Pear Polysaccharide Alleviates Asthma Symptoms in Association With a Ferroptosis‐Related Biomarker Network: An Integrated Bioinformatics and Experimental Study

ABSTRACT

Asthma remains a prevalent global health challenge, necessitating novel therapeutic targets. This study hypothesized that pear polysaccharide alleviates asthma by modulating ferroptosis‐related biomarkers. We integrated transcriptomic data (GSE143303, GSE147878) with database mining (CTD, OMIM, GeneCards, HERB, and FerrDb) to identify genes shared among asthma, pear, and ferroptosis. Machine learning screened core biomarkers, validated in independent datasets (GSE43696 and GSE63142). An ovalbumin‐induced asthma mouse model was used to test the effect of Korla fragrant pear polysaccharide (10 and 40 mg/kg). Bioinformatics analysis identified 146 asthma‐pear shared genes enriched in oxidative stress. Seven biomarkers (AR, CDKN1A, HMOX1, IL1B, IL6, PARP1, and TP53) were screened. In vivo, pear polysaccharide significantly attenuated cytokine release, and reversed the expression of the seven biomarkers in lung tissue and BALF of murine asthma model.Our study demonstrates that pear polysaccharide alleviates asthma pathology, potentially by targeting a hub gene network centered on ferroptosis‐related biomarkers.

Korla fragrant pear polysaccharide alleviates asthma pathology, potentially by targeting a hub gene network centered on ferroptosis‐related biomarkers.

Abbreviations

AR

androgen receptor

AUC

area under the curve

BALF

bronchoalveolar lavage fluid

B‐EOS

blood eosinophil counts

BP

biological process (in GO analysis)

CC

cellular component (in GO analysis)

CCL2

C‐C motif chemokine ligand 2

CTD

Comparative Toxicogenomics Database

CTP

Component‐Target‐Pathology (network)

DEGs

differentially expressed genes

FRGs

ferroptosis‐related genes

GEO

Gene Expression Omnibus

GO

Gene Ontology

H&E

hematoxylin and eosin (staining)

IL‐12

interleukin 12

IL‐1B/IL1B

interleukin 1 Beta

IL‐6/IL6

interleukin 6

IL‐8

interleukin 8

KEGG

Kyoto Encyclopedia of Genes and Genomes

LASSO

least absolute shrinkage and selection operator

MAPK

mitogen‐activated protein kinase

MF

molecular function (in GO analysis)

OMIM

Online Mendelian Inheritance in Man

OVA

ovalbumin

PI3K/AKT

phosphoinositide 3‐kinase/protein kinase B

PPI

protein–protein interaction

RF‐RFE

random forest recursive feature elimination

ROC

receiver operating characteristic

ROS

reactive oxygen species

RT‐qPCR

reverse transcription quantitative polymerase chain Reaction

SHAP

SHapley Additive exPlanations

SVM‐RFE

support vector machine‐recursive feature elimination

TGF‐β

transforming growth factor beta

WGCNA

weighted gene co‐expression network analysis

ZO‐1

zonula occludens‐1

α‐SMA

alpha‐smooth muscle actin

1. Introduction

Asthma is a common chronic respiratory disease. Although the asthma‐related mortality rate has decreased, the absolute number of asthma cases is still increasing due to population growth and environmental risk factors [1, 2, 3]. Although numerous biomarkers of disease activity and treatment response have been identified, including blood eosinophil counts (B‐EOS), their clinical utility remains limited. For instance, B‐EOS is readily accessible, yet its modest predictive value for asthma control and therapeutic response constrains its usefulness in clinical decision‐making [4]. Consequently, future studies must urgently identify novel therapeutic targets and biomarkers to enable more precise and individualized treatment strategies for patients with asthma.

Oxidative stress is considered a key factor in the pathogenesis of asthma, which affects airway smooth muscle function, airway remodeling, and inflammatory responses through multiple pathways [5]. In addition, oxidative stress further exacerbates airway remodeling and inflammatory responses in asthma by influencing the autophagy process [6]. Current research indicates that a diet rich in antioxidants and a healthy lifestyle can reduce the risk of asthma and improve the prognosis of asthma patients [7]. Regulating oxidative stress and antioxidant defense mechanisms may provide new perspectives and strategies for the prevention and treatment of asthma [8, 9].

Pears are one of the most commonly consumed fruits by people and have been used as herbal medicine to relieve cough and moisten the lungs for more than 2000 years [10]. Pears are particularly notable for their antioxidant activity, which benefits from the polyphenols and triterpenoids they are rich in [11]. The components of pears are closely related to oxidative stress, which is reflected in their protective effect against external damage and the regulation of their own ripening [12]. Interestingly, oxidative stress is also a key regulatory factor in the ripening process of pear fruits [13]. This indicates that the beneficial components in pears may themselves originate from their natural ability to adapt to and regulate oxidative stress. Therefore, pear components are potential natural antioxidant components. Therefore, this study hypothesizes that active components in pears can effectively alleviate the oxidative stress state in asthma through their antioxidant activity, thereby reducing asthma symptoms.

Ferroptosis, as a novel form of cell death, has a relationship with oxidative stress and plays an important role in the pathological process of asthma. Studies have shown that oxidative stress leads to lipid peroxidation and intracellular iron accumulation through the excessive generation of reactive oxygen species (ROS), thereby inducing ferroptosis [14]. This mechanism is particularly prominent in respiratory diseases such as asthma. In the pathological process of asthma, ferroptosis not only acts through the oxidative stress pathway but also intertwines with multiple signaling pathways [15, 16].

In this study, we explored the possibility that the bioactive components in pears can exert anti‐asthma effects associated with ferroptosis‐related biomarkers. To verify this hypothesis, this study adopted a comprehensive method combining bioinformatics with multiple machine learning algorithms to screen the potential gene targets of the components in pears for asthma treatment and further verified them in an asthma animal model. This comprehensive method provides basic evidence for the development of pear‐derived therapeutic drugs and is expected to offer new options for asthma treatment.

2. Materials and Methods

2.1. Data Acquisition and Differential Expression Analysis of Asthma‐Related Genes

Gene expression datasets GSE143303 and GSE147878, both related to asthma, were downloaded from the Gene Expression Omnibus (GEO) database. The raw data from both datasets were processed, normalized, and merged. To ensure comparability, batch effects arising from the combination of the two independent studies were removed using the “removeBatchEffect” function from the limma R package. The effectiveness of this data integration and batch correction was visually assessed by generating principal component analysis plots. Subsequently, differential expression analysis was performed on the integrated and batch‐corrected dataset using the limma package. Genes with an adjusted p value < 0.05 and an absolute log2 fold change > 1 were considered statistically significant.

To identify gene modules significantly associated with asthma, a Weighted Gene Co‐expression Network Analysis (WGCNA) was performed on the integrated and batch‐corrected expression dataset. A soft‐thresholding power of 5 was selected to achieve a scale‐free topology network (R ² > 0.85). The minimum module size was set to 50 genes.

2.2. Identification of Pear‐ and Asthma‐related Targets From Online Databases

To identify the potential bioactive targets of pear, the term “pear” was used as a keyword to search the HERB database (https://herb.ac.cn/) [17]. To comprehensively identify known asthma‐associated genes, we integrated information from three major public databases: Online Mendelian Inheritance in Man (OMIM, https://www.omim.org) [18], Comparative Toxicogenomics Database (CTD, https://ctdbase.org) [19], and GeneCards (https://www.genecards.org) [20]. The search terms “asthma” or “bronchial asthma” were applied to each database. The resulting gene lists were downloaded, merged, and deduplicated to generate a consolidated set of asthma‐related targets from existing knowledge bases.

2.3. Acquisition of Ferroptosis‐Related Genes (FRGs)

FRGs were retrieved from the FerrDb database (http://www.zhounan.org/ferrdb/), a manually curated repository specializing in ferroptosis regulators and associated diseases [21].

2.4. Functional Analysis

Gene Ontology (GO) enrichment analysis (covering Biological Process, Molecular Function, and Cellular Component categories) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis were performed using the clusterProfiler package (v4.0) in R. The list of shared genes was used as input, with the background set to all protein‐coding genes in the human genome. Enrichment significance was assessed via the hypergeometric test, and p values were adjusted using the Benjamini‐Hochberg false discovery rate method. Terms and pathways with an adjusted p value (q value) < 0.05 were considered statistically significant.

To explore functional associations, the gene list was submitted to the GeneMANIA web server (version 3.6.0, http://genemania.org) [22]. The resulting network, along with the list of top‐related genes and enriched functions predicted by GeneMANIA, was retrieved for integrative interpretation alongside the GO/KEGG results.

2.5. Protein–Protein Interaction (PPI) Network Construction and Hub Gene Identification

The interactions between proteins were analyzed using STRING (https://string‐db.org), which facilitates the retrieval of interacting genes. The PPI network was constructed by importing shared FRGs into STRING with the species specified as Homo sapiens. The network was then employed for visualization and analysis using Cytoscape.

2.6. Machine Learning Algorithms

To minimize algorithm‐specific bias and reduce false positives arising from heterogeneous data sources, only genes concurrently selected by all three machine learning algorithms were retained as candidate biomarkers for subsequent validation. In order to recognize the potential diagnostic biomarkers in asthma, three machine learning algorithms, least absolute shrinkage and selection operator (LASSO), random forest algorithm combined with recursive feature elimination (RF‐RFE) and support vector machine‐recursive feature elimination (SVM‐RFE), were combined in this study [23]. As a linear regression regularization approach, LASSO introduces the L1 norm as a regularization term to achieve feature selection and model sparsity [24]. RFE combined with Random Forest is a powerful feature selection tool. By recursively removing low‐importance features and evaluating model performance through cross‐validation, it can effectively identify the most valuable features for model prediction, thereby improving the precision and stability of the model [25]. SVM‐RFE utilizes support vector machines for feature selection by recursively training the SVM model and eliminating the least weighted features to select the most important subset of features [26]. By calculating the Shapley values of each gene [27], we systematically revealed its relative importance and biological significance in the model prediction.

2.7. Receiver Operating Characteristic (ROC) Analysis

To assess the diagnostic performance of the identified biomarker genes for asthma, a ROC curve analysis was conducted. The analysis was performed on two independent validation gene expression datasets, GSE43696 and GSE63142, which were downloaded from the GEO database and processed using consistent normalization procedures. For each candidate biomarker, its expression values in the asthma and control samples within each validation set were used to construct the ROC curve. The diagnostic accuracy was quantified by calculating the area under the curve (AUC) value.

2.8. Animal Grouping and Treatment

Female BALB/c mice (6–8 weeks old) were purchased from Medcona Biotech. The animal study was approved by the Ethics Committee of Xinjiang Medical University. All animal experiments complied with the ARRIVE guidelines. All methods were carried out in accordance with relevant guidelines and regulations. Mice were randomly divided into 4 groups: normal group, asthma model group, low‐dose treatment group of Korla fragrant pear polysaccharide (10 mg/kg, polysaccharide‐L), and high‐dose treatment group of Korla fragrant pear polysaccharide (40 mg/kg, polysaccharide‐H), with 6 mice in each group. After 7 days of adaptive feeding in separate cages according to the groups, the experiment was started. Mice in the asthma model group, polysaccharide‐L group, and polysaccharide‐H group were intraperitoneally injected with 1 mL of normal saline solution containing 100 mg of ovalbumin (Sigma‐Aldrich) and 100 mg of aluminum hydroxide on the 1st, 7th, and 14th days of the experiment for sensitization. Mice in the normal group were intraperitoneally injected with an equal volume of normal saline. On the 21st day of the experiment, all mice except those in the control group were nebulized with normal saline containing 1% ovalbumin once a day for 30 min each time for 7 consecutive days. Mice in the polysaccharide‐L group and polysaccharide‐H group were intragastrically administered with polysaccharide 2 h before each ovalbumin challenge, while the normal control group was intragastrically administered with normal saline. The dose of each intragastric administration was 100 µL/10 g.

2.9. Sample Collection

Mice were sacrificed by cervical dislocation after being anesthetized by inhaling excessive isoflurane. The skin and soft tissues were incised to expose the trachea. After making a V‐shaped incision, a lavage catheter connected to a syringe was inserted and fixed. Bronchoalveolar lavage was performed 3 times with 5 mL PBS, and bronchoalveolar lavage fluid (BALF) was collected. After the lung tissues were isolated, 2 lung lobes were fixed in 4% paraformaldehyde, and 3 lung lobes were snap‐frozen in liquid nitrogen for storage.

2.10. RT‐qPCR

Total RNA was isolated from 30 mg of lung tissue using TRIzol Reagent (Invitrogen) according to the manufacturer's instructions. RNA concentration and purity were assessed by measuring the absorbance at 260 and 280 nm using a NanoDrop spectrophotometer (Thermo Fisher Scientific). RNA integrity was verified by 1.5% agarose gel electrophoresis. For each sample, 1 µg of total RNA was reverse‐transcribed into complementary DNA (cDNA) using a PrimeScript RT Reagent Kit with gDNA Eraser (Takara Bio) to eliminate genomic DNA contamination. The qPCR reactions were performed in triplicate for each sample using TB Green Premix Ex Taq II (Takara Bio) on a QuantStudio 5 Real‐Time PCR System (Applied Biosystems). The 20 µL reaction mixture contained 10 µL of TB Green Premix, 0.8 µL of each forward and reverse primer (10 µM), 2 µL of cDNA template (diluted 1:10), and 6.4 µL of nuclease‐free water. The thermal cycling conditions were as follows: initial denaturation at 95°C for 30 s, followed by 40 cycles of 95°C for 5 s and 60°C for 30 s. A melt curve analysis was conducted at the end of each run to confirm the specificity of amplification. All primer sequences are listed in Table 1. GAPDH was used as the internal reference gene for normalization. The relative gene expression level was calculated using the 2^−ΔΔCt method.

TABLE 1.

Primer sequences for RT‐qPCR.

Primer	Sequence
ZO‐1	Forward 5'‐GAGTGGACTATCAAGTGAGCCTAA‐3' Reverse 5'‐ATCCAAGTTGCTCGTCAATCTAA‐3'
E‐Cadherin	Forward 5'‐CAGTTCCGAGGTCTACACCTT‐3' Reverse 5'‐TGAATCGGGAGTCTTCCGAAAA‐3'
Fibronectin	Forward 5'‐GTTATGACGATGGGAAGAC‐3' Reverse 5'‐ATACTGGTTGTAGTTGTGGC‐3'
Collagen I	Forward 5'‐TAAGGGTCCCCAATGGTGAGA‐3' Reverse 5'‐GGGTCCCTCGACTCCTACAT‐3'
Collagen III	Forward 5'‐CTGTAACATGGAAACTGGGGAAA‐3' Reverse 5'‐CCATAGCTGAACTGAAAACCACC‐3'
α‐SMA	Forward 5'‐CCCAGACATCAGGGAGTAATGG‐3' Reverse 5'‐TCTATCGGATACTTCAGCGTCA‐3'
GAPDH	Forward 5'‐TGACCTCAACTACATGGTCTACA‐3' Reverse 5'‐CTTCCCATTCTCGGCCTTG‐3'

2.11. H&E Staining

Lung tissue samples were fixed in 4% paraformaldehyde at 4°C for 16 h, then subjected to standard dehydration through a graded ethanol series, cleared in xylene, and embedded in paraffin wax. Sections were cut at 4–5 µm, mounted onto glass slides, and dried overnight at 37°C. For staining, slides were deparaffinized in xylene and rehydrated through a descending ethanol series to distilled water. Nuclei were stained with Mayer's hematoxylin for 5 min, rinsed in running tap water, differentiated in 1 % acid alcohol for 1–2 s, and blued in Scott's tap water for 30 s. Cytoplasm was counterstained with eosin Y solution for 1 min. Sections were then dehydrated through an ascending ethanol series, cleared in xylene, and mounted with a neutral resinous mounting medium. Stained sections were examined and imaged under a light microscope (Nikon Eclipse Ci‐L).

2.12. Measurement of Cytokine Levels in BALF

The concentrations of CCL2, IL‐8, IL‐12, and TGF‐β in BALF supernatants were determined with commercial ELISA kits (R&D Systems) used strictly according to the manufacturer's instructions. After collection, BALF was centrifuged at 1000 × g for 10 min at 4°C to remove cells and mucus debris; the clarified supernatant was supplemented with complete protease‐inhibitor cocktail (Roche, 1×) and stored in aliquots at −80°C until analysis. Standards and samples were added to antibody‐precoated 96‐well plates, followed by sequential incubation and washing steps. Absorbance was read at 450 nm (correction at 570 nm) on a microplate reader. Cytokine concentrations (pg/mL) were calculated from the respective four‐parameter logistic standard curves.

2.13. Western Blot

Pooled supernatants (1 mL per mouse, 6 mice pooled per group) were first depleted of albumin/IgG with ProteoExtract Albumin/IgG Removal Kit (Merck) and then concentrated 40‐fold with 10‐kDa Amicon Ultra units (Millipore) to a final volume of ∼200 µL. Protein concentration was measured with a BCA kit (Thermo Fisher Scientific). Samples were separated on 12% SDS‐PAGE, transferred to PVDF, and stained with Ponceau S to confirm equal loading; total‐protein lane intensity was used for normalization. Membranes were blocked (5% non‐fat milk in TBST) and incubated overnight at 4°C with primary antibodies: anti‐AR (ab52615, 1:800, Abcam), anti‐CDKN1A (ab109199, 1:1000, Abcam), anti‐HMOX1 (ab82585, 1:1000, Abcam), anti‐IL1B (ab234437, 1:1000, Abcam), anti‐IL6 (ab324449, 1:1000, Abcam), anti‐PARP1 (ab227244, 1:500, Abcam), anti‐TP53 (ab211020, 1:1000, Abcam), and anti‐GAPDH (ab8245, 1:1000, Abcam). After HRP‐secondary antibodies (ab97051, 1:2000, Abcam), bands were visualized with ECL and quantified in ImageJ.

2.14. Statistical Analysis

Statistical analysis was performed using SPSS software (version 26.0) and GraphPad Prism software (version 9.0). Data were collected and analyzed for at least three replicate experiments. A two‐tailed Student's t‐test was used to compare significance between two groups and one‐way analysis of variance (ANOVA) followed by Tukey's test was applied to mark variability in more than three groups. p < 0.05 was considered statistically significant.

3. Results

3.1. Screening and Functional Enrichment of Genes Shared Between Asthma and Pear Targets

Differential expression analysis of the GSE143303 and GSE147878 datasets identified a total of 10,598 DEGs in asthma samples compared to controls, comprising 5,323 upregulated and 5,275 downregulated genes (Figure 1A). To uncover key co‐expressed gene modules associated with asthma, WGCNA was performed, which delineated 12 distinct modules. Among these, the black module demonstrated the highest positive correlation with the asthma phenotype (Figure 1B). By intersecting the 175 genes from the asthma‐correlated black module with the full set of 10,598 DEGs, we obtained 150 high‐confidence asthma‐related genes from the transcriptomic analysis (Figure 1C). To complement this, known asthma‐associated targets were systematically retrieved from public databases, yielding 2,355 unique genes (CTD: 9; OMIM: 40; GeneCards: 2332). The union of these two sources resulted in a consolidated list of 2496 asthma‐related genes for subsequent analysis (Figure 2A). To explore the potential mechanistic link between pear components and asthma, the 2496 asthma‐related genes were cross‐referenced with 709 putative pear‐related targets obtained from the HERB database. This intersection revealed 146 shared genes (Figure 2B). Functional analysis of this shared gene set using GeneMANIA indicated their significant enrichment in biological processes closely related to oxidative stress, including the response to oxidative stress, regulation of ROS metabolic process, fatty acid biosynthetic process, and cellular response to chemical stress (Figure 2C).

FIGURE 1.

Identification of the genes associated with asthma. (A) Limma differential expression analysis volcano plots of the GSE143303 and GSE147878 datasets. (B) Results of WGCNA analysis of the GSE143303 and GSE147878 datasets. (C) Asthma‐related genes identified through the intersection of WGCNA analysis results and DEGs from the GSE143303 and GSE147878 datasets.

FIGURE 2.

Identification and functional analysis of genes shared between asthma and pear targets. (A) Integration of asthma‐associated genes from transcriptomic analysis and public databases. The bar chart (left) shows the number of genes obtained from the CTD, OMIM, and GeneCards. The bar chart (right) shows the number of genes obtained from GSE143303 and GSE147878 datasets and the databases. (B) Venn diagram showing the intersection between the 2496 asthma‐related genes and 709 pear‐related targets from the HERB database, identifying 146 shared genes. (C) Functional association network of the 146 shared genes generated by GeneMANIA. Nodes represent genes, and connecting lines indicate predicted functional interactions.

3.2. Functional Enrichment and Network Analysis of Shared Ferroptosis‐Related Genes (Shared FRGs)

Given the strong functional association of the shared genes with oxidative stress pathways, a hallmark of ferroptosis, we investigated their intersection with ferroptosis. A list of 564 FRGs was acquired from the FerrDb database. A three‐way Venn analysis was performed among the asthma‐related genes (2496), pear‐related targets (709), and FRGs (564). This stringent filtering identified a core set of 25 genes that are concurrently associated with asthma, pear, and ferroptosis (Figure 3A). These were defined as shared FRGs for all downstream analyses. To elucidate the biological functions and pathway involvements of the 25 shared FRGs, a systematic functional analysis was performed. GO enrichment analysis revealed their significant associations with specific biological processes, cellular component (CCs), and molecular functions (Figure 3B). The predominant biological processes (BP) included the positive regulation of programmed cell death (GO:0043068), cellular response to chemical stress (GO:0062197), regulation of inflammatory response (GO:0050727), response to metal ion (GO:0010038), and regulation of lipid localization (GO:1905952). For CCs, these genes were primarily localized to the transcription regulator complex (GO:0005667), endoplasmic reticulum lumen (GO:0005788), perinuclear region of cytoplasm (GO:0048471), nuclear speck (GO:0016607), and nuclear envelope (GO:0005635). Key molecular functions (MF) encompassed oxidoreductase activity (GO:0016702), enzyme activator activity (GO:0008047), carboxylic acid binding (GO:0031406), and sulfur compound binding (GO:1901681).

FIGURE 3.

Functional enrichment analysis and PPI network analysis of shared FRGs. (A) The Venn diagram illustrates the intersection of genes related to pear, ferroptosis, and asthma. (B) Bar chart of GO enrichment analysis for the 25 shared FRGs. (C) Bubble chart of KEGG enrichment analysis for the 25 shared FRGs. (D) The functional associations of the shared FRGs were explored using GeneMANIA. (E) The key genes identified in the PPI network.

KEGG pathway analysis further indicated that these shared FRGs were significantly enriched in several crucial signaling pathways, including Pathways in cancer (hsa05200), Fluid shear stress and atherosclerosis (hsa05418), Ferroptosis (hsa04216), FoxO signaling pathway (hsa04068), and Autophagy‐animal (hsa04140) (Figure 3C).

Complementary functional association analysis using GeneMANIA demonstrated that these genes are closely connected in a network related to oxidative stress response, cellular response to oxidative stress, unsaturated fatty acid biosynthetic process, fatty acid biosynthetic process, ROS metabolic process, and cellular response to oxygen levels (Figure 3D).

To investigate the PPI relationships among the shared FRGs, a network was constructed using the STRING database. Subsequent analysis with the CytoHubba plugin in Cytoscape identified the top 15 hub genes based on connectivity: IL6, IL1B, PPARG, HIF1A, TP53, PARP1, JUN, RELA, PTGS2, TGFB1, CDKN1A, HMOX1, NOS2, AR, and ATG7 (Figure 3E).

3.3. Machine Learning‐Based Identification and Validation of Diagnostic Biomarkers

To identify robust diagnostic biomarkers for asthma from the 15 key hub genes, we employed an integrative approach using three distinct machine learning algorithms (Figure 4A). First, the LASSO regression analysis selected 10 candidate biomarkers: AR, CDKN1A, ATG7, HMOX1, IL1B, IL6, JUN, NOS2, PARP1, and TP53 (Figure 4A,B). Subsequently, the RF‐RFE further refined the list to 9 genes with the highest relevance: PARP1, AR, CDKN1A, IL6, IL1B, HIF1A, PTGS2, HMOX1, and TP53 (Figure 4C). In parallel, SVM‐RFE indicated that the model containing all 15 original hub genes yielded the highest predictive accuracy (Figure 4D,E). By intersecting the results from these three algorithms, we identified a core set of seven consensus genes as potential biomarkers for asthma: AR, CDKN1A, HMOX1, IL1B, IL6, PARP1, and TP53 (Figure 4F). We next validated the expression patterns of these seven genes in two independent asthma cohorts. In the GSE43696 validation set, the expression levels of CDKN1A, PARP1, and TP53 were significantly upregulated, while AR, HMOX1, IL1B, and IL6 were significantly downregulated in asthma samples compared to healthy controls (Figure 4G). These differential expression trends were consistently replicated in the second validation set, GSE63142 (Figure 4H). In two validation datasets, GSE43696 and GSE63142, the AUC for each gene was at least 0.5 (Figure 4I,J), suggesting that these genes have some diagnostic value.

FIGURE 4.

Machine learning and analytical validation. (A) LASSO coefficient path. (B) LASSO regression parameter selection using cross‐validation.(C) RF‐RFE accuracy profile. (D) SVM‐RFE accuracy rate profile. (E) SVM‐RFE error profile. (F) Shared genes between LASSO, RF‐RFE, and SVM‐RFE algorithms. (G, H) Expression levels of the seven potential biomarkers in different samples from the GSE43696 and GSE63142 validation dataset. For all box plots in this figure, the box represents the IQR (25th–75th percentile), the line inside represents the median, and the whiskers represent the minimum and maximum values. (I) The ROC curves for each potential biomarker (AR, CDKN1A, HMOX1, IL1B, IL6, PARP1, and TP53) in the GSE43696 validation dataset. (J) The ROC curves for each potential biomarker (AR, CDKN1A, HMOX1, IL1B, IL6, PARP1, and TP53) in the GSE63142 validation dataset.

To assess the individual predictive importance of the seven identified biomarker genes, we applied SHAP to interpret the XGBoost model. The feature importance ranking revealed the relative contribution of each variable to the model's predictions. PARP1 was identified as the most influential feature, followed by IL6, AR, CDKN1A, HMOX1, TP53, and IL1B (Figure 5A). The SHAP summary plot visualized the distribution of SHAP values for each gene, where red dots indicate high gene expression and blue dots indicate low expression. Analysis of individual gene impacts showed distinct risk profiles. For PARP1, the majority of samples exhibited positive SHAP values (clustered on the right side of the plot), with high‐expression (red) points concentrated in this region. This pattern indicates that high expression of PARP1 is strongly associated with an increased risk of asthma. In contrast, for TP53, high‐expression (red) points were concentrated on the left side of the plot, which corresponds to negative SHAP values. This distribution demonstrates that low expression of TP53 significantly contributes to a higher predicted risk of asthma (Figure 5B).

FIGURE 5.

Explanation of machine learning models using SHAP for 7 gene signature. (A) Feature importance plot. (B) The contribution of each feature in the XGBoost model. When the SHAP value is positive and the future value is predominantly red, this indicates that the feature has a positive correlation with the target variable.

3.4. Pear Polysaccharide Attenuates Asthma Pathology and Modulates Key Biomarkers In Vivo

To investigate the therapeutic potential and mechanism of pear polysaccharide against asthma, we established an ovalbumin‐induced murine asthma model with concurrent low‐ and high‐dose polysaccharide treatment groups. First, we assessed markers of epithelial integrity and fibrotic remodeling in lung tissues. RT‐qPCR analysis revealed that asthmatic mice exhibited a significant downregulation of ZO‐1 and E‐cadherin (genes critical for epithelial barrier integrity), coupled with a marked upregulation of fibronectin, Collagen I, Collagen III, and α‐SMA (key markers of fibrosis and of myofibroblast activation). Both low‐ and high‐dose polysaccharide treatment significantly reversed these expression changes, restoring ZO‐1 and E‐cadherin levels while suppressing the pro‐fibrotic genes (Figure 6A). Histopathological evaluation by H&E staining corroborated these findings. Lung sections from the control group displayed intact airway structure with aligned epithelium and minimal inflammation. In stark contrast, the asthma model group showed substantial inflammatory cell infiltration, bronchial wall thickening, and alveolar septal widening with edema. Both pear polysaccharide treatment regimens effectively mitigated these pathological features, reducing inflammatory infiltration and tissue edema (Figure 6B). We next analyzed the inflammatory milieu. Levels of the pro‐inflammatory cytokines CCL2, IL‐8, IL‐12, and TGF‐β in BALF were significantly elevated in asthmatic mice. Polysaccharide administration at both doses markedly reduced the concentrations of these cytokines (Figure 7A). Finally, Western blot analysis was performed to quantify the protein levels of the seven previously identified hub biomarkers in BALF. The protein expression of CDKN1A, IL1B, IL6, PARP1, and TP53 were significantly increased in the asthma model, and this increase was attenuated by polysaccharide treatment at high dosage. Notably, the expression pattern of HMOX1 was opposite, being downregulated in asthma and upregulated by pear polysaccharide treatment (Figure 7B). Collectively, these results demonstrate that pear polysaccharide alleviates key pathological features of asthma and modulates the expression of core biomarkers identified in our bioinformatic analysis.

FIGURE 6.

Pear polysaccharide alleviates airway remodeling and histological damage in asthmatic mice. (A) mRNA expression levels of epithelial barrier (ZO‐1 and E‐cadherin) and fibrosis‐related genes (Collagen I, Collagen III, and α‐SMA) in lung tissues, measured by RT‐qPCR. Data are presented as mean ± SD (n = 6 mice per group; each sample was run in technical triplicate). *p < 0.05, **p < 0.01, ***p < 0.001 (one‐way ANOVA with Tukey's post‐hoc test). (B) Representative H&E stained lung tissue sections (scale bar = 100 µm), n = 6 mice per group.

FIGURE 7.

Pear polysaccharide modulates inflammatory cytokines and core biomarker proteins in asthmatic mice. (A) Concentrations of the pro‐inflammatory cytokines CCL2, IL‐8, IL‐12, and TGF‐β in BALF, measured by ELISA. Data are presented as mean ± SD (n = 6 mice per group; each sample was assayed in technical triplicate). *p < 0.05, **p < 0.01, ***p < 0.001 (one‐way ANOVA with Tukey's post‐hoc test). (B) BALF supernatants from n = 6 mice per group were pooled and analyzed by Western blot. Representative blot is shown.

4. Discussion

The present study integrated bioinformatics, machine learning algorithms, and experimental validation to systematically investigate the potential of pear polysaccharide in alleviating asthma, with a focus on the underlying mechanism involving ferroptosis. A panel of seven core biomarkers (AR, CDKN1A, HMOX1, IL1B, IL6, PARP1, and TP53) was refined. In an ovalbumin‐induced murine asthma model, we demonstrated that administration of Korla fragrant pear polysaccharide effectively reversed the dysregulated expression of the identified biomarkers related to airway inflammation, fibrosis, and epithelial barrier disruption. Additionally, levels of seven core biomarkers related to ferroptosis can be modulated by pear polysaccharide in vivo. These findings collectively suggest the possibility that pear polysaccharides may exert anti‐asthma effects by regulating the ferroptosis‐related gene network.

A central achievement of this study was the identification and validation of a novel seven‐gene biomarker set (AR, CDKN1A, HMOX1, IL1B, IL6, PARP1, and TP53) that is strongly associated with asthma and appears to be key targets of the therapeutic action of pear polysaccharide. This gene set was distilled from a larger pool of 25 shared FRGs through the application of three distinct and powerful machine learning algorithms: LASSO, RF‐RFE, and SVM‐RFE. The use of multiple, complementary machine learning techniques is a significant strength of the study, as it reduces the risk of algorithm‐specific bias and increases the likelihood of identifying truly robust and predictive features. The intersection of the results from these three methods yielded a consensus set of seven genes, underscoring their collective importance. The AUC values from ROC analysis confirmed their individual diagnostic utility, while SHAP analysis provided deeper insight into the model's predictions, revealing the relative importance of each gene and how their expression levels contribute to the classification of asthma.

Additionally, the bioinformatics analysis revealed that these genes are not merely a random assortment but are enriched in critical biological processes and pathways central to asthma pathogenesis, including the regulation of programmed cell death, inflammatory responses, and cellular responses to oxidative stress and metal ions. The KEGG pathway analysis further implicated these genes in key signaling pathways such as the FoxO signaling pathway and autophagy, both of which are known to intersect with ferroptosis [28, 29, 30]. This suggests that the seven‐gene set may not just be a collection of biomarkers but could represent a core regulatory network that drives the pathological features of asthma. The subsequent in vivo experiments, which demonstrated that pear polysaccharide treatment could reverse the aberrant expression of these specific biomarkers, provide strong evidence that this gene network is a direct target of the therapy, offering a clear and testable mechanism of action.

Furthermore, this study demonstrated the significant therapeutic efficacy of Korla fragrant pear polysaccharides in an OVA‐induced murine asthma model. Histopathological examination revealed that polysaccharide treatment effectively ameliorated pathological changes such as airway inflammatory cell infiltration and bronchial wall thickening. Further molecular analysis indicated that it reduced the levels of key pro‐inflammatory cytokines, including CCL2 and IL‐8, in BALF, exerting a comprehensive anti‐inflammatory effect. Additionally, the polysaccharide upregulated barrier proteins like ZO‐1 and E‐cadherin while suppressing the expression of fibrotic markers such as Collagen I and III. It also exhibited regulatory effects on the seven key genes identified through screening. These findings provide mechanistic evidence that pear polysaccharides alleviate asthma symptoms through multi‐target actions. While studies have demonstrated that polysaccharides can reverse the aberrant expression of biomarkers, the direct molecular targets of these polysaccharides and the specific signaling pathways they activate remain unidentified. It is plausible that polysaccharides exert their effects either by acting on a single upstream target, thereby initiating a cascade that influences the entire network, or by engaging multiple independent targets. Future research should aim to identify the specific receptors or binding partners of pear polysaccharides within airway cells and to map the downstream signaling pathways that lead to the modulation of the seven‐gene network. Additionally, pear polysaccharide is a high‐molecular‐weight biopolymer, it likely exerts therapeutic effects through sustained immunomodulatory activity and epithelial barrier protection rather than through direct, transient ROS scavenging, which may circumvent some of the pharmacokinetic limitations observed with conventional antioxidants. It should also be noted that pear contains other bioactive compounds, including polyphenols and triterpenoids, which may contribute synergistically to the overall anti‐asthma effect; the potential interactions among these components were not explored in the current study and warrant future investigation. This highlights a potential translational challenge, even if the bioactive components are identified.

Despite these findings, our study has several limitations. First, the candidate biomarkers were derived from bioinformatics predictions and machine learning models, an approach increasingly adopted in phytoconstituent screening and immunotoxicity risk assessment [31, 32]. However, the independent GEO validation datasets (GSE43696 and GSE63142) lack detailed clinical phenotypic annotation (e.g., asthma severity, treatment history, comorbidities), constraining clinical translation [32, 33]; these datasets were employed solely to validate statistical discriminative accuracy (ROC/AUC), not to establish clinically actionable thresholds. Although validated in animal models, these findings still require direct experimental confirmation. Second, the acute, allergen‐induced animal model may not fully capture the chronicity and complexity of human asthma, particularly non‐allergic or severe subtypes. Third, potential synergistic effects of other pear bioactive compounds were not explored. Furthermore, antioxidant compounds have proven ineffective as clinical asthma therapeutics due to stability and metabolic issues [9], highlighting a translational challenge. Finally, canonical ferroptosis hallmarks (GPX4, SLC7A11, ACSL4, lipid peroxidation, intracellular iron) were not directly assessed; thus, the current evidence supports an association with ferroptosis‐related biomarkers rather than a direct mechanistic effect, and future prospective clinical and functional validation is warranted.

In summary, this study demonstrates that pear polysaccharide alleviates asthma pathology, potentially by targeting a hub gene network centered on ferroptosis‐related biomarkers. The identified seven‐gene set may serve as novel diagnostic biomarkers and therapeutic targets.

Author Contributions

All authors participated in the design, interpretation of the studies and analysis of the data and review of the manuscript. F. Y. drafted the work and revised it critically for important intellectual content. Q. M. and J. L. were responsible for the acquisition, analysis and interpretation of data for the work. F. Y. and H. T. made substantial contributions to the conception or design of the work. All authors read and approved the final manuscript. 

Funding

The study was supported by 2024 Autonomous Region Key Research and Development Program: Research and Development of Key Technologies for the Deep Processing of Whole Fragrant Pears (Grant No. 2024B04011) and 2024 Urumqi Hongshan Innovative Talent—Young Top‐notch Talent Project: Investigating the Mechanism of Xinjiang Korla Fragrant Pear Polysaccharides in Treating Allergic Asthma Based on Metabolomics and Gut Microbiota (Program No. B241013008).

Ethics Statement

The animal study was approved by the Ethics Committee of Xinjiang Medical University (approval no. 2024‐010). All animal experiments should comply with the ARRIVE guidelines. All methods were carried out in accordance with relevant guidelines and regulations.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File: mnfr70574‐sup‐0001‐blots.zip.

Data Availability Statement

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