ABSTRACT
Background
Despite the remarkable success of immune checkpoint inhibitor (ICI) therapy in solid tumors, immune‐related adverse events (irAEs) have posed great challenges in the whole‐course management of ICI immunotherapy. Reliable biomarkers helping to predict irAEs are still limited and lacking.
Methods
Cancer patients receiving PD‐1/PD‐L1 blockade were enrolled without limiting tumor type, immunotherapy drugs, or affected organs. Blood samples were collected for PBMC isolation. Whole transcriptomic RNA sequencing was used to discover biomarkers in PBMCs of irAEs patients, validated by RT‐PCR in the test set. IrAEs‐free survival analysis and subgroup analysis were performed. LASSO regression prediction classifiers were conducted.
Results
One hundred and two patients were enrolled, with 42 in the training set and 60 in the testing set. The transcriptomic profile of PBMCs distinguished irAEs patients of different grades from non‐irAEs patients. Five candidate biomarkers were identified: FXYD7, SPSB2, SLC35E2A, C2, and SERPING1. Lower expression of FXYD7, SPSB2, and SLC35E2A was associated with shorter irAEs‐free survival [FXYD7, HR = 3.67, p = 0.0004; SLC35E2A, HR = 3.12, p = 0.0001; SPSB2, HR = 3.41, p = 0.0110]. Higher expression of C2 and SERPING1 was associated with shorter irAEs‐free survival in severe cases [C2, HR = 0.30, p = 0.0011; SERPING1, HR = 0.39, p = 0.0082]. Classifiers based on these biomarkers predicted irAEs of all grades (AUC = 0.91) and severe irAEs (AUC = 0.94), with FXYD7, C2, and SERPING1 as key variables.
Conclusions
Together, we revealed crucial circulating immunological determinants in irAEs and provided promising biomarkers to help predict irAEs, which will enlighten future precise management and targeted treatment strategies for irAEs.
Keywords: circulating transcriptome, FXYD7, immune checkpoint inhibitors, immune‐related adverse events
PBMC transcriptomic profiling identifies FXYD7, SPSB2, SLC35E2A, C2, and SERPING1 as predictive biomarkers for irAEs in cancer patients receiving PD‐1/PD‐L1 blockade, enabling noninvasive risk stratification through liquid biopsy‐based classifiers with high accuracy.
1. Introduction
Immune checkpoint inhibitors (ICIs) have demonstrated durable benefits in patients with multiple tumors and have changed the cancer treatment landscape [1, 2, 3]. With the increasing indications of ICIs drugs, problems derived from their complications, so‐called immune‐related adverse events (irAEs), are progressively receiving increasing attention [4, 5]. The burden of irAEs is high because their incidences are above 40% and can even reach 90% [6, 7], and severe irAEs, such as pneumonitis and myocarditis, impact the quality of life and survival outcomes of patients receiving ICIs therapy [8, 9]. In clinical practice, the initial onset of irAEs is undetectable; thus, accurate diagnosis of irAEs is challenging due to their highly variable and often aspecific clinical symptoms. According to the National Comprehensive Cancer Network (NCCN) guidelines, early recognition of symptoms and prompt intervention are key goals for the management of irAEs, which can prevent long‐term morbidity and mortality. Thus, there is an increasing need for reliable and validated biomarkers capable of predicting or signaling irAEs at an early stage.
Tumor tissues used in immune monitoring during the course of ICIs treatment are difficult to obtain for most clinical settings. Thus, blood samples appear to be much more feasible in the clinic and might even mirror the tumor immune environment (TME). ICIs block checkpoint pathways to restore T‐cell anti‐tumor immunity, which in turn kills tumor cells [10]. Simultaneously, global immune homeostasis of the host can also be disturbed to cause a series of irAEs affecting any organs in the body [11]. Blood components are theoretically affected by immune changes throughout the body and serve as their annotation system [12]. Recent studies have reported multiple blood biomarkers for irAEs, including peripheral immune cells such as T cells, eosinophils, and B cells [13, 14], gene variants and gene expression in blood cells [15, 16, 17, 18], protein biomarkers such as cytokines and C‐reactive proteins [19, 20], and plasma metabolites [21]. However, little is known regarding factors that accurately predict the onset of irAEs, and there are still no reliable blood biomarkers used to guide clinical practice.
In peripheral blood, peripheral blood mononuclear cells (PBMCs) are the most accessible and available immune cells. In theory, gene expression regulation in PBMCs is closely related to host systemic immune changes [11]. It has been reported that the expression of immune‐related genes in PBMCs predicts the risk of irAEs [22]. A recent study reported alterations in circulating T cells associated with colon inflammation induced by ICIs immunotherapy [23]. Another report shed light on the baseline activated CD4 memory T‐cell abundance in blood, which could serve as a biomarker of severe irAEs in melanoma [24]. Despite these efforts and advances, reported PBMCs biomarkers in previous studies need further independent validation and often have limited performance. Therefore, we hypothesized that distinct gene expression profiles in pretreatment PBMCs are associated with the subsequent development of irAEs in patients receiving PD‐1/PD‐L1 blockade, and that these transcriptional features can be leveraged to identify candidate predictive biomarkers.
2. Methods
2.1. Patients and Ethics
The flow chart of this study is shown in Figure 1. According to institutional guidelines, all blood samples were derived from residual clinical routine samples and were exempt from the requirement for written informed patient consent. Patients with solid tumors who were admitted to the Oncology Department of West China Hospital of Sichuan University between May 1, 2020 and May 31, 2021 were enrolled in this study; all patients underwent histopathological diagnosis and received initial ICIs immunotherapy, both first‐line and non‐first‐line. Patient follow‐up within 6 months in this study was conducted until irAEs occurrence, tumor progression, or treatment withdrawal. Patients who were lost to follow‐up and those with uncertain irAEs evaluations were excluded. Clinical characteristics, including sex, age, cancer type, stage, pathology, ICIs treatment regimen, treatment line, date of initial ICIs treatment and irAEs occurrence, were recorded.
FIGURE 1.
Flow chart of this study.
This study was approved and monitored by the Biomedical Ethics Committee of West China Hospital, Sichuan University (No. 2019 (1045)), and strictly adhered to the Declaration of Helsinki. As the research solely utilized anonymized data/samples without impacting participant rights, welfare, or privacy, the Ethics Committee of West China Hospital granted a waiver of informed consent in accordance with national ethical guidelines and the Declaration of Helsinki. IrAEs were assessed according to the criteria of the Common Terminology Criteria for Adverse Events 4.0 (CTCAE 4.0). Patients who did not develop irAEs during the follow‐up period were considered the non‐irAEs group. According to the severity of irAEs, patients with G1‐G2 grades of irAEs were regarded as the mild to moderate irAEs group, while those with G3‐G5 grades of irAEs were regarded as the severe irAEs group. All patients who developed G1‐G5 grades of irAEs were considered the irAEs group.
2.2. Sample Collection and Processing
Ethylenediaminetetraacetic acid (EDTA)‐anticoagulated blood samples were collected from enrolled patients after their second cycle of ICIs treatment. Ficoll density gradient centrifugation was used to isolate PBMCs from whole blood according to the manufacturer's instructions (LTS1077, TBD, China). The isolated PBMCs were resuspended in TRIzol Reagent (Invitrogen, USA) and cryopreserved at −80°C. Then, total RNA was extracted by the TRIzol solubilization and extraction method for RNA sequencing and real‐time quantitative PCR (RT‐PCR).
2.3. RNA Sequencing (RNA‐Seq) and Raw Data Processing
RNA library construction was performed using the TruSeq Stranded mRNA LT Sample Prep Kit (Illumina, San Diego, CA, USA). Then, the libraries were sequenced on an Illumina NovaSeq 6000 platform. Then, 150 bp paired‐end raw reads were generated to form fastq format files. Trimmomatic [25] was used for data cleaning and filtering. The clean reads were mapped to the GRCh38 genome. FPKM was calculated to indicate the expression level of each gene.
2.4. RNA‐Seq Data Analysis
In this study, we used three methods, Deseq2, EdgeR, and Limma_voom, to conduct differential analysis among the non‐irAEs group, irAEs group and severe irAEs group. To avoid missing data, a p value < 0.05 and |fold change (FC)| > 1.3 were set as the screening thresholds. Finally, we used the intersection of the three methods to obtain differentially expressed genes (DEGs). Based on the DEGs, Gene Ontology and KEGG pathway enrichment analyses were performed.
2.5. RT‐PCR
The candidate biomarkers were validated via RT‐PCR. The PrimeScript RT reagent Kit with gDNA Eraser (Takara, Japan) was used to remove genomic DNA and conduct reverse transcription reactions. RT‐PCR was performed using a SYBR PrimeScript miRNA RT‐PCR Kit (Takara, Japan) according to the manufacturer's instructions. ACTB was selected as the internal reference gene. The relative expression levels were calculated using the 2−ΔΔCt method. The sequences of the primers used in this study are listed in Supplemental Table S1.
2.6. IrAEs Based Survival Analysis
We conducted irAEs‐based survival analysis in all patients and different subgroups. The occurrence of irAEs was regarded as an event. The survival analyses were conducted using the survival and survminer R packages (version 4.0.2). The log‐rank test was utilized to calculate the p‐value and the hazard ratio (HR) to evaluate the risk of irAEs associated with different variables.
2.7. Least Absolute Shrinkage and Selection Operator (LASSO) Regression Classifier
LASSO regression was used to select variables and establish classifiers. Then, the random three‐fold cross‐validation method was used to avoid bias. Receiver operator characteristic (ROC) curves and confusion matrices were used to assess the performance of the classifiers. Finally, bar plots of variable importance ranking were exported.
2.8. Statistical Analysis
Continuous variables are described as the mean ± standard deviation, and an independent sample Student's t‐test was used for statistical analysis to calculate p values. For discrete variables, frequency and percentage can be used for description, and the chi‐square test was used to statistically compare the differences between groups and calculate p values. A p value ≤ 0.05 was considered statistically significant. Prior to modeling, the overall dataset was divided into training and internal test sets using a random stratified split approach implemented via the createDataPartition function from the R caret package. This method preserves the distribution of the outcome variable in both subsets. The split ratio was set at 60/40. In addition, R (version 4.0.2) and GraphPad Prism (version 6.01) were used to create volcano plots, heatmaps, Venn diagrams, bubble diagrams, and bar graphs.
3. Results
3.1. Patient Characteristics
In total, we enrolled 102 eligible patients who received initial ICIs immunotherapy (Table 1), randomly divided into a training set (N = 42) and a test set (N = 60). In the training set, 13 patients did not suffer from irAEs, 21 patients developed mild to moderate irAEs, and 8 developed severe irAEs. In the test set, 10 patients did not suffer from irAEs, 42 patients developed mild to moderate irAEs, and 8 developed severe irAEs. There were no significant differences among the three groups in either the training or test set in terms of age, gender, cancer type, clinical stage, ICIs drugs, or treatment line (p > 0.05). In the histopathological type, patients in the three irAEs groups in the training set exhibited a significant difference (p = 0.003). A much higher proportion of patients with adenocarcinoma suffered severe irAEs.
TABLE 1.
Patient characteristics at baseline of ICIs therapy.
| Train (N = 42) | Test (N = 60) | |||||||
|---|---|---|---|---|---|---|---|---|
| No irAEs | Mild to moderate | Severe | p | No irAEs | Mild to moderate | Severe | p | |
| (N = 13) | (N = 21) | (N = 8) | (N = 10) | (N = 42) | (N = 8) | |||
| Age | ||||||||
| Mean (SD) | 62.2 (9.52) | 61.3 (11.4) | 67.6 (10.1) | 0.358 | 65.7 (7.26) | 60.7 (10.1) | 60.3 (7.15) | 0.301 |
| Gender | ||||||||
| Male | 11 (84.6%) | 17 (81.0%) | 6 (75.0%) | 0.862 | 8 (80.0%) | 34 (81.0%) | 7 (87.5%) | 0.898 |
| Female | 2 (15.4%) | 4 (19.0%) | 2 (25.0%) | 2 (20.0%) | 8 (19.0%) | 1 (12.5%) | ||
| Cancer type | ||||||||
| NSCLC | 7 (53.8%) | 11 (52.4%) | 6 (75.0%) | 0.208 | 3 (30.0%) | 22 (52.4%) | 5 (62.5%) | 0.915 |
| SCLC | 1 (7.7%) | 5 (23.8%) | 0 (0%) | 3 (30.0%) | 7 (16.7%) | 1 (12.5%) | ||
| EC | 1 (7.7%) | 4 (19.0%) | 2 (25.0%) | 2 (20.0%) | 5 (11.9%) | 1 (12.5%) | ||
| NPC | 2 (15.4%) | 1 (4.8%) | 0 (0%) | 1 (10.0%) | 4 (9.5%) | 0 (0%) | ||
| Other | 2 (15.4%) | 0 (0%) | 0 (0%) | 1 (10.0%) | 4 (9.5%) | 1 (12.5%) | ||
| Clinical stage | ||||||||
| II | 1 (7.7%) | 0 (0%) | 0 (0%) | 0.302 | 1 (10.0%) | 1 (2.4%) | 0 (0%) | 0.323 |
| III | 5 (38.5%) | 4 (19.0%) | 1 (12.5%) | 1 (10.0%) | 3 (7.1%) | 2 (25.0%) | ||
| IV | 6 (46.2%) | 16 (76.2%) | 5 (62.5%) | 8 (80.0%) | 38 (90.5%) | 5 (62.5%) | ||
| Missing | 1 (7.7%) | 1 (4.8%) | 2 (25.0%) | 0 (0%) | 0 (0%) | 1 (12.5%) | ||
| Pathology | ||||||||
| Adenocarcinoma | 0 (0%) | 4 (19.0%) | 6 (75.0%) | 0.003 | 1 (10.0%) | 14 (33.3%) | 3 (37.5%) | 0.480 |
| Squamous carcinoma | 11 (84.6%) | 12 (57.1%) | 2 (25.0%) | 6 (60.0%) | 19 (45.2%) | 3 (37.5%) | ||
| Small cell carcinoma | 1 (7.7%) | 5 (23.8%) | 0 (0%) | 3 (30.0%) | 7 (16.7%) | 1 (12.5%) | ||
| Other | 1 (7.7%) | 0 (0%) | 0 (0%) | 0 (0%) | 1 (2.4%) | 1 (12.5%) | ||
| Missing | 0 (0%) | 0 (0%) | 0 (0%) | 0 (0%) | 1 (2.4%) | 0 (0%) | ||
| Treatment line | ||||||||
| 1 | 5 (38.5%) | 8 (38.1%) | 5 (62.5%) | 0.571 | 8 (80.0%) | 28 (66.7%) | 4 (50.0%) | 0.741 |
| 2 | 6 (46.2%) | 10 (47.6%) | 1 (12.5%) | 2 (20.0%) | 11 (26.2%) | 4 (50.0%) | ||
| 3 | 2 (15.4%) | 2 (9.5%) | 2 (25.0%) | 0 (0%) | 2 (4.8%) | 0 (0%) | ||
| 4 | 0 (0%) | 1 (4.8%) | 0 (0%) | 0 (0%) | 1 (2.4%) | 0 (0%) | ||
| Treatment options | ||||||||
| ICIs monotherapy | 7 (53.8%) | 6 (28.6%) | 2 (25.0%) | 0.256 | 2 (20.0%) | 13 (31.0%) | 1 (12.5%) | 0.486 |
| ICIs plus chemotherapy | 6 (46.2%) | 15 (71.4%) | 6 (75.0%) | 8 (80.0%) | 29 (69.0%) | 7 (87.5%) | ||
| Immunotherapy drugs | ||||||||
| Carrelizumab | 2 (15.4%) | 5 (23.8%) | 4 (50.0%) | 0.288 | 1 (10.0%) | 14 (33.3%) | 0 (0%) | 0.031 |
| Pembrolizumab | 2 (15.4%) | 3 (14.3%) | 2 (25.0%) | 1 (10.0%) | 4 (9.5%) | 2 (25.0%) | ||
| Tislelizumab | 2 (15.4%) | 5 (23.8%) | 0 (0%) | 3 (30.0%) | 3 (7.1%) | 1 (12.5%) | ||
| Atezolizumab | 1 (7.7%) | 5 (23.8%) | 0 (0%) | 2 (20.0%) | 7 (16.7%) | 1 (12.5%) | ||
| Sintilimab | 2 (15.4%) | 1 (4.8%) | 2 (25.0%) | 0 (0%) | 11 (26.2%) | 3 (37.5%) | ||
| Durvalumab | 3 (23.1%) | 1 (4.8%) | 0 (0%) | 1 (10.0%) | 3 (7.1%) | 1 (12.5%) | ||
| Nivolumab | 1 (7.7%) | 1 (4.8%) | 0 (0%) | 2 (20.0%) | 0 (0%) | 0 (0%) | ||
| BMI | ||||||||
| Mean (SD) | 23.3 (3.59) | 22.7 (2.80) | 22.3 (1.76) | 0.770 | 23.7 (2.26) | 22.6 (2.88) | 23.2 (3.27) | 0.505 |
| Smoke | ||||||||
| No | 3 (23.1%) | 6 (28.6%) | 3 (37.5%) | 0.700 | 4 (40.0%) | 26 (61.9%) | 6 (75.0%) | 0.289 |
| Yes | 9 (69.2%) | 15 (71.4%) | 4 (50.0%) | 6 (60.0%) | 16 (38.1%) | 2 (25.0%) | ||
| Missing | 1 (7.7%) | 0 (0%) | 1 (12.5%) | |||||
| Drink | ||||||||
| No | 8 (61.5%) | 13 (61.9%) | 5 (62.5%) | 0.891 | 6 (60.0%) | 32 (76.2%) | 7 (87.5%) | 0.387 |
| Yes | 4 (30.8%) | 8 (38.1%) | 2 (25.0%) | 4 (40.0%) | 10 (23.8%) | 1 (12.5%) | ||
| Missing | 1 (7.7%) | 0 (0%) | 1 (12.5%) | |||||
Abbreviations: EC: esophageal cancer; NPC: nasopharyngeal carcinoma; NSCLC: non‐small cell lung cancer; SCLC: small cell lung cancer.
Then, we depicted the irAEs distribution in subgroups of ICI drugs and cancer types (Figure 2). In patients receiving PD‐1 antibody treatment, 16 did not suffer from irAEs, 47 developed mild to moderate irAEs, and 14 developed severe irAEs (Figure 2a). In patients receiving PD‐L1 antibody treatment, 7 did not suffer from irAEs, 16 developed mild to moderate irAEs, and 2 developed severe irAEs (Figure 2a). In patients with lung cancer, 14 did not suffer from irAEs, 45 developed mild to moderate irAEs, and 12 developed severe irAEs (Figure 2b). In all patients suffering from irAEs, 26 developed multiple organ toxicity (Figure 2c). Among patients suffering from single‐organ irAEs, there were 21 with skin toxicity, 11 with thyroid toxicity, 6 with pulmonary toxicity, 4 with cardiotoxicity, 3 with colitis, and 8 with other organ toxicity (Figure 2c). The median time to irAEs was 96 days in the patients (Figure 2d).
FIGURE 2.
Clinical cohort profile of this study. (a) Distribution of irAEs of different grades in the different subgroups of ICI drugs; (b) Distribution of irAEs of different grades in the different subgroups of cancer type; (c) Composition of irAEs in different organs in the clinical cohort of this study; (d) The curve of time to irAEs in the patients of this study.
3.2. Leveraging Transcriptomic Profiling of PBMCs to Distinguish Patients With irAEs of Different Grades From Those Without irAEs
To explore differential gene expression in PBMCs of patients suffering from irAEs or not, we first identified 78 upregulated genes and 114 downregulated genes in patients with irAEs compared to those without irAEs according to the intersection of three types of algorithms (Figure 3a, Figure S1). In addition, to depict the unique molecular characteristics of patients with severe irAEs, we identified 141 upregulated genes and 208 downregulated genes in patients suffering from severe irAEs compared to those without irAEs (Figure 3b, Figure S1b). The clustering heatmaps showed that the intersecting DEGs performed well in distinguishing patients in different groups (Figure 3a,b). Then, we verified the genes with relatively high expression (FPKM > 3) using RT‐PCR, and found significant correlations between the results of RT‐PCR and RNA‐seq of DEGs (Figure S1c). We found that the TAGLN, PTX3, and C15ORF48 genes showed higher expression in patients suffering from irAEs, while the FXYD7, SOCS2, CPA3, SPSB2, and GCSAML genes showed lower expression (Figure S2a). Compared to patients without irAEs, C2, SERPING1, KIR3DL2, KIR2DL3, PTX3, TAGLN, GBP1, and APOBEC3B genes were significantly higher in patients with severe irAEs, while GATA2, HDC, PTGDR2, SLC35E2A, and SOCS2 were significantly lower (Figure S2b). GO pathway enrichment analysis based on the above DEGs revealed that many of them were enriched in immune response‐regulating signaling pathways (p < 0.001) (Figure 3c).
FIGURE 3.
Screening transcriptomic DEGs of PBMCs to distinguish patients with irAEs of different grades. (a) DEGs of patients with irAEs compared to those without irAEs according to the intersection of three types of algorithms, including Deseq2, EdgeR, and Limma_voom; (b) DEGs of patients with severe irAEs compared to those without irAEs according to the intersection of three types of algorithms; (c) Gene Ontology enrichment analysis of candidate differential genes.
3.3. Identifying Reliable Biomarkers by Internal and External Validations of the Candidate DEGs
To further demonstrate whether the candidate differential genes could serve as reliable predictive biomarkers for irAEs, we performed validation experiments in a patient cohort using RT‐PCR. As shown in Figure 4a, the FXYD7, SPSB2, and SLC35E2A genes were significantly lower in patients with irAEs of all grades in our validation cohort (Figure 4a), with the same trend observed in the training cohort. In addition, C2 and SERPING1 were significantly higher in patients suffering from severe irAEs in the validation cohort (Figure 4b), which is consistent with the trends in the training set. Among them, FXYD7 was downregulated in patients with irAEs of different grades (p < 0.05) and much lower in patients with severe irAEs (p < 0.05). In addition, the time to irAEs onset was shorter in patients with higher expression of C2 and SERPING1 [C2, HR (95% CI): 0.30 (0.09–1.01), log‐rank p = 0.0011; SERPING1, HR (95% CI): 0.39 (0.20–0.76), log‐rank p = 0.0082] (Figure 4c,e), whereas the time to irAEs onset was significantly longer in patients with higher expression of FXYD7, SPSB2, and SLC35E2A [FXYD7, HR (95% CI): 3.67 (2.04–6.59), log‐rank p = 0.00041; SLC35E2A, HR (95% CI): 3.12 (1.66–5.89), log‐rank p = 0.0001; SPSB2, HR (95% CI): 3.41 (1.75–6.69), log‐rank p = 0.011] (Figure 4d,f,g), which indicates that the five genes are able to serve as reliable biomarkers for irAEs.
FIGURE 4.
Identifying reliable biomarkers by validating the candidate DEGs in an independent patient cohort. (a) Differential expression boxplot of potential reliable biomarkers between patients with or without irAEs; (b) Differential expression boxplot of potential reliable biomarkers between patients with severe irAEs and without irAEs; (c), (d), (e), (f), (g), curves of time to irAEs onset based on different expression of C2, FXYD7, SERPING1, SLC35E2A, and SPSB2 genes.
3.4. Investigating the Clinical Value of Candidate Biomarkers Through Further Subgroup Analysis
Next, we further explored the potential clinical value of the five candidate genes in different subgroups of ICIs drugs and cancer types. Because of the sufficient sample size, we conducted subgroup analyses in lung cancer patients and anti‐PD‐1 treatment patients. The results are shown in Figures 5 and 6. In lung cancer patients, FXYD7, SPSB2, and SLC35E2A genes were significantly lower in patients with irAEs (although not significant for SPSB2), and patients with higher expression of FXYD7, SPSB2, and SLC35E2A had significantly increased irAEs‐free survival [FXYD7, HR (95% CI): 5.83 (2.90–11.74), log‐rank p = 0.00053; SLC35E2A, HR (95% CI): 2.92 (1.41–6.08), log‐rank p = 0.00068; SPSB2, HR (95% CI): 1.96 (0.98–3.90), log‐rank p = 0.041] (Figure 5a,d,f,g). In lung cancer patients with severe irAEs, the expression of C2 was higher (although not significant), and patients with higher C2 expression had shorter irAEs‐free survival [HR (95% CI): 0.35 (0.09–1.29), log‐rank p = 0.01] (Figure 5b,c,e).
FIGURE 5.
Validation of the candidate genes in all patients with lung cancer. (a) Differential expression boxplot of potential reliable biomarkers between lung cancer patients with or without irAEs; (b) Differential expression boxplot of potential reliable biomarkers between lung cancer patients with severe irAEs and without irAEs; (c), (d), (e), (f), (g), curves of time to irAEs onset based on different expression of C2, FXYD7, SERPING1, SLC35E2A, and SPSB2 genes in all patients with lung cancer.
FIGURE 6.
Validation of the candidate genes in all patients receiving anti‐PD‐1 treatment. (a) Differential expression boxplot of candidate genes between patients with or without irAEs in the anti‐PD‐1 treatment subgroup; (b) Differential expression boxplot of potential reliable biomarkers between lung cancer patients with severe irAEs and without irAEs in the anti‐PD‐1 treatment subgroup; (c), (d), (e), (f), (g), curves of time to irAEs onset based on different expression of the C2, FXYD7, SERPING1, SLC35E2A, and SPSB2 genes in all patients receiving anti‐PD‐1 treatment.
In patients receiving anti‐PD‐1 blockade, the FXYD7, SPSB2, and SLC35E2A genes were significantly lower in patients with irAEs, and patients with lower levels of FXYD7, SPSB2, and SLC35E2A showed shorter time to irAEs onset [FXYD7, HR (95% CI): 6.50 (2.67–15.81), log‐rank p = 0.03; SLC35E2A, HR (95% CI): 3.16 (1.67–5.99), log‐rank p = 0.0015; SPSB2, HR (95% CI): 3.31 (1.63–6.72), log‐rank p = 0.014] (Figure 6a,d,f,g). In anti‐PD‐1 treatment patients suffering from severe irAEs, the C2 and SERPING1 genes were much higher, and patients with higher C2 and SERPING1 expression had shorter irAEs‐free survival [C2, HR (95% CI): 0.30 (0.07–1.22), log‐rank p = 0.003; SERPING1, HR (95% CI): 0.41 (0.19–0.86), log‐rank p = 0.012] (Figure 6b,c,e).
3.5. Constructing Novel Classifiers to Predict the Onset of irAEs Based on Candidate Biomarkers and Routine Clinical Indicators
Finally, we combined the above candidate genes, including FXYD7, C2, SERPING1, SPSB2, and SLC35E2A, with routine clinical indicators related to irAEs, such as neutrophil percentage (NEUT), hemoglobin (HGB), C‐reactive protein (CRP), adrenocorticotropic hormone (ACTH), alanine aminotransferase (ALT), high‐density lipoprotein cholesterol (HDLC), and calcium (Ca), to construct novel classifiers using LASSO regression to predict the occurrence of all irAEs and severe irAEs. Early identification of irAEs of all grades: The ROC area under the curve (AUC) was 0.91 (95% CI: 0.80–1.00) (Figure 7a). The confusion matrix for predicting irAEs of all grades showed that the accuracy was 0.88, the sensitivity was 0.89, and the specificity was 0.86 (Figure 7a). Among all variables, FXYD7 was the most important in predicting irAEs (Figure 7a). In addition, the AUC value for the early identification of severe irAEs was 0.94 (95% CI: 0.84–1.00) (Figure 7b). The confusion matrix for predicting severe irAEs showed that the accuracy was 0.85, the sensitivity was 1.00, and the specificity was 0.84 (Figure 7b). Among all variables, C2 and SERPING1 were much more important in predicting severe irAEs (Figure 7b).
FIGURE 7.
LASSO regression models predict the occurrence of irAEs in patients receiving anti‐PD‐1 treatment. (a) ROC curve (left), confusion matrix (middle), and ranking of variable importance (right) in predicting irAEs of all grades. (b) ROC curve (left), confusion matrix (middle), and ranking of variable importance (right) in predicting severe irAEs.
4. Discussion
At present, irAEs have posed great challenges in the whole course management of cancer patients receiving ICIs immunotherapy. However, biomarkers helping to predict the risks of irAEs are still limited and lacking. In this study, we utilized transcriptome analysis of PBMCs to search for the specific circulating immune signatures that preceded the progression of irAEs during the course of ICIs treatment. To our knowledge, this study is the first to identify reliable transcriptomic biomarkers of PBMCs and validate them in a test set. Importantly, we addressed five promising blood biomarkers for irAEs, including FXYD7, C2, SERPING1, SLC35E2A, and SPSB2, and the classifiers based on these genes performed well, although their clinical value still needs to be validated in large‐scale, multicenter clinical studies.
The FXYD7 gene is located on chromosome 1 and is expressed in various types of immune cells [26]. FXYD7 downregulates intracellular oxidative phosphorylation and tricarboxylic acid cycle metabolic pathways [27]. Therefore, PBMCs with low expression of FXYD7 may have enhanced oxidative respiration and energy metabolism, which may reflect an activated immune state. This is consistent with our results that patients with lower FXYD7 expression showed a higher risk of irAEs as well as severe fatal irAEs, which suggests that the energy metabolism of immune cells plays an important role in the response of patients to ICI immunotherapy. Immune cells in cancer patients receiving PD‐1/PD‐L1 blockade fight energy‐expensive battles against not only cancer cells [28] but also host normal cells, such as myocardial cells and thyroid cells, during the emergence of irAEs. Therefore, efficient and fast metabolism regulation is crucial to fuel immune cells [29] and varies among different organs. In PBMCs, we found common changes in FXYD7 genes in patients suffering from irAEs. In this study, we propose the important concept that energy metabolism regulation is crucial and more abundant in irAEs than in a single antitumor immune response. Nevertheless, the specific function and mechanisms of FXYD7 in immune cells still need to be confirmed by further basic and clinical studies.
In the classifier predicting irAEs of all grades, FXYD7 was the most crucial, whereas in the classifier predicting severe irAEs, C2 and SERPING1 were much more important. The C2 gene is localized in the MHC‐III gene of the chromosome 6p region and encodes complement 2, a glycoprotein that is part of the classical pathway of the complement system. C2 is associated with the development of autoimmune diseases, possibly in the form of upregulation in the early stage and downregulation in the late stage [30], and corresponding complementary therapy promotes the remission of autoimmune disease [31]. In our study, early elevation of C2 was positively correlated with the risk of irAEs, which may be related to autoimmune disease‐like changes in irAEs after PD‐1/PD‐L1 blockade. In addition, SERPING1 also participates in the regulation of the complement system. SERPING1 is an important molecule in innate immunity and can inactivate C1r and C1s [32, 33]. However, the role of SERPING1 as a favorable biomarker of irAEs is uncertain in this study. We suggest that the regulation of the complement system plays an important role in irAEs.
This study has several limitations; first, its single‐center, retrospective, real‐world design inherently carries the risk of uncontrolled confounding factors, despite our use of cohort balancing methods and subgroup analyses (including in lung cancer patients and those receiving PD‐1 inhibitors) to address this concern. The high clinical heterogeneity of the cohort (encompassing various cancer types, stages, and treatments) further limits the granularity and generalizability of the findings. Second, while focusing on early treatment time points provides the practical advantage of identifying convenient candidate biomarkers for early clinical application and reduces interference from later confounding events, it is important to clarify that this single early timepoint design, while clinically useful for risk stratification, differs from a baseline prediction model. Furthermore, the limited sample size prevented more detailed analyses (e.g., by irAE‐involved organ). Therefore, our conclusions should be interpreted as hypothesis‐generating within this complex, real‐world context. Future studies with larger, prospective, multicenter cohorts are warranted to validate our findings and to explore the biomarker dynamics in organ‐specific irAEs.
In conclusion, within the constraints of a retrospective, heterogeneous real‐world cohort, our analysis identifies C2, SERPING1, FXYD7, SPSB2, and SLC35E2A as candidate early circulating biomarkers for irAEs. The classifiers developed based on these genes showed promising performance in our study population. These findings generate a hypothesis that must be validated in prospective, multicenter studies. Our data and biomarkers may serve as a resource to inform future research into the precise management of irAEs.
Author Contributions
Liting You: investigation, resources, formal analysis, writing – original draft. Jianzhao Zhai: investigation, visualization, writing – original draft. Zhaodan Xin: investigation, validation. Feifei Na: investigation, validation. Yang Wen: investigation. Jin Li: investigation. Jiajia Song: investigation. Ling Bai: investigation. Xiaohan Zhou: investigation. Binwu Ying: conceptualization, writing – review and editing. Juan Zhou: conceptualization, resources, writing – review and editing. All authors have read and approved the final manuscript.
Funding
This work was supported by the National Natural Science Foundation of China (No. 82372331), the Project of Postdoctoral Foundation of Sichuan University (grant No. 20826041G4067), and the Young Scientists Fund of the National Natural Science Foundation of China (No. 82302623).
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Figure S1: Identification of differentially expressed genes (DEGs) in patients with irAEs. (a) Venn diagram showing the intersection of DEGs identified by three algorithms in patients with irAEs vs. those without irAEs. (b) Venn diagram showing the intersection of DEGs in patients with severe irAEs versus those without irAEs. (c) Correlation analysis between the results of RT‐PCR and RNA‐seq for the selected DEGs.
Figure S2: Validation of differentially expressed genes by RT‐PCR. (a) Relative expression levels of a panel of genes in patients with irAEs compared to those without irAEs. (b) Relative expression levels of a panel of genes in patients with severe irAEs compared to those without irAEs.
Table S1: Primer sequences used for RT‐PCR validation of candidate biomarkers.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
References
- 1. Ribas A. and Wolchok J. D., “Cancer Immunotherapy Using Checkpoint Blockade,” Science 359, no. 6382 (2018): 1350–1355. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Robert C., “A Decade of Immune‐Checkpoint Inhibitors in Cancer Therapy,” Nature Communications 11, no. 1 (2020): 3801. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Li A.‐Q. and Fang J.‐H., “Anti‐Angiogenic Therapy Enhances Cancer Immunotherapy: Mechanism and Clinical Application,” Interdisciplinary Medicine 2, no. 1 (2024): e20230025. [Google Scholar]
- 4. Conroy M. and Naidoo J., “Immune‐Related Adverse Events and the Balancing Act of Immunotherapy,” Nature Communications 13, no. 1 (2022): 392. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Johnson D. B., Nebhan C. A., Moslehi J. J., and Balko J. M., “Immune‐Checkpoint Inhibitors: Long‐Term Implications of Toxicity,” Nature Reviews Clinical Oncology 19, no. 4 (2022): 254–267. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Ramos‐Casals M., Brahmer J. R., Callahan M. K., et al., “Immune‐Related Adverse Events of Checkpoint Inhibitors,” Nature Reviews Disease Primers 6, no. 1 (2020): 38. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Zhang L., Reynolds K. L., Lyon A. R., Palaskas N., and Neilan T. G., “The Evolving Immunotherapy Landscape and the Epidemiology, Diagnosis, and Management of Cardiotoxicity: JACC: CardioOncology Primer,” JACC CardioOncol 3, no. 1 (2021): 35–47. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Wang D. Y., Salem J. E., Cohen J. V., et al., “Fatal Toxic Effects Associated With Immune Checkpoint Inhibitors: A Systematic Review and Meta‐Analysis,” JAMA Oncology 4, no. 12 (2018): 1721–1728. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Salem J. E., Manouchehri A., Moey M., et al., “Cardiovascular Toxicities Associated With Immune Checkpoint Inhibitors: An Observational, Retrospective, Pharmacovigilance Study,” Lancet Oncology 19, no. 12 (2018): 1579–1589. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Chen D. S. and Mellman I., “Oncology Meets Immunology: The Cancer‐Immunity Cycle,” Immunity 39, no. 1 (2013): 1–10. [DOI] [PubMed] [Google Scholar]
- 11. Verheijden R. J., van Eijs M. J. M., May A. M., van Wijk F., and Suijkerbuijk K. P. M., “Immunosuppression for Immune‐Related Adverse Events During Checkpoint Inhibition: An Intricate Balance,” npj Precision Oncology 7, no. 1 (2023): 41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Griffiths J. I., Wallet P., Pflieger L. T., et al., “Circulating Immune Cell Phenotype Dynamics Reflect the Strength of Tumor‐Immune Cell Interactions in Patients During Immunotherapy,” Proceedings of the National Academy of Sciences of United States of America 117, no. 27 (2020): 16072–16082. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Giommoni E., Giorgione R., Paderi A., et al., “Eosinophil Count as Predictive Biomarker of Immune‐Related Adverse Events (irAEs) in Immune Checkpoint Inhibitors (ICIs) Therapies in Oncological Patients,” Immuno 1, no. 3 (2021): 253–263. [Google Scholar]
- 14. Das R., Bar N., Ferreira M., et al., “Early B Cell Changes Predict Autoimmunity Following Combination Immune Checkpoint Blockade,” Journal of Clinical Investigation 128, no. 2 (2018): 715–720. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Refae S., Gal J., Ebran N., et al., “Germinal Immunogenetics Predict Treatment Outcome for PD‐1/PD‐L1 Checkpoint Inhibitors,” Investigational New Drugs 38, no. 1 (2020): 160–171. [DOI] [PubMed] [Google Scholar]
- 16. Shahabi V., Berman D., Chasalow S. D., et al., “Gene Expression Profiling of Whole Blood in Ipilimumab‐Treated Patients for Identification of Potential Biomarkers of Immune‐Related Gastrointestinal Adverse Events,” Journal of Translational Medicine 11 (2013): 75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Friedlander P., Wood K., Wassmann K., Christenfeld A. M., Bhardwaj N., and Oh W. K., “A Whole‐Blood RNA Transcript‐Based Gene Signature Is Associated With the Development of CTLA‐4 Blockade‐Related Diarrhea in Patients With Advanced Melanoma Treated With the Checkpoint Inhibitor Tremelimumab,” Journal for Immunotherapy of Cancer 6, no. 1 (2018): 90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Xin Z., You L., Na F., et al., “Immunogenetic Variations Predict Immune‐Related Adverse Events for PD‐1/PD‐L1 Inhibitors,” European Journal of Cancer 184 (2023): 124–136. [DOI] [PubMed] [Google Scholar]
- 19. Jia X. H., Geng L. Y., Jiang P. P., et al., “The Biomarkers Related to Immune Related Adverse Events Caused by Immune Checkpoint Inhibitors,” Journal of Experimental & Clinical Cancer Research 39, no. 1 (2020): 284. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Lim S. Y., Lee J. H., Gide T. N., et al., “Circulating Cytokines Predict Immune‐Related Toxicity in Melanoma Patients Receiving Anti‐PD‐1‐Based Immunotherapy,” Clinical Cancer Research 25, no. 5 (2019): 1557–1563. [DOI] [PubMed] [Google Scholar]
- 21. Dubin K., Callahan M. K., Ren B., et al., “Intestinal Microbiome Analyses Identify Melanoma Patients at Risk for Checkpoint‐Blockade‐Induced Colitis,” Nature Communications 7 (2016): 10391. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Giunta E. F., Barra G., De Falco V., et al., “Baseline IFN‐γ and IL‐10 Expression in PBMCs Could Predict Response to PD‐1 Checkpoint Inhibitors in Advanced Melanoma Patients,” Scientific Reports 10, no. 1 (2020): 17626. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Luoma A. M., Suo S., Williams H. L., et al., “Molecular Pathways of Colon Inflammation Induced by Cancer Immunotherapy,” Cell 182, no. 3 (2020): 655–671. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Lozano A. X., Chaudhuri A. A., Nene A., et al., “T Cell Characteristics Associated With Toxicity to Immune Checkpoint Blockade in Patients With Melanoma,” Nature Medicine 28, no. 2 (2022): 353–362. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Bolger A. M., Lohse M., and Usadel B., “Trimmomatic: A Flexible Trimmer for Illumina Sequence Data,” Bioinformatics 30, no. 15 (2014): 2114–2120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Brownlie D., Scharenberg M., Mold J. E., et al., “Expansions of Adaptive‐Like NK Cells With a Tissue‐Resident Phenotype in Human Lung and Blood,” Proceedings of the National Academy of Sciences of United States of America 118, no. 11 (2021): e2016580118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Jin M., Zhang H., Yang J., Zheng Z., and Liu K., “Expression Mode and Prognostic Value of FXYD Family Members in Colon Cancer,” Aging (Albany NY) 13, no. 14 (2021): 18404–18422. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Schuster S., Ewald J., and Kaleta C., “Modeling the Energy Metabolism in Immune Cells,” Current Opinion in Biotechnology 68 (2021): 282–291. [DOI] [PubMed] [Google Scholar]
- 29. Pearce E. L., “Metabolism as a Driver of Immunity,” Nature Reviews. Immunology 21, no. 10 (2021): 618–619. [DOI] [PubMed] [Google Scholar]
- 30. Kaye S., Lokki A. I., Hanttu A., et al., “Upregulation of Early and Downregulation of Terminal Pathway Complement Genes in Subcutaneous Adipose Tissue and Adipocytes in Acquired Obesity,” Frontiers in Immunology 8 (2017): 545. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Thurman J. M. and Yapa R., “Complement Therapeutics in Autoimmune Disease,” Frontiers in Immunology 10 (2019): 672. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Sanfilippo C., Cambria D., Longo A., et al., “SERPING1 mRNA Overexpression in Monocytes From HIV+ Patients,” Inflammation Research 66, no. 12 (2017): 1107–1116. [DOI] [PubMed] [Google Scholar]
- 33. Chen S., Li L. Y., Wu Z. M., et al., “SerpinG1: A Novel Biomarker Associated With Poor Coronary Collateral in Patients With Stable Coronary Disease and Chronic Total Occlusion,” Journal of the American Heart Association 11, no. 24 (2022): e027614. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1: Identification of differentially expressed genes (DEGs) in patients with irAEs. (a) Venn diagram showing the intersection of DEGs identified by three algorithms in patients with irAEs vs. those without irAEs. (b) Venn diagram showing the intersection of DEGs in patients with severe irAEs versus those without irAEs. (c) Correlation analysis between the results of RT‐PCR and RNA‐seq for the selected DEGs.
Figure S2: Validation of differentially expressed genes by RT‐PCR. (a) Relative expression levels of a panel of genes in patients with irAEs compared to those without irAEs. (b) Relative expression levels of a panel of genes in patients with severe irAEs compared to those without irAEs.
Table S1: Primer sequences used for RT‐PCR validation of candidate biomarkers.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.