Introduction
Immune checkpoint inhibitors (ICIs) have revolutionized the therapeutic landscape of numerous advanced solid and haematological malignancies over the past decade, emerging as a cornerstone of modern cancer therapy.1 ICIs are monoclonal antibodies targeting inhibitory receptors on the surface of T cells or tumor cells, primarily including cytotoxic T-lymphocyte-associated antigen-4 (CTLA-4) inhibitors (i.e., ipilimumab), programmed death protein 1 (PD-1) inhibitors (i.e., nivolumab and pembrolizumab), and programmed death protein ligand 1 (PD-L1) inhibitors (i.e., atezolizumab, durvalumab).2 By blocking these inhibitory pathways, ICIs can activate T-cell responses and thereby amplify antitumor activity. Currently, ICIs have demonstrated remarkable efficacy in a broad spectrum of cancers, including melanoma, non-small cell lung cancer (NSCLC), renal cell carcinoma, hepatocellular carcinoma (HCC), and esophageal squamous-cell carcinoma.3–7 Compared to traditional chemotherapy, ICIs offer superior efficacy, enhanced tolerability, and improved prognosis for patients with advanced malignancies.8 In recent years, ICIs have garnered global attention as a revolutionary approach in cancer treatment (Table 1).
Table 1Immune checkpoint inhibitors and their indications
| Target | Drug | Indications | Time to market |
|---|
| CTLA-4 | Ipilimumaba,b | Melanoma, RCC, MSI-H or dMMR CRC, HCC, NSCLC, MPM, ESCC | 2011 |
| PD-1 | Nivolumaba,b | Melanoma, NSCLC, MPM, RCC, cHL, HNSCC, UC, MSI-H or dMMR CRC, HCC, ESCC, GC/GEJC/EAC | 2014 |
| Pembrolizumaba,b | Melanoma, NSCLC, HNSCC, cHL, PMBCL, UC, MSI-H or dMMR CRC, ESCC, HCC, MCC, RCC, TNBC, BTC, CC, EC, GC/GEJC/EAC TMB-H solid tumors, cSCC | 2014 |
| Cemiplimaba | cSCC, BCC, NSCLC | 2018 |
| Toripalimaba,b | Melanoma, NPC, UC, NSCLC, SCLC, RCC, ESCC, TNBC | 2018 |
| Sintilimaba,b | cHL, NSCLC, HCC, ESCC, GC/GEJC/EAC | 2018 |
| Camrelizumaba,b | cHL, HCC, NSCLC, ESCC, NPC | 2019 |
| Tislelizumaba,b | cHL, UC, NSCLC, SCLC, HCC, MSI-H or dMMR solid tumor, ESCC, NPC, GC/GEJC/EAC | 2019 |
| Penpulimabb | cHL, NSCLC | 2021 |
| Zimberelimabb | cHL | 2021 |
| Serplulimabb | MSI-H or dMMR solid tumor, NSCLC, SCLC, ESCC | 2022 |
| Pucotenlimabb | Melanoma, MSI-H or dMMR solid tumor | 2022 |
| Enlonstobartb | CC | 2024 |
| PD-L1 | Atezolizumaba,b | Melanoma, UC, NSCLC, SCLC, HCC | 2017 |
| Avelumaba | MCC, UC, RCC | 2017 |
| Durvalumaba,b | NSCLC, SCLC, HCC, BTC, dMRR EC | 2017 |
| Envafolimabb | MSI-H or dMMR CRC | 2021 |
| Sugemalimabb | NSCLC, ENKTL, ESCC, GC/GEJC/EAC | 2021 |
| Adebrelimabb | SCLC | 2023 |
| Socazolimabb | CC | 2023 |
| Benmelstobartb | SCLC | 2024 |
| PD-L1/CTLA-4 | Cadonilimabb | CC | 2022 |
| Iparomlimab and tuvonralimabb | CC | 2024 |
| PD-1/VEGF | Ivonescimabb | NSCLC | 2024 |
However, the widespread adoption of ICIs for advanced cancers has been accompanied by unexpected immunological and inflammatory complications, known as immune-related adverse events (irAEs), arising from an overactive immune response that targets normal tissues or organs.9 These irAEs can influence a range of organ systems, particularly the skin, gastrointestinal tract, endocrine glands, and liver.10 Among these, immune-mediated hepatotoxicity (IMH), a liver-related irAE, has represented a clinically significant challenge.11 IMH typically manifests as asymptomatic elevations in aspartate aminotransferase (AST), alanine aminotransferase (ALT), and/or alkaline phosphatase (ALP), with reported incidences ranging from 1% to 15% in clinical trials.12,13 Although most IMH cases are relatively mild, sometimes they can be life-threatening, leading to severe hepatitis, liver failure, and even fatalities, necessitating treatment discontinuation.14,15 According to data from the World Health Organization, 20.2% (124 out of 613) of fatal ICI-related toxic events were attributed to IMH, highlighting the critical need for early detection and effective management.16
Currently, the precise mechanisms of IMH development are not fully elucidated, with proposed roles for T-helper cell expansion, monocyte/macrophage activation, and regulatory T (Treg)-cell depletion,17 making therapy targeting specific molecular pathways challenging. Consequently, management strategies for IMH primarily rely on expert consensus guidelines derived from clinical trial protocols, which recommend discontinuing ICIs and initiating glucocorticoids or immunosuppressants.18,19 However, this reactive approach, which is often initiated only after toxicity manifests, fails to prevent the occurrence of severe IMH and is ineffective in patients who show a poor response or resistance to corticosteroid therapy. More critically, data on risk factors and biomarkers of IMH are limited, resulting in the lack of reliable predictive tools that can accurately identify high-risk patients before treatment initiation, which eventually complicates treatment decisions for clinicians, potentially causing treatment delays or permanent discontinuation. Therefore, identifying the risk factors and biomarkers of IMH in patients receiving ICI therapy is urgently needed. This information will allow clinicians to identify high-risk individuals early, optimize immunotherapy regimens, and initiate timely interventions, ultimately mitigating the occurrence of IMH.
Evidence suggests that several clinical risk factors, such as female sex, pre-existing non-alcoholic fatty liver disease (NAFLD), and the use of combination ICIs (e.g., anti-CTLA-4 plus anti-PD-1/PD-L1), are associated with an elevated risk of IMH, which allows for early identification of high-risk populations.20–22 Concurrently, emerging investigations into potential biomarkers, including circulating blood cells, autoantibodies, cytokines, specific immune cell subsets (e.g., CD8+ T cells), and genomic signatures, have demonstrated promise for risk stratification and early intervention of irAEs, suggesting their potential as markers for IMH prediction.23,24 However, the available evidence on the risk factors and biomarkers that predict IMH occurrence remains limited, and most studies are retrospective, not specifically focused on IMH. Therefore, this review aims to comprehensively synthesize the current landscape of the risk factors and emerging biomarkers associated with IMH during ICI therapy, which will provide promise for early identification and individualized management of IMH, thereby promoting the safety and efficacy of tumor immunotherapy.
Potential mechanism of IMH
As previously described, the central mechanism by which ICIs exert anti-tumor effects is the activation of cytotoxic T cells (CTLs) by inhibiting the CTLA-4 and PD-1/PD-L1 signaling pathways. Although these two pathways function as negative regulators of T-cell activation, they have different roles in immune regulation. CTLA-4 functions primarily during the initial priming phase of T-cell activation within secondary lymphoid organs, such as lymph nodes.25,26 On activated T cells and Treg cells, CTLA-4 competitively binds to the co-stimulatory molecules B7-1 (CD80) and B7-2 (CD86) on antigen-presenting cells (APCs) with a higher affinity than the activating receptor CD28, thereby inhibiting the initial clonal expansion and proliferation of T cells (Fig. 1A). CTLA-4 inhibitors block this interaction, which prevents inhibitory signaling and enhances T-cell activation by promoting CD28-mediated co-stimulation. In contrast, the PD-1 pathway is primarily engaged during the effector phase within peripheral tissues and the tumor microenvironment. PD-1 is expressed on activated T cells and binds to its ligand PD-L1, which is often upregulated on tumor cells and various host cells.26,27 This engagement suppresses downstream T-cell receptor and CD28 signaling cascades, leading to the functional impairment or exhaustion of T-cell effector functions and facilitating tumor immune evasion (Fig. 1B). PD-1/PD-L1 inhibitors also block this interaction, reinvigorating exhausted T cells and restoring their cytotoxic and proliferative potential. This fundamental mechanistic divergence underpins the differential immune effects observed with ICI therapies: CTLA-4 blockade predominantly enhances early CD4+ T-cell clonal expansion and promotes T-cell trafficking to tumor sites, whereas PD-1/PD-L1 blockade primarily reverses the exhausted CD8+ T cells within tissues.26 Consequently, this complementary biology provides a rationale for the use of both monotherapy and combination regimens to achieve synergistic anti-tumor immunity.
While the precise mechanisms driving IMH following ICI therapy are not yet fully understood, the liver’s unique immunological properties are recognized as central to its pathophysiology. Under homeostatic conditions, the liver maintains a state of immunotolerance achieved through the anti-inflammatory functions of both parenchymal and non-parenchymal cells, as well as the constitutive expression of immune checkpoint molecules by various cell subsets.17 Notably, a cornerstone of liver immunotolerance involves PD-L1 expressed on diverse cell types, including hepatic stellate cells, Kupffer cells, liver sinusoidal endothelial cells, hepatocytes, as well as liver-resident or infiltrating immune cells such as macrophages, along with CTLA-4 expressed on CD4+ Treg cells.17 By suppressing CD8+ T-cell activation and function, these checkpoint molecules help protect the liver from antigen-driven autoimmune responses in various inflammatory contexts. However, due to the use of ICIs blocking these key modulatory pathways, immune tolerance of the liver can be broken, rendering it more susceptible to inflammatory damage triggered by drug exposure, underlying neoantigens, or concurrent microbial stimuli.13
Current evidence points to a multifactorial effect in IMH pathogenesis, involving a complex interplay between adaptive and innate immunity, which ultimately disrupts hepatic immune tolerance. The primary mechanisms include direct activation of CTLs and epitope spreading, as well as indirect effects on T-helper cells, forkhead box P3–positive Tregs, B cells, the inflammatory cytokine milieu, and activation of innate immunity (Fig. 1C–F).28 Firstly, ICI blockades can overcome immune tolerance by stimulating the proliferation of CD8+ cytotoxic T lymphocytes, thereby inducing tumor cell death. This is accompanied by an alteration in their transcriptional profile, leading to the upregulation of proliferative and cytotoxic genes such as interferon (IFN)-γ, granzyme, and granulysin.17,26 Concurrently, epitope spreading refers to the diversification of the initial T-cell response against novel epitopes and neoantigens that differ from the originally targeted ones.29 In this process, lysed tumor cells release substantial amounts of self-antigens or neoantigens into the microenvironment, where APCs capture, process, and cross-present these antigens, thereby triggering a secondary immune response, leading to an immune attack toward hepatocytes that share overlapping epitopes.17 Secondly, the expansion of T-helper cells, particularly Th1 and Th17 cells, results in an increase in pro-inflammatory cytokines, including interleukin (IL)-2, IFN-γ, tumor necrosis factor (TNF)-α, and IL-17, which can activate CTLs, natural killer cells, and monocyte-derived macrophages.30,31 Concurrently, ICI therapy can impair the suppressive function of Treg cells, often accompanied by reduced expression of the transcription factor forkhead box P3 and decreased production of anti-inflammatory cytokines such as IL-10, IL-35, and TGF-β.17,32 Thirdly, B cells play a crucial role in the anti-tumor immune response by engaging in crosstalk with CD8+ T cells, which involves co-stimulatory signaling through CD27/CD70 interactions, and promotes CTL survival and proliferation independently of antigen presentation.33 CTLA-4 and PD-1 are also expressed in B cells, and their blockade by ICIs causes excessive activation and proliferation of B cells, leading to increases in CD21lo B cells, plasmablasts, and pro-inflammatory cytokines that correlate with the occurrence of irAEs and IMH.34,35 Overactivated and dysregulated B cells may produce antibodies targeting self-antigens (i.e., autoantibodies), which can potentially mediate liver damage by antibody-dependent mechanisms and trigger inflammation and hepatocyte injury; however, the pathophysiological relevance of these antibodies and the precise role of B cells in irAEs require further investigation.36 In addition, ICIs promote the activation of monocytes and CD8+ T lymphocytes, leading to increased secretion of pro-inflammatory cytokines (e.g., IL-1β, IL-6, IFN-γ, IL-12p70, and TNF-α) to form the inflammatory microenvironment driving hepatotoxicity.17,30,37 These cytokines, in turn, help to activate an innate immune response by recruiting natural killer cells and macrophages, which contribute to the pathogenesis of liver injury. Liver biopsies from patients and mouse models reveal that CD8+ T cells and CCR2+ macrophages colocalize in damaged areas, with their crosstalk activating the NLRP3 inflammasome to promote hepatocyte apoptosis.30,38 Importantly, macrophage activation and recruitment to the liver can occur independently of CD8+ T cells, highlighting the complexity and redundancy of the inflammatory networks driving IMH.
Predictive biomarkers for IMH
While the risk factors outlined above provide a clinical foundation for identifying high-risk populations susceptible to IMH, they often lack the sensitivity and specificity required for individualized risk prediction. To advance toward personalized immunotherapy management, there is an urgent need for reliable biomarkers capable of accurately forecasting IMH prior to its clinical onset. The development of such biomarkers would facilitate earlier detection, enable timely intervention, and ultimately enhance both the safety and efficacy of ICI therapy.
The current diagnosis of IMH primarily relies on a combination of clinical presentation, temporal association with ICI administration, abnormal liver function tests, and exclusion of other potential causes.18,19 Liver biopsy, while considered the diagnostic gold standard, is invasive, prone to sampling variability, and unsuitable for repeated assessments. Therefore, the development of sensitive, specific, and minimally invasive biomarkers is of paramount importance. While several studies have explored potential biomarkers for predicting irAEs during ICI treatment, this section focuses on synthesizing the current evidence on promising biomarkers specifically associated with IMH, including circulating blood cell profiles, autoantibodies, cytokines, genetic and human leukocyte antigen (HLA) markers, and the gut microbiome, as illustrated in Figure 2.
Circulating blood cell-based biomarkers
Blood cell counts and ratios are attractive biomarkers due to their routine availability, low cost, and straightforward interpretation. While evidence suggests that some cellular predictors, including monocyte and eosinophil counts, lymphocytes, white blood cell (WBC) counts, neutrophil-to-lymphocyte ratio (NLR), and platelet-to-lymphocyte ratio, have emerged as potential biomarkers for IMH,19,23,25,38,50,65 their predictive value for IMH remains inconsistent across studies, as shown in Table 3.23,30,39,40,45,49,69–76
Table 3Potential blood cell-based biomarkers for predicting IMH
| Potential biomarker | Study design | N | Cancer type | Timing of measurement | Association | Significance | References |
|---|
| Circulating blood cell |
| Absolute monocyte counts (AMCs) | Prospective | 95 | Melanoma | Before ICI treatment | Elevated AMCs were associated with IMH | p < 0.0005 | Wolffer et al.23 |
| Absolute eosinophil count (AEC) | Retrospective | 533 | Solid tumors | Before ICI treatment | AEC ≥130/µL was associated with high risk of grade ≥ 2 IMH | HR = 3.01; 95% CI: 1.27–7.12; p = 0.012 | Yoshikawa et al.69 |
| Absolute lymphocyte count (ALC) | Retrospective | 1,096 | Solid tumors | Before ICI treatment | Higher ALCs were associated with risk for IMH | p = 0.040 | Miah et al.39 |
| Retrospective | 1,086 | Solid tumors | Before ICI treatment | Higher ALCs were associated with risk for IMH | HR = 1; 95% CI: 1.000–1.001; p = 0.013 | Kawano et al.49 |
| Lymphocyte (%) | Retrospective | 226 | Solid tumors | Before ICI treatment and post-treatment | Significantly lower percentage change in lymphocyte count was associated with grade 3 IMH | p < 0.05 | Haraguchi et al.72 |
| Neutrophil count, Neutrophil to lymphocyte Ratio (NLR) | Retrospective | 432 | Melanoma Renal cancer | Before ICI treatment | Low absolute neutrophil counts and NLR were associated with high risk of IMH | p < 0.05 | Atallah et al.40 |
| NLR | Retrospective | 1,096 | Solid tumors | Before ICI treatment | Low NLR was associated with high risk of IMH | p = 0.048 | Miah et al.39 |
| Meta-analysis | 10,344 | Solid tumors | Before ICI treatment | Higher NLR was associated with a reduced risk of IMH | HR = 0.89; 95% CI 0.82–0.96; p = 0.002 | Madiar et al.45 |
| Retrospective | 249 | HCC | Before ICI treatment | NLR ≥3.0 was associated with high risk of any grade and grade ≥3 IMH | Any grade, p = 0.017; grade ≥3, p = 0.008 | Tada et al.70 |
| Meta-analysis | 11,491 | Solid tumors | Before ICI treatment | Higher NLR was associated with an increased risk of IMH | OR = 2.44; 95% CI 1.23–4.84; p = 0.010 | Zhang et al.71 |
| WBC, Platelet-to-lymphocyte ratio (PLR) | Retrospective | 274 | NSCLC | Before ICI treatment and post-treatment | Low baseline WBC (≤11.0×109/L) was an independent predictor for IMH development, and elevated WBC and PLR were associated with grade 3/4 IMH | p < 0.05 | Yang et al.73 |
| Cell surface markers |
| CD4+ T and B | Retrospective | 68 | HCC | Before ICI treatment and post-treatment | Patients who experienced grade3/4 IMH had a lower decrease in the levels of CD4+ T lymphocytes and B lymphocytes upon irAE onset | p < 0.05 | Yu et al.74 |
| CD8+ T cell | Cohort | 20 | Melanoma | Before ICI treatment and post-treatment | Increased frequency of EMRA CD8+ T cells before and after dual ICI initiation, and Ki67+CD8+ T cells after starting the dual ICI treatment, was associated with severe IMH | p < 0.05 or p < 0.01 | Muller et al.75 |
| Prospective | 48 | Melanoma | Before ICI treatment and post-treatment | IMH is associated with an activation of peripheral monocytes and enhanced effector phenotype of CD8+ T cells | p < 0.05 or p < 0.01 | Gudd et al.30 |
| Ki-67+ regulatory T cells | Prospective | 144 | Melanoma NSCLC | Before ICI treatment and post-treatment | An early expansion of Ki-67+ regulatory T cells was associated with increased risk of IMH in melanoma patients | p < 0.05 | Nunez et al.76 |
Studies evaluating pre-treatment cellular counts have yielded divergent results. A prospective study in 95 melanoma patients identified a higher absolute monocyte count at ICI initiation as significantly increasing IMH risk, with no correlation found for neutrophil, lymphocyte, or eosinophil counts.23 In contrast, a larger retrospective analysis of 533 patients treated with ICIs for various malignancies established a baseline eosinophil count ≥130/µL (HR = 3.01 for <130; p = 0.012) as an independent risk factor for grade ≥2 IMH.69 Another study of 1,086 patients further reported a significant association between higher baseline lymphocyte counts and IMH, particularly in hepatocellular-injury-type cases.49 These discrepancies likely arise from differences in population characteristics, study design, and methodology. Variations in cancer types, for instance, the distinct tumor immune microenvironment in melanoma compared to other solid tumors, may alter baseline immune profiles and their interaction with ICIs. Inconsistent diagnostic criteria and grading for IMH, along with variable definitions of “elevated” cell counts, hinder direct comparison of effect estimates. Furthermore, these studies relied solely on pre-treatment baseline counts, ignoring the potential predictive value of on-treatment dynamic changes in immune cells. This underscores that single baseline cell counts are unlikely to serve as reliable biomarkers for IMH prediction.
The predictive utility of baseline cell ratios, particularly the NLR, has been studied with conflicting conclusions. A meta-analysis of 1,096 patients on ICIs revealed that low NLR (<3) (OR = 2.63, 95% CI 1.63 to 4.26, p < 0.001) was significantly linked to any-grade irAEs, including IMH.77 This association between lower baseline NLR and IMH development was also observed in two larger retrospective studies, though multivariable analysis was not performed or did not establish NLR as an independent predictor.39,40 Conversely, studies focused on HCC populations consistently reported opposite findings. A multicenter study by Tada et al. indicated that higher baseline NLR (≥3.0) was associated with an increased risk of any-grade and grade ≥3 IMH in patients with unresectable HCC treated with atezolizumab plus bevacizumab.70 Similarly, another large meta-analysis of 11,491 cancer patients found that high NLR (>5) was specifically correlated with a higher incidence of IMH.71 This discrepancy may stem from constitutively elevated baseline NLR in HCC patients with chronic liver inflammation and cirrhosis, the lack of a standardized NLR cutoff value, and differences in ICI regimens, which may fundamentally alter the immunological context and subsequent irAE profiles.
Beyond baseline values, on-treatment hematological dynamics may provide crucial predictive insights. Haraguchi et al. compared lymphocyte counts before treatment and at the onset of irAEs, reporting a markedly lower percentage reduction in lymphocyte counts from baseline to irAE onset in patients with grade 3 IMH, coinciding with a significant rise in NLR.72 This suggests that dynamic changes in cellular biomarkers better reflect evolving immune dysregulation. A retrospective cohort of 274 NSCLC patients receiving ICIs demonstrated that low baseline WBC (≤11.0 × 109/L) (p < 0.001) and high albumin (≥35 g/L) (p < 0.001) were independent predictors for IMH development, and elevated WBC (p = 0.003) and platelet-to-lymphocyte ratio (p = 0.017) were associated with grade 3/4 IMH compared to those with grade 1/2 events.73 This indicates that although lower baseline inflammatory status may predispose to IMH, severe cases are characterized by exaggerated post-treatment immune activation. Therefore, it is imperative to observe the dynamics of these markers for IMH development in future prospective research.
T and B lymphocyte subsets have emerged as potential biomarkers for IMH severity. In a retrospective study of 67 HCC patients treated with ICIs combined with tyrosine kinase inhibitors, Yu et al. found that the baseline levels of lymphocyte subsets did not differ between AE and non-AE groups; however, upon irAE onset, CD4+ T lymphocyte, CD8+ T lymphocyte, and B lymphocyte counts decreased (p < 0.05), and the decrease in CD4+ T and B lymphocyte levels was significantly greater in patients with grade 3/4 IMH compared to those with milder events (p < 0.05).74 A small cohort study by Gudd et al. reported that IMH patients exhibited activated peripheral monocytes and an enhanced effector phenotype of CD8+ T cells.30 Another two-cohort study of melanoma patients receiving PD-1 blocking monotherapy or dual ICI therapy noted an increased frequency of EMRA CD8+ T cells before and after dual ICI initiation, as well as a rise in Ki67+CD8+ T cells post-treatment, which could predict severe irAEs, including IMH.75 Additionally, early expansion of Ki-67+ Treg cells was significantly correlated with IMH risk in melanoma patients.76
In summary, circulating blood cells may represent accessible tools for IMH prediction, but their implementation requires consideration of tumor type, baseline liver condition, and treatment regimen. The transition from baseline to dynamic on-treatment changes of these markers offers a promising direction for IMH. Future research should focus on standardizing measurements, validating findings in large prospective cohorts, and integrating these cellular biomarkers into multi-factor prediction models to optimize their clinical utility in managing IMH.
Serum proteins
Beyond cellular biomarkers, some serum proteins have emerged as biomarkers for IMH, offering insights into the underlying inflammatory processes and tissue damage. C-reactive protein (CRP), an acute-phase protein, has been found to correlate with the risk of IMH. A retrospective study revealed that levels of CRP were significantly higher in patients experiencing grade 3/4 IMH upon irAE onset, with CRP ≥ 8.2 mg/L identified as a potential independent predictor for IMH development; when patients recovered, elevated levels of CRP returned to baseline.74 Alpha-fetoprotein (AFP), traditionally a tumor marker, has also been utilized to evaluate liver damage associated with ICIs. A multicenter retrospective study examined the CRAFITY score, a combination of CRP and AFP, in HCC patients treated with atezolizumab and bevacizumab.78 The scoring system assigned 0 points for AFP <100 ng/mL and CRP <10 mg/L, 1 point for either AFP ≥ 100 ng/mL or CRP ≥ 10 mg/L, and 2 points for both AFP ≥ 100 ng/mL and CRP ≥ 10 mg/L. Patients with a CRAFITY score of 2 had a significantly higher incidence of grade ≥3 IMH compared to those with scores of 0 or 1. Additionally, a recent retrospective study reported that low serum albumin, indicative of systemic inflammation, was also linked to higher IMH incidence.79 Reduced drug binding to plasma proteins due to lower albumin levels leads to slower drug elimination and a longer half-life, consequently increasing the body’s exposure to toxicity.
However, these serum proteins, particularly CRP and albumin, are susceptible to interference from non-IMH factors, reducing their specificity for IMH prediction. Future large-scale, multicenter prospective studies are warranted to validate the predictive value of these markers across various cancer types and ICI regimens. Moreover, exploring their dynamic changes in IMH and integrating them with other biomarkers could improve predictive accuracy.
Autoantibodies
Autoantibodies represent one of the most extensively investigated predictive biomarkers for irAEs. Pre-existing autoantibodies, including antinuclear antibodies (ANA), rheumatoid factor, and antithyroid antibodies, have been associated with increased incidence and severity of organ-specific irAEs following ICI treatment.80,81 This has prompted investigation into their potential role as biomarkers for predicting or diagnosing IMH.
Several studies demonstrate correlations between baseline autoantibody profiles and subsequent IMH development. A retrospective study of 252 NSCLC patients identified ANA positivity as a significant predictor for IMH.82 Notably, the predictive strength varied substantially between different ICI agents, with a markedly higher odds ratio observed for pembrolizumab (OR = 7.834) than for nivolumab (OR = 2.133). This suggests that the predictive value of autoantibodies may be significantly influenced by the specific ICI regimen. Similarly, Ghosh et al. reported differential ANA positivity rates between IMH subtypes, with 18% (15/85) in hepatitis-pattern injury and 42% (5/12) in cholangitic-pattern injury, suggesting potential phenotypic variation in autoantibody associations.83 The spectrum of relevant autoantibodies may also extend beyond classical targets, as evidenced by Zheng et al., who identified thyroid peroxidase antibodies as a prognostic biomarker for liver injury in patients treated with sintilimab.79 The underlying mechanisms linking these antibodies to IMH remain unclear. One hypothesis is that PD-1 expression is regulated through both T cell-independent and T cell-dependent pathways, resulting in high levels on activated B cells, facilitating autoantibody production upon PD-1/PD-L1 blockade, thereby contributing to an increased incidence of IMH.
Conversely, some studies have failed to demonstrate significant associations between liver-specific autoantibodies and IMH development.84 A prospective cohort study of 131 patients specifically found no correlation between IMH development and various autoantibodies, including ANA, anti-smooth muscle antibody, anti-mitochondrial antibody, and anti-liver-kidney microsomal antibodies, suggesting that liver autoantibodies may not serve as reliable predictors for IMH.85 This further implies that patients with pre-existing liver autoantibodies do not exhibit an elevated risk of IMH during ICI therapy.
Overall, evidence on autoantibodies predicting IMH risk is inconsistent and limited, making them seem unreliable as predictors for IMH when used alone. Future research should prioritize the validation of defined autoantibody panels in large prospective cohorts and explore their integration with other biomarker classes to develop composite risk scores.
Cytokines
Cytokines, as central mediators of immune activation and inflammation, have emerged as promising biomarkers for predicting irAEs.86 Given that the pathogenesis of IMH is closely related to T cell activation and inflammatory cytokines, changes in cytokine levels may serve as a valuable tool for IMH prediction.
Several pro-inflammatory cytokines have been implicated in IMH development. Among these, the IL-1 family, particularly IL-1β, acts as a central driver of liver inflammation by regulating networks of pro-inflammatory cytokine and immune-regulatory gene expression.87 Elevated IL-1β RNA expression has been observed in patients with grade ≥3 IMH compared to those without irAEs, underscoring its potential as a susceptibility biomarker for IMH occurrence.88 Similarly, IL-6, another pivotal cytokine, demonstrates significant associations with organ-specific irAEs, including IMH. While baseline IL-6 levels did not differ significantly between irAE and non-irAE groups, a marked rise in IL-6 was observed upon irAE onset. Specifically, HCC patients experiencing grade 3/4 IMH exhibited a more pronounced increase in IL-6 compared to those with grade 1/2 AEs, and IL-6 levels subsequently returned to baseline following resolution of hepatitis.74 Another small cohort also found that IL-6 levels after irAEs were significantly higher compared with before.89 This suggests that IL-6 may be a more reliable marker of active, severe inflammation than a baseline predictor, warranting further validation in large-scale cohorts. Furthermore, IL-23, a key regulator of Th17 cell differentiation, has also been linked to severe irAEs, and its blockade ameliorates liver inflammation in preclinical models, positioning it as a compelling IMH biomarker and therapeutic target.90
Currently, the limitations of single-cytokine measurements have prompted a shift towards combining multiple cytokine and chemokine panels, which could enhance the accuracy of irAE prediction. Lim et al. investigated circulating cytokines in 98 patients with melanoma using the 65-plex cytokine discovery assay, identifying that elevated baseline and early-treatment levels of 11 cytokines (G-CSF, GM-CSF, Fractalkine, FGF-2, IFN-α2, IL-12p70, IL-1α, IL-1β, IL-1RA, IL-2, and IL-13) were significantly associated with severe irAEs, including IMH.91 The CYTOX score, which integrated these 11 cytokines, could help in the early management of severe, potentially life-threatening immune-related toxicity. In a separate report, Moi et al. reported high baseline levels of pro-inflammatory cytokines (e.g., IFN-γ, IL-6, CXCL9, CXCL10, CXCL11, CXCL13) and anti-inflammatory cytokines (e.g., IL-10, IL-1RA) in three consecutive patients with IMH, although these findings require validation in larger cohorts.92 Recently, Farooqi et al. also noted that high baseline CXCL10 and increased TNF-α during treatment were linked to IMH risk, while elevated CCL27 levels at baseline and during treatment may reduce IMH risk, suggesting a protective role, further illustrating the complexity of the cytokine network.93
The timing of biomarker measurement is paramount, as cytokine levels can exhibit significant fluctuations during treatment. Zeng et al. analyzed plasma cytokine profiles at three key time points: baseline, IMH onset (IMH-D1), and seven days post-onset (IMH-D7).88 They found that 12 pro-inflammatory cytokines, namely CCL11, CCL4, CXCL1, CXCL10, CXCL12, IFN-γ, IL-10RA, IL-18, IL-1α, IL-1β, IL-7, and IL-8, were significantly lower at baseline in the ≥G3 IMH group than in the non-irAE group. Interestingly, higher IL-1α levels at IMH onset were associated with resolution of grade ≥3 IMH in the subgroup, while elevated levels of nine cytokines (including CCL11, CCL3, CCL5, CXCL1, CXCL12, IL-10RA, IL-18, IL-7, and TNF-α) at IMH-D7 were linked to IMH-related mortality.88 Recently, another prospective study of 134 solid tumor patients receiving PD-(L)1 inhibitors identified that the highest levels of CXCL9, CXCL10, CXCL11, IL-18, and IL-10 were observed at the onset of IMH, with no baseline differences between groups.94 Notably, cytokine levels tended to be higher in severe IMH compared to mild irAEs and those without irAEs; however, this difference did not reach statistical significance due to an insufficient sample size. This underscores that cytokine profiles are not static; longitudinal monitoring is essential to capture their predictive and prognostic value. Future studies should employ standardized detection methods to evaluate dynamic changes in multiple cytokines across different time points in large-scale populations, thereby identifying which cytokines are most suitable for predicting the occurrence of IMH.
Other cytokines/chemokines may also predict IMH risk. For instance, elevated serum levels of soluble CD163 (sCD163) have been observed in IMH patients, suggesting the CD163/sCD163 axis as a potential biomarker.30 Additionally, molecules implicated in DILI, such as IL-33, growth factors (EGF, HGF), metalloproteinases, tissue inhibitors of metalloproteinases, and damage-associated molecular patterns, may hold potential as IMH biomarkers given the pathological similarities between these conditions.95 Large prospective cohorts will be needed to develop and validate multi-omics models that combine cytokine profiles with clinical and immunophenotypic data to improve IMH prediction and mechanistic insight.
Genetic and HLA markers
Genetic predisposition, particularly single-nucleotide polymorphisms (SNPs) in immune-associated genes and HLA profiles, is increasingly recognized as a determinant of IMH. Recent research has highlighted the role of SNPs and HLA in predicting IMH occurrence, as shown in Table 4.23,85,96–101
Table 4Gene-related biomarkers for predicting IMH development
| Gene-related biomarkers | Study design | N | Cancer type | Timing of measurement | Association | Significance | References |
|---|
| Single-nucleotide polymorphisms (SNPs) |
| PTPN11 333–223A>G | Retrospective | 322 | NSCLC | Before ICI treatment | PTPN11 333–223A>G was associated with an increased risk of IMH in the exploration cohort | OR = 2.42; 95% CI: 1.061–5.523; p = 0.036 | Bins et al.96 |
| EDIL3,SEMA5A,GABRP, SLCO1B1,SMAD3 | Prospective/ retrospective | 57 | Melanoma Lung cancer | Before ICI treatment | rs1862167 in EDIL3, rs35719165 in SEMA5A, rs73800947 in GABRP, rs34234515 in SLCO1B1 and rs12913535 in SMAD3 were strongly linked to IMH | OR = 2.08–2.4; p < 0.01 | Fontana et al.97 |
| GABRP, PACRG, RGMA | Retrospective | 69 | Melanoma | Before ICI treatment | GABRP rs11743438, GABRP rs11743735, and PACRG rs55733913 were linked to a higher risk of IMH, while RGMA rs4778080 seemed to protect against this adverse event | OR = 6.16–9.17; p < 0.05 | Rodriguez-Pinas et al.98 |
| CD274, SLCO1B1 | Prospective | 95 | Melanoma | Before ICI treatment | CNVs on CD274 and SLCO1B1 were significantly linked to IMH | p < 0.05 | Wolffer et al.23 |
| Human leukocyte antigen (HLA) genotypes |
| HLA-DRB1*04:01; HLA-DRB1*15:01-DQB1*06:02 | Prospective | 131 | Melanoma NSCLC | Before ICI treatment | HLA-DRB1*04:01 and the haplotype DRB1*15:01–DQB1*06:02 were significantly associated with IMH | p < 0.05 | Purde et al.85 |
| HLA-A homozygosity | Prospective | 95 | Melanoma | Before ICI treatment | HLA-A homozygosity was significantly associated with IMH | p = 0.015 | Wolffer et al.23 |
| HLA-DR4 | Retrospective | 132 | Melanoma | Before ICI treatment | HLA-DR4 was significantly associated with IMH | p < 0.01 | Akturk et al.99 |
| HLA-A*26:01 | Retrospective | 530 | Solid tumors | Before ICI treatment | HLA-A*26:01 was significantly associated with IMH | OR = 2.67; 95% CI: 0.92–8.31; p = 0.037 | Jiang et al.100 |
| HLA-B*27:04 | Retrospective | 117 | Solid tumors/ hematologic malignancies | Before ICI treatment | HLA-B27:04 was associated with grade 3 IMH | p = 0.007 | Titmuss et al.101 |
| HLA DR4, HLA-DRB1*15:01-DQB1*06:02 | Prospective/ retrospective | 57 | Melanoma Lung cancer | Before ICI treatment | No association between two HLA alleles and IMH | p >0.05 | Fontana et al.97 |
A retrospective study of 322 nivolumab-treated patients assessed the association with irAEs for seven specific SNPs in PDCD1, PTPN11, ZAP70, and IFNG genes via TaqMan allelic discrimination assays.96 Specifically, PTPN11 333–223A>G was associated with an increased risk of IMH in the exploration cohort; however, this association was not replicated in a validation cohort, highlighting the challenges of population-specific genetic effects and the necessity for large-scale validation. The study by Fontana et al. selected candidate gene variants associated with IMH risk in 57 high-causality IMH cases from the Drug-Induced Liver Injury Network.97 They investigated 25 candidate genes and target SNPs in 5 candidate genes using an Illumina MiSeq platform, finding that rs1862167 in EDIL3, rs35719165 in SEMA5A, rs73800947 in GABRP, rs34234515 in SLCO1B1, and rs12913535 in SMAD3 were strongly linked to IMH compared to population controls. Recently, Rodríguez-Piñas et al. conducted a multi-center study to explore tumor SNPs associated with the risk of IMH using a MassARRAY platform.98 Significant associations were identified between IMH risk and 4 of the 20 SNPs. Three SNPs— GABRP rs11743438, GABRP rs11743735, and PACRG rs55733913—were linked to a higher risk of IMH, whereas RGMA rs4778080 appeared to protect against this adverse event. These findings suggest that SNPs could serve as useful biomarkers to predict IMH risk, requiring confirmation in larger patient cohorts.
Beyond SNPs, other genetic alterations, such as small sequence variations and copy number variations, have been explored as potential biomarkers for IMH. A prospective study by Wolffer et al. identified several genes, including SMAD3, PRDM1, IL1RN, CD274, SLCO1B1, TSHR, and FAN1, as being associated with irAEs, particularly organ-specific events.23 Notably, copy number variations in CD274 (p = 0.043) and SLCO1B1 (p = 0.010) were significantly linked to hepatitis, further underscoring the role of genetic variations in IMH risk. Additionally, specific gene expression profiling has also emerged as a valuable tool for identifying IMH. For example, Zeng et al. demonstrated significant upregulation of the IL-1β gene and other inflammation-related genes in tumor samples, such as HLA-C, IL-2, TNFRSF14, and VEGFRB, in patients with grade ≥3 IMH compared to those without irAEs.88
HLA profiles have been extensively linked to susceptibility to immune-mediated diseases and cancer.102 Evidence suggests that specific HLA genotypes are associated with organ-specific irAEs, including IMH. In a prospective cohort study of 131 cancer patients assessing the association between HLA and IMH, Purde et al. observed that two HLA alleles, DRB1*04:01 and the haplotype DRB1*15:01–DQB1*06:02, were nominally significantly associated with the risk of IMH development in NSCLC patients.85 However, this association was absent in the overall patient cohort or after correction for multiple comparisons, highlighting the need for validation in larger studies. Another prospective study by Wolffer et al. found that HLA-A class I homozygosity was significantly linked to the occurrence of IMH in melanoma patients.23 Similarly, Akturk et al. conducted a case-control study to evaluate the association between the presence of HLA-DR alleles and irAEs in advanced melanoma patients treated with ICIs and found that HLA-DR4 was significantly associated with IMH.99 Additionally, a large cohort study involving 530 cancer patients identified several HLA types associated with organ-specific irAEs, including a significant link between HLA-A*26:01 and elevated bilirubin levels.100 Recently, Titmuss et al. analyzed 117 patients who received ICI treatment through the ongoing Personalized OncoGenomics program, reporting that MHC class I alleles in the HLA-B27 family were associated with grade 3 IMH (p = 0.007).101 Collectively, these findings suggest that pre-treatment HLA profiling could help identify patients at risk for specific irAEs, particularly IMH, following ICI therapy.
However, not all studies have confirmed these associations. For instance, Fontana et al. reported no significant associations between the overall HLA DR4 or HLA-DRB1*15:01–DQB1*06:02 haplotype and the occurrence of IMH.97 This discrepancy may be attributed to several factors, including relatively small sample sizes, heterogeneity in patient cohorts, divergent study methodologies, and lack of comprehensive genome-wide data. To clarify the role of HLA genotypes in IMH development, future studies should involve larger, well-defined cohorts of patients receiving uniform treatment regimens for the same tumor type. Direct comparisons between ICI-treated patients who develop IMH and those who do not will be essential to validate the potential of these genetic variants as predictive biomarkers.
Gut microbiome
The gut microbiome is a key regulator of immune homeostasis, and its composition has emerged as a promising predictor for both the efficacy and toxicity of ICIs.103 For instance, patients treated with ipilimumab who exhibited a microbiome enriched with Faecalibacterium and Firmicutes at baseline were found to have a higher risk of ICI-related colitis.104 However, this specific association between the gut microbiome and IMH has not yet been directly established.
Evidence from other organ systems may provide a rationale for a potential link. The gut-liver axis, for instance, demonstrates how intestinal microbes can influence extra-intestinal immunity. Alterations in the gut microbiome have been implicated in hepatocyte injury and immune-mediated liver dysfunction, such as in autoimmune hepatitis.105,106 Furthermore, specific microbes, like Veillonella (a member of the Firmicutes phylum), are frequently enriched in liver diseases and cancers.107,108 Recently, a study by Ryan et al. found a significant correlation between Veillonella abundance and the severity of ICI-related hepatotoxicity, suggesting its potential as a microbial biomarker for hepatic irAEs.109 This mechanistic insight provides a basis for the hypothesis that gut-derived bacterial signals, known to influence systemic inflammation, could also influence the development of IMH. Therefore, future high-quality studies are urgently needed to directly determine whether specific microbial markers can reliably assess IMH risk prior to or during ICI therapy.
Future perspectives
With the increasing use of ICIs in various cancers, IMH has emerged as a significant clinical challenge due to its potential impact on treatment efficacy and patient survival. This review synthesized current evidence on risk factors and potential biomarkers for IMH prediction. While factors such as specific demographic characteristics, pre-existing conditions, particular cancer types, and combination ICI regimens have been associated with increased IMH risk, and various biomarkers, including circulating blood cell counts, autoantibodies, cytokines, and genetic profiles, demonstrate promise in predicting IMH, none of the proposed biomarkers can currently be applied in clinical practice to accurately predict its occurrence.
The current evidence base exhibits significant limitations that hinder clinical application. First, most studies are retrospective and suffer from substantial heterogeneity in patient populations, ICI regimens, and detection methods, leading to inconsistent findings. Second, the underlying mechanisms of IMH remain poorly understood, and inconsistent diagnostic and grading criteria for IMH across different studies hinder the rational selection of biomarkers. Furthermore, existing studies focus on single categories of biomarkers, with a lack of integration of multi-dimensional, cross-omics approaches, thereby limiting the development of robust prediction models. Crucially, while an ideal biomarker should enable both pre-treatment risk stratification and dynamic monitoring during therapy, most studies to date have focused on static pre-treatment assessment. The absence of longitudinal data limits the clinical translation and predictive utility of biomarkers.
It is unlikely that a single risk factor or biomarker will be specific or sensitive enough to predict irAE development accurately. Given that several mechanisms are involved in IMH, a combination of multiple biomarkers, such as blood cell counts, autoantibodies, cytokine levels, genetic markers, and microbiome, is essential to identify risk stratification and personalize monitoring strategies to prevent the occurrence of IMH. In a retrospective study, Zheng et al. developed a clinical risk score to predict immune-mediated liver injury caused by sintilimab; multi-factor prediction models that integrate clinical characteristics, blood cell counts, and liver function tests to predict IMH have also been established.79,110 Artificial intelligence and machine learning algorithms could be employed to improve the accuracy of predictive models in this setting. Advancements in multi-omics technologies, including genomics, transcriptomics, proteomics, and metabolomics, hold promise for uncovering novel biomarkers and elucidating the molecular mechanisms of irAEs.111
To address these gaps, well-designed prospective studies in large, multi-center cohorts are imperative. Future studies should focus on developing and validating prediction models for IMH that integrate multi-omics biomarkers and comprehensive clinical data using artificial intelligence and machine learning. It is also essential to implement longitudinal monitoring of biomarkers from baseline through treatment, onset, and resolution of IMH to capture their dynamic profiles. Standardized protocols for biospecimen collection and uniform detection methods must be established to ensure data comparability across studies. Furthermore, stratified analyses accounting for variables such as tumor type, ICI regimen, and baseline clinical characteristics are necessary to enhance model accuracy and clinical applicability. Additionally, further investigation into the immunopathological mechanisms of IMH should be pursued to identify novel therapeutic targets and strategies, which may simultaneously yield new predictive biomarkers. By leveraging multi-dimensional data through advanced technologies and validating them in large, prospective clinical cohorts, it will be possible to identify high-risk populations before treatment initiation, guide personalized monitoring strategies during therapy, and ultimately reduce the incidence and severity of IMH, thereby enhancing the safety and efficacy of cancer immunotherapy.