Metabolomics Reveals Potential Biomarkers for Early Detection of irAEs in ICI-Treated Patients

Abstract

Purpose

To identify predictive metabolic biomarkers for immune-related adverse events (irAEs) in cancer patients treated with immune checkpoint inhibitors (ICIs) using metabolomics, supporting early detection and intervention.

Patients and Methods

Fifty-five ICI-treated cancer patients from Changxing People’s Hospital were divided into irAEs-positive (34 cases) and irAEs-negative (21 cases) groups after 12-month follow-up. Pretreatment serum samples were analyzed by untargeted UPLC-MS metabolomics. PCA, OPLS-DA, and KEGG pathway enrichment were used to screen differential metabolites and key pathways.

Results

Seventy significant differential metabolites (eg, ursodeoxycholic acid, uric acid) were identified. β-Alanine metabolism, pentose phosphate pathway, and coenzyme A biosynthesis were closely correlated with irAEs. Five metabolites showed AUC 0.7–0.9 with favorable predictive performance.

Conclusion

Metabolomics reveals specific metabolites and metabolic pathways linked to ICI-induced irAEs, providing potential biomarkers for early prediction and new insights into irAEs pathogenesis to optimize clinical management.

Introduction

In recent years, cancer immunotherapy, particularly ICIs, has achieved significant advancements in the clinical treatment of malignancies.1 ICIs, which include cytotoxic T-lymphocyte-associated antigen 4 inhibitors, programmed cell death protein 1 inhibitors, and programmed death-ligand 1 inhibitors, have expanded their application due to their superior clinical efficacy and safety profile.2 However, while these therapies inhibit tumor growth, they may also trigger excessive activation of immune cells, leading to immune tolerance imbalance and the emergence of irAEs, exhibiting an occurrence frequency spanning from 15% to 90%.3,4

The delayed onset and inflammatory nature of irAEs pose unique challenges in their management and treatment. These adverse events primarily manifest in organs such as the colon, liver, lungs, pituitary gland, thyroid, and skin, but can also affect rarer sites like the heart and nervous system.5 Due to the complexity of irAEs, presently, early prognostic indicators are insufficiently identified, which limits the effective prevention and management of these adverse events in clinical settings.

Metabolomics, as an emerging omics discipline, offers new perspectives for disease diagnosis and treatment by analyzing the metabolic responses of organisms to pathophysiological stimuli. Since its first proposal in 1999, metabolomics technology has been extensively applied in pharmacotoxicology research and the analysis of pathophysiological differences.6,7 In the context of ICI therapy, metabolomics can provide a comprehensive assessment of the patient’s cellular state, considering genetic regulation, changes in enzymatic activity, and metabolic reactions, thus identifying potential risks for irAEs.8

Given the current gap in research on predictive biomarkers for irAEs, this study aims to utilize metabolomics technology to screen for specific predictive biomarkers in tumor immunotherapy. We hope that early prediction of these biomarkers will provide an assessment of potential adverse event risks during clinical use of ICIs, thereby enabling dose adjustments, proactive prevention, and early interventions to avoid or mitigate the occurrence of irAEs. This research not only aids in advancing personalized treatment management practices but may also reduce the incidence and severity of irAEs, ultimately enhancing patients’ life quality.

Materials and Methods

Study Population

Patients diagnosed with malignant tumors and treated with ICIs at Changxing People’s Hospital, Zhejiang Province, from April 2023 to August 2024 were included in this study. Inclusion criteria were: confirmed pathological diagnosis, no prior use of ICIs and no contraindications for ICIs, planned treatment with a regimen containing ICIs, normal blood routine and liver and kidney functions at admission, and an expected survival period of more than 3 months. Exclusion criteria were: patients who developed infectious diseases or received glucocorticoid therapy during treatment, patients who died during the treatment course, patients who discontinued ICIs due to intolerance or poor efficacy, and patients who refused to participate in the study. Patients were enrolled in the study for a 12-month follow-up period starting from the initiation of ICI therapy. The intensity and classification of adverse events were evaluated based on the Common Terminology Criteria for Adverse Events (CTCAE) version 5.0. Throughout the study, efforts were made to minimize attrition by conducting regular follow-ups with patients and maintaining a supportive relationship with the clinical team to encourage continued participation. The final analysis included all patients who completed the 12-month follow-up period, and the reasons for dropout were documented and analyzed to assess their potential impact on the study results. Based on the occurrence of immune-related adverse events (irAEs) identified during the follow-up, the enrolled patients were categorized into two groups: the irAEs group and the control group. The study population included both male and female patients, and the potential influence of sex on the development of irAEs was considered in the analysis, ensuring a balanced representation across both groups to account for any gender-related differences in treatment response. Additionally, we meticulously documented the age and body weight of enrolled patients, along with other relevant demographic data, to ensure a comprehensive representation of the patient population’s characteristics that might influence the study outcomes.

Informed consent forms were signed by all participants. The study protocol received approval by the Medical Ethics Committee of Changxing People’s Hospital (Ethics No.: 2023-KY-053) and carried out in alignment with the tenets of the Declaration of Helsinki. The study has been registered and made public with the Chinese Clinical Trial Registry Center, ensuring transparency and accessibility of the trial’s methods and findings.

In this observational study, subjects were assigned to the irAEs group or the control group based on post-treatment outcomes, and without randomization, to reflect the natural course of immune-related adverse events following ICI therapy. Furthermore, to reduce potential bias, a blinded approach was implemented, where both investigators and subjects were kept unaware of the group allocation throughout the study conduct and analysis phases.

Specimen Collection and Processing

During the treatment cycle, 2.5 mL of peripheral blood was collected from each patient prior to the first dose of ICI therapy. Serum Specimens were subjected to centrifugation at 3000 rpm for a duration of 10 minutes at 4°C. A total of 500 μL of supernatant was aspirated and stored at −80°C until analysis.

Untargeted Metabolomics Materials and Methods

Experimental Reagents and Instruments

Experimental reagents and instruments are listed in Table 1 and Table 2.

Table 1.

List of Experimental Reagents

Name	CAS	Purity	Brand
Methanol	67-56-1	LC-MS grade	CNW Technologies
Acetonitrile	75-05-8	LC-MS grade	CNW Technologies
Ammonium acetate	631-61-8	LC-MS grade	SIGMA-ALDRICH
Ammonium hydroxide	1336-21-6	LC-MS grade	CNW Technologies
Ultra-pure water (ddH2O)	–	–	Watsons
Acetic acid	64-19-7	LC-MS grade	SIGMA-ALDRICH
2-Propanol	67-63-0	LC-MS grade	CNW Technologies

Table 2.

List of Experimental Instruments

Instrument	Model	Brand
Ultra-high-performance liquid chromatography	Vanquish	Thermo Fisher Scientific
High-resolution mass spectrometer	Orbitrap Exploris 120	Thermo Fisher Scientific
Centrifuge	Heraeus Fresco17	Thermo Fisher Scientific
Balance	BSA124S-CW	Sartorius
Ultrasonic cleaner	PS-60AL	Shenzhen Leadsonic Co., Ltd.
Homogenizer	JXFSTPRP-24	Shanghai Jingxin Technology Co., Ltd.
Freeze dryer	LGJ-10C	Beijing Four-ring Furi Technology Development Co., Ltd.

Sample Preparation

Metabolite Isolation: Transfer 50 μL of the sample into an EP tube, then add 200 μL of extraction solvent (methanol to acetonitrile ratio of 1:1,V/V), which contains an isotope-labeled internal standard. Vortex the mixture for 30 seconds, followed by sonicated for 10 minutes in an ice-water bath. Let the mixture stand at −40°C for 1 hour, then centrifuge at 4°C, 12000 rpm (centrifugal force 13800×g, radius 8.6 cm) for a duration of 15 minutes. The extraction solvent contained a commercially available, proprietary isotope-labeled internal standard mixture (provided by BiotreeDB, V3.0) to monitor the reproducibility of the sample preparation and instrumental analysis. The supernatant is collected into sample vials for subsequent analysis.

Preparation of Quality Control Samples: All samples are combined to create a quality control (QC) specimen, which is analyzed alongside the study samples to monitor system stability and the dependability of experimental outcomes.

Chromatography and Mass Spectrometry Conditions

Chromatography Conditions: Ultra-performance liquid chromatography was employed using a Waters ACQUITY UPLC BEH Amide column (2.1 mm × 50 mm, 1.7 μm) for the chromatographic separation of target compounds. The mobile phase A consisted of an aqueous solution containing 25 mmol/L ammonium acetate and 25 mmol/L ammonium hydroxide, wherein the mobile phase B constituted acetonitrile. The autosampler’s thermal condition was calibrated to 4°C, with an aliquot volume of 2 μL per injection. QC specimens were intermittently integrated within the analytical sequence to surveil and appraise the apparatus’s constancy as well as the veracity of the experimental metrics.

Mass Spectrometry Conditions: Mass spectrometry was conducted utilizing an Orbitrap Exploris 120 mass spectrometer, orchestrated by Xcalibur 4.4 software (Thermo Fisher Scientific) for the acquisition of both MS1 and MS2 spectra. The specific parameters were delineated as follows: sheath gas flux at 50 Arb, auxiliary gas flux at 15 Arb, capillary thermic condition at 320°C, full MS discernment at 60,000, MS/MS discernment at 15,000, collision energy SNCE at 20/30/40, and ionization potential at +3.8 kV (positive polarity) or −3.4 kV (negative polarity).

Data Processing and Statistical Analysis Methods

Related Metrics

The group size was determined based on a priori power calculations that ensured a sample size sufficient to detect a clinically meaningful difference in the prevalence of irAEs between the groups, with an alpha level of 0.05 and a power of 0.80. Data were organized and calculated using Excel 2023, with grouped data expressed as MEAN ± SD. Statistical analysis was conducted using GraphPad Prism 9. The independent sample t-test was employed to evaluate significant differences, with P < 0.05 considered statistically significant.

Untargeted Metabolomics Data Processing and Analysis

Raw Data Preprocessing

To attenuate the influence of systemic discrepancies and better highlight biological significance, a series of preprocessing steps were performed on the raw data, including outlier filtering, missing value handling, and data normalization.9 The unprocessed data were transmuted to mzXML format via ProteoWizard 3.0, generating data files with matched m/z ratios, retention times, and peak intensities. Metabolite identification was conducted using the R program with the BiotreeDB (V3.0) database.

Differential Metabolite Analysis

The data files were imported into SIMCA (V16.0.2) software for Principal Component Analysis (PCA), which assessed the stability of the instrumentation methods based on QC sample clustering, as well as the distribution trends and outliers among the samples. Subsequently, OPLS-DA was performed, and the model’s reliability and applicability were evaluated using a 200-fold permutation test. Biomarkers were identified based on precise molecular weights using the BiotreeDB (V3.0) database. Metabolite identification levels were annotated according to the Metabolomics Standards Initiative (MSI):9 Level 1, metabolites in the sample match standards in MS1, MS2, and RT; Level 2, metabolites match public database entries in MS1 and MS2; Level 3, metabolites match compounds in MS1, MS2, and RT; Level 4, unknown compounds.

Based on the VIP values obtained from the OPLS-DA model, which reflect the contribution of each variable to the grouping, significant differential metabolites with biological relevance were identified. The study used VIP > 1.0, FC < 0.67 or > 1.50, and P < 0.05 as criteria to screen for differential metabolites. Univariate Analysis (UVA) was further employed to validate whether the metabolites showed significant differences. The results were visualized using a volcano plot. Metabolic pathway enrichment was performed using MetaboAnalyst 4.0, and Receiver Operating Characteristic (ROC) curves were used to finalize the screening and evaluation of biomarkers.

Results

Clinical Baseline Characteristics

This investigation encompassed 55 patients diagnosed with malignancies, each of whom underwent a minimum of one cycle of immune checkpoint inhibitor (ICI) therapy at Changxing People’s Hospital, Zhejiang Province. The cohort was dichotomized into two groups: 21 individuals who did not manifest any immune-related adverse events (irAEs) (control group) and 34 individuals who experienced at least one irAE (irAEs group). Among the 34 patients who developed irAEs, the most common manifestations were thyroid dysfunction (24 events, 70.6%), dermatitis (9 events, 26.5%), colitis/diarrhea (3 events, 8.8%), and pneumonitis (2 events, 5.9%). If a patient had multiple events, they were counted in each category. According to CTCAE v5.0, 30 patients experienced Grade 1–2 events, and 4 patients experienced Grade 3–4 events. Statistical scrutiny disclosed no significant disparities in baseline demographics, including mean age, gender distribution, oncological categorization, and staging between the two cohorts (Table 3).

Table 3.

Fundamental Attributes of Patients

Clinical Data	Control Group (n=21)	irAEs Group (n=34)	P-value
Age (years)	67.81 (52–81)	68.22 (41–92)	0.885
Gender (Male %)	71.4%	75%	0.773
Lipase	36.92±13.59	43.74±34.03	0.410
Globulin	30.03±6.09	29.30±6.48	0.697
CRP	20.61±35.78	18.12±24.80	0.774
Neutrophils	4.33±2.37	4.61±3.19	0.734
Lymphocytes	1.23±0.54	1.43±1.81	0.627
NLR	4.10±3.10	4.46±3.82	0.715
Liver Function
ALT (U/L)	17.50±12.06	17.72±9.39	0.940
AST (U/L)	24.70±14.81	23.77±10.75	0.789
Total Bilirubin (μmol/L)	11.73±5.22	13.71±7.19	0.281
Direct Bilirubin (μmol/L)	2.49±1.92	3.41±4.03	0.335
Renal Function
BUN (mmol/L)	6.32±1.94	6.01±1.91	0.562
Scr (μmol/L)	83.98±27.37	83.42±24.10	0.937
Type of ICIs			0.296
Pembrolizumab	2	4
Sintilimab	12	14
Toripalimab	2	1
Camrelizumab	-	6
Durvalumab	2	3
Tislelizumab	3	8
Cancer Type			0.545
Lung Cancer	13	19
Gastric Cancer	3	4
Liver Cancer	0	3
Others	5	10
Tumor Stage			0.697
Stage I	1	3
Stage II	2	1
Stage III	3	6
Stage IV	15	26

Untargeted Metabolomic Analysis

Metabolomic Data Reliability

The raw data comprised 7 QC samples and 57 experimental specimens, yielding an extraction of 12,621 peaks; after preprocessing, 11,030 peaks were retained. The stability of the detection was assessed by evaluating the variation in peak height of the internal standard across QC samples. As demonstrated in Figure 1a and b, the retention time and response intensity of the internal standard in QC samples exhibited minimal variation, indicating robust stability in instrument data acquisition. An unguided Principal Component Analysis (PCA) was utilized to evaluate the dispersion of the QC specimens. The QC samples clustered tightly in the PCA score scatter plot, with internal standard responses across different samples displaying stability, nearly all within two standard deviations (Figure 1c). These findings confirm that the analytical methodology is stable and reproducible, ensuring the reliability of the generated data.

Figure 1.

(a) Extracted Ion Chromatogram (EIC) of the Internal Standard in Negative Ion Mode Across All QC Samples; (b) EIC of the Internal Standard in Positive Ion Mode Across All QC Samples; (c) One-Dimensional PCA-X Distribution of QC Samples.

Screening of Differential Metabolites

After excluding outlier samples from the raw data, 21 cases remained in the control group, and 34 cases in the irAEs group. Figure 2 illustrates the PCA score scatter plot encompassing all specimens, inclusive of QC samples. There was no apparent separation between the sample points of the control and irAEs groups, indicating that the PCA model was insufficient for distinguishing between the two groups, necessitating a reanalysis using OPLS-DA.

Figure 2.

The PCA score scatter diagram encompassing the entirety of samples, QC specimens included.

The data were log-transformed and UV-scaled using SIMCA (V16.0.2) software before OPLS-DA modeling of the first principal component. To validate the model’s robustness, a 7-fold cross-validation was employed. The model’s quality was then assessed using the cross-validated R²Y (the model’s explanatory power for the categorical variable Y) and Q² (the model’s predictive power). Finally, the model’s validity was further verified through a permutation test by randomly rearranging the order of the categorical variable Y repeatedly to derive diverse random Q² values.

As illustrated in Figure 3, the OPLS-DA analysis revealed a significant distinction in plasma metabolic profiles of the categorical variable Y iteratively to generate a range of stochastic Q² values, with an R²Y value of 0.847 and a Q² value of 0.215. This model demonstrates excellent fit and predictive capacity, suggesting that there are discernible differences in metabolic profiles between patients who develop irAEs following treatment with immune checkpoint inhibitors and those who do not.

Figure 3.

The OPLS-DA model score scatter plot comparing the control and irAEs groups.

To identify potential metabolic biomarkers, VIP scores from the OPLS-DA score plot were utilized to assess each metabolite’s contribution to intergroup differentiation. Metabolites with VIP values >1.0 and P-values <0.05 were selected as candidate biomarkers, with the overall distribution of these differential metabolites depicted in Figure 4. In this figure, the FC represents the magnitude of alteration in the baseline levels of metabolites between the irAEs and control groups. The volcano plot, constructed based on FC values, P-values, and VIP scores, provides a visual representation of the overall distribution of differential metabolites between the groups. Each point in the volcano plot corresponds to a specific metabolite detected in this study. The horizontal axis represents the logarithmic fold change (log2FC) between the groups, while the vertical axis denotes the P-value from the Student’s t-test, transformed to a negative logarithm base 10. The size of each scatter point correlates with the VIP value from the OPLS-DA model, where larger points indicate higher VIP values. Metabolites with significant upregulation are highlighted in red, those with marked downregulation in blue without significant differences are shown in gray. The preliminary screening identified 1231 candidate differential metabolites, with 761 being upregulated and 470 downregulated.

Figure 4.

Volcano Plot of Differential Metabolite Screening between Control and irAEs Groups.

Further refinement of differential metabolites was conducted using an FC criterion exceeding 1.5 or falling below 0.67, with an identification confidence level exceeding Level 3. This process ultimately identified 70 significantly altered metabolic biomarkers. The top 30 of these differential metabolites are listed in Table 4.

Table 4.

Differential Metabolites List (Top 30)

NO	Compound Name	Molecular Formula	Ion Mode	m/z	RT	Regulation
1	Ursocholic acid	C24H40O5	NEG	407.2796	138.8	↑
2	Allocholic acid	C24H40O5	NEG	407.2796	138.8	↑
3	Uric acid	C5H4N4O3	NEG	167.0207	193.1	↑
4	Cholic acid	C24H40O5	NEG	407.2796	138.8	↑
5	Glyceric acid	C3H6O4	NEG	105.0191	182.1	↑
6	Biliverdin	C33H34N4O6	NEG	581.2403	149.5	↑
7	Uracil	C4H4N2O2	NEG	111.0198	35.7	↑
8	Glutamate	C5H9NO4	NEG	146.0456	227.7	↑
9	Pantothenic acid	C9H17NO5	NEG	218.103	162.9	↑
10	Threonic acid	C4H8O5	NEG	135.0296	186.1	↑
11	Bilirubin	C33H36N4O6	NEG	583.2558	27.7	↑
12	Glucuronic acid	C6H10O7	NEG	193.0349	222.2	↑
13	Xanthine	C5H4N4O2	NEG	151.0258	124.7	↑
14	cis-4-Decenoylcarnitine	C17H31NO4	POS	314.2315	123.3	↑
15	2-Hydroxyethanesulfonic acid	C2H6O4S	NEG	124.9911	86	↑
16	Pentaethylene glycol	C10H22O6	POS	239.148	30.3	↑
17	N1,N8-Diacetylspermidine	C11H23N3O2	POS	230.1856	210	↑
18	Hyocholic acid	C24H40O5	NEG	407.2796	138.8	↑
19	alpha-Muricholic acid	C24H40O5	NEG	407.2796	138.8	↑
20	beta-Muricholic acid	C24H40O5	NEG	407.2796	138.8	↑
21	omega-Muricholic acid	C24H40O5	NEG	407.2796	138.8	↑
22	Oxypurinol	C5H4N4O2	NEG	151.0258	124.7	↑
23	3-Succinoylpyridine	C9H10ClNO3	NEG	178.0505	110.1	↑
24	Homovanillic acid sulfate	C9H10O7S	NEG	261.0071	99.5	↑
25	(S)-3,4-Dihydroxybutyric acid (lithium hydrate)	C4H8O4	NEG	119.0347	164.2	↑
26	Mintsulfide	C15H24S	POS	237.1688	21.9	↑
27	1-(10Z-Heptadecenoyl)-sn-glycero-3-phospho-(1′-rac-glycerol)	C23H45O9P	NEG	495.2778	15.3	↓
28	SM(d17:1/24:1(15Z))	C46H92N2O6P	POS	799.6682	56.9	↓
29	(1R,2R,4R)-4-(2-Hydroxypropan-2-yl)-1-methylcyclohexane-1,2-diol	C10H20tO3	POS	189.1479	24.9	↑
30	5-(2-Methylbutan-2-yl)-4,5,6,7-tetrahydro-2H-indazole-3-carbohydrazide	C13H22N4O	POS	251.1846	18.9	↑

Notes: ↑ indicates up-regulation in the irAEs group; ↓ indicates down-regulation in the irAEs group.

KEGG Pathway Enrichment and Metabolic Pathway Analysis of Differential Metabolites

Employing MetaboAnalyst 4.0, the delineated differential metabolites underwent KEGG pathway enrichment scrutiny, culminating in the discernment of 15 distinct biosynthetic and catabolic routes (Figure 5). The abscissa delineates the proportion of annotated differential metabolites within each discrete pathway relative to the cumulative annotated differential metabolites. The ordinate enumerates the appellations of the enriched KEGG metabolic cascades. As depicted, the enriched biosynthetic pathways encapsulate: histidine catabolism, oncogenic central carbon flux, glyoxylate and dicarboxylate turnover, pentose phosphate flux, biliary excretion pathway, proteolysis and absorption pathway, carbon flux, ABC transporter-mediated translocation, pantothenate and coenzyme A biosynthetic routes, taurine and hypotaurine catabolism, β-alanine turnover, retrograde endocannabinoid signaling pathway, neurodegenerative pathways, and the FoxO signal transduction cascade.

Figure 5.

KEGG classification of the differential metabolites.

KEGG enrichment merely annotates the pathways involving these differential metabolites; however, to comprehensively understand the impact of experimental conditions on the alteration of metabolites and their corresponding pathways, further analysis is required to elucidate the influence of these metabolites on their associated pathways. Through an integrated analysis of differential metabolites within their respective pathways, encompassing both enrichment and topological assessments, we can refine the pathway selection process to identify the key pathways most closely correlated with metabolic discrepancies.10

At the outset, we aligned the differential metabolites with established metabolic databases, including KEGG and PubChem. Upon obtaining the matching information for these metabolites, we conducted pathway searches and metabolic pathway analyses specific to Homo sapiens. The results of the metabolic pathway analysis are visualized in a bubble plot (Figure 6).

Figure 6.

Pathway analysis through a bubble plot.

In the bubble plot, each sphere symbolizes a metabolic pathway. The horizontal axis and the size of the bubble denote the impact factor derived from the topological analysis, with larger bubbles indicating greater impact factors. The vertical axis and bubble color correspond to the P-value from the enrichment analysis (expressed as the negative natural logarithm, -ln(P)), where deeper colors indicate smaller P-values and, consequently, more significant enrichment. The findings reveal a significant correlation between the occurrence of irAEs induced by immune checkpoint inhibitors and the β-alanine metabolism pathway (P=0.0055816), pentose phosphate pathway (P=0.0081515), and pantothenate and CoA biosynthesis pathway (P=0.048889), with all P-values being less than 0.05.

The outcomes of the metabolic pathway analysis are also illustrated using a treemap (Figure 7). In this treemap, each rectangle signifies a metabolic pathway, with the size representing the impact factor from the topological analysis, where larger rectangles indicate greater impact factors. The color of each rectangle reflects the P-value from the enrichment analysis, with deeper hues indicating smaller P-values and, thus, more significant enrichment.

Figure 7.

Pathway Analysis Treemap.

Prognostic Framework for Immunologically-Mediated Adverse Events in Tumor Immunotherapy

To evaluate the potential of these differential metabolites as predictive biomarkers for irAEs, we generated ROC curves and calculated the AUC values. Among the 70 identified differential metabolites, five exhibited AUC values ranging between 0.7 and 0.9, surpassing random chance and indicating predictive value. These metabolites include (S)-3,4-dihydroxybutyric acid (lithium hydrate) (Supplementary Figure 1), 1-(10Z-heptadecenoyl)-sn-glycero-3-phospho-(1′-rac-glycerol) (Supplementary Figure 2), (1R,2R,4R)-4-(2-hydroxypropan-2-yl)-1-methylcyclohexane-1,2-diol (Supplementary Figure 3), 5-(2-methylbutan-2-yl)-4,5,6,7-tetrahydro-2H-indazole-3-carbohydrazide (Supplementary Figure 4), and mintsulfide (Supplementary Figure 5). Representative MS/MS spectra of the five identified metabolites are provided in Supplementary Figures 1–5. Figure 8 presents the ROC curves for these differential metabolites with AUC values exceeding 0.7. These findings suggest potential biomarkers for the early diagnosis and management of irAEs.

Figure 8.

ROC Curves for Differential Metabolites with AUC > 0.7.

Discussion

With the expanding clinical application of ICIs in oncological therapies, the incidence of irAEs has become increasingly prevalent, affecting various organ systems and presenting a wide spectrum of clinical manifestations, thereby significantly impacting cancer treatment outcomes.11 The etiology of irAEs associated with ICIs is likely multifactorial, potentially involving the activation of reactive T cells, cytokines, and antibodies through various immunological pathways.12 Currently, there is no consensus within existing scholarly discourse pertaining to hazard determinants or prognostic biomarkers for irAEs, particularly those related to immune-mediated thyroid adverse events. According to existing studies, irAEs have been correlated with factors such as age,13 gender,14,15 body mass index (BMI),16 ethnicity,17 comorbidities,18 choice of ICIs,19 and cumulative dosage.20 In this study, baseline characteristics such as age and gender were comparable between groups, suggesting that the metabolic alterations observed are more likely attributable to the biological response to ICIs rather than demographic confounders.

In this study, we employed an untargeted metabolomics approach to characterize the plasma metabolic profiles associated with irAEs in patients receiving ICI therapy. Our analysis revealed significant alterations in specific metabolites linked to immune cell activation and inflammatory responses.21 While we observed perturbations in well-known inflammatory mediators such as uric acid (UA) and cholic acid, which corroborate existing literature on immune activation,11,22,23 the primary focus of this discussion is directed towards a novel set of five differential metabolites. These five metabolites, identified through ROC curve analysis with AUC values ranging from 0.7 to 0.9, demonstrate superior predictive performance and offer deeper mechanistic insights into the pathogenesis of irAEs.

Beyond the analysis of individual well-characterized metabolites, our ROC curve analysis identified five differential metabolites with AUC values ranging from 0.7 to 0.9, demonstrating promising predictive performance for irAEs. A thorough examination of the biological relevance of each metabolite is warranted, as these compounds may serve not only as predictive biomarkers but also as windows into the underlying pathophysiological mechanisms of irAEs.

The first biomarker, (S)-3,4-dihydroxybutyric acid (lithium salt hydrate), is a short-chain hydroxy acid belonging to the family of hydroxylated fatty acid derivatives. This compound is structurally related to intermediates in the β-oxidation and ω-oxidation pathways of fatty acid metabolism, as well as to metabolites generated during oxidative stress-induced lipid peroxidation. Dihydroxybutyric acid and its analogues have been implicated in the regulation of cellular redox homeostasis, as the presence of two hydroxyl groups confers both pro-oxidant and antioxidant properties depending on the microenvironment. In the context of ICI therapy, immune hyperactivation leads to the generation of excessive reactive oxygen species (ROS), which in turn accelerates the oxidative degradation of polyunsaturated fatty acids and generates downstream hydroxylated metabolites. Ferroptosis, an iron-dependent form of regulated cell death driven by lipid peroxidation, has been increasingly recognized as a key mechanism in tumor immunotherapy responses,24,25 and the accumulation of lipid peroxidation-derived metabolites may reflect heightened oxidative stress preceding the clinical manifestation of irAEs. Furthermore, short-chain hydroxy acids have been shown to modulate the function of G-protein-coupled receptors (GPCRs), particularly hydroxycarboxylic acid receptors (HCARs), which are expressed on immune cells including macrophages and dendritic cells. The hydroxycarboxylic acid receptor 2 (HCAR2) has been demonstrated to mediate anti-inflammatory effects in experimental autoimmune encephalomyelitis,26 and HCAR2 signaling has been shown to suppress pro-inflammatory cytokine production in multiple immune cell types.27 Through this receptor-mediated mechanism, (S)-3,4-dihydroxybutyric acid may directly influence immune cell polarization and cytokine secretion profiles, thereby contributing to the dysregulated immune responses observed in irAEs. Additionally, oxidative stress-mediated disruption of antioxidant defense in T cells has been linked to autoimmune pathogenesis, as demonstrated by Hisada et al who showed that impaired antioxidant defense in CD4⁺ T cells promotes excessive interleukin-2 production in systemic lupus erythematosus.28 The favorable AUC value of this metabolite suggests that pre-treatment oxidative stress status, as captured by lipid peroxidation-derived metabolites, may be a valuable predictor of susceptibility to irAEs.

The second biomarker, 1-(10Z-heptadecenoyl)-sn-glycero-3-phospho-(1′-rac-glycerol), is a lysophosphatidylglycerol (LPG) species containing a mono-unsaturated C17 fatty acyl chain. Lysophospholipids are bioactive lipid mediators generated through the enzymatic action of phospholipases, particularly phospholipase A2 (PLA2), on membrane phospholipids, and have been comprehensively reviewed as critical mediators in health and disease.29 These mediators play critical roles in intercellular signaling, immune cell recruitment, and inflammatory responses. The glycerol headgroup of this particular LPG species distinguishes it from the more widely studied lysophosphatidic acid (LPA) and lysophosphatidylcholine (LPC), and suggests a role in mitochondrial membrane remodeling, as phosphatidylglycerol and its lysoderivatives are enriched in mitochondrial membranes, particularly as cardiolipin biosynthetic intermediates. In the setting of ICI-induced immune activation, T cells undergo profound metabolic reprogramming, shifting from oxidative phosphorylation to aerobic glycolysis, a process that necessitates extensive remodeling of mitochondrial membranes, as demonstrated by Baixauli et al who showed that an LKB1-mitochondria axis is essential for controlling T helper 17 effector function.30 The accumulation of LPG species may reflect enhanced cardiolipin turnover and mitochondrial membrane restructuring during T-cell activation. Moreover, lysophospholipids have been demonstrated to modulate CD8⁺ T cell immunosurveillance and metabolism to impair anti-tumor immunity, as shown by Turner et al who found that LPA signaling impairs CD8⁺ T cell function through metabolic reprogramming.31 Similarly, Konen et al demonstrated that autotaxin, a key lysophospholipid-generating enzyme, suppresses cytotoxic T cells via LPAR5 to promote anti-PD-1 resistance in non-small cell lung cancer.32 Hisano and Hla further reviewed the roles of bioactive lysolipids in cancer and angiogenesis, highlighting their multifaceted functions in the tumor microenvironment.33 The presence of an unusual odd-chain fatty acyl group (C17:1) is also noteworthy, as odd-chain fatty acids have been increasingly recognized as biomarkers of mitochondrial dysfunction and altered gut microbiota metabolism.34 Wei et al demonstrated that the odd-chain fatty acid pentadecanoic acid, produced by gut microbiota, can suppress non-alcoholic steatohepatitis.35 The elevated baseline levels of this LPG species in irAE-prone patients may therefore indicate a pre-existing state of mitochondrial stress or altered lipid signaling that predisposes individuals to excessive immune activation upon ICI exposure.

The third biomarker, (1R,2R,4R)-4-(2-hydroxypropan-2-yl)-1-methylcyclohexane-1,2-diol, is a chiral terpenoid-derived diol. The cyclohexane-diol scaffold with a hydroxyisopropyl substituent is characteristic of metabolites derived from the biotransformation of monoterpene or sesquiterpene precursors, potentially originating from both endogenous terpenoid metabolism and exogenous dietary or microbial sources. Terpenoid metabolites are known to possess diverse biological activities, including anti-inflammatory, immunomodulatory, and antioxidant properties. The specific stereochemistry (1R,2R,4R) of this compound suggests enzymatic origin, as enantioselective biosynthesis is a hallmark of terpene-metabolizing enzymes such as cytochrome P450 monooxygenases and epoxide hydrolases, which are abundantly expressed in hepatic and intestinal tissues. The observed alteration of this metabolite in the irAEs group may reflect changes in hepatic biotransformation capacity induced by ICI therapy. ICIs can trigger immune-mediated hepatotoxicity, one of the most common irAEs, which involves infiltration of the liver by activated T cells and subsequent hepatocyte damage. Such hepatic immune injury may alter the activity of drug-metabolizing enzymes and terpenoid-metabolizing pathways, leading to the accumulation of specific biotransformation intermediates. Additionally, terpenoid-derived metabolites have been shown to interact with nuclear receptors such as pregnane X receptor (PXR) and constitutive androstane receptor (CAR), which regulate not only xenobiotic metabolism but also immune responses through cross-talk with NF-κB and STAT signaling pathways. Zhou et al demonstrated that mutual repression between PXR and NF-κB signaling pathways links xenobiotic metabolism and inflammation,36 and Pascussi et al further elucidated the complex crosstalk among nuclear receptors controlling xenobiotic metabolism and transport.37 More recently, Wang et al showed that a ginger-derived terpenoid compound alleviates gut inflammation specifically through PXR activation,38 reinforcing the relevance of terpenoid–PXR interactions in immune-mediated tissue injury. The differential levels of this terpenoid diol may thus serve as a proxy for the interplay between hepatic metabolic capacity and immune activation status, providing a mechanistic link between ICI-induced hepatotoxicity and systemic metabolic perturbations.

The fourth biomarker, 5-(2-methylbutan-2-yl)-4,5,6,7-tetrahydro-2H-indazole-3-carbohydrazide, is a heterocyclic compound featuring an indazole core with a branched alkyl substituent and a carbohydrazide functional group. The indazole pharmacophore is widely recognized in medicinal chemistry for its anti-inflammatory and kinase-inhibitory properties, and endogenous indazole-containing metabolites have been identified in human biofluids as products of tryptophan and indole metabolism. The carbohydrazide moiety is of particular interest, as hydrazide-containing compounds can act as chelators of transition metals and as inhibitors of metal-dependent enzymes, including matrix metalloproteinases (MMPs) and myeloperoxidase (MPO), both of which are key mediators of inflammatory tissue damage. In the context of irAEs, excessive MMP activity contributes to tissue remodeling and barrier breakdown in affected organs such as the colon (colitis), liver (hepatitis), and lungs (pneumonitis). The presence of this indazole-carbohydrazide metabolite at differential levels in irAEs-prone patients may reflect altered tryptophan catabolism through the kynurenine pathway, which is a well-established immunoregulatory axis in cancer immunotherapy. Indoleamine 2,3-dioxygenase (IDO), the rate-limiting enzyme of the kynurenine pathway, is frequently upregulated in the tumor microenvironment and in immune-activated states, and its activity generates a cascade of downstream metabolites with immunosuppressive or immunostimulatory properties depending on the specific branch of the pathway that is activated. Bari et al demonstrated that microbial metabolism of tryptophan is associated with resistance to ICB therapy in renal cell carcinoma,39 and Zakharia et al reported clinical trial results combining the IDO pathway inhibitor indoximod with checkpoint inhibitors for melanoma treatment,40 underscoring the clinical relevance of this metabolic axis. Perturbations in the kynurenine cascade, as reflected by the altered levels of indazole-derived metabolites, may indicate a shift in the balance between immune tolerance and immune activation, thereby influencing the likelihood of irAE development. The branched alkyl substituent (2-methylbutan-2-yl) further suggests that this metabolite may be a product of gut microbial metabolism, as branched-chain alkyl groups are characteristic of microbial amino acid fermentation products, reinforcing the emerging role of the gut microbiome–metabolome axis in modulating ICI responses and toxicity.41,42

The fifth biomarker, mintsulfide, is a sulfur-containing organic compound. Sulfur metabolites occupy a central position in cellular defense mechanisms, as they are integral components of the glutathione (GSH) synthesis pathway, taurine and hypotaurine metabolism, and the transsulfuration pathway that links methionine metabolism to cysteine biosynthesis. Cirino et al have comprehensively reviewed the physiological roles of hydrogen sulfide (H₂S) in mammalian cells, tissues, and organs, highlighting its multifaceted functions in vascular homeostasis, inflammation regulation, and cytoprotection.43 Sulfur-containing metabolites, including various thioethers, thioesters, and sulfoxides, have been increasingly recognized as important regulators of immune cell function and inflammatory responses. The thiol-disulfide balance within cells critically determines the redox state of the intracellular milieu, which in turn governs the activation status of redox-sensitive transcription factors such as NF-κB, AP-1, and Nrf2—all of which are central to the regulation of innate and adaptive immune responses. In the specific context of ICI therapy, the activation of cytotoxic T lymphocytes leads to a burst of intracellular ROS production, which must be counterbalanced by enhanced synthesis of sulfur-containing antioxidants. Dahabieh et al recently demonstrated that the prostacyclin receptor PTGIR is an NRF2-dependent regulator of CD8⁺ T cell exhaustion, linking redox-regulatory pathways directly to the functional state of anti-tumor T cells.44 An altered level of mintsulfide in the pre-treatment serum of irAE-prone patients may therefore reflect an inherent imbalance in sulfur amino acid metabolism or a compromised antioxidant defense capacity, rendering these individuals more susceptible to oxidative stress-mediated tissue injury during immune hyperactivation. Furthermore, sulfur metabolites have been shown to modulate the gut microbiome composition, as H₂S, a product of cysteine catabolism by both host and microbial enzymes, acts as a signaling molecule that influences intestinal barrier integrity and mucosal immune homeostasis. Oh et al demonstrated that spleen-targeted H₂S donor delivery can achieve effective systemic immunomodulation and treatment of inflammatory bowel disease,45 supporting the therapeutic relevance of sulfur metabolite pathways in immune-mediated conditions. Disruption of sulfur metabolite homeostasis may thus contribute to gut barrier dysfunction, facilitating the translocation of microbial antigens that can trigger or exacerbate immune-mediated adverse events, particularly colitis and hepatitis, which are among the most prevalent irAEs observed in clinical practice.42

Collectively, these five metabolites span diverse biochemical categories—hydroxylated fatty acid derivatives, lysophospholipids, terpenoid diols, indazole heterocycles, and sulfur compounds—reflecting the multi-system nature of metabolic perturbations that precede irAE onset. The convergence of oxidative stress signaling, lipid mediator dysregulation, hepatic biotransformation changes, tryptophan–kynurenine pathway alterations, and sulfur amino acid metabolism underscores the complexity of irAE pathogenesis and supports the utility of a multi-metabolite panel, rather than a single biomarker, for risk stratification. The complementary predictive information offered by these five metabolites may enable the construction of a composite predictive model with enhanced sensitivity and specificity for irAE risk assessment in clinical settings.

Our KEGG pathway enrichment analysis additionally revealed significant associations between irAEs and the β-alanine metabolism pathway, the pentose phosphate pathway (PPP), and the pantothenate and CoA biosynthesis pathway. The PPP, particularly its non-oxidative branch, plays a pivotal role in maintaining α-ketoglutarate levels and modulating DNA methylation states in regulatory T cells (Treg cells).46 Deficiency in transketolase (TKT) enzyme leads to reduced α-KG levels in Treg cells, resulting in hypermethylation of DNA, which restricts the expression of genes associated with active regulatory T cells and their suppressive activity, potentially increasing the risk of irAEs. The CoA biosynthesis pathway may intersect with cellular redox states and immune cell activation, as CoA significantly impacts the function of Th17 cells and CD8+ T cells.47,48 These pathway-level findings provide a broader metabolic context that complements and is consistent with the individual biomarker alterations described above, collectively supporting the notion that irAEs arise from systemic metabolic vulnerabilities that pre-exist before ICI exposure.

In this study, by constructing ROC curves, we successfully validated the five differential metabolites described above with strong diagnostic performance, all demonstrating AUC values ranging between 0.7 and 0.9. These findings underscore the potential clinical utility of these metabolites in predicting irAEs. The detection of these metabolic indicators provides clinicians with a promising tool for early detection of irAEs, which is critical for implementing timely preventive measures. By monitoring these biomarkers, clinicians can preemptively adjust drug dosages, initiate immunomodulatory treatments, or closely observe high-risk patients, thereby mitigating the burden of adverse reactions and enhancing patient quality of life. Furthermore, the innovative aspect of this study lies in the application of non-targeted metabolomics, which enabled a comprehensive screening of potential biomarkers associated with irAEs during ICIs therapy. This approach elucidates the biological significance of these biomarkers and lays a solid foundation for future validation studies in larger patient cohorts.

Naturally, several limitations of this study must be acknowledged. First, the relatively small sample size and the single-center design warrant caution in interpreting our findings. Future studies in larger, multi-center cohorts are necessary to validate the predictive value of these five metabolites. Second, as our current analysis is primarily based on correlations, we plan to conduct in vitro cell models or animal experiments in the future to further elucidate the causal roles of these metabolites in the pathogenesis of irAEs. Lastly, given that this study was designed as a cross-sectional investigation, we lack longitudinal tracking of the dynamic processes of metabolic changes. Consequently, the dynamic monitoring of these biomarkers and their correlation with the severity of irAEs and therapeutic response will be a crucial focus for future research. Additionally, considering the variability and complexity of irAEs, future studies should incorporate multi-omics integrative analyses, combining genomics and transcriptomics data, to more comprehensively delineate the molecular landscape of irAEs.49–52 Such multidimensional data will facilitate a more profound comprehension of the molecular pathways driving irAEs and may reveal new therapeutic targets, thus enabling more personalized and precise treatment strategies. Further exploration of how these metabolites behave in different patient populations, and how they are influenced by genetic, environmental, and lifestyle factors, will augment our understanding of irAEs’ development and progression, providing more accurate clinical guidance.

Conclusions

In conclusion, this study employed untargeted metabolomics to characterize the serum metabolic profiles of patients undergoing immune checkpoint inhibitors (ICIs) therapy, aiming to identify early predictors of immune-related adverse events (irAEs). We successfully identified a distinct metabolic signature comprising five key metabolites: (S)-3,4-dihydroxybutyric acid, 1-(10Z-heptadecenoyl)-sn-glycero-3-phospho-(1′-rac-glycerol), (1R,2R,4R)-4-(2-hydroxypropan-2-yl)-1-methylcyclohexane-1,2-diol, 5-(2-methylbutan-2-yl)-4,5,6,7-tetrahydro-2H-indazole-3-carbohydrazide, and mintsulfide. These findings suggest that perturbations in oxidative stress, lipid remodeling, and gut microbiota-host co-metabolism may underlie the pathophysiology of irAEs. However, given the clinical heterogeneity of irAEs and the retrospective nature of this cohort, these metabolic alterations should be interpreted as potential risk indicators rather than definitive diagnostic markers for specific irAE subtypes. Future multicenter studies with larger sample sizes are warranted to validate the specificity of this metabolic panel across different irAE phenotypes, ultimately facilitating their translation into precision clinical management.

Funding Statement

This study was supported by grants from the Medicine and Health Science and Technology plan projects in Zhejiang Province (2024KY428) and the Hospital Pharmacy Special Research Fund of Zhejiang Pharmaceutical Association (2023ZYY05).

Data Sharing Statement

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Ethics Approval and Consent to Participate

The study protocol received approval by the Medical Ethics Committee of Changxing People’s Hospital (Ethics No.: 2023-KY-053) and carried out in alignment with the tenets of the Declaration of Helsinki. Informed consent forms were signed by all participants.

Consent for Publication

All participants provided their written informed consent for the publication of this research and any potentially identifiable data included within it. The consent procedure explained the purpose of the study, how the data would be used, and that the information would be accessible to the public upon publication. Participants were assured that their personal details would remain confidential and any identifying information would be removed or modified to protect their privacy. We confirm that we have obtained consent from all participants and that this consent is in accordance with the editorial policies of the journal.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare that they have no competing interests.