Multiomics and Machine Learning Reveal Distinct Immune-Metabolic Signatures and Diagnostic Biomarkers for Asthma Inflammatory Endotypes

Abstract

Background: bronchial asthma is a highly heterogeneous chronic inflammatory disease, with eosinophilic (EA) and neutrophilic (NEA) asthma subtypes exhibiting distinct pathological mechanisms and treatment responses. However, the protein-metabolite interactions underlying the pathogenesis of asthma remain poorly understood, limiting the development of precision diagnostics and therapeutics. Methods: this study integrated proteomics, metabolomics, and bioinformatics analyses to perform multiomics profiling of sputum samples from 53 asthma patients (27 EA, 20 NEA) and 53 healthy controls. Differential expression analysis, weighted gene coexpression network analysis (WGCNA), and CIBERSORT immune cell infiltration analysis were systematically applied to characterize the molecular features of EA and NEA. Machine learning approaches (LASSO regression, SVM-RFE, and Random Forest) were further employed to screen biomarkers and construct diagnostic models. Results: the study systematically characterized proteins and metabolites in asthma patients and healthy controls. We constructed comprehensive interaction networks that reveal critical protein–metabolite relationships and dysregulated pathways, elucidate potential molecular mechanisms underlying asthma heterogeneity, including inflammatory signaling cascades, oxidative stress responses, and metabolic reprogramming. The study also revealed that EA is characterized by Th2-type immune responses and dysregulated histidine/purine metabolism, whereas NEA exhibits neutrophil-mediated oxidative stress (HIF-1 signaling activation, arginine metabolism disturbance) and predominant innate immunity. Multiomics network analysis identified reactive oxygen species (ROS) metabolism and cytokine production as key features in EA, while hypoxia response and NF-κB pathway enrichment were prominent in NEA. A diagnostic model based on 7 core proteins (ASS1, C4B, etc.) effectively distinguished asthma from healthy controls (AUC = 0.944), while 2 overlapping biomarkers (CLCA1, NAMPT) accurately stratified EA and NEA (AUC = 0.903). Conclusion: this study comprehensively delineated the distinct immune-metabolic signatures of asthma patient and healthy controls, as well as EA and NEA through multiomics integration and to establish high-accuracy noninvasive diagnostic models. These findings provide novel targets for molecular subtyping and precision treatment of asthma, particularly offering a theoretical foundation for hypoxia-pathway-targeted interventions in NEA.

1. Introduction

Bronchial asthma is a prevalent and debilitating chronic respiratory disorder characterized by recurrent episodes of wheezing, breathlessness, chest tightness, and coughing, affecting over 300 million individuals worldwide. , This complex disease manifests through persistent airway inflammation, bronchial hyperresponsiveness, and reversible airflow obstruction, with pathogenesis involving intricate interactions between genetic predisposition and environmental triggers. , The underlying mechanisms encompass multiple pathological processes, including Th2-mediated eosinophilic inflammation, Th17-driven neutrophilic activation, oxidative stress-induced airway remodeling, and impaired epithelial barrier function. Notably, emerging evidence highlights the crucial role of metabolic reprogramming and microbiome dysbiosis in modulating immune responses and disease severity. ,

Despite significant advances in asthma management, substantial challenges persist in clinical practice. Eosinophilic asthma (EA) and neutrophilic asthma (NEA) represent two predominant inflammatory endotypes with distinct pathogenic mechanisms and clinical characteristics. Studies have demonstrated that EA is primarily driven by Th2-mediated immune responses and eosinophilic airway inflammation, which typically responds well to corticosteroid therapy. In contrast, NEA is predominantly mediated through Th17/IL-23 axis activation, featuring neutrophil-dominant inflammation and oxidative stress bursts, with approximately 40–50% of patients exhibiting corticosteroid resistance. Current therapeutic strategies, primarily relying on inhaled corticosteroids and bronchodilators, demonstrate limited efficacy in approximately 30–50% of patients, particularly those with neutrophilic subtypes. This therapeutic gap stems largely from the remarkable heterogeneity of asthma, which encompasses diverse endotypes and phenotypes with distinct molecular signatures and treatment responses. This therapeutic gap underscores the urgent need for precise endotyping strategies to identify NEA-specific molecular signatures and therapeutic targets.

With the rapid advancement of omics technologies, our research capabilities in understanding asthma pathogenesis have undergone revolutionary improvements. The systems biology approach integrating proteomics and metabolomics has become a powerful tool in biomedical research, , particularly in biomarker discovery and precise phenotyping of complex diseases. Through comprehensive analysis of the interrelationships between metabolites and proteins, we can understand biological processes from holistic, dynamic, and multidimensional perspectives, thereby facilitating the identification of robust molecular signatures for disease classification. This integrated research methodology has been widely applied in the field of traditional medicine and has demonstrated remarkable potential in screening diagnostic biomarkers and establishing molecular subtyping systems.

Regarding biomarker development, the Global Initiative for Asthma (GINA 2023) guidelines have recommended blood eosinophil counts (≥300 cells/μL) and fractional exhaled nitric oxide (FeNO ≥50 ppb) as auxiliary diagnostic indicators for EA. However, the specificity of these markers significantly decreases in patients with comorbid allergic diseases. For NEA diagnosis, although biomarkers such as YKL-40 and IL-17 have demonstrated moderate predictive value in Western populations, validation studies in Asian cohorts showed significantly reduced diagnostic performance. Therefore, deciphering the molecular signatures of these distinct endotypes and establishing reliable biomarkers for early diagnosis are critically important for precision asthma management.

This study systematically elucidates immunometabolic signature differences between bronchial asthma and healthy controls by integrating large-scale proteomic, metabolomic, and machine learning approaches. Comparisons between asthma patients and healthy controls provide foundational context for understanding core molecular alterations, while endotype-specific analyses identify key immune-inflammatory and metabolic pathways distinguishing EA and NEA. Our machine learning-derived diagnostic model successfully discriminates between these endotypes, offering a robust tool for precise subtyping and laying the groundwork for personalized asthma management.

This study employs an integrated multiomics approach to investigate asthma patients and healthy controls, particular further differentiating between eosinophilic (EA) and neutrophilic (NEA) asthma endotypes. Through systematic protein and metabolite characterization, differential expression analysis, and computational modeling, we constructed comprehensive interaction networks that reveal critical protein–metabolite relationships and dysregulated pathways. These networks elucidate potential molecular mechanisms underlying asthma heterogeneity, including inflammatory signaling cascades, oxidative stress responses, and metabolic reprogramming. The work ultimately aims to identify key molecular signatures and validate potential biomarkers for improved asthma subtyping and personalized therapeutic strategies.

2. Methods

2.1. Study Design and Participant Selection

This single-center observational study was conducted to investigate proteomic and metabolomic in bronchial asthma. A total of 53 bronchial asthma and 53 healthy controls were recruited from The First Affiliated Hospital of Guangzhou Medical University between May 2024 and May 2025. Asthma diagnosis followed the GINA guidelines, according to fulfilment of 1 or more of the following criteria: (1) an increase >12% in forced expiratory volume in 1 s (FEV1) and an increase >200 mL after inhalation of 400 μg of albuterol; (2) a 20% reduction in FEV1 in response to a provocative concentration of inhaled methacholine (PC20 < 10 mg/mL); (3) and/or an increase >20% in FEV1 after 2 weeks of treatment with systemic oral corticosteroids (OCS) or inhaled corticosteroids (ICS). (4) Participants were aged between 18 and 70 years. (5) All study participants attended a single visit that encompassed assessments of lung function, asthma symptoms, asthma-specific quality of life, and medication use, with no history of smoking. (6) Patients were initially diagnosed with mild-to-moderate asthma exacerbation or persistent asthma according to GINA guidelines. (7) Patients had been routinely treated with fluticasone or an equivalent dose of ICS for over 6 months but had experienced more than 1 acute exacerbations in the past half year. Exclusion criteria: history of chronic lung diseases or any other diagnosed conditions, ongoing immunotherapy, malignancy diagnosis, recent respiratory infection, or antibiotic use in past 4 weeks.

Healthy controls were age- and sex-matched individuals without respiratory symptoms, recent infections, or chronic respiratory diseases. All controls demonstrated normal pulmonary function tests and had no history of systemic antibiotic use within the preceding 4 weeks. A schematic representation of the study can be found in Figure .

1.

Integrated multi-omics analysis for bronchial asthma investigation.

Ethical approval was obtained from the Ethics Committee of The First Affiliated Hospital of Guangzhou Medical University (approval codes: 2022 No. 121 and 2024 No. G-007). Written informed consent was obtained from all participants before study enrollment, and the study was conducted in accordance with the Declaration of Helsinki.

2.2. Samples Collection

Sputum induction involved isotonic saline with a bronchodilator, processed as described in ref . Asthma patients were classified as EA or NEA based on sputum cell counts: EA had ≥3% eosinophils; NEA had <3% eosinophils and neutrophil thresholds of ≥75.57% (<20 years), ≥61.61% (20–40 years), ≥63.25% (40–60 years), or ≥67.25% (>60 years). Samples with >10% squamous cells were excluded. All participants underwent spirometry per ERS guidelines to assess lung function and airway obstruction severity.

2.3. Liquid Chromatography–Mass Spectrometry Analysis for Proteomics

Prior to LC–MS analysis, protein samples (5 μL) underwent denaturation in 8 M urea/100 mM ammonium bicarbonate (Sigma-Aldrich) in a 50 μL final volume. Sequential reduction (10 mM DTT, 37 °C, 30 min) and alkylation (50 mM IAA, RT, dark) preceded overnight tryptic digestion (1:30 enzyme/substrate, 37 °C; Promega). Digested peptides were purified using C18 StageTips (Thermo Fisher) and quantified via Nanodrop One (A280). Chromatographic separation employed a 15 cm × 75 μm C18 column (2 μm particles; Dr. Maisch) with a 60 min 2–90% ACN/0.1% FA gradient at 300 nL/min flow rate.

Mass spectrometric analysis data were acquired using a timsTOF Pro2 (Bruker, Germany) in parallel accumulation-serial fragmentation (PASEF) mode. The diaPASEF scan covered an m/z range of 400–1200, ion mobility 0.7–1.43 V s/cm², with a 32 × 25 Th window scheme and 100 ms ramp time. Raw data were processed using Spectronaut 18 (Biognosys), targeting the UniProtKB human protein database. Applying the following search criteria: tryptic specificity (≤2 missed cleavages), carbamidomethylation (fixed), oxidation/acetylation (variable), and strict mass accuracy thresholds. MS/MS spectra were interference-corrected, with FDR controlled at <1% for both PSM and protein identifications.

2.4. UHPLC–MS/MS Analysis for Metabolomics

The metabolomic profiling protocol commenced with sample preparation wherein 100 μL aliquots were combined with 400 μL of chilled methanol (Sigma-Aldrich, USA) at a 1:4 (v/v) ratio. Following thorough vortex mixing, complete metabolite extraction was achieved through 60 min ice-bath sonication. Subsequent protein precipitation was performed via 1 h incubation at −20 °C, followed by centrifugation (14,000g, 20 min, 4 °C) to obtain protein-free supernatants. These supernatants were subsequently concentrated to dryness using a Thermo Fisher Scientific vacuum concentrator prior to LC–MS analysis.

The metabolomic analysis was performed on an ultrahigh-performance liquid chromatography–electrospray ionization-quadrupole-Orbitrap mass spectrometry (UHPLC–ESI-Q-Orbitrap-MS) system, specifically utilizing a Shimadzu Nexera X2 LC-30AD high-performance liquid chromatography (HPLC) system (Shimadzu Corporation, Japan) coupled with a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific, USA). Chromatographic separation was achieved using an ACQUITY UPLC HSS T3 column (2.1 × 100 mm, 1.8 μm particle size, Waters Corporation, USA), with a mobile phase of 0.1% formic acid in water (Sigma-Aldrich, USA) (A) and 100% acetonitrile (Sigma-Aldrich, USA) (B), flowing at 0.3 mL/min. The gradient program started at 0% B, increased to 48% B over 4 min, then to 100% B over the next 4 min, held at 100% B for 2 min, and returned to 0% B in 0.1 min, followed by a 3 min re-equilibration phase.

For heated electrospray ionization (HESI), the optimized settings were: spray voltage at 3.8 kV for positive mode and 3.2 kV for negative mode, capillary temperature at 320 °C, sheath gas flow at 30 arbitrary units, auxiliary gas flow at 5 arbitrary units, probe heater temperature at 350 °C, and an S-Lens RF level of 50. Full MS scans covered an m/z range of 70–1050 Da, with a resolution of 70,000 for full scans and 17,500 for MS/MS scans at m/z 200. Stepped normalized collision energies (20, 30, 40) were used for fragmentation.

Quality control (QC) samples, made from pooled aliquots of all specimens, and blank samples (75% acetonitrile in water, Sigma-Aldrich, USA) were injected every six samples to ensure data integrity and normalization consistency. MS-DIAL software was used for data analysis, including peak alignment, retention time correction, and peak area extraction. Metabolite identification was based on exact mass (tolerance <10 ppm) and MS/MS spectral data (tolerance <0.02 Da), using databases such as Human Metabolome Database (HMDB), MassBank, and a proprietary metabolite standard library. Only metabolites detected in over 50% of samples in any group were considered for further analysis, ensuring the reliability and significance of detected metabolites.

2.5. Construction of Weighted Gene Coexpression and Immune Infiltration Analysis

To explore the complex genetic interactions and identify potential biomarkers or therapeutic targets for asthma, we employed the Weighted Gene Coexpression Network Analysis (WGCNA) framework using the “WGCNA” package in R software (version 4.4.2). The input data comprised a proteomic expression matrix derived from LC–MS analysis. To ensure robust analysis and minimize spurious correlations, the data underwent rigorous preprocessing. Low-abundance proteins with expression variance in the lowest 25% quartile were filtered out to prioritize biologically informative proteins. Proteins with missing values in >50% of all samples were first filtered out to ensure data reliability. For the remaining missing values, median imputation was applied, where each missing value was replaced by the median expression level of the corresponding protein across all samples. The ‘goodSamplesGenes̀’ function confirmed data quality. A soft-thresholding power (β = 18) was selected using ‘pickSoftThreshold̀’, achieving a scale-free topology fit (R ² > 0.85) while balancing network connectivity. A topological overlap matrix (TOM) was constructed from a Pearson’s correlation-based adjacency matrix, and modules were identified via hierarchical clustering with a minimum size of 30 proteins. The gray module, containing proteins with weak coexpression, was excluded from downstream functional analyses to ensure biological interpretability. Gene modules were identified using average linkage hierarchical clustering based on TOM-based dissimilarity, and modules significantly associated with asthma were selected for further exploration.

For immune cell infiltration analysis, we performed computational deconvolution of proteomic data using established R packages. The source code of the R package was obtained from the official CIBERSORTx Web site (https://cibersortx.stanford.edu/). We employed the CIBERSORT algorithm to quantify the relative abundances of 22 immune cell subsets in sputum proteomic profiles, comparing asthma patients versus healthy controls, and EA versus NEA subgroups.

2.6. Proteomics Statistical Analysis

Statistical analysis of proteomics data was performed using Perseus software and R software. The analysis encompassed hierarchical clustering at both the protein and peptide levels, with sequence annotation supported by databases such as UniProtKB/Swiss-Prot, Kyoto Encyclopedia of Genes and Genomes (KEGG), and Gene Ontology (GO). Enrichment analyses for GO terms and KEGG pathways were conducted using Fisher’s exact test, with multiple comparisons controlled via false discovery rate (FDR) correction. GO terms were categorized into biological processes (BP), molecular functions (MF), and cellular components (CC), with pathways deemed statistically significant at a q-value threshold of <0.05.

Additionally, furthermore, protein–protein interaction (PPI) networks were constructed using the STRING database and visualized with Cytoscape software to elucidate functional relationships among differentially expressed proteins (DEPs). Hub protein analysis was performed within the Cytoscape environment, identifying the top 10 hub proteins through the minimum concave cost (MCC) algorithm and the top 8 hub proteins using the molecular complex detection (MCODE) algorithm. Then we employed a Venn diagram approach that combined hub proteins identified by WGCNA, DEPs, MCC, and MCODE algorithms, enabling systematic prioritization of key hub proteins for subsequent immune infiltration analysis.

2.7. Metabolomics Statistical Analysis

To distinguish between the metabolic profiles of subjects, we constructed models based on orthogonal partial least squares-discriminant analysis (OPLS-DA). The robustness and predictive accuracy of these models were assessed using permutation tests (n = 200) to ensure against overfitting. Model quality was evaluated based on R2X (cumulative) and R2Y (cumulative) values, which reflect the descriptive power of the model, and Q2 (cumulative) values, which indicate predictive reliability. Ideal models achieved values close to 1 for these metrics. Importantly, for a model to be considered valid, the R ² and Q ² values obtained from permutation tests must not exceed those of the original, nonpermuted model, confirming that predictive capabilities are not due to chance.

The identification of discriminating metabolites was based on variable importance in projection (VIP) values from the OPLS-DA model, coupled with a two-tailed Student’s t-test applied to normalized data for univariate analysis. Additionally, p-values were determined using one-way analysis of variance (ANOVA) for comparisons involving multiple groups. Fold-change (FC) values were calculated based on the mean values for each patient group, with selection criteria for differentially expressed metabolites (DEMs).

Visualization tools such as volcano plots, heat maps, and boxplots were employed to represent these findings, using the ‘heatmap’ and ‘ggplot2’ packages in R software (version 4.4.2). Further, KEGG pathway analysis was performed to explore the biological implications of the identified DEMs, utilizing the Metaboanalyst6.0 web portal (http://www.metaboanalyst.ca/).

2.8. Statistical Analysis

2.8.1. Statistical Analysis of Clinical Data

Statistical analyses of clinical indices were conducted using SPSS software (version 26.0). Continuous variables were expressed as mean ± standard deviation (SD) for normally distributed data or as median with interquartile range (IQR) for non-normally distributed data. Normality was assessed using the Shapiro–Wilk test. Between-group comparisons were conducted using independent t tests for normally distributed variables and Wilcoxon rank-sum tests for non-normally distributed variables, and a p-value of <0.05 was set as the threshold for statistical significance.

2.8.2. Receiver Operating Characteristic Curve Validation

The diagnostic performance of each identified hub protein was evaluated using ROC curve analysis with the “pROC” package in R software. This analysis provided estimates of the area under the curve (AUC) and 95% confidence intervals (CI). Biomarkers with an AUC greater than 0.8 were considered diagnostically relevant, serving as a reliable measure of each gene’s capacity to distinguish between disease states.

2.8.3. Machine Learning Analysis

Three machine learning approaches were employed for predictive modeling and biomarker discovery: (1) random forest (RF) for feature importance evaluation and variable selection, (2) support vector machine (SVM) for high-dimensional classification, and (3) least absolute shrinkage and selection operator (LASSO) regression to identify the most predictive protein associations strongly linked to asthma pathogenesis. The data set was split into a training set (75%, n = 80, including 40 asthma and 40 healthy controls) and a validation set (25%, n = 26, including 13 asthma and 13 healthy controls) using stratified random sampling to maintain balanced representation of asthma and control groups. For EA versus NEA stratification, the asthma cohort (n = 47) was similarly split into a training set (75%, n = 35, including 20 EA and 15 NEA) and a validation set (25%, n = 12, including 7 EA and 5 NEA). LASSO regression was performed with 5-fold cross-validation repeated 10 times to select features with nonzero coefficients. SVM-RFE employed 5-fold cross-validation repeated 10 times during the recursive elimination process to rank features until the root-mean-square error (RMSE) was minimized. For the RF model, hyperparameters were optimized through 10-fold cross-validation repeated 5 times, using 500 trees (ntree), unlimited tree depth, and 3 features considered at each split (mtry = 3), and ranking features by mean decrease in Gini impurity. The intersection of features identified by all three methods was used to construct logistic regression models. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, with the area under the curve (AUC) and 95% confidence interval (CI) reported for both the training and validation sets.

3. Results

3.1. Characteristics of all Participants

The baseline characteristics and pulmonary function metrics of the bronchial asthma cohort (n = 53) and healthy controls (HC) (n = 53) are summarized in Table . No statistically significant differences were observed in age or body mass index (BMI) between the two groups. However, asthmatic patients exhibited significantly elevated serum total IgE levels compared to HC (p < 0.001). Blood cell analysis revealed markedly higher absolute counts of leukocytes, eosinophils, and lymphocytes in the asthma group. Lung function tests demonstrated significant intergroup disparities: the HC group showed superior performance in FEV1% predicted, FEV1/FVC ratio, and FVC % predicted compared to asthma patients (all p < 0.001). The baseline characteristics and pulmonary function metrics of EA and neutrophilic asthma NEA patients are summarized in Table . Hematological analysis revealed significantly elevated absolute eosinophil counts and eosinophil percentages in EA compared to NEA (p < 0.001). Serum total IgE levels were markedly higher in the EA group (p < 0.001). However, no significant differences were observed in pulmonary function parameters, including FEV1% predicted, FEV1/FVC ratio, and FVC % predicted, between the two asthma subtypes (all p > 0.05).

1. Basic Information of Bronchial Asthma in Discovery Cohort .

	healthy controls	bronchial asthma	P-value
N	53	53
age/year	48.04 ± 15.33	47.23 ± 12.97	0.769
BMI (kg/m 2 )	23.44(20.42, 25.73)	23.26(20.44, 25.64)	0.835
Blood Cell Detections
WBC, 109/L	5.47(4.61, 6.50)	6.53(5.28, 7.83)	0.005
EOS, 109/L	0.11(0.065, 0.18)	0.25(0.125, 0.56)	<0.001
NEU, 109/L	3.34(2.4, 4.28)	3.74(2.84, 4.24)	0.057
LYM, 109/L	1.63(1.295, 1.995)	1.8(1.445, 2.39)	0.026
HGB, 109/L	130 ± 14.59	133.86 ± 17.379	0.218
RBC, 109/L	4.45 ± 0.52	4.65 ± 0.60	0.069
EOS %	2.3(1.3, 3.05)	4.3(2.15, 10.1)	<0.001
NEU %	59.87 ± 8.72	56.71 ± 10.61	0.097
Lung Function
FEV1/FVC	102.27(88.39, 109.55)	73.05(58.15, 90.50)	<0.001
FEV1	2.75 ± 0.65	1.88 ± 0.60	<0.001
FVC	3.26(3.05, 3.67)	2.68(2.30, 3.37)	<0.001
Inflammatory Indicators
IgE, mg/L	35.73(3.42, 45.86)	127.39(51.8, 381.30)	<0.001

^aWBC: white blood cell; EOS: eosinophils; NEU: neutrophil; LYM: lymphocyte; HGB: glycosylated hemoglobin; RBC: red blood cell; FEV1: forced expiratory volume in 1 s; FVC: forced vital capacity; IgE: total IgE.

2. Basic Information of Eosinophilic and Neutrophilic Subtypes in Discovery Cohort .

	eosinophilic asthma (EA)	neutrophils asthma (NEA)	P-value
N	27	20
age/year	44.96 ± 12.63	49.05 ± 13.72	0.296
BMI (kg/m 2 )	22.60 ± 3.02	24.25 ± 3.65	0.098
Blood Cell Detections
WBC, 109/L	6.76 ± 1.94	7.10 ± 2.33	0.597
EOS, 109/L	0.42(0.22, 0.93)	0.16(0.073, 0.412)	0.001
NEU, 109/L	3.57 ± 1.01	4.53 ± 2.12	0.075
LYM, 109/L	1.96 ± 0.48	1.87 ± 0.76	0.650
HGB, 109/L	136.22 ± 17.13	130.20 ± 17.82	0.248
RBC, 109/L	4.73 ± 0.46	4.58 ± 0.78	0.417
EOS %	7.9(3.80, 13.50)	2.25(1.35, 5.43)	<0.001
NEU %	53.32 ± 7.55	62.00 ± 11.73	0.04
Sputum
neutrophils %	37(15.50, 46.25)	85.50(70.00, 85.50)	<0.001
eosinophils %	34(17.50, 63.50)	0.50(0.00, 1.88)	<0.001
macrophages %	16.50(5.50, 34.0)	9.25(3.31, 24.63)	0.203
lymphocytes %	1.00(0.00, 1.50)	0.38(0.00, 3.50)	0.335
Lung Function
FEV1/FVC	73.56 ± 21.88	69.15 ± 20.13	0.483
FEV1	1.87 ± 0.60	1.84 ± 0.63	0.888
FVC	2.87 ± 1.01	2.87 ± 0.76	0.974
Inflammatory Indicators
IgE, mg/L	126.83(72.19, 502.29)	64.06(14.52, 169.46)	0.035

^aWBC: white blood cell; EOS: eosinophils; NEU: neutrophil; LYM: lymphocyte; HGB: glycosylated hemoglobin; RBC: red blood cell; FEV1: forced expiratory volume in 1 s; FVC: forced vital capacity; IgE: total IgE.

3.2. Identification Differentially Expressed Proteins in Proteomic Data between Bronchial Asthma and Healthy Controls

In our proteomic analysis of sputum from a discovery cohort of 53 bronchial asthma and 53 healthy controls, DEPs were identified using the following screening criteria: |log₂ FC| > 1.5 and q-value <0.05. We identified 1151 DEPs: 1026 were upregulated, and 125 were downregulated (Figure A, Supporting Information Table S1). Through comprehensive GO and KEGG pathway analyses, we identified significant enrichment of DEPs in critical biological processes and molecular pathways associated with asthma pathogenesis, which enriched in immune defense mechanisms, particularly bacterial defense responses, Fc receptor-mediated signaling, and mast cell degranulation. Oxidative stress regulation pathways were prominently represented, with cellular components localized to azurophil granules and autophagic vesicle. KEGG analysis further identified significant enrichment in the Fc epsilon RI, VEGF, and HIF-1 signaling pathways, demonstrating the critical involvement of immune hyperresponsiveness, oxidative stress, and hypoxia-related pathways in asthma pathogenesis (Figure B, Supporting Information Table S2). A hierarchical clustering dendrogram of the proteomic data effectively segregated the 106 study samples, a soft-thresholding power of β = 14 was selected to construct a biological relevant, scale-free coexpression network. Pearson method was used to establish the association between function and modules (Figure C,D). Using dynamic tree-cutting analysis, we partitioned the coexpression network into 7 distinct modules. Among these, 3 modules demonstrated significant enrichment of coexpressed proteins with strong functional coherence, suggesting their potential involvement in biologically (Figure E).

2.

Differential proteomic analysis reveals key molecular signatures associated with asthma. (A) Volcano plot of DEPs between bronchial asthma (AS) and healthy controls (control). (B) GO/KEGG analysis categories and pathways of DEPs. (C) Threshold selection for WGCNA network construction (β = 14). (D) WGCNA network module classification with a minimum module size of 30. (E) A protein clustering dendrogram displays the modular organization of proteins, with color coding indicating specific modules. The first row in each block represents the correlation coefficient, and the second row in parentheses represents the p-value. Red indicates positive correlation, and blue indicates negative correlation.

3.3. Functional Enrichment Analysis of Weighted Gene Coexpression Network Analysis Modules and Differentially Expressed Proteins in Asthma

Functional analysis revealed three distinct module-asthma associations. The gray module showed negative correlation (r = −0.32) with asthma, enriched in immune pathways (leukocyte-mediated immunity) and microbial defense (Staphylococcus aureus infection) (Figure A, Supporting Information Table S3). The blue module showed a strong positive correlation with asthma and was enriched for fatty acid degradation, phagosome, and glutathione metabolism pathways, suggesting its involvement in metabolic-immune interactions underlying asthma pathogenesis. KEGG pathway revealed coordinated upregulation of stress–response pathways including oxidative phosphorylation and antigen presentation, indicating a cellular state of chronic inflammation and metabolic dysregulation (Figure B, Supporting Information Table S4). The yellow module showed significant positive correlations with key immune–metabolic processes, particularly complement activation, cholesterol metabolism and PPAR signaling. KEGG pathway revealed coordinated upregulation of humoral immunity (immunoglobulin-mediated responses), acute inflammation, and lipid homeostasis pathways, along with extracellular matrix components and vesicular transport systems. These findings suggest the module represents an integrated immune-lipid metabolic network potentially involved in vascular inflammation and pathogen defense mechanisms (Figure C, Supporting Information Table S5).

3.

Functional enrichment analysis of WGCNA modules and DEPs in asthma. (A) GO and KEGG pathway enrichment of the gray module (negatively correlated with asthma). (B) GO and KEGG pathway enrichment of the blue module (positively correlated with asthma). (C) GO and KEGG pathway enrichment of the yellow module (positively correlated with asthma). (D) Integration analysis of asthma-associated proteins: (left) Venn diagram showing the overlap between DEPs and WGCNA-identified asthma-positive modules, (right) chord diagram displaying GO/KEGG pathway enrichment of the overlapping gene set. (E) Functional annotation of asthma-negative protein network: chord diagram illustrating GO/KEGG pathway enrichment for genes shared between DEPs and WGCNA-identified asthma-negative modules.

Further functional characterization of the gene set derived from the intersection of DEPs and WGCNA positively correlated modules, a total of 17 differential genes was performed through GO and KEGG analyses. GO enrichment revealed their pivotal roles in lipid homeostasis, including sterol/cholesterol metabolic processes and cholesterol efflux; energy metabolism, notably the tricarboxylic acid (TCA) cycle and immune-related trafficking. KEGG pathway analysis further identified: central metabolic pathways (carbon metabolism, citrate cycle); lipid modulation pathways (propanoate metabolism, valine/leucine/isoleucine degradation) and immune-metabolic crosstalk (neutrophil extracellular trap formation) (Figure D, Supporting Information Table S6). On the other hand, functional analysis of the 64 DEPs derived from the intersection of DEPs and WGCNA negatively correlated modules revealed distinct biological themes through GO and KEGG analyses, which were enriched in immune regulation (leukocyte migration), muscle function (sarcomere organization), and cellular structure (intermediate filaments), suggesting potential compensatory mechanisms in disease progression. These findings suggest that the negatively correlated modules may play compensatory roles in immune modulation and structural maintenance during disease progression (Figure E, Supporting Information Table S7).

3.4. Immune Landscape in Bronchial Asthma

The CIBERSORT algorithm was used to assess the characteristics of immune cell infiltration. Figure A presents a heatmap illustrating the proportional distribution of 21 immune cell types across samples from both the bronchial asthma and healthy controls. In comparison to healthy controls, the bronchial asthma group exhibited an elevated proportion of NK cells activated and a diminished presence of macrophages MO and macrophages M2 (Figure B, Supporting Information Table S8).

4.

Immunoinfiltration analysis and correlation of hub DEPs identified by PPI network screening. (A) A heatmap showing the relative proportions of immune cell infiltration in samples from asthma (AS) versus HC controls (CON) assessed using CIBERSORT. (B) A violin plot displaying the differential analysis results of immune cells in samples from AS versus CON. *p < 0.05; **p < 0.01; ***p < 0.001. (C) Correlation analysis between immune cells. (D) Venn diagram showing the overlap between DEPs, WGCNA-derived modules, and high-confidence interactors from MCC/MCODE analyses. (E) PPI network constructed from hub proteins using STRING. (F–M) Bar plot showing correlation between hub DEPs and infiltrating immune cell subtypes.

Further analysis of immune cell interactions environment demonstrated a synchronized increase in eosinophils and NK cells resting, as well as T cells CD4 memory activated cells, suggesting coordinated immune activation in response to the bronchial asthma attack (Figure C, Supporting Information Table S9). Furthermore, PPI network was constructed via the STRING database using 81 overlapping genes identified by both DEPs and WGCNA analyses. Network topology was subsequently analyzed using both MCC and MCODE algorithms to identify highly interconnected protein clusters. This multialgorithm integration significantly improved the reliability of hub protein identification for downstream functional analyses (Figure D,E).

Subsequently, spearman correlation analysis between core hub proteins and immune cells identified several key proteins that were significantly negatively correlated with macrophages M0 or M2 levels, including IDH2, APOB, APOL1, ACADS, ALDH6A1, SAA4, and PON1. Conversely, ACADS, ALDH6A1, APOB, IDH2, PON1, SUCLG1 displayed a consistent positive correlation with naive dendritic cells resting, SSA4 displayed a consistent positive correlation with eosinophils, suggesting distinct immune regulatory dynamics in the respiratory tract environment (Figure F–M, Supporting Information Table S10).

3.5. Sputum Metabolomic Insights into Bronchial Asthma

Building upon our proteomic findings, we performed comprehensive metabolomic profiling to elucidate the biochemical alterations associated with bronchial asthma. We identified 336 significant metabolites using the following stringent criteria: VIP scores >1, |log₂ FC| > 0.585, and FDR-adjusted q-value <0.05 (Figure A–C, Supporting Information Table S11). Pathway enrichment analysis revealed significant disruptions in key metabolic pathways, including beta-alanine metabolism, arginine and proline metabolism, steroid hormone biosynthesis, glutathione metabolism and histidine metabolism. These findings provide insights into the biochemical underpinnings of bronchial asthma and offer potential targets for therapeutic intervention (Figure D,E, Supporting Information Table S12).

5.

Multiomics insights into bronchial asthma. (A) OPLS-DA differentiation of healthy controls and bronchial asthma. (B) Volcano plots reveal DEMs between the two groups. (C) A Venn diagram filters for DEMs fulfilling specified criteria. (D,E) KEGG analyses of DEMs. (F) A network diagram showcases the intricate interactions between metabolites and proteins implicated in bronchial asthma, illustrating their interactions and highlighting key proteins and metabolites. (G) GO and KEGG pathway enrichment of the key proteins. (H) Identification of key metabolic pathways.

By innovatively constructing a multiomics network integrating DEPs and DEMs from bronchial asthma samples, we established a comprehensive molecular landscape of this disease. The resulting network revealed intricate interactions between proteins and metabolites, demonstrating significant pathophysiological alterations (Figure F, Supporting Information Table S13). GO enrichment analysis of network proteins highlighted two particularly prominent biological processes: reactive oxygen species metabolic process and leukocyte-mediated immunity. Concurrently KEGG pathway analysis of network metabolites identified significant disruptions in amino acid metabolism pathways and infection-related pathways (Figure G, Supporting Information Table S14). These findings collectively suggest the activation of oxidative stress response and immunometabolic reprogramming as potential host defense strategies in bronchial asthma pathogenesis.

Functional enrichment analysis of the five key metabolites (spermine, butyric acid uracil, spermidine and creatine) within the interaction network revealed significant enrichment in several critical metabolic pathways, most notably spermidine and spermine biosynthesis, β-alanine metabolism, and fatty acid biosynthesis. These findings collectively suggest profound dysregulation of polyamine homeostasis and lipid metabolism in bronchial asthma pathogenesis.

3.6. Identification Differentially Expressed Proteins in Proteomic Data between Eosinophilic Asthma and Neutrophils Asthma

While previous studies have delineated the protein-metabolite interaction networks in bronchial asthma compared to healthy controls, the clinical management of asthma remains challenging due to the divergent therapeutic responses across inflammatory phenotypes. Among these, EA and NEA represent two predominant but mechanistically distinct subtypes, the underlying protein-metabolite crosstalk remain poorly elucidated. We performed an integrated multiomics analysis to unravel the dysregulated molecular networks and interaction mechanisms distinguishing EA from NEA, aiming to identify phenotype-specific diagnostic markers.

In our proteomic profiling of sputum samples from 53 bronchial asthma patients, we stratified the asthma cohort into distinct inflammatory endotypes, identifying 27 cases of EA and 20 cases of NEA based on sputum cell counts. DEPs were identified using the following screening criteria: |log₂ FC| > 1 and q-value <0.05. We identified 248 DEPs: 67 were upregulated, and 181 were downregulated (Figure A, Supporting Information Table S15). Through comprehensive GO and KEGG pathway analyses, DEPs in critical biological processes significantly enriched in core neutrophil functions (chemotaxis, degranulation, bacterial killing), Fcγ receptor-mediated phagocytosis, reactive oxygen species (ROS) generation. KEGG analysis further identified significant enrichment in the Fc epsilon RI, VEGF, and HIF-1 signaling pathways, demonstrating the critical involvement of immune hyperresponsiveness, oxidative stress, and hypoxia-related pathways in asthma pathogenesis (Figure B,C, Supporting Information Table S16). A hierarchical clustering dendrogram of the proteomic data effectively segregated the 47 study samples, scale-free coexpression network. Pearson method was used to establish the association between function and modules (Figure D,E). Through dynamic tree-cutting analysis, the coexpression network was systematically partitioned into 8 distinct modules. Notably, one module exhibited significant coexpression patterns of proteins. Through integrative analysis of DEPs and WGCNA-derived modules significantly associated with NEA, we identified 23 overlapping genes (Figure F). Functional characterization demonstrated that these genes enriched in oxidative stress modulation, metabolic rewiring, and oxygen-sensing mechanismsall hallmarks of neutrophilic airway inflammation. The specific involvement of arginine metabolism and HIF-1 signaling provides mechanistic insight into NEA’s distinct pathophysiology compared to EA (Figure G, Supporting Information Table S17). The convergence of these pathways points to potential therapeutic targets for modulating neutrophilic airway inflammation.

6.

Differential proteomic analysis reveals key molecular signatures associated with NEA. (A) Volcano plot of DEPs between EA and NEA. (B,C) GO/KEGG enrichment analysis categories and pathways of DEPs. (D) WGCNA network module classification with a minimum module size of 30. (E) A protein clustering dendrogram displays the modular organization of proteins, with color coding indicating specific modules. The first row in each block represents the correlation coefficient, and the second row in parentheses represents the p-value. Red indicates positive correlation, and blue indicates negative correlation. (F) Venn diagram of overlapping genes between DEPs and NEA-associated WGCNA modules. (G) GO/KEGG enrichment analysis categories and pathways of 23 overlapping genes.

3.7. Immune Landscape in Eosinophilic Asthma and Neutrophils Asthma

The CIBERSORT algorithm was also used to assess the characteristics of immune cell infiltration patterns across in asthma subtypes. Figure A displays a heatmap illustrating the proportional distribution of 21 immune cell types across samples from both EA and NEA. In comparison to EA, the NEA group exhibited an elevated proportion of macrophages M0, Mast cells activated and NK Cells. Whereas activated NK cells were significantly reduced compared to EA (Figure B, Supporting Information Table S18).

7.

Immunoinfiltration analysis and correlation of hub DEPs identified by PPI network screening in EA and NEA. (A) Heatmap showing the relative proportions of immune cell infiltration in samples from EA and NEA assessed using CIBERSORT. (B) A violin plot displaying the differential analysis results of immune cells in samples from EA and NEA. *p < 0.05; **p < 0.01; ***p < 0.001. (C) Correlation analysis between immune cells subtypes, highlighting EA/NEA associated patterns. (D) PPI network constructed from hub proteins using STRING. (E–L) Bar plot showing correlation between hub DEPs and infiltrating immune cell subtypes.

Further analysis of immune cell interactions environment demonstrated a synchronized increase in neutrophils and mast cells activated, as well as NK cells resting, macrophages M0 cells, suggesting a tightly regulated inflammatory response (Figure C, Supporting Information Table S19). Additionally, we constructed a PPI network from the intersection of DEGs and WGCNA-identified hub genes using the STRING database and Cytoscape software. Subsequent application of both MCC and MCODE algorithms enabled the identification of 8 core hub proteins that are potentially critical to disease pathogenesis (Figure D).

Furthermore, spearman correlation analysis between 8 core hub proteins and immune cells identified several key proteins that were significantly negatively correlated with macrophages M0 levels, including ALPL, CEACAM5, CLCA1, ITLN1, NOS2, and NOS3. Conversely, ALPL, NOS2 and NOS3 displayed a consistent positive correlation with NK cells activated, suggesting distinct immune regulatory dynamics in the respiratory tract environment (Figure E–L, Supporting Information Table S20).

3.8. Sputum Metabolomic Insights into Eosinophilic Asthma and Neutrophils Asthma

Building upon our proteomic findings, we performed comprehensive metabolomic profiling to elucidate the biochemical alterations associated with NEA. We selected 452 significant metabolites based on VIP scores >1.5 (Figure A, Supporting Information Table S21). Pathway enrichment analysis revealed significant disruptions in key metabolic pathways, including butanoate metabolism, histidine metabolism, alanine, aspartate and glutamate metabolism, glyoxylate and dicarboxylate metabolism, arginine and proline metabolism. These findings provide insights into the biochemical underpinnings of EA and offer potential targets for therapeutic intervention (Figure B,C, Supporting Information Table S22).

8.

Multiomics insights into eosinophilic asthma and neutrophils asthma. (A) OPLS-DA differentiation of EA and NEA. (B,C) KEGG analyses of significant metabolites. (D) A network diagram showcases the intricate interactions between metabolites and proteins implicated in EA and NEA, illustrating their interactions and highlighting key proteins and metabolites. (E,F) GO and KEGG pathway enrichment of the key proteins. (G,H) Identification of key metabolic pathways.

By innovatively constructing a multiomics network integrating DEPs and DEMs from 47 asthma samples, we established a comprehensive molecular landscape of this disease. The resulting network revealed intricate interactions between proteins and metabolites, demonstrating significant pathophysiological alterations (Figure D, Supporting Information Table S23). Functional enrichment analysis revealed distinct molecular signatures between EA and NEA. GO term analysis demonstrated significant enrichment of reactive oxygen species metabolic process and cellular response to oxidative stress in NEA, consistent with neutrophil-mediated oxidative bursts, while humoral immune response and cytokine production were predominantly associated with EA, reflecting Th2-driven eosinophilic inflammation. KEGG pathway analysis identified marked dysregulation in arginine and proline metabolism and HIF-1 signaling pathway in NEA, suggesting altered nitric oxide biosynthesis and hypoxia-mediated responses. Notably, canonical NF-kappaB signaling was enriched in both subtypes, though with differential activation patterns (Figure E,F, Supporting Information Table S24). These findings highlight fundamental mechanistic divergences between EA and NEA, with eosinophilic asthma characterized by adaptive immune activation and neutrophilic asthma dominated by innate immune and metabolic stress pathways-particularly showing enrichment of proline metabolism and HIF-1 signaling pathway in NEA.

Functional enrichment analysis of the seven key metabolites (l-glutamic acid, butyric acid, taurine, hypoxanthine, creatinine, pyridine, xanthine) within the interaction network revealed significant enrichment in several critical metabolic pathways, butanoate metabolism, purine metabolism, nitrogen metabolism, taurine and hypotaurine metabolism, arginine biosynthesis. These findings provide important mechanistic insights into the distinct pathophysiological characteristics of EA and NEA (Figure G,H).

3.9. Diagnostic Models for Differentiating Bronchial Asthma and Inflammatory Phenotypes

To address the diagnostic and stratification challenges posed by the clinical heterogeneity of bronchial asthma, we employed machine learning approaches to identify hub genes and construct two diagnostic models: one to distinguish bronchial asthma from healthy controls, and another to stratify patients into eosinophilic asthma (EA) and noneosinophilic asthma (NEA) subtypes. We applied three machine learning algorithms for feature selection. LASSO regression with 10-fold cross-validation identified 11 candidate proteins (Figure A), while SVM-RFE analysis indicated that the root-mean-square error (RMSE) was minimized when 51 proteins were included (Figure B). The random forest algorithm screened 36 potential biomarkers (Figure C). By integrating the results from all three methods, seven hub proteinsASS1, C4B, CHAD, CKM, MSLN, NAMPT, and PEBP4were ultimately selected (Figure D,E). A logistic regression model based on these biomarkers demonstrated strong discriminatory performance, with an AUC of 0.944 (95% CI: 0.884–0.979) in the training set and 0.928 in the validation set (Figure F).

9.

Machine learning-based screening of hub genes as potential diagnostic biomarkers for asthma endotypes. (A) LASSO regression selected hub genes (λ ∼ min = 0.0018). (B) SVM–RFE selected hub genes. (C) Random forest selected hub genes. (D) Venn diagram of key genes selected by three machine learning methods. (E) The differential expression of the 7 overlapping genes in bronchial asthma vs healthy controls and their ROC curves in the discovery cohort, showing excellent performance (training AUC = 0.944, validation AUC = 0.928). (F) The differential expression and ROC curve (AUC = 0.835) of the 7 overlapping genes in the external validation cohort. (G) LASSO regression selected hub genes (λ ∼ min = 0.088). (H) SVM–RFE selected hub genes. (I) Random forest selected hub genes. (J) Venn diagram of key genes selected by three machine learning methods. (K) Differential expression of the 2 overlapping genes and their ROC curves in the discovery cohort, showing excellent performance in differentiating EA from NEA (training AUC = 0.903, validation AUC = 0.86). (L) Differential expression of the 2 overlapping genes and the validation ROC curve (AUC = 0.92) in the external cohort.

Similarly, we applied LASSO, SVM–RFE, and random forest to identify biomarkers for differentiating EA from NEA (Figure G–I). Two overlapping proteins, CLCA1 and NAMPT, were selected from the intersection of the three methods (Figure J,K). The diagnostic model for EA vs NEA achieved an AUC of 0.903 (95% CI: 0.768–0.953) in the training set and 0.86 in the validation set (Figure L). Detailed performance metrics of all three machine learning models are provided in Supporting Information Tables S25 and S26.

To further validate the generalizability of these biomarkers, we performed additional sequencing on an external cohort comprising 20 asthma patients (10 EA and 10 NEA) and 10 healthy controls. Demographic characteristics of this validation cohort are summarized in Supporting Information Tables S27 and S28. Differential expression of the seven biomarkers between asthma patients and healthy controls was confirmed in this cohort (Figure F). The logistic regression model based on these seven biomarkers exhibited an AUC of 0.835 in the external validation, indicating good predictive performance. Similarly, the expression differences of the two biomarkers (CLCA1 and NAMPT) between EA and NEA were also validated (Figure L), and their logistic model achieved an AUC of 0.92, demonstrating excellent stratification capability.

4. Discussion

This study provides a comprehensive characterization of the molecular signatures distinguishing bronchial asthma patients from healthy controls, as well as between EA and NEA asthma subtypes, through an integrative multiomics approach combining proteomics, metabolomics, and machine learning. We constructed comprehensive interaction networks that reveal critical protein–metabolite relationships and dysregulated pathways. A well-constructed diagnostic model was established to discern asthma from healthy individuals and to distinguish EA from NEA. These findings not only confirm established aspects of asthma pathophysiology but reveal novel immunometabolic networks and potential therapeutic targets that could transform precision medicine approaches for this complex disease.

4.1. Molecular Signatures and Immunometabolic Networks in Bronchial Asthma

Our proteomic analysis identified 1286 DEPs, with significant enrichment in oxidative phosphorylation and antigen presentation pathways. These findings align with emerging evidence of mitochondrial dysfunction and metabolic reprogramming in asthmatic airways. , The WGCNA revealed that the blue module (positively correlated with asthma) was enriched for glutathione metabolism and phagosome pathways, suggesting that impaired antioxidant defenses may exacerbate oxidative damage through ROS accumulation. , Metabolomic profiling further demonstrated that disturbances in β-alanine and histidine metabolism were closely associated with enhanced Th2 inflammation, where histamine metabolites may amplify immune responses via mast cell-epithelial cell axis activation. These findings collectively support the emerging paradigm of “immune-metabolic crosstalk” in asthma pathogenesis.

The comparative analysis between EA and NEA uncovered more profound mechanistic differences. EA demonstrates a Th2-polarized phenotype characterized by dysregulated histamine metabolism, where histidine decarboxylase upregulation drives a self-amplifying inflammatory loop through histamine-mediated Th2 cell proliferation and mast cell-derived IL-4/IL-13 production. This histaminergic axis may explain the suboptimal response to current biologics in approximately 30% of EA patients. In contrast, NEA exhibits hypoxia-driven pathophysiology, with HIF-1α activation promoting neutrophil survival while simultaneously inducing corticosteroid resistance via NF-κB crosstalk. , Concurrent disturbances in arginine metabolism further exacerbate airway hyperresponsiveness through altered nitric oxide homeostasis, highlighting fundamental differences in the molecular mechanisms between these asthma endotypes.

4.2. Immune Microenvironment and Cellular Crosstalk

CIBERSORT analysis revealed endotype-specific immune landscapes. The overall asthma phenotype showed increased NK cell activation and reduced M2 macrophages, indicating dysregulated innate immunity that may impair inflammation resolution. The coordinated elevation of eosinophils and activated CD4+ T cells in EA reinforces the central role of the “Th2-eosinophil axis”, while NEA demonstrated predominant M0 macrophage/mast cell accumulation, reflecting innate immunity-driven pathology.

Notably, we identified significant correlations between hub proteins and immune cells. IDH2 and APOB showed negative correlations with M0 macrophages, suggesting their potential role in modulating macrophage polarization through lipid metabolism regulation. In NEA, the positive correlation between ALPL, NOS2/3 and activated NK cell simplifies possible involvement of oxidative stress-vascular tone dysregulation. These findings open new avenues for therapeutic strategies targeting immunometabolic networks.

4.3. Clinical Implications of Multiomics Networks

The protein–metabolite interaction network derived from our multiomics analysis reveals several clinically actionable therapeutic targets that address the fundamental pathophysiological mechanisms underlying different asthma endotypes. The consistent dysregulation of FcεRI signaling and glutathione metabolism across asthma subtypes supports the expanded use of anti-IgE therapies and antioxidant strategies in broader asthma populations, with FcεRI expression levels potentially serving as a predictive biomarker for treatment response.

For EA, our network analysis identifies histidine metabolism as a pivotal regulatory node, representing a novel therapeutic target. Preclinical studies have demonstrated that pharmacological inhibition of histidine decarboxylase significantly attenuates eosinophilic inflammation and airway hyperresponsiveness, while showing synergistic therapeutic effects when combined with existing anti-IL-13 biologics. This finding suggests promising opportunities for developing combination therapies targeting both the histaminergic and IL-13 pathways in EA.

In NEA, our data highlight the therapeutic potential of simultaneously targeting the HIF-1/NF-κB/arginase axis. Experimental evidence indicates that combined inhibition of HIF-1 and NF-κB signaling pathways markedly reduces neutrophilic inflammation and restores corticosteroid sensitivity in NEA models. Furthermore, modulation of arginine metabolism through supplementation has been shown to provide additional therapeutic benefits, supporting the development of multitarget intervention strategies for this treatment-resistant asthma phenotype.

These findings collectively establish a molecular framework for advancing precision medicine approaches in asthma management. By targeting the distinct pathogenic networks operative in different asthma endotypes, these strategies promise to overcome the limitations of current uniform treatment paradigms and improve outcomes for patients with difficult-to-treat asthma. Future clinical trials should focus on validating these targeted approaches and developing companion diagnostics to identify patients most likely to benefit from specific therapeutic interventions.

4.4. Diagnostic Model Performance and Limitations

Our machine learning diagnostic models demonstrated excellent performance (AUC >0.9), outperforming conventional biomarkers like blood eosinophil counts. The integration of LASSO, SVM–RFE, and random forest algorithms identified robust biomarkers with strong clinical translation potential. C4B, as a key component of the complement system, exhibits altered levels that may reflect increased susceptibility to respiratory infections in asthma patients. Furthermore, the two novel biomarkers (CLCA1 and NAMPT) identified by our EA/NEA stratification model demonstrated strong mechanistic links to fundamental pathogenic pathwaysCLCA1’s association with histamine-mediated inflammatory cascades and NAMPT’s involvement in hypoxia-response signaling networks, thereby providing clinically actionable tools for endotype-specific asthma diagnosis and management.

However, sample size limitations and metabolite stability issues require validation in larger multicenter cohorts. We fully acknowledge that the cohort size in our study is relatively modest compared to certain previous publications, which may affect the generalizability of the findings. Nonetheless, the robust performance of our diagnostic modelseven with this initial cohorthighlights the potential clinical relevance of the identified biomarkers. Future large-scale, multicenter studies are essential to further validate and refine these molecular signatures. Building upon our findings, several critical research avenues warrant further investigation. Experimental validation of the HIF-1/arginine axis in preclinical models could elucidate its therapeutic potential for neutrophilic asthma, while parallel efforts to develop rapid point-of-care assays for identified biomarkers like ALPL and APOB would facilitate clinical implementation. , Importantly, the distinct molecular signatures we identified strongly support the need for precision clinical trials specifically designed to target different asthma endotypes. Collectively, this work not only provides the most comprehensive molecular characterization of asthma heterogeneity to date but also establishes a robust foundation for advancing personalized therapeutic strategies in asthma care.

5. Conclusion

This study leverages an integrative multiomics approach, combining proteomic, metabolomic, and immune profiling, to delineate the molecular pathogenesis of bronchial asthma, with a focus on EA and NEA endotypes. While confirming established paradigmssuch as Th2-mediated inflammation with elevated IgE and eosinophil activation in EA and HIF-1 signaling with neutrophilic inflammation in NEAour analyses reveal novel insights into protein-metabolite interactions, including dysregulated histidine metabolism in EA and arginine/proline metabolism in NEA. Machine learning algorithms identified robust diagnostic biomarkers (CLCA1, NAMPT) with exceptional discriminatory capacity (AUC > 0.9) for asthma classification and endotype stratification. These findings advance the molecular understanding of asthma heterogeneity, uncover novel immunometabolic networks, and provide a foundation for precision medicine through targeted therapeutic strategies and biomarker-driven diagnostics.

Glossary

Abbreviations

AUC

area under the curve

BMI

body mass index

BP

biological process

CC

cellular component

MF

molecular function

CI

confidence interval

DEMs

differentially expressed metabolites

DEPs

differentially expressed proteins

EA

eosinophilic asthma

NEA

neutrophilic asthma

FC

fold change

FDR

false discovery rate

FEV1

forced expiratory volume in 1 s

FVC

forced vital capacity

GINA

Global Initiative for Asthma

GO

Gene Ontology

ICS

inhaled corticosteroids

IgE

immunoglobulin E

IQR

interquartile range

KEGG

Kyoto Encyclopedia of Genes and Genomes

LASSO

least absolute shrinkage and selection operator

RF

random forest

SVM–RFE

support vector machine

LC–MS

liquid chromatography–mass spectrometry

MCC

minimum concave cost

MCODE

molecular complex detection

OCS

oral corticosteroids

OPLS-DA

orthogonal partial least squares-discriminant analysis

PPI

protein–protein interaction

QC

quality control

RBC

red blood cell

ROC

receiver operating characteristic

TCA

tricarboxylic acid

VIP

variable importance in projection

WBC

white blood cell

WGCNA

weighted gene co-expression network analysis

All processed omics data sets supporting the conclusions of this study, including differentially expressed metabolites (DEMs) and differentially expressed proteins (DEPs) are provided in the Supporting Information. Additionally, the R language software is used for both proteomic and metabolomic analyses has been uploaded to Zenodo (DOI: 10.5281/zenodo.16264766). All tools and data sets are open source and freely accessible.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c07657.

• Lists of differentially expressed proteins and metabolites, WGCNA coexpression module analyses, protein-metabolite interaction networks, immune infiltration profiling, corresponding GO and KEGG pathway enrichment results and additional experimental methodological details (PDF)

#.

S.L., Z.L., and J.Z. should be regarded as co-first authors. Conception and design of the research: B.S., Y.J. Drafting the manuscript: S.L., Z.L. Article structure design: Z.L. and J.Z.; statistical analysis: S.L. and T.L.; samples collection and detection: L.Y. and Q.Y.; acquisition of data: Z.H.; experimental management: S.L. and B.S. All authors read and approved the final version of the manuscript.

The Chinese National Natural Science Foundation (81960023), National Key Research and Development Program of China (2023YFF1203800), Natural Science Foundation of Guangdong Province (2023A1515012917), Natural Science Foundation of Guangdong Province (2024A1515012830), State Key Laboratory Project (SKLRD-Z-202305), Major clinical research project of Guangzhou Medical University (GMUCR2024-02009, GMUCR2025-02014), Guangdong Provincial Clinical Research Center for Laboratory Medicine (2023B110008).

Ethical approval for the inclusion of human subjects was granted by the Ethics Committee of The First Affiliated Hospital of Guangzhou Medical University, with approval codes 2022 No. 121 and 2024 No. G-007. Written informed consent was secured from all study participants prior to their inclusion in the research, and the study was conducted in accordance with the Declaration of Helsinki.

The authors declare no competing financial interest.