6.7 months, P value = 0.0029; 7.2 vs. 5.9 months, P value < 0.0001; 7.2 vs. 6.7 months, P value = 0.0038, respectively) and PFS (6.7 vs. 5.4 months, P value = 0.0013; 6.7 vs. 5.9 months, P value = 0.0016; 6.7 vs. 4.5 months, P value < 0.0001; 6.8 vs. 5.5 months, P value = 0.0006, respectively) than high NLR (≥ 5), PLR (≥ 140), or SII (≥ 970) or low LMR (< 1.8). The study conducted by Jeon et al. ²⁹ measured the frequency of classical monocytes (cMonocyte) and PD-L1-expressing classical monocytes (cMonocyte-PDL1) in blood of 45 HCC patients treated with PD-1 ICI (nivolumab) and calculated the on-treatment monocyte index by dividing cMonocyte_D7/D0 by cMonocyte-PDL1_D7/D0, in which cMonocyte_D7/D0 and cMonocyte-PDL1_D7/D0 were defined as the fold change in the frequency of cMonocyte and cMonocyte-PDL1 at day 7 over day 0 after treatment initiation, respectively. It was shown that high monocyte index (≥ 1) predicted DCR (sensitivity/specificity, 83%/65%) and better PFS (P value = 0.008) than low monocyte index (< 1). Moreover, the study conducted by Mao et al. ³⁰ analyzed the pre-treatment composition of gut microbiota in stool of 30 HCC patients treated with PD-1 ICI and ascertained that patients with a high abundance of Erysipelotrichaceae bacterium-GAM79 or low abundance of Veillonellaceae had longer median OS (NR vs. 15.9 months, P value = 0.041; NR vs. 7.8 months, P value = 0.03, respectively) and PFS (15.9 vs. 5.5 months, P value = 0.021; 10.8 vs. 3.6 months, P value = 0.005, respectively) than patients with a low abundance of Erysipelotrichaceae bacterium-GAM79 or high abundance of Veillonellaceae.

Predictive biomarkers of ICI combination therapy with other ICIs or TKIs for HCC

The predictive value of many biomarkers in HCC patients receiving ICI combination therapy with other ICIs or TKIs have been validated in multiple lines of studies (Table 2). The study conducted by Ng et al. ³¹ investigated the pre-treatment expression of CD38 in tumor tissues of 49 HCC patients treated with PD-1 or PD-L1 ICI combined with or without CTLA-4 ICI and showed that high CD38-positive (CD38⁺) cell proportion (≥ 5%) was associated with higher ORR (43% vs. 4%, P value = 0.019) and longer median OS (19.1 vs. 9.6 months, P value = 0.0295) and PFS (8.2 vs. 1.6 months, P value = 0.0065) than low CD38⁺ cell proportion (< 5%). The study conducted by Muhammed et al. ³² measured the pre-treatment prognostic nutritional index (PNI), NLR, and PLR in blood of 362 HCC patients treated with PD-1 ICI (such as nivolumab and pembrolizumab) or PD-L1 ICI (such as atezolizumab, avelumab, and durvalumab) combined with or without CTLA-4 ICI (ipilimumab) and revealed that high PNI (≥ 45) predicted higher DCR (66% vs. 52%, P value = 0.014) and longer median OS (17.7 vs. 10.8 months, P value = 0.018) than low PNI (< 45); low NLR (< 5) predicted higher ORR (22% vs. 12%, P value = 0.034) and longer median OS (17.6 vs. 7.7 months, P value = 0.0001) and PFS (3.8 vs. 2.1 months, P value = 0.025) than high NLR (≥ 5); low PLR (< 300) predicted longer median OS (16.5 vs. 6.4 months, P value < 0.0001) and PFS (3.7 vs. 1.8 months, P value = 0.0006) than high PLR (≥ 300). In addition, the study conducted by Shao et al. ³³ detected the on-treatment change in AFP levels in blood of 43 HCC patients treated with PD-1 ICI combined with or without CTLA-4 ICI and identified that high AFP decrease (> 20%) predicted higher ORR (73% vs. 14%, P value < 0.001) and DCR (80% vs. 46%, P value = 0.033) and longer median OS (28.0 vs. 11.2 months, P value = 0.048) and PFS (15.2 vs. 2.7 months, P value = 0.002) than low AFP decrease (≤ 20%). The studies conducted by Lee et al. ³⁴ and Hsu et al. ³⁵ also evaluated the on-treatment AFP change in blood of patients. The former study confirmed that high AFP decrease (> 10%) predicted higher ORR (64% vs. 10%, P value < 0.001) and DCR (82% vs. 14%, P value < 0.001) and longer median OS (24.7 vs. 6.9 months, P value = 0.009) than low AFP decrease (≤ 10%) in 75 HCC patients treated with PD-1 ICI (such as nivolumab and pembrolizumab) combined with or without TKI (such as sorafenib, regorafenib, and lenvatinib); the latter study verified that high AFP decrease (> 15%) predicted higher ORR (46% vs. 10%, P value < 0.001) and DCR (80% vs. 29%, P value < 0.001) and longer median OS (21.9 vs. 5.6 months, P value < 0.001) and PFS (7.5 vs. 2.3 months, P value < 0.001) than low AFP decrease (≤ 15%) in 95 HCC patients treated with PD-1 ICI (such as nivolumab and pembrolizumab) combined with or without locoregional therapy (such as SBRT, TACE, and radiofrequency ablation (RFA)) or TKI (such as sorafenib, regorafenib, and lenvatinib). The study conducted by Sun et al. ³⁶ analyzed the post-treatment change in AFP and protein induced by vitamin K absence or antagonist-II (PIVKA-II) levels in blood of 235 HCC patients treated with PD-1 ICI (such as nivolumab, toripalimab, sintilimab, camrelizumab, and pembrolizumab) combined with or without locoregional therapy (TACE) and/or TKI and demonstrated that high AFP decrease (> 50%) or PIVKA-II decrease (> 50%) was correlated with higher ORR (53% vs. 18%, P value < 0.001; 50% vs. 18%, P value = 0.003, respectively) and longer median OS (NR vs. 13.7 months, P value = 0.003; NR vs. 14.4 months, P value = 0.006, respectively) and PFS (13.1 vs. 4.5 months, P value < 0.001; 10.9 vs. 4.5 months, P value = 0.021, respectively) than low AFP decrease (≤ 50%) or PIVKA-II decrease (≤ 50%). The study conducted by Li et al. ³⁷ established a nomogram based on 7 pre-treatment clinical parameters including Eastern Cooperative Oncology Group performance status (ECOG PS), TACE, extrahepatic metastasis (EHM), Child-Pugh score, alanine aminotransferase (ALT), AFP, and PLR to stratify 258 HCC patients treated with PD-1 ICI (such as sintilimab and camrelizumab) combined with or without TKI (such as sorafenib, regorafenib, and lenvatinib) and ascertained that patients with a low-risk score (≤ 182.7) had the longest median OS (53.2 vs. 17.5 vs. 7.6 months, P value < 0.0001) compared with patients with a medium-risk (> 182.7 to ≤ 240.3) or high-risk (> 240.3) score. The study conducted by Guo et al. ³⁸ combined the pre-treatment PIVKA-II level and metastasis to stratify 191 HCC patients treated with PD-1 ICI (such as nivolumab, toripalimab, sintilimab, pembrolizumab, and tislelizumab) combined with TKI (lenvatinib) into low-risk (PIVKA-II < 600 mAU/mL and without metastasis), medium-risk (PIVKA-II > 600 mAU/mL or with metastasis), and high-risk (PIVKA-II > 600 mAU/mL and with metastasis) subgroups and found that patients with a low-risk score had the longest median OS (24.0 vs. 17.7 vs. 12.1 months, P value < 0.001) compared with patients with a medium- or high-risk score.

Moreover, the study conducted by Xu et al. ³⁹ examined the pre-treatment levels of circulating tumor DNA (ctDNA) tumor mutation burden (TMB) and maximum somatic allele frequency (MSAF) in blood of 107 HCC patients treated with PD-1 ICI (camrelizumab) combined with TKI (apatinib). It was shown that low ctDNA TMB (≤ 4) predicted higher DCR (90% vs. 64%, P value = 0.002) and longer OS (P value = 0.019) than high ctDNA TMB (> 4); low ctDNA MSAF (≤ 0.027) predicted higher DCR (90% vs. 72%, P value = 0.043) and better OS (P value = 0.002) and PFS (P value = 0.004) than high ctDNA MSAF (> 0.027). The study conducted by Su et al. ⁴⁰ determined the pre-treatment counts of PD-L1⁺ circulating tumor cells (CTCs) in blood of 47 HCC patients treated with PD-1 ICI (such as sintilimab, camrelizumab, and tislelizumab) combined with locoregional therapy (intensity-modulated radiotherapy (IMRT)) and TKI (such as sorafenib, regorafenib, lenvatinib, apatinib, and anlotinib) and unraveled that low PD-L1⁺ CTC count (< 2) predicted higher ORR (sensitivity/specificity, 77%/67%; 57% vs. 17%, P value = 0.007) and longer median OS (NR vs. 10.8 months, P value = 0.001) than high PD-L1⁺ CTC count (≥ 2). The study conducted by Lee et al. ⁴¹ monitored the pre-treatment composition of gut microbiota in stool of 74 HCC patients treated with PD-1 ICI (such as nivolumab and pembrolizumab) combined with or without TKI (41 patients as derivation cohort and 33 patients as validation cohort) and stratified patients into good-signature (coexistence of Prevotella 9 depletion and Lachnoclostridium enrichment), fair-signature (coexistence of both two bacteria depletion or both two bacteria enrichment), and poor-signature (coexistence of Prevotella 9 enrichment and Lachnoclostridium depletion) subgroups. In the derivation cohort, patients with a good-signature microbiota had the longest median OS (22.8 vs. 8.0 vs. 4.8 months, P value = 0.007) compared with patients with a fair- or poor-signature microbiota; in the validation cohort, patients with a good-signature microbiota had the highest ORR (53% vs. 20% vs. 0%, P value = 0.06) and DCR (95% vs. 90% vs. 0%, P value < 0.001) and the longest median OS (NR vs. 11.1 vs. 6.5 months, P value < 0.001) and PFS (8.8 vs. 7.6 vs. 1.8 months, P value < 0.001) compared with patients with a fair- or poor-signature microbiota.

Predictive biomarkers of ICI combination therapy with VEGF inhibitors for HCC

Multiple lines of studies have validated the predictive value of many biomarkers in HCC patients receiving ICI combination therapy with VEGF inhibitors (Table 3). The studies conducted by Zhu et al. ⁴² and Kuzuya et al. ⁴³ evaluated the on-treatment AFP change in blood of patients. The former study identified that high AFP decrease (≥ 75%) or low AFP increase (≤ 10%) predicted ORR (sensitivity/specificity, 59%/86%; 77%/44%, respectively) and longer median OS (NR vs. 14.2 months, P value < 0.001; 23.7 vs. 10.6 months, P value < 0.001, respectively) and PFS (13.2 vs. 6.7 months, P value < 0.001; 9.9 vs. 5.5 months, P value < 0.001, respectively) than low AFP decrease (< 75%) or high AFP increase (> 10%) in 150 HCC patients treated with PD-L1 ICI (atezolizumab) combined with VEGF inhibitor (bevacizumab); the latter study verified that low AFP ratio (< 1.4) predicted DCR (sensitivity/specificity, 89%/88%) and longer median PFS (30 vs. 6 weeks, P value = 0.0003) than high AFP ratio (≥ 1.4) in 50 HCC patients treated with PD-L1 ICI (atezolizumab) combined with VEGF inhibitor (bevacizumab). The study conducted by Chon et al. ⁴⁴ measured the pre-treatment PIVKA-II level and NLR and on-treatment change in AFP and PIVKA-II levels and NLR in blood of 121 HCC patients treated with PD-L1 ICI (atezolizumab) combined with VEGF inhibitor (bevacizumab) and confirmed that high AFP decrease (≥ 30%) or PIVKA-II decrease (≥ 50%) or low NLR (< 2.5) predicted higher ORR (43% vs. 22%, P value < 0.05; 50% vs. 26%, P value < 0.05; 39% vs. 19%, P value < 0.05, respectively) than low AFP decrease (< 30%) or PIVKA-II decrease (< 50%) or high NLR (≥ 2.5); high NLR decrease (≥ 10%) predicted better OS than low NLR decrease (< 10%); low PIVKA-II level (< 186 mAU/mL) or NLR (< 2.5) predicted better OS and PFS than high PIVKA-II level (≥ 186 mAU/mL) or NLR (≥ 2.5). In addition, the study conducted by Campani et al. ⁴⁵ combined the pre-treatment ALBI grade and on-treatment AFP change to stratify 70 HCC patients treated with PD-L1 ICI (atezolizumab) combined with VEGF inhibitor (bevacizumab) and showed that patients with low ALBI grade (grade 1) and high AFP decrease (≥ 20%) had the longest median OS (NR vs. 16.6 vs. 11.8 vs. 5.7 months, P value = 0.046) and PFS (NR vs. 8.6 vs. 5.6 vs. 2.3 months, P value = 0.012), followed by patients with high ALBI grade (grade 2) and high AFP decrease (≥ 20%), patients with low ALBI grade (grade 1) and low AFP decrease (< 20%), and patients with high ALBI grade (grade 2) and low AFP decrease (< 20%). The study conducted by Hatanaka et al. ⁴⁶ combined the pre-treatment modified albumin-bilirubin (mALBI) grade and AFP level to stratify 426 HCC patients treated with PD-L1 ICI (atezolizumab) combined with VEGF inhibitor (bevacizumab) (255 patients as derivation cohort and 171 patients as validation cohort) and revealed that patients with low mALBI grade (1/2a) and low AFP level (< 100 ng/mL) had the highest OS rate (derivation cohort, 83% vs. 62% vs. 25%, P value < 0.001; validation cohort, 94% vs. 62% vs. 46%, P value < 0.001) and the longest median PFS (derivation cohort, 9.5 vs. 6.6 vs. 3.8 months, P value < 0.001; validation cohort, 9.3 vs. 6.7 vs. 4.7 months, P value = 0.018), followed by patients with high mALBI grade (2b/3) or high AFP level (≥ 100 ng/mL) and patients with high mALBI grade (2b/3) and high AFP level (≥ 100 ng/mL). Moreover, the study conducted by Yang et al. ⁴⁷ detected the pre-treatment levels of several cytokines in blood of 165 HCC patients treated with PD-L1 ICI (atezolizumab) combined with VEGF inhibitor (bevacizumab) (84 patients as derivation cohort and 81 patients as validation cohort) and found that low interleukin-6 (IL-6) level (< 18.49 pg/mL) predicted higher ORR (derivation cohort, 38% vs. 0%; validation cohort, 29% vs. 7%) and longer OS (derivation cohort, P value = 0.021; validation cohort, P value < 0.001) and PFS (derivation cohort, P value = 0.003; validation cohort, P value = 0.018) than high IL-6 level (≥ 18.49 pg/mL). The study conducted by Giovannini et al. ⁴⁸ analyzed the pre-treatment percentage of PD-1⁺ granulocytes in blood of 34 HCC patients treated with PD-L1 ICI (atezolizumab) combined with VEGF inhibitor (bevacizumab) and ascertained that low PD-1⁺ granulocyte percentage (< 13%) was associated with longer mean time to progression (TTP) (NR vs. 3.2 months, P value < 0.0001) than high PD-1⁺ granulocyte percentage (≥ 13%). The study conducted by Balcar et al. ⁴⁹ examined the post-treatment immunoglobulin G (IgG) change in blood of 72 HCC patients treated with PD-1 ICI (such as nivolumab and pembrolizumab) or PD-L1 ICI (atezolizumab) combined with or without VEGF inhibitor (bevacizumab) and demonstrated that low IgG increase (< 14%) was correlated with longer median OS (15.9 vs. 6.4 months, P value = 0.001) and PFS (7.9 vs. 2.9 months, P value = 0.011) than high IgG increase (≥ 14%).

Conclusions

This review comprehensively summarizes the evidence from the literature published so far which validate the predictive significance of a variety of biomarkers at different treatment time points (including pre-treatment, on-treatment, and post-treatment time points) in different sample sources (including tissue, blood, and stool samples) for the treatment responses and outcomes of HCC patients receiving different categories of ICI-based therapies (including ICI mono-therapy and combination therapy with other ICIs or TKIs or VEGF inhibitors) (Table 4). Among the current predictive biomarkers, most are derived from the blood and stool samples of HCC patients, supporting the convenience advantages of the use of noninvasive sampling methods in clinical application. Moreover, the clinical applicability varies among the predictive biomarkers. Certain biomarkers are selective for one category of ICI therapy at one specific treatment time point, such as tumoral PD-L1 expression level, intratumoral CD38⁺ cell proportion, ctDNA TMB or MSAF, PD-L1⁺ CTC count, PD-1⁺ granulocyte percentage, SII, PNI, and IgG change; in contrast, some biomarkers show predictive value for 2 or 3 different categories of therapies at 2 or 3 different treatment time points, such as NLR, PLR, AFP change, and PIVKA-II change. It should be carefully noted that even the same biomarker may have different cut-off values when applied for predicting HCC patients receiving different categories of therapies at different treatment time points. Besides, as a potential way to overcome the inter-patient heterogeneity among HCC patients, several biomarkers combine 2 different factors or even more factors in a nomogram to stratify the patients into more subgroups for prediction, such as CRP level and AFP level, PIVKA-II level and metastasis, ALBI grade and age, ALBI grade and AFP change, and mALBI grade and AFP level. In addition, many predictive biomarkers are based on immune cells or inflammatory cytokines, such as NLR, PLR, LMR, PD-1⁺ granulocyte percentage, SII, monocyte index, TGF-β, IL-6, and CRP, reflecting the clinical implication of tumor immune microenvironment in the efficacy of ICI therapy for HCC. Additionally, the predictive significance of the composition of gut microbiota, such as Erysipelotrichaceae, Veillonellaceae, Prevotella 9, and Lachnoclostridium, has been validated in HCC patients receiving ICI mono-therapy and combination therapy with TKIs. Considering the impact of gut microbiota-derived metabolites on ICI therapy for cancer ^50-52, evaluation of the predictive significance of microbial metabolites in the blood and/or stool samples of HCC patients receiving ICI-based mono- and combination therapies may hold great promise to discover novel predictive biomarkers. Furthermore, non-coding RNAs such as microRNAs and long non-coding RNAs have been closely implicated in cancer and ICI therapy ^{53, 54}. Whether non-coding RNAs can also serve as predictive biomarkers for ICI therapy in HCC patients is worth further investigation. Last but not the least, since the HCC patient cohorts evaluated in different studies may have different clinicopathological features and receive ICI-based therapies with different drugs (even though sharing the same molecular targets) at different treatment dosages, doses, and dosing intervals, it is quite important to take this issue into consideration when applying the predictive biomarkers to select the most suitable patient for the most suitable treatment for better therapeutic responses and outcomes.