Metabolite‐Based Endotypes of Asthma Reveal Distinct Clinical Characteristics and Immune Cell Signatures

ABSTRACT

Background

Asthma control is commonly defined by symptom burden and recent exacerbation history. However, symptom‐based clinical stability does not necessarily reflect biological quiescence, and heterogeneity in lung function and airway structure may persist despite apparent clinical control. This study aimed to determine whether blood‐based metabolomic profiling can discriminate biologically distinct subgroups within a clinically stable asthma population.

Methods

We conducted a prospective observational study of adults with clinically stable asthma, defined by sustained symptom control and absence of recent exacerbations under maintenance inhaled corticosteroid‐based therapy. Untargeted plasma metabolomic profiling was performed using liquid chromatography–tandem mass spectrometry, and metabolite‐derived subgroups were identified by consensus clustering. Lung function, airway structure, small‐airway physiology, and peripheral immune cell profiles were compared across clusters, with associations assessed using regression analyses.

Results

Three metabolite‐derived subgroups were identified based on patterns in the relative abundance of selected representative metabolites. Symptom control and exacerbation history were comparable across clusters, whereas lung function, airway wall thickness, small‐airway physiology, and immune cell profiles differed substantially. Cluster 1 (C1; Remodeling‐prone), characterized by greater relative abundance of glycerophospholipid‐related metabolites, showed lower post‐bronchodilator FEV₁, thicker airway walls, and increased innate lymphoid cells. Cluster 2 (C2; Biologically stable) had preserved airway structure, with thinner airway walls and higher post‐bronchodilator FEV₁. Cluster 3 (C3; T2‐high) exhibited metabolite patterns associated with type 2 inflammatory features, including higher FeNO levels and blood eosinophil counts, while maintaining preserved airway structure.

Conclusions

Metabolomic profiling reveals biologically distinct subgroups among clinically stable asthma, indicating that symptom‐based stability does not necessarily reflect biological homogeneity.

This study determines whether blood‐based metabolomic profiling can discriminate biologically distinct subgroups within a clinically stable asthma population. Blood‐based metabolomic profiling identified three biologically distinct subgroups among adults with clinically stable asthma: Remodeling‐prone, biologically stable, and T2‐high, differing in lung function, airway structure, and immune profiles despite comparable symptom control. These findings demonstrate that clinical stability does not reflect biological homogeneity. FeNO, fractional exhaled nitric oxide; ICS, inhaled corticosteroid; IL, interleukin; ILC, innate lymphoid cell.

Abbreviations

ACT

Asthma Control Test

AUC

area under the curve

AWT Pi10

square‐root wall thickness standardized to an internal perimeter of 10 mm

BEC

blood eosinophil count

CD

cluster of differentiation

CI

confidence interval

CT

computed tomography

DEM

differentially expressed metabolite

EDTA

ethylenediaminetetraacetic acid

FeNO

fractional exhaled nitric oxide

FEV₁

forced expiratory volume in 1 second

FVC

forced vital capacity

GINA

Global Initiative for Asthma

ICS

inhaled corticosteroid

IFN‐γ

interferon gamma

IgE

immunoglobulin E

IL

interleukin

ILC

innate lymphoid cell

LC–MS/MS

liquid chromatography–tandem mass spectrometry

PCA

principal component analysis

PLA2

phospholipase A₂

ROC

receiver operating characteristic

VIP

variable importance in projection

WA

wall‐area percentage

1. Introduction

Asthma is a heterogeneous airway disease that is managed based on symptom burden and exacerbation history [1]. However, symptom‐based control does not accurately reflect the underlying biological activity of the disease [2]. Longitudinal studies demonstrated that even patients with well‐controlled asthma can experience progressive lung function impairment or structural airway changes, suggesting that clinical control reflects a phenotypic state rather than biological quiescence [3, 4, 5]. To address this discrepancy, recent studies have focused on inflammatory endotypes, particularly those driven by type 2 immune pathways [6]. While type 2 endotyping has enabled the development of targeted biologics, it does not fully account for variations in airway structure, small airway dysfunction, or long‐term progression [7, 8].

Peripheral blood transcriptomic clustering has identified distinct asthma subgroups with divergent immune endotypes, uncovering biological heterogeneity that is not apparent from clinical assessment alone [9]. Building upon this, circulating metabolites represent integrated downstream products of environmental exposure, immune activation, and tissue remodeling, and therefore provide a systems‐level view of disease biology [10].

Metabolomic profiling has shown associations with asthma severity, lung‐function impairment, and inflammatory phenotypes [11, 12, 13]. However, it remains unclear whether biologically distinct metabolic signatures persist in clinically stable patients irrespective of the symptom burden, challenging the assumption that clinical stability reflects biological stability [14]. Therefore, this study used blood‐based metabolomic profiling to identify distinct subgroups among patients with clinically stable asthma and evaluate their associations with lung function, airway structure, small‐airway physiology, and immune cell profiles.

2. Methods

This study adhered to the principles outlined in the STROBE statement, with additional alignment to the STROBE‐ME extension to maintain methodological clarity and comprehensive reporting of biospecimen‐based components [15].

2.1. Study Design and Participants

This study was conducted as a prospective observational investigation at a university‐affiliated teaching hospital from July 2023 to August 2024. Participants with clinician‐diagnosed asthma were identified according to the Global Initiative for Asthma (GINA) recommendations based on variable respiratory symptoms and objective evidence of reversible airflow limitation. Eligible participants were required to have used the same inhaled corticosteroid (ICS)‐based regimen for at least 3 months before screening, adherence greater than 80% based on medication possession ratio, and well‐controlled asthma defined by an Asthma Control Test (ACT) score of ≥ 20 without a history of exacerbation during the 4 weeks before screening. None of the participants were receiving biologic therapies (e.g., anti‐IgE, anti‐IL‐5/IL‐5Rα, or anti‐IL‐4Rα agents) during the study period. Exclusion criteria included other chronic respiratory diseases (e.g., COPD, bronchiectasis), rare genetic airway disorders (e.g., alpha‐1‐antitrypsin deficiency and primary ciliary dyskinesia), and systemic conditions associated with marked eosinophilia (e.g., ABPA, EGPA). Additionally, patients with significant systemic comorbidities or those receiving medications that modify inflammatory or metabolic pathways (e.g., systemic corticosteroids, immunosuppressive agents, antidiabetic therapies, or lipid‐lowering medications) were excluded to minimize confounding.

2.2. Study Setting and Procedures

Participants underwent a three‐month screening process to confirm clinical stability under consistent ICS therapy (Figure 1). Following reassessments of symptom control and adherence, baseline evaluations including spirometry, impulse oscillometry (IOS), chest computed tomography (CT), and blood sampling were performed at month 3. To ensure an unbiased characterization, participants were not pre‐stratified by asthma phenotype or endotype prior to metabolomic analysis.

FIGURE 1.

Study design.

2.3. Data Collection

Demographic and clinical variables were collected at the baseline visit through a structured clinical assessment. Asthma severity was assessed at the three‐month enrollment visit to allow sufficient time to determine the treatment intensity required for disease control, thereby reflecting both intrinsic disease activity and therapeutic responsiveness.

Standardized spirometry (Vmax 22) and IOS (MasterScreen) were performed following international guidelines. Forced expiratory volume in one second (FEV₁), forced vital capacity (FVC), and the FEV₁/FVC ratio were obtained, and reference values and z‐scores were derived from the 2022 Global Lung Function Initiative equations. Impulse oscillometry (MasterScreen IOS, CareFusion, Germany) was used to assess airway resistance and reactance, including R5, R20, R5–R20, X5, resonant frequency, and the reactance area.

High‐resolution chest computed tomography was conducted to evaluate airway structure. Quantitative CT parameters (AWT Pi10, WA%) were analyzed using Aview software (Coreline Soft, Seoul, Korea) and included the square‐root wall thickness standardized to an internal perimeter of ten millimeters (AWT Pi10) and the airway wall‐area percentage (WA). Fractional exhaled nitric oxide (FeNO) was measured using a chemiluminescence analyzer (NIOX VERO, Circassia, Sweden).

Peripheral venous blood samples were collected at the baseline visit under non‐fasting conditions as part of routine outpatient assessment. Specific types of dietary intake were not recorded. Laboratory measurements included complete blood cell counts with differential profiles, total immunoglobulin E (IgE), serum chemistry testing, and C‐reactive protein levels.

2.4. Immune Cell Profiling

Peripheral blood mononuclear cells were isolated from ethylenediaminetetraacetic acid–treated blood by density‐gradient centrifugation. Flow‐cytometric analyses were performed to characterize adaptive and innate immune populations (Figure S1). Each subset was quantified as both proportions of CD45⁺ leukocytes and absolute counts calculated from complete blood cell counts. Detailed experimental procedures, including antibody information and gating strategies, are provided in Supporting Information.

2.5. Metabolite Profiling

Plasma metabolomic profiling was performed using untargeted LC–MS/MS, followed by data preprocessing, metabolite annotation, normalization, and differential metabolite analyses. Detailed experimental procedures and analytical parameters are provided in Supporting Information.

2.6. Consensus Clustering

Consensus clustering was performed to derive metabolite‐based subgroups using a resampling‐based clustering framework [16]. Metabolite concentrations were scaled and centered before analysis. Clustering was conducted using the R package “ConsensusClusterPlus” with resampling‐based consensus procedures.

For each candidate solution of two or three clusters, 1000 resampling iterations were performed. In each iteration, 80% of participants and 80% of metabolites were randomly sampled without replacement, and clustering was carried out using a partitioning‐around‐medoids algorithm with a Euclidean distance metric.

Consensus matrices were generated to summarize the probability that pairs of samples were assigned to the same cluster across resampled datasets, thereby providing a measure of cluster stability and reproducibility. The optimal number of clusters was determined by examining the cumulative distribution function of consensus values, the relative change in the area under the distribution curve, and the stability of the consensus matrix across candidate solutions. Each sample was assigned to the cluster with the highest consensus membership probability.

Principal component analysis (PCA) was used to visualize the separation of metabolite‐defined subgroups, and clinical, physiologic, imaging, immunologic, and metabolic characteristics were compared across clusters to evaluate biological plausibility and phenotypic distinctions.

2.7. Statistics

Baseline characteristics were summarized using descriptive statistics. Continuous variables are presented as means with standard deviations or medians with interquartile ranges, as appropriate. Comparisons across metabolite‐derived clusters were performed using one‐way analysis of variance for normally distributed variables and the Kruskal–Wallis test for non‐normally distributed variables. Categorical variables were compared using the chi‐squared test or Fisher's exact test, as appropriate. For metabolomic analyses, individual metabolite concentrations were standardized using log transformation and Pareto scaling prior to clustering analyses. Associations between DEMs and clinical variables including spirometric indices, impulse oscillometry measures, radiologic airway‐wall parameters on chest computed tomography (CT), and inflammatory markers were examined using univariable linear regression models. Regression coefficients were examined to indicate the direction and magnitude of associations.

Receiver operating characteristic analyses were performed to assess separation of each metabolite derived cluster from the remaining clusters based on the selected metabolite signatures. All analyses were performed using R software version 4.5.0 and GraphPad Prism version 10.0.3. A two‐sided p < 0.05 was considered statistically significant.

2.8. Ethics

The study adhered to the ethical principles of the Declaration of Helsinki and met all applicable regulatory standards. The study was approved by the Institutional Review Boards of Seoul Metropolitan Government–Seoul National University Boramae Medical Center (IRB No. 30‐2023‐14) and the Seoul National University Institutional Review Board (IRB No. E2308/003‐009). Written informed consent was obtained from all participants.

3. Results

3.1. Identification of Metabolite‐Derived Clusters

Untargeted LC–MS/MS analysis of plasma from 56 clinically stable asthma patients identified 2045 normalized metabolic features. Consensus clustering of the processed metabolite dataset identified three metabolite‐derived clusters, designated C1 (n = 22), C2 (n = 21), and C3 (n = 13). The consensus matrix and PCA showed clear separation, indicating high within‐cluster stability and distinct metabolic profiles across subgroups (Figure 2A,B).

FIGURE 2.

Metabolomic clustering of asthma patients. (A) Consensus matrix for the three‐cluster solution (K = 3) showing high within‐cluster consensus and clear separation between clusters. (B) Principal component analysis plot demonstrating partial separation of metabolite‐defined clusters in reduced dimensional space. PC, principal component; PCA, principal component analysis.

3.2. Baseline Characteristics Across the Clusters

Symptom control and exacerbation history were comparable across the clusters (Table 1). C1 was characterized by airflow limitation and structural changes, with the lowest post‐bronchodilator FEV₁ and a higher prevalence of FEV₁/FVC < 0.7 (Remodeling‐prone cluster). This cluster also showed greater airway wall thickness on CT compared to C2, along with numerically higher small airway dysfunction indices (R5–R20, Fres, and AX), though these did not reach statistical significance. In contrast, C2 presented the most preserved physiological and structural profile, with the highest FEV₁ and minimal airway wall thickening (Biologically stable cluster). C3 was distinguished by prominent type 2 inflammatory signatures, including the highest FeNO levels and blood eosinophil counts, and a higher proportion of severe asthma (T2‐high cluster). Lung function and structural parameters in C3 remained relatively preserved compared to C1, despite this inflammatory profile.

TABLE 1.

Baseline characteristics according to metabolite‐derived clusters.

	C1 (n = 22)	C2 (n = 21)	C3 (n = 13)	p
Age, mean (SD)	62.91 (15.19)	61.81 (22.00)	46.25 (15.53)	0.031
Age > 65, n (%)	10 (45.5)	12 (57.1)	1 (8.3)	0.022
Male, n (%)	11 (50.0)	12 (57.1)	6 (50.0)	0.876
BMI, mean (SD)	23.83 (4.12)	23.61 (4.00)	25.65 (3.50)	0.329
Smoking status				0.923
Never smoker, n (%)	10 (45.5)	9 (42.9)	5 (41.7)
Ex‐smoker, n (%)	7 (31.8)	6 (28.6)	5 (41.7)
Current smoker, n (%)	5 (22.7)	6 (28.6)	2 (16.7)
Pack years among ever smokers
> 10, n (%)	11 (50.0)	10 (47.6)	6 (50.0)	0.985
Respiratory symptoms
ACT, mean (SD)	22.50 (1.74)	22.52 (1.83)	22.00 (1.86)	0.686
Chronic sputum production, n (%)	22 (100.0)	16 (76.2)	9 (75.0)	0.044
Previous AE history
Severe, n (%)	7 (31.8)	3 (14.3)	2 (16.7)	0.337
Comorbidities
CCI ≥ 1, n (%)	9 (40.9)	5 (23.8)	1 (8.3)	0.113
Chronic rhinitis, n (%)	18 (81.8)	13 (61.9)	8 (66.7)	0.333
Chronic sinusitis, n (%)	8 (36.4)	5 (23.8)	3 (25.0)	0.623
Nasal polyp, n (%)	5 (22.7)	2 (9.5)	1 (8.3)	0.371
Asthma severity				0.030
Mild, n (%)	9 (40.9)	11 (52.4)	2 (16.7)
Moderate, n (%)	9 (40.9)	10 (47.6)	5 (41.7)
Severe, n (%)	4 (18.2)	0 (0.0)	5 (41.7)
Treatment				0.715
Inhaled therapy
ICS, n (%)	1 (4.5)	0 (0.0)	0 (0.0)
ICS/LABA, n (%)	18 (81.8)	17 (81.0)	9 (75.0)
ICS/LABA/LAMA, n (%)	3 (13.6)	4 (19.0)	3 (25.0)
Oral therapy
PDE‐4 inhibitor, n (%)	2 (9.1)	0 (0.0)	0 (0.0)	0.211
LTRA, n (%)	13 (59.1)	6 (28.6)	5 (41.7)	0.129
Spirometry
PostBDR FEV1, L, median (IQR)	1.94 [1.31, 2.53]	2.42 [1.64, 2.95]	2.41 [2.24, 2.79]	0.065
PostBDR FEV1 (%), median (IQR)	88.00 [68.00, 110.00]	98.00 [91.00, 111.00]	88.00 [87.00, 95.75]	0.051
PostBDR FEV1 < 80%, n (%)	10 (45.5)	2 (9.5)	2 (16.7)	0.019
PostBDR FVC, L, median (IQR)	3.32 [2.13, 3.78]	3.38 [2.85, 3.89]	3.11 [2.99, 3.54]	0.463
PostBDR FVC (%), median (IQR)	95.50 [79.25, 109.50]	98.00 [87.00, 108.00]	90.50 [81.25, 97.50]	0.400
PostBDR FEV/FVC (%), median (IQR)	67.61 [55.21, 71.93]	74.59 [59.93, 80.13]	77.19 [73.34, 82.02]	0.029
PostBDR FEV/FVC < 0.7, n (%)	14 (63.6)	8 (38.1)	2 (16.7)	0.025
Impulse oscillometry
R5, kPa/(L/s), mean (SD)	0.48 (0.20)	0.38 (0.11)	0.40 (0.11)	0.079
R20, kPa/(L/s), mean (SD)	0.30 (0.07)	0.27 (0.06)	0.28 (0.05)	0.199
R5–R20, kPa/(L/s), mean (SD)	0.18 (0.15)	0.11 (0.07)	0.12 (0.08)	0.141
R5–R20 > 0.10 kPa/(L/s), n (%)	16 (72.7)	10 (47.6)	6 (50.0)	0.201
X5, kPa/(L/s), mean (SD)	−0.18 (0.11)	−0.14 (0.07)	−0.17 (0.09)	0.285
Fres, Hz, mean (SD)	20.58 (6.20)	16.93 (4.53)	16.95 (6.41)	0.075
AX, kPa/L, mean (SD)	1.81 (1.63)	0.96 (0.69)	1.18 (1.09)	0.076
Chest CT parameters
AWT Pi10, mm	4.63 [4.35, 5.16]	3.84 [3.71, 4.54]	4.58 [4.00, 4.97]	0.005
Wall area, %	68.16 [64.95, 73.02]	59.40 [57.27, 66.59]	69.65 [59.88, 71.60]	0.004
FeNO, ppb, median (IQR)	17.50 [12.25, 45.25]	39.00 [18.00, 70.00]	62.00 [31.50, 110.50]	0.023
FeNO > 25 ppb, n (%)	8 (36.4)	13 (61.9)	9 (81.8)	0.035
Blood tests
WBC, /μL, median (IQR)	6955 [5245, 8370]	6530 [5350, 7370]	7375 [5210, 8178]	0.834
Neutrophil, (%), median (IQR)	54.65 [52.33, 60.70]	53.00 [48.50, 64.30]	58.45 [46.73, 61.80]	0.841
Lymphocyte, (%), median (IQR)	30.65 [21.92, 34.75]	30.00 [23.90, 36.80]	28.70 [26.18, 34.97]	0.995
Eosinophil, (%), median (IQR)	2.15 [1.00, 3.92]	4.30 [1.70, 6.00]	5.45 [3.58, 8.45]	0.040
Eosinophil, /μL, median (IQR)	143.04 [52.18, 267.16]	297.36 [140.03, 352.80]	403.45 [222.24, 596.07]	0.046
Eosinophil > 300/μL, n (%)	5 (22.7)	10 (47.6)	7 (58.3)	0.085
Eosinophil < 100/μL, n (%)	9 (40.9)	5 (23.8)	1 (8.3)	0.113
IgE, IU/mL, median (IQR)	143.00 [47.42, 820.00]	125.00 [60.00, 452.00]	317.00 [43.50, 460.00]	0.905
Protein, g/dL, mean (SD)	6.65 (1.48)	6.87 (1.43)	6.58 (2.01)	0.844
Albumin, g/dL, mean (SD)	4.49 (0.35)	4.54 (0.26)	4.65 (0.25)	0.333
BUN, mg/dL, mean (SD)	15.45 (6.78)	16.67 (8.08)	18.17 (23.73)	0.837
Creatinine, mg/dL, mean (SD)	0.87 (0.23)	0.90 (0.25)	0.83 (0.16)	0.712
CRP, mg/dL, median (IQR)	0.11 [0.03, 0.56]	0.19 [0.08, 0.58]	0.13 [0.07, 0.28]	0.644

Note: Values are presented as median (IQR) or number (%) for categorical variables.

Abbreviations: ACT, Asthma Control Test; AE, exacerbation; AWT Pi10, square‐root–transformed airway wall thickness standardized to an internal perimeter of 10 mm; AX, area of reactance; BMI, body mass index; BUN, blood urea nitrogen; CCI, Charlson Comorbidity Index; CRP, C‐reactive protein; FeNO, fractional exhaled nitric oxide; FEV₁, forced expiratory volume in 1 second; Fres, resonant frequency; FVC, forced vital capacity; ICS, inhaled corticosteroid; LABA, long‐acting β₂‐agonist; LAMA, long‐acting muscarinic antagonist; LTRA, leukotriene receptor antagonist; PDE‐4, phosphodiesterase‐4; PostBDR, post‐bronchodilator; R5, resistance at 5 Hz; R20, resistance at 20 Hz; WBC, white blood cell count; X5, reactance at 5 Hz.

3.3. Differential Metabolomic Features and Cluster Annotation

Pairwise comparisons of metabolite levels revealed distinct and class‐specific metabolic differences across clusters (Figure 3A–D). C3 was distinguished by a higher abundance of metabolites classified as carboxylic acids and derivatives, which accounted for the majority of differential features. Fatty acyls, benzene, and substituted derivatives, along with organonitrogen compounds, also showed a predominance of higher levels in C3. In contrast, C1 was characterized by an enrichment of glycerophospholipids and steroids relative to C2.

FIGURE 3.

Differentially expressed metabolites across metabolomic clusters. (A–C) Volcano plots show metabolite differences for each pairwise cluster comparison. Upregulated (red) and downregulated (blue) differentially expressed metabolites are shown. (D) Distribution of differentially expressed metabolites by metabolite class for each pairwise cluster comparison.

DEMs identified in each pairwise cluster comparison are summarized in Tables S1–S3. From these, eleven representative metabolites showing significant associations with clinical, spirometric, or inflammatory parameters were selected (Figure 4). Higher levels of glycerophospholipids were associated with more pronounced small airway dysfunction, increased airway wall thickness, and lower post‐bronchodilator FEV₁. Conversely, carboxylic acids and benzene derivatives showed opposite associations, with higher levels linked to less small airway dysfunction and higher FEV₁, a pattern also observed for fatty acyl metabolites. Hydroxypropionylcarnitine was an exception, correlating with worse small airway dysfunction, diminished FEV₁, and reduced FeNO and blood eosinophil levels. Carboxylic acids and benzene‐related metabolites were positively associated with type 2 inflammatory markers, including FeNO.

FIGURE 4.

Associations of circulating metabolites with clinical, spirometric, and inflammatory parameters. AWT, airway wall thickness; FEV₁, forced expiratory volume in 1 second; FeNO, fractional exhaled nitric oxide; FVC, forced vital capacity; IgE, immunoglobulin E; Pi10, square‐root–transformed airway wall thickness standardized to an internal perimeter of 10 mm; R5–R20, difference between resistance at 5 and 20 Hz.

As shown in Figure S2, each cluster had distinct patterns in the relative abundance of the eleven selected representative metabolites. Based on these metabolite‐associated patterns, clusters were descriptively characterized as follows. C1 showed relative predominance of glycerophospholipid‐related metabolites and was therefore designated as a glycerophospholipid‐associated metabolic pattern. C2 showed no clear predominance of a specific metabolite class and was characterized by intermediate metabolite distributions, consistent with a metabolically intermediate pattern. In contrast, C3 showed relatively elevated levels across multiple metabolite classes without a single predominant metabolic feature and was designated as a multi‐class–elevated metabolic pattern. Receiver operating characteristic (ROC) analyses based on the selected metabolite signatures are shown in Figure S3. The metabolite panel showed differential separation of C1 and C2 from the remaining clusters, with areas under the curve (AUC) of 0.937 (95% CI, 0.867–1.000) and 0.944 (95% CI, 0.886–0.993), respectively. In contrast, separation of C3 from the other clusters was modest (AUC, 0.766; 95% CI, 0.506–0.932).

3.4. Immune Cell Profiles and Its Association With DEMs

ILC profiles demonstrated significant cluster‐specific differences, with the highest proportion and absolute count of total ILCs observed in C1 and the lowest in C2 (p = 0.004 and p = 0.023, respectively) (Table S4). However, the relative distribution of functional ILC subsets, including IFN‐γ⁺, IL‐5⁺, IL‐13⁺, and IL‐17A⁺ ILCs, as well as ILC1/ILC2 and ILC3/ILC2 ratios did not differ across clusters.

Adaptive immune profiling revealed that C2 had a lower CD4⁺ T cell proportion and a higher CD8+ T cell proportion, resulting in a reduced CD4/CD8 ratio (p = 0.002) compared to C1 (Table S5). C1 showed the highest CD4/CD8 ratio. B cells and cytokine‐producing CD4⁺ T cell subsets, including IFN‐γ⁺, IL‐5⁺, IL‐13⁺, and IL‐17A⁺ T cells, did not differ among clusters.

Associations between the DEMs and immune cells showed that glycerophospholipid metabolites were associated with higher total ILCs and IL‐5⁺ ILCs and with lower proportions of IL‐13⁺ CD4⁺ T cells (Figure 5). In contrast, fatty acyl metabolites showed opposite or heterogeneous patterns. L‐hexanoylcarnitine and oleamide were associated with lower IL‐5⁺ ILCs, and oleamide was additionally associated with lower CD4⁺ T cell proportions and higher IL‐13⁺ CD4⁺ T and IL‐17⁺ CD4⁺ T cells. Metabolites classified as carboxylic acids, benzene derivatives, and phenols were generally associated with lower IL‐5⁺ ILCs and CD4⁺ T cells. By contrast, the phenol class metabolite showed positive associations with IL‐13⁺ CD4⁺ T cells.

FIGURE 5.

Immune cell profiles and 11 differentially expressed metabolites correlation. Color intensity represents the slope coefficient derived from simple linear regression, with red and blue denoting positive and negative associations, respectively. Statistical significance was determined using two‐tailed Spearman correlation analysis and is indicated by asterisks (*p < 0.05) Abbreviations: CD, cluster of differentiation; IFN‐γ, interferon gamma; IL, interleukin; ILC, innate lymphoid cell.

4. Discussion

In this prospective study of patients with clinically stable asthma, we identified that symptom‐based clinical stability does not necessarily correspond to biological homogeneity. Using untargeted plasma metabolomics, we identified three distinct metabolite‐derived clusters with distinct physiological, radiological, and immunological signatures. The remodeling‐prone profile of C1, characterized by glycerophospholipid predominance, correlates with impaired FEV₁ and structural changes, with lower type 2 inflammatory markers suggesting a non–type 2 endotype potentially encompassing neutrophilic or paucigranulocytic asthma profiles. C2 demonstrated relatively preserved airway structure and physiology, consistent with a more stable biological profile. C3 exhibited metabolite patterns associated with type 2 inflammation, including elevated FeNO levels and BECs, suggesting that this subgroup may reflect a type 2 inflammatory endotype even among patients classified as clinically stable.

The central contribution of this study lies not in the identification of novel metabolites per se, but in the repositioning of clinical stability as a descriptor of observed clinical control rather than an indicator of biological inactivity. Current asthma management frameworks implicitly treat sustained symptom control and absence of recent exacerbations as indicators of low biological risk. Our data indicate that, even under uniform maintenance therapy and stable clinical presentation, metabolically distinct biological states persist, some of which are associated with airway remodeling and impaired lung function. This conceptual reframing provides a foundation for understanding why structural progression and lung function decline may occur despite apparent clinical control. Accordingly, the present analysis was designed as an exploratory metabolomic profiling study aimed at identifying biologically distinct subgroups within clinically stable asthma rather than developing a predictive classification model.

C1 was uniquely associated with remodeling‐prone features, including lower post‐bronchodilator FEV₁ and increased airway wall thickness, and higher circulating ILC abundance. In addition, phospholipid‐derived mediators participate in innate immune signaling pathways [17], and ILC2s have been implicated in airway remodeling through cytokine‐mediated inflammatory responses [18]. Given that glycerophospholipid metabolism provides substrates for bioactive lipid mediators including lysophospholipids and eicosanoids [19], perturbations in lipid metabolism may contribute to structural airway changes even in clinically stable patients [20]. In our analysis, glycerophospholipid‐related metabolites were associated with higher proportions of IL‐5⁺ ILCs, supporting a potential link between lipid metabolism and innate immune activation in this cluster. However, IL‐13⁺ T cells were not increased, indicating that the observed immune pattern may reflect selective innate immune activation rather than a generalized type 2 cytokine response.

In contrast, C2 serves as a biological reference for stability, characterized by minimal innate immune activity and preserved structure [21, 22, 23]. The lower abundance of ILCs suggests reduced innate immune activity, in keeping with the preservation of airway structure and lung function observed in this cluster [24, 25]. These findings indicate that distinct metabolite‐associated profiles may correspond to differences in underlying biological activity despite similar clinical stability. This subgroup provides a biological reference within the clinically stable population, suggesting that biological stability is present but heterogeneously distributed among patients classified as clinically stable.

C3, characterized by distinct patterns in selected metabolites previously linked to inflammatory and environmental exposure pathways, exhibited elevated FeNO levels and BECs, consistent with a type 2 inflammatory profile despite clinical stability. Recent studies have shown that metabolite signatures related to aromatic and phenolic compounds are associated with eosinophilic airway inflammation in asthma [26]. Despite a higher proportion of severe asthma in this cluster, lung function remained relatively preserved, and quantitative CT indices showed limited structural airway alterations. This dissociation between inflammatory activity and structural airway changes suggests that inflammatory activation may occur independently of measurable airway remodeling in clinically stable patients. The observed metabolite patterns may reflect differences in inflammatory regulation or environmental exposure, although causal relationships cannot be established within the present study design.

Metabolomic patterns were associated more strongly with innate ILC profiles than with adaptive T cells. Specifically, C1, characterized by relative predominance of selected glycerophospholipid‐related metabolites, showed higher proportions of total ILCs and increased frequencies of IL‐5⁺ ILCs. This aligns with the distinct metabolic and phospholipid requirements of ILC2s [27]. In contrast, C2 and C3, which demonstrated different metabolite‐associated profiles, showed lower circulating ILC proportions. Although differences in adaptive immune cell subsets were observed, these were less consistently aligned with metabolite‐associated patterns. The findings suggest that circulating metabolite patterns more closely reflect variations in innate immune activity than adaptive immune responses in clinically stable patients.

A notable finding in our study was the discordance between metabolite‐associated immune signatures and conventional type 2 markers. C3 showed higher FeNO levels and BECs but did not present with a clear increase in IL‐5⁺ or IL‐13⁺ CD4⁺ T cells. Similarly, C1 showed relatively lower BECs and FeNO levels but was associated with higher proportions of IL‐5⁺ ILCs. The apparent dissociation across clusters likely reflects the biological reality of clinically stable asthma patients under maintenance therapy. In C3, the absence of elevated IL‐5⁺ or IL‐13⁺ lymphoid cells, despite its metabolic association with carboxylic acids and benzene, is most plausibly explained by ICS‐mediated suppression of cytokine production [28, 29], while the existing eosinophil pool persists due to its longer half‐life, creating a temporal dissociation. In C1, the enlarged ILC pool without specific subset enrichment suggests that innate immune activation may preferentially drive structural remodeling rather than eosinophilic inflammation. Furthermore, the glycerophospholipid‐rich environment may functionally restrain ILC2 cytokine output through PGI₂ and LXA₄ signaling, further decoupling ILC numbers from downstream eosinophilopoiesis [30]. These findings suggest that metabolomic profiling may capture stable endotypic signatures that persist despite therapeutic modulation of downstream inflammatory outputs.

In addition, regulatory immune pathways were not assessed in the present study. Evaluation of regulatory markers, such as IL‐10 production in T cells and ILCs and FOXP3 expression in T cells, may provide further insight into the immunological heterogeneity observed across metabolite‐defined clusters. Previous human studies have shown that reduced FOXP3 expression in T cells is associated with impaired regulatory immune responses in asthma [31] and that IL‐10‐producing regulatory ILC can attenuate T helper responses in human allergic disease [32], supporting a potential role for regulatory immune pathways in shaping the immunological heterogeneity observed across metabolite‐defined clusters. Future studies incorporating these parameters may help to better delineate the balance between inflammatory and regulatory immune responses in clinically stable asthma.

Several limitations should be acknowledged. First, the cross‐sectional design of this study precludes determination of temporal relationships between metabolite‐associated patterns and airway remodeling. Although metabolomic profiles may show relative stability at the endotype level, individual metabolic features can vary over the disease course [33]. Second, the modest sample size limits the precision of subgroup‐specific estimates. The limited size potentially affects the statistical stability of subgroup comparisons and the robustness of cluster‐specific estimates. As this study was designed as an exploratory study rather than developing a predictive classification model, a larger, independent cohort for external validation will be needed to determine whether the identified metabolite signatures can serve as reproducible biomarkers. Third, plasma samples were used, which may not fully reflect compartment‐specific airway biology. Induced sputum analysis, which is the reference method for airway inflammatory phenotyping, was not available for this study. Fourth, residual confounding by unmeasured environmental, lifestyle, and microbiome‐related factors cannot be excluded. Samples were collected under non‐fasting conditions without dietary records, which may introduce postprandial metabolic variability. Importantly, smoking status and cumulative smoking exposure were evenly distributed across the identified clusters, suggesting that differences in tobacco exposure were unlikely to account for the observed metabolomic clustering patterns. Nevertheless, other environmental factors not captured in this study may still contribute to metabolic heterogeneity. Finally, the identification of monocyte subsets was based on a gating strategy that did not include CD56 in the lineage exclusion panel. Therefore, a small degree of contamination by NK cells, CD56⁺CD16⁺ subset cannot be completely excluded. This may have influenced the population of non‐classical monocytes.

Despite these limitations, this study identifies that clinically stable asthma encompasses biologically divergent states with distinct structural and immunologic profiles. From a clinical perspective, these findings suggest that reliance on symptom control alone may underestimate future risk in a subset of patients. Metabolomic profiling, in conjunction with imaging and immune markers, may offer a complementary approach for identifying patients with ongoing biological activity despite apparent clinical stability and may provide a basis for future biomarker development aimed at stratifying inflammatory endotypes and treatment responsiveness.

In conclusion, among patients with clinically stable asthma, metabolomic profiling identified distinct subgroups with similar symptom control but differing lung function, airway structure, and immune cell profiles. Biological heterogeneity persists in clinically stable asthma, necessitating metabolomic profiling to identify underlying risks.

Author Contributions

Y.J.P., N.K., and H.W.L. conceived and designed the study. Y.J.P. and H.W.L. acquired the clinical samples and immune cell data. Y.J.P., N.K., J.C., D.‐J.P., K.Y., D.Y.L., H.W.L., and C.‐H.Y. performed the data analysis and contributed to the interpretation of the findings. Y.J.P., N.K., and H.W.L. drafted the manuscript. D.K.K., H.W.L., and C.‐H.Y. critically revised the manuscript for important intellectual content. H.W.L. and C.‐H.Y. supervised the study. All authors read and approved the final manuscript.

Funding

This work was supported by the Research Grant from Seoul National University (0525‐20240127), the Ministry of Science and ICT, Republic of Korea (RS‐2023‐00218476, RS‐2024‐00454619), Ministry of Health & Welfare, Republic of Korea (RS‐2024‐00507606), and the BK21 FOUR Program of the Department of Agricultural Biotechnology, Seoul National University.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Pairwise comparisons of differentially abundant metabolites between Cluster1 and Cluster2.

Table S2: Pairwise comparisons of differentially abundant metabolites between Cluster1 and Cluster3.

Table S3: Pairwise comparisons of differentially abundant metabolites between Cluster 2 and Cluster 3.

Table S4: Innate immune cell profiles according to metabolite‐derived clusters.

Table S5: Adaptive immune cell profiles according to metabolite‐derived clusters.

Figure S1: Flow cytometric gating strategy for immune cells in human PBMC.

Figure S2: Eleven differentially expressed metabolites.

Figure S3: ROC analysis of cluster discrimination using representative differentially expressed metabolites.

Data Availability Statement

The data supporting this study's findings are available upon reasonable request from the corresponding author. Restrictions may apply due to ethical considerations.