Content area
Background
Checkpoint inhibitor-related pneumonitis (CIP) represents a highly lethal immune-related adverse event. Early diagnosis of CIP is crucial for timely intervention and improved prognosis; however, the absence of precise and effective diagnostic techniques often leads to underdiagnosis and misdiagnosis. This study aims to identify microRNA (miRNA) features from serum and extracellular vesicles (EVs) for the early CIP detection and prognosis.
Methods
Small RNA sequencing identified candidate miRNAs in 27 serum-derived EV samples from persons with lung cancer and CIP (CIP group) and those without, including immunotherapy-treated persons with lung cancer without CIP (immune checkpoint inhibitor, ICI group) and patients with infectious pneumonia (PNE group). These miRNAs were validated in EV samples in a discovery cohort (n=48) using a quantitative reverse transcription-PCR (qRT-PCR). Diagnostic models for the biomarkers were developed using a training cohort (ICI:47, PNE:28, CIP:31) and validated in a separate validation cohort (ICI:32, PNE:19, CIP:21) using qRT-PCR in both EV and serum samples, and logistic regression. Using a Cox regression model, we built a prognostic risk stratification for patients with CIP based on three miRNAs.
Results
Sequencing analysis initially screened and identified 13 overexpressed miRNAs in patients with CIP. Subsequently, qRT-PCR demonstrated that three miRNAs (EVs miR-193a-5p, serum miR-193a-5p, and serum miR-378a-3p) effectively distinguished CIP from non-CIP individuals (training cohort: area under the curve (AUC)=0.870; validation cohort: AUC=0.837). Notably, this miRNA signature was equally robust in differentiating CIP from ICI (training cohort: AUC=0.823; validation cohort: AUC=0.845) and PNE groups (training cohort: AUC=0.892; validation cohort: AUC=0.907). Furthermore, when combined with lymphocyte levels, the miRNA signature significantly enhanced the overall diagnostic accuracy in distinguishing CIP from the non-CIP group (training cohort: AUC=0.900; validation cohort: AUC=0.932), and maintained its robustness in distinguishing CIP from the ICI group (training cohort: AUC=0.898; validation cohort: AUC=0.946) and the PNE group (training cohort: AUC=0.938; validation cohort: AUC=0.959). Additionally, the three-miRNA panel was independently and significantly associated with overall survival in patients with CIP (HR: 2.827; p=0.040).
Conclusions
Our circulating miRNA-based signature represents a non-invasive and robust diagnostic tool for patients with CIP and could accurately predict their prognosis. This signature may facilitate early detection and personalized management of these patients.
Full text
Correspondence to Dr Lifu Wang; [email protected]; Dr Chengzhi Zhou; [email protected]; Dr Yao Liao; [email protected]
WHAT IS ALREADY KNOWN ON THIS TOPIC
MicroRNAs (miRNAs) have been used as diagnostic and prognostic biomarkers for cancers and inflammatory diseases, attributed to their notable abundance in blood. Although limited research suggests that circulating miRNAs exhibit differential expression in immune-related adverse events, their role in the diagnosis or prognosis of checkpoint inhibitor pneumonitis (CIP) remains inadequately understood.
WHAT THIS STUDY ADDS
This study illustrates that the integration of miRNA signatures from extracellular vesicles and serum effectively distinguishes CIP from non-CIP conditions and, more critically, differentiates CIP from infectious pneumonia. Furthermore, the combination of miRNA signatures with traditionally assessed lymphocyte levels enhances diagnostic accuracy. Additionally, miRNA-based risk stratification may serve as a predictive tool for CIP prognosis.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
The findings underscore the potential of circulating miRNAs in the diagnostic and prognostic evaluation of CIP. This signature may facilitate early detection and personalized management of these patients.
Introduction
Immune checkpoint inhibitors (ICIs) have the capacity to enhance antitumor T-cell activity, thereby exerting sustained antitumor effects against cancer. However, immune system activation by ICIs can also lead to immune-related adverse events (irAEs) across various organs and tissues.1 Severe irAEs can be fatal, with ICI-related pneumonitis (CIP) potentially causing extensive respiratory symptoms and parenchymal abnormalities, which may progress to respiratory failure and even death.2 A previous study reported a significantly higher incidence of severe irAEs in the respiratory system, with a mortality rate reaching 26.3%.3 Real-world data indicate that the incidence of CIP ranges from 9% to 19%.4–6
Early recognition and timely intervention are crucial for the effective management of CIP. Unfortunately, there is no established gold standard for diagnosing CIP, as it presents challenges due to the non-specific nature of its symptoms, imaging findings, laboratory results, and pathological features. Currently, clinicians diagnose CIP by evaluating the patient’s symptoms and conducting various tests, following the exclusion of other diseases and incorporating discussions within a multidisciplinary team.7 In critically ill patients, diagnostic tools such as chest CT, alveolar lavage, and pathological biopsy may not be feasible. A missed diagnosis or delayed treatment of CIP can lead to disease progression and potentially fatal outcomes. Conversely, empirical treatment of CIP may adversely affect the efficacy of immunotherapy.8 Consequently, there is a critical need to develop non-invasive auxiliary diagnostic biomarkers to enhance the identification of CIP. Previous studies have identified correlations between peripheral blood markers (such as lymphocytes (LYM), the neutrophil-to-lymphocyte ratio (NLR), albumin, and cytokines IL-6, IL-10, IL-17A, and IL-35) and the diagnosis of CIP.6 9 However, these markers lack the necessary sensitivity and specificity, and they are unable to differentiate CIP from infectious pneumonia. Furthermore, CIP may coexist with pulmonary infections, a condition we refer to as ‘mixed type’, which complicates the diagnostic process for CIP.10
In recent years, the advent of high-throughput molecular profiling techniques has catalyzed extensive research using blood-based biomarkers, including circulating proteins, DNA, and various RNA molecules, to differentiate diseases. MicroRNAs (miRNAs), which are small non-coding RNA molecules involved in gene expression regulation, have gained prominence as potential diagnostic and prognostic tools due to their high stability and abundance in tissues, blood, and other body fluids.11 12 Extracellular vesicles (EVs), secreted into the bloodstream by diverse cell types, encapsulate biologically active molecules that mirror their cellular origin and physiological state.13 EVs have been implicated in the initiation and progression of cancers, inflammatory conditions, and autoimmune disorders.14 15 Notably, EVs derived from pulmonary and extrapulmonary tissues have been reported to influence the physiological and pathological processes of lung tissues by facilitating communication between alveolar cells within the lung microenvironment.16 EVs may be associated with CIP, but limited literature has evaluated this association. Torasawa et al revealed that serum EV proteomic signatures exhibit predictive value for durvalumab-induced pneumonitis in locally advanced non-small cell lung cancer, yet show no diagnostic utility for CIP.17 EVs containing miRNAs are recognized as some of the most abundant and stable molecules within EVs.18 Consequently, EVs offer a promising framework for enhanced specificity in the detection of miRNA biomarkers in blood samples. Circulating EV-miRNAs have been implicated in the diagnostic processes of various cancers, autoimmune disorders, and inflammatory lung diseases.19 20 Garcia-Martin et al demonstrated significant differences between miRNAs derived from EVs and those from cells.21 Considering that blood miRNAs exhibit high sensitivity and that EV-miRNAs demonstrate strong tissue specificity, their combination can offer an optimal combination of sensitivity and specificity.22 23 A recent study revealed that combined miRNA panels from peripheral blood and EVs exhibit a superior area under the curve (AUC) value (AUC=0.98) for diagnosing pancreatic ductal adenocarcinoma compared with panels derived solely from peripheral blood (AUC=0.90) or EVs (AUC=0.97).23
In the present study, we conducted an analysis of serum and EV miRNA profiles between patients with CIP and non-CIP subjects to identify a novel diagnostic miRNA signature for the early detection of CIP. We subsequently conducted a rigorous evaluation and validation of this non-invasive circulating signature within clinical cohorts to determine its diagnostic efficacy for early CIP detection. Additionally, we developed a prognostic signature based on the same miRNAs, capable of predicting clinical outcomes in patients with CIP.
Materials and methods
Patient enrollment and sample collection
This case–control observational study included a CIP group and a non-CIP group, with the non-CIP group comprising the ICI group (patients who received Programmed Cell Death 1 (PD-1)/Programmed Cell Death-Ligand 1 (PD-L1) inhibitor without developing CIP) and the PNE group (patients with infectious pneumonia). The inclusion and exclusion criteria for the CIP group mandated that patients with lung cancer (stages I–IV) had received at least one dose of PD-1/PD-L1 inhibitor, irrespective of treatment line, for anticancer treatment and subsequently developed CIP, while those with pre-existing interstitial lung disease (ILD) or with missing detailed clinical data were excluded. CIP was diagnosed according to the guidelines of the National Comprehensive Cancer Network, the American Society for Clinical Oncology, and the European Society for Medical Oncology.24–26 The inclusion and exclusion criteria for the ICI group were defined as follows: persons with lung cancer (stages I–IV) who received at least one dose of PD-1/PD-L1 inhibitor as first-line therapy without developing CIP were included; patients with concurrent other cancers, pre-existing ILD, radiation pneumonitis, those who had experienced other serious irAEs, or those with incomplete clinical data were excluded. For the PNE group, the inclusion criteria comprised patients with infectious pneumonia, including bacterial pneumonia, fungal pneumonia, tuberculous pneumonia, and inflammatory pulmonary nodules, while exclusion criteria included patients with cancer, pre-existing ILD, or missing clinical data.
The first phase of the study included a pilot cohort and a discovery cohort from the National Center for Respiratory Medicine between June 1, 2020 and December 31, 2022. In the pilot cohort, small RNA sequencing was performed in 27 serum EV specimens, consisting of nine patients with CIP and 18 patients without CIP (nine from each of the ICI and PNE groups). In the discovery cohort, quantitative real-time reverse transcription-PCR (qRT-PCR) assays were performed to examine the expression levels of candidate EV-miRNAs in 48 specimens from 16 patients with CIP and 32 non-CIP controls (16 cases each in ICI and PNE groups).
In the second phase of the study, which included training and validation cohorts from the National Center for Respiratory Medicine, Collaborative Innovation Center for Cancer Medicine, and Dongguan People’s Hospital between June 1, 2020 and August 31, 2024, qRT-PCR assays were conducted to assess the expression levels of serum and EV-miRNAs. A total of 291 patients were initially recruited for this study. However, individuals with a history of ILD (n=42), those with incomplete baseline information (n=43), and those with severe hemolysis (n=28) were excluded, resulting in the inclusion of 178 participants. To minimize potential bias between the groups, the two cohorts were combined and randomly divided into two cohorts in a 6:4 ratio (training cohort, n=106; validation cohort, n=72). The training cohort comprised 31 patients with CIP, 47 patients in the ICI group, and 28 patients in the PNE group. The clinical validation cohort comprised 21 patients diagnosed with CIP, 32 patients in the ICI group, and 19 patients in the PNE group. The whole study design is illustrated in online supplemental figure S1.
Blood samples from both the PNE and CIP groups were required to be collected within 72 hours following the onset of symptoms for inclusion in the study. A BD Vacutainer Serum Tube was employed to collect 6 mL of peripheral blood. To obtain serum, the samples were centrifuged at 3000 rpm for 10 min, after which the serum was isolated and aliquoted into four 1.5 mL cryovials for storage at −80°C until further analysis. Clinical characteristics were documented at the time of blood collection, encompassing variables such as age, sex, smoking history, tumor histology, TNM stage, and treatment information, as detailed in online supplemental table S1. Routine peripheral blood parameters were also collected concurrently with blood sampling. Overall survival (OS) was calculated from the date of clinical diagnosis of CIP.
EV isolation
EVs were isolated following the standard protocol established by our group.27 Briefly, 1 mL of thawed serum per participant was centrifuged at 300×g for 10 min at 4°C. The supernatants were then subjected to sequential centrifugation at 2000×g for 10 min at 4°C and 10000×g for 60 min at 4°C to remove cellular debris. Thereafter, the supernatant was centrifuged at 120000×g for 90 min at 4°C using 4 mL Open-Top Tickwall Polycarbonate Tubes (Beckman Coulter) in an Optima XE-100 tabletop ultracentrifuge (Swinging bucket rotor, model SW60 Ti, Optima XE-100, Beckman Coulter, USA). Finally, EV pellets were resuspended in 100 µL phosphate-buffered saline (PBS) after another round of ultracentrifugation with PBS. The isolated EVs were stored at −80°C before use.
Western blot analysis
The isolated EVs underwent lysis, and A549 cell lysate in RIPA buffer (P0013B, Beyotime, China) supplemented with protease inhibitor PMSF (ST507, Beyotime, China). The total protein content was quantified employing a BCA protein assay kit (P0010, Beyotime, China). For protein expression analysis, rabbit polyclonal antibodies against CD9 (1:1000, ab236630, Abcam, UK), CD81 (1:1000, ab109201, Abcam, UK), Alix (1:1000, ab275377, Abcam, UK), and Calnexin (1:5000, 81 938-1-RR, Proteintech, USA) were used as the primary antibodies. Horseradish peroxidase-conjugated goat anti-rabbit IgG antibody (511203, Zenbio; 1:5000) was used as the secondary antibody.
Nanoflow analysis
Particle concentration and size of EVs were analyzed using Nano flow cytometry (NanoFCM, China). Briefly, the samples were diluted 500–1000 fold in PBS to optimize particle visibility. Data acquisition and analysis were performed using NanoFCM Professional Suite V.1.8 software.
Transmission electron microscopy
We deposited 10 microlitres of EV solution onto a copper grid and incubated it for 1 min at room temperature. Subsequently, the EVs were rinsed with sterile distilled water and contrasted using a uranium acetate solution for 1 min. The samples were then dried under an incandescent lamp for 2 min. The copper mesh was observed and photographed using a table-mounted low-voltage transmission electron microscope (TEM) (JEM-1400PLUS, Japan).
Small RNA sequencing
We assessed serum-derived EVs from the CIP (n=9), ICI (n=9), and PNE (n=9) groups. The same cohort of EVs was pooled for small RNA sequencing. Total RNA was extracted from EVs using TRIzol Reagent (Invitrogen) following the manufacturer’s protocol. RNA integrity was assessed using the Agilent 5300 Bioanalyzer system, while quantification was performed with a NanoDrop ND-2000 spectrophotometer. Sequencing library preparation was exclusively conducted using RNA specimens meeting stringent quality criteria: optical density ratios (OD260/280=1.8–2.2; OD260/230≥2.0), RNA quality number (RQN≥6.5), ribosomal ratio (28S:18S≥1.0), and minimum quantity (>1 µg).
RNA processing workflow was conducted at Chengqi Biotechnology (Shenzhen, China) under standardized protocols. The procedure comprised four phases: (1) Total RNA extraction (1 µg input/sample) using miRNeasy Mini Kit (Qiagen, Cat# 217004); (2) Library preparation with QIAseq miRNA Library Kit (Qiagen, Cat# 331505) involving: 5'/3' adapter ligation (T4 RNA ligase, Thermo Scientific), reverse transcription (SuperScript IV, Thermo Fisher Scientific), and PCR amplification (11 cycles) with Q5 High-Fidelity DNA Polymerase (NEB); (3) Quality control through fragment selection (6% TBE-PAGE, Bio-Rad) and quantification (Qubit 4.0 Fluorometer, Thermo Fisher); and (4) Sequencing execution on Illumina NovaSeq X Plus platform (150 bp single-end reads).
Quality control and reads mapping
Raw data (raw reads) in fastq format were first processed using fastp with default parameters. Then, clean reads were obtained by removing 3' end adaptors, reads containing poly-N, low-quality bases (Sanger base quality <20) of the 3' end, and sequencing adaptors from the raw data using the fastx toolkit software. All identical sequences between 18 and 35 nt were counted and eliminated from the initial dataset. Bowtie was used to annotate the chromosomal location against the reference genome data.
Identification of miRNAs and differential expression analysis
The mapped small RNA tags were first used to identify known miRNA using the miRBase database (http://www.mirbase.org/) as a reference. Then, the remaining tags were aligned with the Rfam database and Repbase database to exclude ribosomal RNA, transfer RNA, small nuclear RNA (snRNA), small nucleolar RNA, and other ncRNA and repeats. The unannotated tags were predicted and identified as novel miRNAs using the mirdeep2 software according to the tag positions in the genome and their hairpin structures. The expression level of each miRNA was calculated using the counts per million (CPM) method. Differentially expressed miRNAs (DEMs) with |log2FC|≥1 and p values <0.05 were considered as significantly differentially expressed genes.
RNA extraction and qRT-PCR
RNA from serum-derived EVs was extracted using HiPure Exosome RNA Midi Kit (Magen, China), which includes an essential Proteinase K digestion step (55°C, 15 min). Total circulating RNA from serum was isolated using HiPure Liquid RNA/miRNA Kit (Magen, China), following the manufacturer’s instructions. RNA concentration was assessed with a NanoDrop ND-2000 spectrophotometer (Thermo Scientific, USA). For reverse transcription, cDNA was synthesized using a miRNA first-strand cDNA synthesis kit (Accurate Biology, China). qRT-PCR analysis was conducted on a QuantStudio (TM) 6 Flex System (Thermo Fisher Scientific, UK) using the SYBR Green Premix Pro Taq HS qPCR Kit (Accurate Biology, China) with the miRNAs primer set. miRNA expression levels were calculated using the 2−ΔΔCT method, with U6 snRNA employed as the reference miRNA for normalization.28 The primers used are detailed in online supplemental table S2.
Statistical analysis
Statistical analyses and graphical representations were performed using SPSS V.25.0 (IBM, R (V.4.0.3, https://cran.r-project.org/), and GraphPad Prism V.8.0 (GraphPad Software, California, USA). Differences in continuous variables between the two groups were analyzed using a two-tailed unpaired Student’s t-test or the Mann‒Whitney U test. The Kruskal-Wallis H test was used to analyze the differences in age across the three groups. The χ² or Fisher’s exact test was used to analyze categorical variables. The Pearson correlation coefficient (r) was employed to assess the correlation between miRNA expression in EVs and serum components, and the associations of miRNAs with clinical laboratory parameters. The correlation between miRNA expression and clinical baseline data was assessed using rank correlation. A miRNA diagnostic model was constructed using multivariate logistic regression analysis in the training cohort. A stepwise backward selection method was applied in the training cohort to incorporate laboratory data into the miRNA model. The validation cohort served to validate the diagnostic models. AUC values derived from the receiver operating characteristic (ROC) curves were used to assess the predictive performance of the diagnostic models. Optimal cut-off values for the ROC curves were determined based on the Youden index. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy of the miRNA-based models were estimated across all the cohorts. Calibration plots were used to evaluate the performance of the models, using 1000 bootstrap resamples. Decision curve analysis (DCA) and clinical impact curve (CIC) evaluated the models’clinical applicability.
To predict risk scores, we used a Cox proportional hazards regression model. After fitting the model using the expression data of three miRNAs and survival information from 36 patients diagnosed with CIP, a risk score was calculated for each patient. The median risk score was used as the threshold for classifying patients into either low-risk or high-risk groups. OS was estimated from the time of clinical diagnosis of CIP. OS curves were generated using the Kaplan-Meier method and compared using a log-rank test. A Cox proportional hazards model was employed to identify prognostic factors associated with OS in the CIP group through multivariable survival analysis, including those variables with p values <0.1 in the univariate analysis. HRs for both univariate and multivariate analyses were calculated, and 95% CIs values were provided. Statistical significance was determined using two-tailed p values <0.05.
Results
Identification of EVs and particle miRNA markers in CIP
Circulating EVs were successfully isolated from patients' serum samples and characterized using TEM, nanoflow analysis, and western blot analysis. TEM revealed multiple isolated vesicles, which were characteristically round, cup-shaped, and exhibited a double closed membrane structure (figure 1A–C). Nanoflow analysis indicated that EV concentrations were 3.77×1012, 3.16×1012, and 1.55×1012, particles/mL in the ICI, PNE, and CIP groups, respectively. Additionally, the mean diameters of the collected particles were 100.31 nm, 92.6 nm, and 92.89 nm in the ICI, PNE, and CIP groups, respectively (figure 1D–F). Western blot analysis confirmed the positive expression of the EV markers CD9, CD81, and Alix, and absence of the negative EV marker Calnexin, in vesicles isolated from serum samples (figure 1G and online supplemental file 2). These findings suggest that there were no significant differences in morphological, size variation, and amount between the controls and CIP EVs.
Figure 1. Characterization of serum-derived extracellular vesicles (EVs) and identification of EV-associated microRNA (miRNA) expression profiles. (A–C) Representative transmission electron microscopy images of serum-derived EVs from ICI, PNE, and CIP groups. (D–F) Nanoflow analysis determined the size distribution and concentration of the isolated particles. (G) Immunoblot characterization of EVs from serum depicted presence of EV markers CD9, CD81, and Alix, and absence of other negative EV marker such as Calnexin. (H–I) Volcano plots show the distribution of upregulated and downregulated miRNAs by at least fourfold and p values <0.05. (J) Venn diagram depicting the common and unique miRNAs in two comparison groups. (K–N) The bar charts displayed four filtering methods: the counts per million (CPM) of ICI and PNE groups is zero, sorted by CPM of the CIP group; the CPM of the ICI group is zero, sorted by CPM of the CIP group; comparing the PNE and ICI groups, the CIP group was upregulated, sorted by CPM of the CIP group; comparing the ICI group, the CIP group was upregulated, and sorted by CPM of the CIP group (13 candidate miRNAs highlighted in red). CIP, checkpoint inhibitor-related pneumonitis; ICI, immune checkpoint inhibitor; miRNAs, microRNA; PNE, pneumonia.
The primary objective of this study was to identify clinically relevant EV-associated miRNAs (EV-miRNAs) as biomarkers for diagnosing patients with CIP. As illustrated in the overall workflow (online supplemental figure S1), RNA-seq analysis was conducted on the pilot cohort. We screened EV-miRNAs that were distinct between patients with CIP and patients without CIP. We selected the DEMs determined according to the stringent statistical threshold (|log2FC|≥2 and p values <0.05) between the CIP and ICI groups or the CIP and PNE groups. A total of 592 miRNAs were found to differ between the CIP and ICI groups, and 659 miRNAs differed between the CIP and PNE groups, with 266 miRNAs being common differentials (figure 1H–J). Four screening methods were employed: (1) miRNAs with a CPM of 0 in both the ICI and PNE groups were sorted by their CPM in the CIP group; (2) miRNAs with a CPM of 0 in the ICI group were sorted by their CPM in the CIP group; (3) when comparing the PNE and ICI groups, miRNAs upregulated in the CIP group were sorted by their CPM in the CIP group; and (4) when comparing the ICI group, miRNAs upregulated in the CIP group were sorted by their CPM in the CIP group (figure 1K–N). Through this process, we selected 13 EV-miRNAs specific for patients with CIP that are distinct from the controls.
Identification of miRNA profiles in the discovery and training cohorts
We obtained 13 candidate EV-miRNAs from RNA-seq analysis and validated them using qRT-PCR analysis on the EV samples from the discovery cohort. One miRNA was excluded due to a low detection rate (<75%).29 We found four EV-miRNAs (miR-542-3p, miR-6838-3p, miR-193a-5p, and miR-378a-3p) exhibited significantly different expression levels between the CIP and ICI groups or CIP and PNE groups (online supplemental figure S2). Further validation in the training cohort using qRT-PCR confirmed that three EV-miRNAs (miR-6838-3p, miR-193a-5p, and miR-378a-3p) were significantly upregulated in the CIP group compared with the ICI group or the PNE group (figure 2A–D). In addition, qRT-PCR analysis of serum samples from the training cohort indicated that miR-193a-5p and miR-378a-3p were upregulated in the CIP group, compared with the ICI group or the PNE group (figure 2E–G). Based on these findings, we ultimately screened EVs miR-193a-5p, serum miR-193a-5p, and serum miR-378a-3p as potential biomarkers for distinguishing patients with CIP from both ICI and PNE groups. Subsequently, to elucidate the expression pattern of miR-193a-5p in total serum and EVs, we conducted a correlation analysis of miR-193a-5p between these two components. Interestingly, miR-193a-5p exhibited no significant correlation between the two components (r=0.18, p=0.1) (figure 2H). This observation suggests that the two types of miRNA markers originate from distinct sources; therefore, the integration of the two may enhance diagnostic performance.
Figure 2. Detection of differentially expressed miRNAs in serum and serum-derived extracellular vesicles (EVs) in the training cohort. (A–D) The expression level of miR-542-3 p, miR-6838–3 p, miR-193a-5p, and miR-378a-3p in serum-derived EVs in ICI (n=47), PNE (n=28), and CIP (n=31) groups, as quantified by qRT-PCR. (E–G) The expression level of serum miR-6838–3 p, miR-193a-5p, and miR-378a-3p was measured by qRT-PCR (ICI, n=47; PNE, n=28; and CIP, n=31). (H) The scatter plot indicated no significant correlation between serum miR-193a-5p and EVs miR-193a-5p. The data are shown as the means±SDs (ns, not significant, *p<0.05, **p<0.01, ***p<0.001). CIP, checkpoint inhibitor-related pneumonitis; ICI, immune checkpoint inhibitor; PNE, pneumonia; qRT-PCR, quantitative reverse transcription PCR.
Construction of a serum and EV-miRNA model for CIP diagnosis
We performed ROC analysis using three preselected serum and EV miRNAs to estimate the accuracy of the CIP diagnostic test. To further evaluate the performance of these miRNAs as combinatorial panels, logistic regression was applied to all possible combinations of the three miRNAs, and ROC curves were generated to determine the optimal miRNA combination for diagnosing CIP. Our findings indicate that the three-miRNA signature exhibited greater accuracy (AUC=0.870; 95% CI: 0.778 to 0.962) compared with other combinations (figure 3A,B). The equation for the miRNA combination model (miRsig1) is the following: miRsig1=−3.983+(0.736×EV-miR-193a-5p)+(0.615×serum-miR-193a-5p)+(0.458×serum-miR-378a-3p). Notably, the miRsig1 model demonstrated superior overall diagnostic accuracy (87.7%), PPV (92.3%), and NPV (86.7%) (online supplemental table S3). A calibration plot indicated that the predicted probabilities from the miRNA combination were consistent with the observed probabilities (figure 3C).
Figure 3. Evaluation and validation of miRNA diagnostic signature in the training and validation cohorts. (A) The area under the ROC curves (AUCs) in individual miRNAs and combination sets in distinguishing between CIP and non-CIP groups in the training cohort. (B) ROC curve analysis for the three-miRNAs combination panel CIP versus non-CIP (miRsig1 model) in the training cohort. (C) Calibration curve analyses to evaluate the performance of the miRsig1 model in the training cohort. (D) The AUCs in individual miRNAs and combination sets in distinguishing between CIP and ICI groups in the training cohort. (E) ROC curve analysis for the three-miRNAs combination panel CIP vs ICI (miRsig2 model) in the training cohort. (F) Calibration curve analysis to evaluate the performance of the miRsig2 model in the training cohort. (G) The AUCs in individual miRNAs and combination sets in distinguishing between CIP and PNE groups in the training cohort. (H) ROC curve analyses for the three-miRNAs combination panel CIP vs PNE (miRsig3 model) in the training cohort. (I) Calibration curve analysis to evaluate the performance of the miRsig3 model in the training cohort. (J) The expression level of serum-derived EVs miR-193a-5p (left), serum miR-193a-5p (center), and serum miR-378a-3p (right) measured by qRT-PCR in the validation cohort (ICI, n=32; PNE, n=19; and CIP, n=21). (K) ROC curve analyses for the three-miRNAs combination panel, assessing its discriminatory efficacy between CIP and non-CIP (left), CIP and ICI (center), and CIP and PNE (right) in the validation cohort, respectively. The data are shown as the means±SDs (ns, not significant, *p<0.05, **p<0.01, ***p<0.001). AUC, area under the curve; CIP, checkpoint inhibitor-related pneumonitis; EVs, extracellular vesicles; ICI, immune checkpoint inhibitor; miRNAs, microRNA; PNE, pneumonia; ROC, receiver operating characteristic.
To further differentiate between patients in the CIP and ICI groups, we developed a combination of miRNAs and identified that the model incorporating these three miRNAs yielded the highest AUC values (figure 3D,E). The equation for this model, termed miRsig2, is as follows: miRsig2=−3.648+(0.898×EV-miR-193a-5p)+(0.720×serum-miR-193a-5p)+(0.200×serum-miR-378a-3p). The miRsig2 model exhibited an AUC value of 0.823 (95% CI: 0.695 to 0.950), with a sensitivity of 60%, specificity of 100%, PPV of 100%, and NPV of 84.3%, indicating robust diagnostic performance (figure 3F and online supplemental table S3).
Next, we analyzed the differential diagnosis between the CIP group and the PNE group. By integrating these three miRNAs, we achieved an AUC of 0.892 (95% CI: 0.794 to 0.989), with a sensitivity of 75.0%, specificity of 88.9%, PPV of 88.2%, and NPV of 76.2%, indicating good diagnostic performance (figure 3G–I and online supplemental table S3). The equation for the miRNA combination model (miRsig3) is as follows: miRsig3=−3.925+(0.640×EV-miR-193a-5p)+(0.469×serum-miR-193a-5p)+(1.578×serum-miR-378a-3p). Collectively, these findings concerning miRNA biomarkers support the initial hypothesis, while both serum and EVs miRNA markers demonstrate considerable robustness. Their combined analysis offers a superior method for the accurate identification of patients with CIP, whether distinguishing ICI or PNE groups.
Successful validation of the circulating miRNA diagnostic signature in the validation cohort
Next, the diagnostic potential of miRNA analysis was validated through qRT-PCR analysis of serum and EV specimens from 21 patients with CIP and 51 controls within the validation cohort. The expression levels of EV-miR-193a-5p, serum-miR-193a-5p, and serum-miR-378a-3p were consistently elevated in the CIP group compared with the ICI or PNE groups (figure 3J). In this validation effort, the logistic regression equations, coefficients, and constants for each miRNA were applied as established in the training cohort model, and risk scores were subsequently calculated. Consistent with findings from the training cohort, the miRsig1 model exhibited superior diagnostic performance with an AUC of 0.837 (95% CI: 0.712 to 0.962), sensitivity of 72.2%, specificity of 93.0%, PPV of 81.3%, and NPV of 88.9% in this validation cohort of CIP and non-CIP groups (figure 3K and online supplemental table S3). Likewise, as was the case in the training cohort, the diagnostic potential of the miRsig2 model (AUC=0.845, sensitivity of 76.5%, specificity of 89.3%, PPV of 81.3%, and NPV of 86.2%) and the miRsig3 model (AUC=0.907, sensitivity of 77.8%, specificity of 93.3%, PPV of 93.3%, and NPV of 77.8%) remained comparable, even within this validation cohort of CIP versus ICI or PNE (figure 3K and online supplemental table S3). These findings highlighted that while individual serum and EV-miRNAs exhibited considerable robustness, their combined yielded superior diagnostic performance in distinguishing CIP from either the ICI or PNE groups.
Circulating miRNA signature and LYM levels in blood significantly improve diagnostic accuracy for CIP
In clinical practice, patients routinely undergo laboratory tests, and previous studies have indicated that these results may be pertinent to the diagnosis of CIP.6 9 Consequently, we collected symptoms and laboratory test data from the training cohort. Our analysis revealed that the proportion of cough, shortness of breath, fever, and chest tightness was higher among patients with CIP compared with the ICI group (figure 4A). However, these symptoms were not effective in differentiating patients with CIP from those in the PNE group (figure 4A). Within the training cohort, the CIP group exhibited lower levels of LYM and elevated NLR, and hypersensitive C reactive protein (hsCRP) levels compared with both the ICI and PNE groups (figure 4B–I). Subsequently, we validated these peripheral blood markers in the validation cohort. In the validation cohort, the NLR increased in the CIP group, while the LYM decreased, compared with both the ICI and PNE groups (figure 4J). The hsCRP did not exhibit significant differences between the CIP and PNE groups in the validation cohort (figure 4J).
Figure 4. The performance evaluation of the miRNA signature combined with lymphocyte (LYM) in the training and validation cohorts. (A) Comparisons of clinical symptoms were from the CIP versus ICI group (top) and the CIP versus PNE (bottom). (B–I) The differential expressions observed in routine blood tests (B), neutrophil-to-lymphocyte ratio (NLR) (C), platelet-to-lymphocyte ratio (D), albumin (E), lactate dehydrogenase (F), cytokines (G), procalcitonin (H), and hypersensitive C reactive protein (hsCRP) (I), within the training cohort. (J) The differential expressions of LYM (left), NLR (center), and hsCRP (right) in the validation cohort. (K–P) ROC curve analyses for the miRNA signature combined with LYM levels, validating its discrimination efficiency in the training and validation cohorts, respectively. The data are shown as the means±SDs (ns, not significant, *p<0.05, **p<0.01, ****p<0.0001). CIP, checkpoint inhibitor-related pneumonitis; EOS, eosinophils; hsCRP, hypersensitive C reactive protein; ICI, immune checkpoint inhibitor; miRNA, microRNA; NEU, neutrophils; PLT, platelet; PNE, pneumonia; ROC, receiver operating characteristic; WBC, white blood cell.
The potential enhancement of diagnostic performance through the integration of a miRNA-based model with LYM/NLR was investigated. Logistic regression was employed to conduct a stepwise screen of miRsig1, NLR, and LYM, ultimately identifying miRsig1 and LYM for model development. Notably, the AUC of miRsigs was significantly higher than that for LYM in the training cohort (figure 4K–M). It was particularly encouraging to observe that combining LYM with miRsig1 significantly enhanced diagnostic performance, as demonstrated by a superior AUC of 0.900 (95% CI: 0.808 to 0.993) in the training cohort (sensitivity, 85.7%; specificity, 92.3%; PPV, 81.8%; NPV, 94.1%, and accuracy, 90.4%; figure 4K and online supplemental table S3). More notably, this diagnostic performance was equally remarkable when miRsig2 was combined with LYM (AUC of 0.898; sensitivity of 76.2%; specificity of 100.0%), effectively distinguishing CIP from the ICI group (figure 4L and online supplemental table S3). Similarly, the combination of miRsig3 with LYM also demonstrated a higher AUC of 0.938 (sensitivity, 95.0%; specificity, 88.2%), confirming its ability to discriminate between CIP and PNE groups (figure 4M and online supplemental table S3).
Similarly, conducting this analysis on patients within the validation cohort resulted in a marked enhancement of diagnostic sensitivity and specificity for distinguishing CIP from non-CIP, ICI, or PNE groups (figure 4N–P and online supplemental table S3). In this validation cohort, the AUC of the miRsig1+LYM model was further improved up to 0.932 (sensitivity, 72.2%; specificity, 100.0%; PPV, 100.0%; NPV, 89.1%, and accuracy, 91.5%; figure 4N and online supplemental table S3). Likewise, the combined miRsig2 and LYM exhibited a superior diagnostic performance with an AUC value of 0.946 (sensitivity of 82.4%, specificity of 100.0%, PPV of 100.0%, NPV of 89.7%, and accuracy of 93.0%; figure 4O and online supplemental table S3) in this validation cohort. This optimized combination transcriptomic assay performed remarkably well for the identification of CIP and PNE (AUC, 0.959; sensitivity, 94.4%; specificity, 86.7%; PPV, 89.5%; NPV, 92.9%; and accuracy, 90.9%; figure 4P and online supplemental table S3).
Our results further demonstrated a positive correlation between the levels of EV-miR-193a-5p and markers such as white cell counts, neutrophils (Neu), NLR, and hsCRP. In addition, serum miR-193a-5p levels showed a positive correlation with hsCRP levels (online supplemental figure S3). Notably, these three miRNAs did not exhibit significant correlations with age, sex, and smoking status (online supplemental table S4). Collectively, these findings underscore that while the miRNA-based assay was quite robust independently, its combination with LYM levels substantially enhances overall diagnostic accuracy, highlighting its promising potential for clinical application in the early detection of CIP.
An miRNA-based model offers a significant benefit for the early detection of patients with CIP
In current clinical practice, the diagnosis of patients with CIP is achieved through symptoms, tests, imaging, and pathology, followed by corticosteroid treatment. Accordingly, based on current clinical practice, false-positive or false-negative results could have detrimental effects on individuals undergoing such screening. To evaluate the clinical utility of the miRsig1+LYM combination model, DCA, CIC, and calibration curve analysis were performed. The DCA indicated that the miRsig1+LYM combination model provided a higher net benefit across a broad range of threshold probabilities compared with the strategies of diagnosing all or none of the patients with CIP (figure 5A). Further examination using the CIC revealed that when the threshold probability exceeded 0.4, the model’s diagnostic predictions closely aligned with actual occurrences, demonstrating high clinical efficacy (figure 5B). These findings suggested that the combined model offers significant clinical benefits for identifying patients with CIP. In addition, the calibration curve demonstrated a good agreement between observed and predicted probabilities (online supplemental figure S4), highlighting the robust diagnostic potential of the combination model for identifying patients with CIP.
Figure 5. Diagnostic potential evaluation of the combination model and construction of miRNA prognostic risk model in patients with CIP. (A) Decision curve analysis and (B) Clinical impact curve to evaluate the performance of the miRNA signature in combination with lymphocyte (LYM) levels. (C) Prognostic risk score model analysis with risk score, survival status distribution, and heat maps of three prognostic miRNAs in patients with low-risk and high-risk CIP (n=36). (D) Kaplan-Meier survival curves of the low-risk (n=18) and high-risk groups (n=18). (E) Cox proportional hazard regression analysis of overall survival in prespecified subgroup. ALB, albumin; CIP, checkpoint inhibitor-related pneumonitis; EVs, extracellular vesicles; LDH, lactate dehydrogenase; miRNA, microRNA; Neu, neutrophils; NLR, neutrophil-to-lymphocyte ratio; TNF-α, tumor necrosis factor-α; TPS, tumor proportion score; WBC, white blood cell.
Prediction of CIP prognosis through the combination of circulating miRNA panel
To develop a comprehensive prognostic risk model, we analyzed risk scores, survival years, and miRNA expression levels. The final follow-up time was August 31, 2024. The total follow-up period was 68 months. The median follow-up time was 35.6 months (95% CI: 29.3 to 42.0 months). A total of 36 patients with follow-up data were enrolled in the survival analysis. These patients were stratified into low-risk (n=18) and high-risk groups (n=18) according to the median risk score derived from previously assessed miRNAs (EV-miR-193a-5p, serum-miR-193a-5p, and serum-miR-378a-3p) (figure 5C). Analysis of OS revealed that individuals in the high-risk group exhibited a significantly worse clinical prognosis and an elevated mortality risk compared with those in the low-risk group (HR=2.594, 95% CI: 1.085 to 6.206; p=0.032) (figure 5D). We developed a univariate Cox proportional hazards regression model, which demonstrated that albumin, tumor necrosis factor-α (TNF-α), and miRNA risk stratification were significantly associated with OS (online supplemental table S5 and figure 5E). In the multivariate Cox proportional hazards regression model, only miRNA risk stratification (high-risk vs low-risk: HR=2.827, 95% CI: 1.051 to 7.603; p=0.040) remained significantly and independently correlated with OS in patients with CIP (online supplemental table S5). These findings indicated that the three-miRNA signature serves as a reliable prognostic marker of OS in patients with CIP.
Discussion
CIP is characterized by a high severity and lethality rate, with its incidence on the rise.3 30 Early detection and intervention are crucial for reducing mortality rates. However, there is currently no non-invasive diagnostic test available for the early detection of CIP. This study is the first to identify and validate a miRNA signature for the non-invasive identification of patients with CIP. Our findings reveal that CIP exhibits a distinct circulating EV miRNA signature, which is different from those of two other comparison groups. We developed the models that integrate miRNA from serum and EVs, optimizing the sensitivity and specificity of miRNA signatures in a liquid biopsy assay for CIP diagnosis. Notably, these models distinguish CIP not only from patients without CIP undergoing immunotherapy but also from patients with infectious pneumonia. Furthermore, a risk stratification model based on three specific miRNAs effectively categorized patients with CIP into high-risk and low-risk groups, with significant differences in OS. The differentially expressed circulating miRNAs identified in this study may also serve as potential therapeutic targets in future research.
Diagnosing CIP presents a considerable challenge. Current diagnostic approaches for CIP predominantly focus on imaging, cytokine analysis in bronchoalveolar lavage fluid or blood, and routine laboratory tests.31 While some studies have been conducted to develop predictive models for CIP,32 33 established CIP-specific diagnostic models remain scarce. A study analyzing tissue immune profiles between patients with and without CIP constructed a diagnostic model achieving an AUC of 0.755.34 However, a biopsy may not be feasible for some severely ill patients, and preliminary studies have explored the correlation between EVs and irAEs. Previous studies have demonstrated that EVs miR-146a expression is diminished in grade 3–4 irAEs compared with grade 1–2 irAEs.35 36 These studies primarily focus on the severity of irAEs rather than on diagnostic markers, and they do not specifically address CIP. In this study, although the serum and EV-miRNA biomarker panels individually exhibited remarkable performance, the combined miRNA signature demonstrated superior diagnostic efficacy, with AUC values of 0.870 in the training cohort and 0.837 in the validation cohort for identifying patients with CIP. Notably, the same three miRNAs were utilized to construct miRNA models that exhibited excellent diagnostic performance in distinguishing CIP from ICI groups or lung infections. The tests presented here provide enhanced comfort, convenience, and acceptability compared with existing options. Importantly, circulating miRNA-based assays facilitate repeat sampling and offer more comprehensive disease information. Consequently, our findings underscore the diagnostic potential of circulating miRNAs as novel biomarkers for CIP.
Our previous research indicated that absolute LYM values have the potential for the detection of CIP.6 However, this indicator demonstrated poor specificity. In the current study, the final diagnostic model, with AUC values of 0.900 in the training cohort and 0.932 in the validation cohort, exhibited significantly superior diagnostic performance compared with the LYM or miRsig model alone for identifying patients with CIP from non-CIP. Early diagnosis of CIP is essential for facilitating timely treatment, which is critical for enhancing patient prognosis. For patients suspected of having CIP, non-invasive and cost-effective circulating miRNA and routine LYM testing can serve as auxiliary diagnostic tools. If an individual is identified as “high-risk” by this diagnostic model, further testing can be recommended to guide clinical management and intervention.
The distinction between patients with CIP and those with infectious pneumonia proves more complex than distinguishing CIP from patients who have undergone immunotherapy without developing CIP. Consequently, our study uniquely included patients with lung infections as control subjects. Our findings indicate that symptoms alone are insufficient to differentiate between CIP and infectious pneumonia, highlighting the urgent need for novel diagnostic biomarkers. A multimodal study integrating imaging features and hematological biomarkers demonstrated 80% accuracy, 52% sensitivity, and 90% specificity in distinguishing CIP from pneumonia.34 Another study evaluating patients with CIP vs non-COVID-19 pneumonia achieved an AUC of 0.79 (95% CI: 0.72 to 0.87) in the training cohort, with a validation cohort AUC of 0.71 (95% CI: 0.53 to 0.86).37 Our integrated miRNA biomarker signature demonstrated robust discriminative performance, achieving AUC values of 0.892 (training cohort) and 0.907 (validation cohort). When combined with LYM, the combination model showed enhanced diagnostic accuracy, with AUCs increasing to 0.938 in the training cohort and 0.959 in the validation cohort. In addition, the lung infections in this study encompassed a diverse array of types, including bacterial pneumonia, fungal pneumonia, tuberculous pneumonia, and inflammatory lung nodules, without preselection, thereby mirroring real-world clinical scenarios. These findings highlighted that miRNA signature-based models hold promising potential for diverse clinical applications.
Effective clinical management of CIP requires a combination of early diagnosis and risk-stratified interventions, and this study demonstrated that these three miRNAs can be used for both early diagnosis and stratification. The grading system has remained central to prognostication and treatment guidance for CIP.24–26 38 However, subjective CIP grading strategies have been found inadequate in providing precise information for personalized treatment. Notably, significant variations in clinical outcomes and prognosis exist among patients with CIP, even within the same CIP grade. Our prognostic stratification model offers improved stratification of patient prognosis and enhances the accuracy of survival predictions.
According to our results, our miRNA profiles demonstrate robustness in diagnostic and OS prognostic capabilities. Several studies have identified miR-193a-5p as a tumor suppressor.39 40 A previous research study has reported significantly elevated serum levels of miR-193a-5p in children with severe pneumonia.41 Furthermore, miR-193a-5p in serum has been associated with the grading and fibrosis stage of non-alcoholic fatty liver disease.42 Another study demonstrated a link between serum-derived EV miR-193a-5p and liver fibrosis, showing its role in upregulating the expression of α-smooth muscle actin, collagen 1a1, and tissue inhibitor of metalloproteinases 1 in human hepatic stellate cell lines.43 Regarding miR-378a-3p, it has been found to be upregulated in patients with asthma,44 as well as in the small intestine following intraperitoneal irradiation.45 Collectively, these findings, along with our miRNA signature, suggest potential involvement in various inflammatory processes and fibrosis progression, warranting further investigation to elucidate the specific functions of each miRNA.
Several limitations of this study are worth mentioning. First, the limited size of our two cohorts necessitates the inclusion of a larger number of participants and collaboration with additional research centers to facilitate validation of the models. Second, the PNE group exhibited a younger demographic composition and fewer smokers compared with the CIP and ICI groups. Although no significant correlation was observed between the expression levels of the three miRNAs and age or smoking status, usingg age-matched and smoking status-matched controls would be ideal for the development of CIP diagnostic strategies. Third, although these miRNAs demonstrated the ability to differentiate between CIP and PNE, as well as between CIP and ICI, it is essential to validate the applicability of our models to distinguish infectious pneumonia without concurrent CIP from CIP specifically among cancer immunotherapy patients. Such validation could potentially enhance the model’s diagnostic robustness.
Conclusions
In conclusion, based on the expression levels of serum and serum-derived EVs miRNAs, we have established and validated non-invasive miRNA-based signatures for a liquid biopsy assay for the early detection of patients with CIP. This three-miRNA signature also demonstrated the capacity to accurately predict the prognosis of patients with CIP. To enhance the diagnostic accuracy of the three-miRNA panels, further validation in a larger cohort is necessary, along with a deeper understanding of the pathophysiologic roles of these miRNAs.
HD, YY, YY and YL contributed equally.
Contributors HD, YiY, YilY, YLiao, CZ and LifW were involved in conception, and study design. HD, YiY, YilY, FW, LY, KM, JM, ZC, JW, YLiu and JS performed the experiments. HD, YilY, YLiang, LiqW, NS, WG, XL and XX collected clinical sample. SS, YX, ZW and XW were involved in acquisition of data. HD, YiY, YilY, YLiang, YLiao, CZ and LifW analyzed the clinical data and samples. HD, YiY and YilY wrote the paper and LifW revised the paper. CZ and LifW acquired funding and supervised the study. All authors read and approved the final manuscript. The guarantor of the study is LifW.
Funding This work was supported by the Science and Technology Plan Project of Guangzhou (No. 2023A04J0559), the Guangzhou Basic and Applied Basic Research Foundation (No. 2024A04J5143), the Postdoctoral Fellowship Program of China Postdoctoral Science Foundation (No. GZC20230601), the Department of Education of Guangdong Province (No. 2023ZDZX2048), the National Natural Science Foundation of China (No. 81902081), the Natural Science Foundation of Guangdong Province (Nos. 2020A1515011573, 2023A1515220167), and Open Project of Guangzhou Medical University, Guangzhou Science and Technology Fund (No. 2024A03J0791) to LifW, and the National Natural Science Foundation of China (No. 82570007) and Beijing Health Alliance Charitable Foundation (No.1-37) to CZ.
Disclaimer The study sponsors had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.
1 Postow MA, Sidlow R, Hellmann MD. Immune-Related Adverse Events Associated with Immune Checkpoint Blockade. N Engl J Med 2018;378:158–68. doi:10.1056/NEJMra1703481
2 Lin MX, Zang D, Liu CG, et al. Immune checkpoint inhibitor-related pneumonitis: research advances in prediction and management. Front Immunol 2024;15:1266850. doi:10.3389/fimmu.2024.1266850
3 Ruste V, Goldschmidt V, Laparra A, et al. The determinants of very severe immune-related adverse events associated with immune checkpoint inhibitors: A prospective study of the French REISAMIC registry. Eur J Cancer 2021;158:217–24. doi:10.1016/j.ejca.2021.08.048
4 Suresh K, Voong KR, Shankar B, et al. Pneumonitis in Non-Small Cell Lung Cancer Patients Receiving Immune Checkpoint Immunotherapy: Incidence and Risk Factors. J Thorac Oncol 2018;13:1930–9. doi:10.1016/j.jtho.2018.08.2035
5 Atchley WT, Alvarez C, Saxena-Beem S, et al. Immune Checkpoint Inhibitor-Related Pneumonitis in Lung Cancer: Real-World Incidence, Risk Factors, and Management Practices Across Six Health Care Centers in North Carolina. Chest 2021;160:731–42. doi:10.1016/j.chest.2021.02.032
6 Lin X, Deng H, Yang Y, et al. Peripheral Blood Biomarkers for Early Diagnosis, Severity, and Prognosis of Checkpoint Inhibitor-Related Pneumonitis in Patients With Lung Cancer. Front Oncol 2021;11:698832. doi:10.3389/fonc.2021.698832
7 Zhou C, Deng H, Yang Y, et al. Cancer therapy-related interstitial lung disease. Chin Med J 2025;138:264–77. doi:10.1097/CM9.0000000000003149
8 Bai X, Hu J, Betof Warner A, et al. Early Use of High-Dose Glucocorticoid for the Management of irAE Is Associated with Poorer Survival in Patients with Advanced Melanoma Treated with Anti-PD-1 Monotherapy. Clin Cancer Res 2021;27:5993–6000. doi:10.1158/1078-0432.CCR-21-1283
9 Wang YN, Lou DF, Li DY, et al. Elevated levels of IL-17A and IL-35 in plasma and bronchoalveolar lavage fluid are associated with checkpoint inhibitor pneumonitis in patients with non-small cell lung cancer. Oncol Lett 2020;20:611–22. doi:10.3892/ol.2020.11618
10 Lin X, Deng H, Chen L, et al. Clinical types of checkpoint inhibitor-related pneumonitis in lung cancer patients: a multicenter experience. Transl Lung Cancer Res 2021;10:415–29. doi:10.21037/tlcr-20-1258
11 Mohr AM, Mott JL. Overview of microRNA biology. Semin Liver Dis 2015;35:3–11. doi:10.1055/s-0034-1397344
12 Ho PTB, Clark IM, Le LTT. MicroRNA-Based Diagnosis and Therapy. Int J Mol Sci 2022;23:7167. doi:10.3390/ijms23137167
13 van Niel G, D’Angelo G, Raposo G. Shedding light on the cell biology of extracellular vesicles. Nat Rev Mol Cell Biol 2018;19:213–28. doi:10.1038/nrm.2017.125
14 Urabe F, Kosaka N, Ito K, et al. Extracellular vesicles as biomarkers and therapeutic targets for cancer. Am J Physiol Cell Physiol 2020;318:C29–39. doi:10.1152/ajpcell.00280.2019
15 Robbins PD, Morelli AE. Regulation of immune responses by extracellular vesicles. Nat Rev Immunol 2014;14:195–208. doi:10.1038/nri3622
16 Hu Q, Zhang S, Yang Y, et al. Extracellular vesicles in the pathogenesis and treatment of acute lung injury. Mil Med Res 2022;9:61. doi:10.1186/s40779-022-00417-9
17 Torasawa M, Horinouchi H, Yagishita S, et al. Exploratory analysis to predict pneumonitis during durvalumab consolidation therapy for patients with locally advanced non-small cell lung cancer from proteomic profiling of circulating extracellular vesicles. Thorac Cancer 2023;14:2909–23. doi:10.1111/1759-7714.15077
18 Ge Q, Zhou Y, Lu J, et al. miRNA in plasma exosome is stable under different storage conditions. Molecules 2014;19:1568–75. doi:10.3390/molecules19021568
19 Kumar MA, Baba SK, Sadida HQ, et al. Extracellular vesicles as tools and targets in therapy for diseases. Signal Transduct Target Ther 2024;9:27. doi:10.1038/s41392-024-01735-1
20 Gon Y, Shimizu T, Mizumura K, et al. Molecular techniques for respiratory diseases: MicroRNA and extracellular vesicles. Respirology 2020;25:149–60. doi:10.1111/resp.13756
21 Garcia-Martin R, Wang G, Brandão BB, et al. MicroRNA sequence codes for small extracellular vesicle release and cellular retention. Nature New Biol 2022;601:446–51. doi:10.1038/s41586-021-04234-3
22 San Lucas FA, Allenson K, Bernard V, et al. Minimally invasive genomic and transcriptomic profiling of visceral cancers by next-generation sequencing of circulating exosomes. Ann Oncol 2016;27:635–41. doi:10.1093/annonc/mdv604
23 Nakamura K, Zhu Z, Roy S, et al. An Exosome-based Transcriptomic Signature for Noninvasive, Early Detection of Patients With Pancreatic Ductal Adenocarcinoma: A Multicenter Cohort Study. Gastroenterology 2022;163:1252–66. doi:10.1053/j.gastro.2022.06.090
24 Puzanov I, Diab A, Abdallah K, et al. Managing toxicities associated with immune checkpoint inhibitors: consensus recommendations from the Society for Immunotherapy of Cancer (SITC) Toxicity Management Working Group. J Immunother Cancer 2017;5:95. doi:10.1186/s40425-017-0300-z
25 Brahmer JR, Lacchetti C, Schneider BJ, et al. Management of Immune-Related Adverse Events in Patients Treated With Immune Checkpoint Inhibitor Therapy: American Society of Clinical Oncology Clinical Practice Guideline. JCO 2018;36:1714–68. doi:10.1200/JCO.2017.77.6385
26 Haanen J, Obeid M, Spain L, et al. Management of toxicities from immunotherapy: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up. Ann Oncol 2022;33:1217–38. doi:10.1016/j.annonc.2022.10.001
27 Wang L, Zhu Z, Liao Y, et al. Host liver-derived extracellular vesicles deliver miR-142a-3p induces neutrophil extracellular traps via targeting WASL to block the development of Schistosoma japonicum. Mol Ther 2022;30:2092–107. doi:10.1016/j.ymthe.2022.03.016
28 Moustafa S, Burn M, Mamillapalli R, et al. Accurate diagnosis of endometriosis using serum microRNAs. Am J Obstet Gynecol 2020;223:557. doi:10.1016/j.ajog.2020.02.050
29 Li L, Zhang Q, Feng Y, et al. A Novel Serum Exosomal MicroRNA Signature in the Early Prediction of Persistent Organ Failure in Patients With Acute Pancreatitis. Ann Surg 2025;282:93–9. doi:10.1097/SLA.0000000000006229
30 Wang DY, Salem J-E, Cohen JV, et al. Fatal Toxic Effects Associated With Immune Checkpoint Inhibitors: A Systematic Review and Meta-analysis. JAMA Oncol 2018;4:1721–8. doi:10.1001/jamaoncol.2018.3923
31 Wang W, Wang Q, Xu C, et al. Chinese expert consensus on the multidisciplinary management of pneumonitis associated with immune checkpoint inhibitor. Thorac Cancer 2022;13:3420–30. doi:10.1111/1759-7714.14693
32 Zhang Y, Zhang L, Cao S, et al. A nomogram model for predicting the risk of checkpoint inhibitor‐related pneumonitis for patients with advanced non‐small‐cell lung cancer. Cancer Med 2023;12:15998–6010. doi:10.1002/cam4.6244
33 Jia X, Zhang Y, Liang T, et al. Comprehensive nomogram models for predicting checkpoint inhibitor pneumonitis. Am J Cancer Res 2023;13:2681–701.
34 Cui P, Li J, Tao H, et al. Deciphering pathogenic cellular module at single-cell resolution in checkpoint inhibitor-related pneumonitis. Oncogene 2023;42:3098–112. doi:10.1038/s41388-023-02805-4
35 Ivanova E, Asadullina D, Rakhimov R, et al. Exosomal miRNA-146a is downregulated in clear cell renal cell carcinoma patients with severe immune-related adverse events. Noncoding RNA Res 2022;7:159–63. doi:10.1016/j.ncrna.2022.06.004
36 Ivanova E, Asadullina D, Gilyazova G, et al. Exosomal MicroRNA Levels Associated with Immune Checkpoint Inhibitor Therapy in Clear Cell Renal Cell Carcinoma. Biomedicines 2023;11:801. doi:10.3390/biomedicines11030801
37 Hindocha S, Hunter B, Linton-Reid K, et al. Validated machine learning tools to distinguish immune checkpoint inhibitor, radiotherapy, COVID-19 and other infective pneumonitis. Radiother Oncol 2024;195:110266. doi:10.1016/j.radonc.2024.110266
38 Tone M, Izumo T, Awano N, et al. High mortality and poor treatment efficacy of immune checkpoint inhibitors in patients with severe grade checkpoint inhibitor pneumonitis in non-small cell lung cancer. Thorac Cancer 2019;10:2006–12. doi:10.1111/1759-7714.13187
39 Tang W, Rao Y, Pi L, et al. A review on the role of MiR-193a-5p in oncogenesis and tumor progression. Front Oncol 2025;15:1543215. doi:10.3389/fonc.2025.1543215
40 Khordadmehr M, Shahbazi R, Sadreddini S, et al. miR-193: A new weapon against cancer. J Cell Physiol 2019;234:16861–72. doi:10.1002/jcp.28368
41 Su X, Zhu L, Zhuo J, et al. Diagnostic value of miR-193a-5p in severe pneumonia and its correlation with prognosis. J Cardiothorac Surg 2025;20:87. doi:10.1186/s13019-024-03256-y
42 Johnson K, Leary PJ, Govaere O, et al. Increased serum miR-193a-5p during non-alcoholic fatty liver disease progression: Diagnostic and mechanistic relevance. JHEP Rep 2022;4:100409. doi:10.1016/j.jhepr.2021.100409
43 Wang S, Chen Y, Lei G, et al. Serum Exosome-Derived microRNA-193a-5p and miR-381-3p Regulate Adenosine 5’-Monophosphate-Activated Protein Kinase/Transforming Growth Factor Beta/Smad2/3 Signaling Pathway and Promote Fibrogenesis. Clin Transl Gastroenterol 2024;15:e00662. doi:10.14309/ctg.0000000000000662
44 Wang Q, Hong L, Chen M, et al. Targeting M2 Macrophages Alleviates Airway Inflammation and Remodeling in Asthmatic Mice via miR-378a-3p/GRB2 Pathway. Front Mol Biosci 2021;8:717969. doi:10.3389/fmolb.2021.717969
45 Chu X, Zheng W, Wang J, et al. CDK6 inhibition targeted by miR-378a-3p protects against intestinal injury induced by ionizing radiation. Biochem Biophys Res Commun 2020;531:328–34. doi:10.1016/j.bbrc.2020.07.093
© Author(s) (or their employer(s)) 2025. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ Group.. This work is licensed under the Creative Commons Attribution – Non-Commercial License http://creativecommons.org/licenses/by-nc/4.0/ (the "License"). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.