2.1. COPD datasets acquisition

Relevant gene expression profiles were identified using the keyword “COPD” in the GEO database. The array datasets GSE37768, GSE38974, GSE212331, GSE148004, and GSE1650 were downloaded for analysis. Specifically, GSE37768 includes 18 COPD lung tissue samples and 9 non-smoking lung tissue samples (with samples from smoking patients excluded), while GSE38974 comprises 23 COPD samples and 9 healthy control samples (platform: Agilent-014850 Whole Human Genome Microarray 4x44K G4112F) [14]. Meanwhile, The GSE212331 dataset contains 72 COPD patient samples and 15 healthy control samples [15], GSE148004 includes 7 COPD patient samples and 10 healthy control samples [16], and GSE1650 consists of 18 COPD patient samples and 12 healthy control samples [17]. These three datasets were used as independent cohorts to evaluate the performance of 113 machine learning models.

2.2. Data preprocessing and merging

The raw gene expression matrix from the The Gene Expression Omnibus (GEO) dataset was preprocessed and sorted using Perl to ensure accurate mRNA data. The R package “limma” (version 3.54.2) was employed to normalize and process the mRNA expression matrix [18,19]. Additionally, the “SVA” package was utilized to integrate the GSE37768 and GSE38974 datasets and eliminate batch effects, further enhancing data consistency.

2.3. Identification of DEGs in COPD

The R package “limma” was utilized to identify DEGs from the merged GSE37768 and GSE38974 datasets. For the microarray expression data, genes were classified as DEGs if they met the criteria of |log2 FoldChange| ≥ 0.585 and an adjusted p-value < 0.05 [20]. Visualization of DEGs, including volcano plots and heatmaps, was generated using the R packages “ggplot2” and “pheatmap,” respectively.

2.4. Weighted gene co-expression network analysis (WGCNA)

The WGCNA algorithm was employed to categorize genes and assess the relationship between gene modules and clinical traits [21]. WGCNA was conducted on the variable genes from the merged GSE37768 and GSE38974 datasets. Module connectivity was determined based on gene correlation within each module, allowing for the exploration of co-expression similarity. To investigate the association between gene expression modules and clinical traits, correlation coefficients and p-values were calculated, with modules considered significant if the p-value was less than 0.05. A heatmap was used to visualize these relationships.

Additionally, the “venn” package in R was used to generate Venn diagrams comparing DEGs with co-expression module genes. Genes shared between these sets were identified as potential key contributors to COPD.

2.5. Functional enrichment analysis

Functional enrichment analysis encompassed both GO [22,23]and KEGG [24] pathway analyses. GO analysis was conducted across three domains: biological processes, molecular functions, and cellular components, while KEGG pathways were used to interpret metabolic pathways and assess gene and genomic functions. Both approaches play a crucial role in understanding gene function during enrichment analysis.

The R software package “org.hs.egg.db” (version 3.16.0) was used as a reference background for the GO and KEGG pathway enrichment analysis process. In addition, we used the R package “clusterProfiler” (version 4.6.2) to obtain functional enrichment results. If the GO term and the KEGG pathway meet the p < 0.05 condition, this result is considered significantly enriched.

2.6. Integrated machine learning framework to identify and construct characteristic genes and diagnostic models of COPD

Using an integrated machine learning approach, we identified and selected pivotal genes associated with COPD to construct a molecular diagnostic model that differentiates between control subjects and COPD cases. The development of the COPD diagnostic model was based on a training cohort derived from the merged datasets GSE37768 and GSE38794, employing a framework that integrates twelve machine learning algorithms: Least Absolute Shrinkage and Selection Operator (LASSO), ridge regression, elastic network (Enet), support vector machine (SVM), gradient boosting with component-wise linear models (glmBoost), stepwise generalized linear model (Stepglm), linear discriminant analysis (LDA), random forest (RF), gradient boosting machine (GBM), partial least squares regression for generalized linear models (plsRglm), eXtreme gradient boosting (XGBoost), and Naive Bayes [13,25]. Detailed descriptions of these algorithms are summarized in the S1 File.

Within this integrated computational framework, four feature selection algorithms—Lasso, RF, Stepglm, and glmBoost—were employed to benchmark the model, while the remaining algorithms were utilized for model fitting, resulting in a total of 113 machine learning combinations.

To avoid overfitting and data leakage, strict separation between training and external test cohorts was implemented. The integrated machine learning framework was constructed exclusively on the merged training cohort (GSE37768 and GSE38974). 10-fold cross-validation (10-CV) was applied during model training and hyperparameter optimization, with feature selection procedures (Lasso, RF, Stepglm, glmBoost) nested within each cross-validation fold. All preprocessing steps including batch correction and normalization were conducted only within the training subset of each fold. Three completely independent datasets (GSE212331, GSE148004, and GSE1650) were used as external test sets and were not involved in gene screening, feature selection, model construction, or parameter tuning.

The R package “pROC” was employed to evaluate the predictive performance of the diagnostic model and calculate the area under the receiver operating characteristic curve (AUC). The machine learning combination achieving the highest average AUC in both the training and testing cohorts was designated as the optimal model, referred to as MS. The experimental setup of the integrated machine learning framework, including the use of R packages, cross-validation, and hyperparameter optimization, is detailed in the supplementary materials.

2.7. Transcription Factor (TF)-gene regulatory network construction

The JASPAR database (http://jaspar.genereg.net/) [26], accessed via the NetworkAnalyst 3.0 platform [27], was utilized to construct a co-regulatory network of transcription factors (TFs) associated with COPD diagnostic biomarkers. Visualization of this network was performed using Cytoscape software. Based on twelve identified COPD diagnostic biomarkers, we identified transcription factors from the JASPAR database that regulate COPD-related pathways and gene expression levels, leading to the development of a TF gene regulatory network.

2.8. Immune infiltration analysis

We employed the CIBERSORT algorithm to identify the immune cell composition within the COPD gene expression matrix. Using the CIBERSORT software, we calculated the proportions of 22 immune cell types between the COPD and control groups, visualizing the differential expression of immune cells through bar and box plots. The sum of the percentages of the 22 immune cell types in both groups was constrained to 1, with 1,000 simulations conducted and significance set at p < 0.05 [28].

2.9. Mendelian randomization (MR) analysis

We conducted Mendelian Randomization (MR) analysis to identify pivotal genes for COPD, assessing their potential as pathogenic and therapeutic candidates to enhance the model’s therapeutic efficacy [13]. The instrumental variables (SNPs) for ITGB2 and HNRNPAB protein expression were derived from large-scale human plasma proteome quantitative trait locus (pQTL) data from the UK Biobank and deCODE genetics proteome database. The R package “TwoSampleMR” [29] facilitated a two-sample MR analysis to investigate the causal relationship between gene expression (exposure) and COPD risk (outcome). We retrieved the deCODE database (https://www.decode.com/summarydata/) protein quantitative trait loci (pQTL) as exposure data, this study selected FinnGen alliance of COPD outcomes data (R10 release, on October 13, 2024) [30]. The analysis focused on the “COPD” phenotype, which comprised 166,401 Finnish adult participants, including 20,066 cases and 338,303 controls. Adjustments were made for sex, age, the first ten principal components, and genotyping batches.

In accordance with MR assumptions [31], we selected single nucleotide polymorphisms (SNPs) significantly associated with protein expression, applying a threshold of p < 5 × 10^ − 8. These SNPs served as instrumental variables (IVs) for the two-sample MR analysis. To mitigate linkage disequilibrium (LD), we excluded SNPs with an LD-R^2 greater than 0.01 within a 10,000 Kb window. The exposure and outcome data were subsequently harmonized. Five methods were employed to assess the causal relationship between protein expression and COPD risk: MR Egger, weighted median, inverse variance weighted (IVW), simple mode, and weighted mode, with the IVW method as the focal approach. We also utilized MR-pleiotropy residual sum and outlier (MR-PRESSO) to detect any biased SNPs. Additionally, heterogeneity and pleiotropy tests were performed on the results. A p-value < 0.05 was deemed statistically significant.

2.10. Gene Set Variation Analysis (GSVA)

Gene Set Variation Analysis (GSVA) is an unsupervised, non-parametric method that evaluates the enrichment of gene sets based on pathway activity [32]. We utilized the gene set “c2.cp.kegg.symbols” from the Molecular Signatures Database (MSigDB) as a reference for gene pathways. To evaluate the association between ITGB2/HNRNPAB expression and pathway activity, linear regression was performed, and t-values were obtained to indicate the direction and strength of the correlation. The positive t-value indicates a positive correlation between gene expression and pathway activity, while the negative t-value indicates a negative correlation between gene expression and pathway activity. The R package GSVA was employed to score pathway enrichment in COPD and control samples. P-values were calculated to assess statistical significance. Pathways with p < 0.05 were considered significantly correlated. All results were presented with t-value and p-value for complete interpretability.

2.11. Preparation of CSE and cell preparation

The CSEr9-050m basic culture medium 1640 was prepared by the principle of negative pressure filtration and placed in a conical flask. Ten Diamond brand cigarettes were placed in a smoke generator. The smoke produced by cigarettes is filtered through the culture medium in a conical flask by vacuum negative pressure filtration. The resulting solution is CSE. The pH value is adjusted to 7.0, and after filtration and sterilization, it is stored at low temperature. This CSE is defined as 100%CSE with a concentration of 1. MLE-12 was purchased from the National Experimental Cell Resource Sharing Platform. The cells were cultured in a 5%CO cell incubator with MLE-12 medium, 10% fetal bovine serum and 1% double antibody. Passage was performed when the cells grew to the logarithmic phase.

2.12. Establishment of an animal model of COPD

Mice were randomly divided into a control group and a COPD model group. The COPD model was established in the model group by combining exposure to cigarette smoke with intratracheal instillation of lipopolysaccharide (LPS): daily smoke exposure for 2 hours (30 cigarettes per session), 6 times per week, for 8 consecutive weeks; LPS (50 μg/mouse) was administered via intratracheal instillation on days 1, 14, and 28 of the modeling period. Compared with the control group, the COPD mouse model must meet the following core criteria to be considered successfully established [1–3]: ① Behavioral characteristics: lethargy, dull and yellowish coat; rapid breathing with audible wheezing; significantly reduced activity; decreased appetite; heightened stress response (irritability); ② Body weight changes: Significantly slowed weight gain; body weight at the end of the modeling period was markedly lower than that of the control group; ③ Pathological changes in lung tissue: Demonstrated typical pathological features of emphysema and chronic bronchitis, specifically manifested as alveolar wall rupture, alveolar fusion, and alveolar space enlargement (significant increase in alveolar cross-sectional area); thickened airway walls and narrowed lumens; increased mucus secretion within the airways; ④ Pulmonary function parameters: Measurements using a small-animal spirometer showed that ventilatory function parameters, including forced expiratory volume in 0.1 seconds (FEV0.1), forced expiratory volume in 0.05 seconds (FEV0.05), forced vital capacity (FVC), and vital capacity (VC), were all significantly reduced. Using impaired lung function and pathological changes associated with emphysema as core indicators can validate the effectiveness of the model.

2.13. Lung function test

MeMice were anesthetized via intraperitoneal injection of 1.25% tribromoethanol (Avertin) at a dose of 0.2 mL/10 g body weight. Following the induction of anesthesia, the mice were secured in a supine position on the operating table. The fur on the neck was shaved, and the skin was disinfected with povidone-iodine. A longitudinal incision was made along the midline of the neck, and the subcutaneous tissue and sternocleidomastoid muscle were bluntly dissected. After exposing the trachea, a horizontal incision was made between the tracheal cartilage rings. A tracheal tube was inserted and secured, and the distal end of the tube was connected to a flow-pressure sensor and a small animal ventilator. Using the BUXCO small animal pulmonary function testing system, the following parameters were measured to comprehensively evaluate airway ventilation function: tidal volume (TV), minute ventilation (MV), airway resistance (RL), dynamic lung compliance (Cdyn), forced vital capacity (FVC), forced expiratory volume in 50 ms (FEV50), forced expiratory volume in 100 ms (FEV100), FEV50/FVC, FEV100/FVC, peak expiratory flow (PEF), and maximum mid-expiratory flow (MMEF). The forced expiratory parameters were obtained using the system’s built-in negative pressure suction.

2.14. HE staining

Sections were dewaxed, rehydrated, and subjected to heat-induced antigen retrieval with EDTA buffer. Following blocking, sections were incubated with primary antibody overnight and fluorophore-conjugated secondary antibody. Nuclei were counterstained with DAPI, autofluorescence was quenched, and sections were mounted with anti-fade medium for fluorescence microscopy.

2.15. Real-Time Fluorescent PCR

Total RNA was extracted from a COPD model prepared using TRIzol reagent (EZB-RN4, Suzhou, China) from mouse alveolar epithelial cells (MLE-12). One microgram of total RNA was reverse transcribed into cDNA using a reverse transcription kit (A0010CGQ, Suzhou, China). Mouse β-actin served as the internal control, with primer sequences shown in Table 1. Finally, the SYBR qPCR mixture was used for qRT-PCR analysis on a fluorescence quantitative PCR analyzer (QG-9600, Hangzhou, China). Amplification conditions were as follows: Hot start enzyme activation at 95°C for 5 min; PCR reaction at 95°C for 10 s, 60°C for 30 s, repeated for 40 cycles. Relative mRNA expression levels were calculated using the comparative CT method (2^^-ΔΔCt).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Primer sequence.

https://doi.org/10.1371/journal.pone.0349338.t001

2.16. Immunohistochemistry

Sections underwent dewaxing, rehydration, and heat-induced antigen retrieval with citrate buffer. After blocking endogenous peroxidase and serum, sections were incubated with primary antibody overnight followed by HRP-conjugated secondary antibody. Staining was visualized with DAB, counterstained with hematoxylin, dehydrated, cleared, and mounted.

2.17. Immunofluorescence

Sections were dewaxed, rehydrated, and subjected to heat-induced antigen retrieval with EDTA buffer. Following blocking, sections were incubated with primary antibody overnight and fluorophore-conjugated secondary antibody. Nuclei were counterstained with DAPI, autofluorescence was quenched, and sections were mounted with anti-fade medium for fluorescence microscopy.

2.18. Statistical analysis

R software version 4.2.2 was used to conduct all statistical analyses in this study, and a p-value less than 0.05 was considered statistically significant. Data were analyzed using GraphPad Prism 7.0 (GraphPad Software, San Diego, CA, USA). Differences between groups were compared using one-way analysis of variance (ANOVA), assuming data were normally distributed and had equal variances. Statistical significance was set at P < 0.05.

2.19. Ethics statement

This research was conducted using publically available datasets. No potentially identifiable human images or data is presented in this study.

3.1. Identification of differentially expressed genes (DEGs) for COPD

In this study, DEGs associated with COPD were identified by combining the datasets GSE37768 and GSE38974, and the boxplot and PCA plot after removing batch effects and merging the two datasets were shown in S1 Fig. Using the thresholds of |log FoldChange| ≥ 0.585 and adjusted p-value<0.05, 292 DEGs were identified, comprising 181 upregulated and 111 downregulated genes (Fig 2).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Differential expression analysis of GSE37768 and GSE38974 combined data setst.

(A) The asymptotic volcano plot of gene expression in the combined dataset of GSE37768 and GSE38974 showed the distribution of all DEGs, with the top ten DEGs specially marked. (B) The heatmap of DEGs in the combined GSE37768 and GSE38974 datasets (n  =  292, adjusted p-value  <  0.05, |log₂FoldChange|  ≥  0.585) illustrated the expression patterns of the top 50 DEGs.

https://doi.org/10.1371/journal.pone.0349338.g002

3.2. Identification of COPD associated gene modules by Weighted gene co-expression network analysis (WGCNA)

In this study, we employed the WGCNA package (version 1.72–1) in R to construct a co-expression network based on the combined datasets GSE37768 and GSE38974. A soft threshold of 7 was selected to fit a scale-free network with maximum average connectivity, achieving a scale-free R² of 0.95 (Fig 3A). Using the dynamictreecut method, we identified six co-expression modules, each containing over 60 genes (Fig 3B). Among these six significant modules (Fig 3C), the yellow, blue, brown, and gray modules exhibited a positive correlation with COPD, while the green and turquoise modules demonstrated a negative correlation. Notably, only the turquoise module (p = 0.02) met the criterion of p < 0.05, leading us to consider its genes as significantly relevant for further analysis. A significant moderate positive correlation was observed between module membership and gene significance in the turquoise module (r = 0.5, p < 1e-40), confirming its critical role in the potential pathogenesis of COPD (Fig 3D). Ultimately, we identified a total of 753 co-expressed genes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Identification of significant modules and genes of GSE37768 and GSE38974 combined datasets by WGCNA.

(A) Network topology analysis with different soft thresholds. (B) A cluster dendrogram in specific colors reveals six co-expressed gene modules, each with over 60 genes. (C) Correlation between disease groupings and gene modules. (D) Correlation between module membership and gene significance in the turquoise module.

https://doi.org/10.1371/journal.pone.0349338.g003

Subsequently, we intersected the differentially expressed genes from the combined datasets GSE37768 and GSE38974 with the genes from the co-expression modules, resulting in the identification of 112 common genes associated with COPD (S2 Fig).

3.3. An integrated framework based on machine learning to predict COPD

To predict COPD using a machine learning-based integrative framework, we employed the combined GSE37768 and GSE38974 datasets as the training cohort, while GSE212331, GSE148004, and GSE1650 served as independent test cohorts for validation. A total of 12 machine learning algorithms were applied, resulting in 113 model combinations (Fig 4A). This approach identified 12 key COPD genes: CD300C, GNG13, HNRNPAB, ITGB2, LDB2, LTC4S, MCOLN3, NAIP, NOV, PDAP1, S100B, and SMAD7 (S1 Table), with corresponding gene expression boxplots shown in Fig 4B. Among these, the combination of glmBoost and plsRglm demonstrated robust diagnostic performance, with an average AUC score of 0.789 and an overall AUC value of 0.989 (Fig 4A, 4C), indicating strong predictive capabilities.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Performance of Machine Learning Models in Predicting COPD and Expression Patterns of Key Biomarkers.

(A) Comparison of AUC scores for various machine learning model combinations across training and test cohorts (GSE148004, GSE1650, and GSE213331), with the glmBoost+plsRglm combination achieving the highest AUC of 0.989 (95% CI: 0.966–1.000). (B) Expression boxplots of 12 key COPD biomarkers (CD300C, GNG13, HNRNPAB, ITGB2, LDB2, LTC4S, MCOLN3, NAIP, NOV, PDAP1, S100B, SMAD7) between control and COPD groups. These biomarkers were identified as significant in differentiating between the two groups. (C) ROC curve showing the diagnostic performance of the best machine learning model (glmBoost+plsRglm) with an AUC of 0.989.

https://doi.org/10.1371/journal.pone.0349338.g004

3.4. Gene Ontology (GO), KEGG Pathway Analysis

We performed GO and KEGG pathway analyses of COPD-related molecular mechanisms using the “clusterProfiler” package in R. Clustering analysis of the common genes associated with COPD revealed results for three GO categories: biological processes (BP), cellular components (CC), and molecular functions (MF), as well as KEGG pathways. The top 10 terms for each GO category and KEGG pathway are summarized in Table 2. BP terms were primarily enriched in processes such as regulation of long-term synaptic depression, regulation of dendritic cell differentiation, negative regulation of molecular mediator production in immune response, cellular defense response, and tissue migration. In the CC category, actin-based cell projections, filopodia, collagen trimers, cell leading edges, and basement membranes were significantly associated with COPD common genes. MF analysis highlighted amyloid-beta binding, GTPase inhibitor activity, inhibitory MHC class I receptor activity, myosin heavy chain binding, and S100 protein binding as key molecular functions linked to COPD (Fig 5A, 5B).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Top 10 GO category, GO and KEGG pathways and their corresponding P-values and common genes of COPD.

https://doi.org/10.1371/journal.pone.0349338.t002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Functional enrichment analysis of COPD common gene sets.

(A) Ring diagram of GO enrichment analysis. (B) Dot plot of GO enrichment analysis. Higher p value indicated a higher number of genes involved in this GO ontology. (C) Identification of KEGG enrichment analysis results.

https://doi.org/10.1371/journal.pone.0349338.g005

Notably, the top five pathways identified in the KEGG analysis of this study were the Relaxin signaling pathway, Tryptophan metabolism, Cytoskeletal regulation in muscle cells, AGE-RAGE signaling pathway in diabetic complications, and Glycosphingolipid biosynthesis—ganglio series. These pathway results are illustrated in Fig 5C.

3.5. Construction of Transcription factor (TF)-gene regulatory network

Using the JASPAR database’s TF binding site profiles, we constructed a TF-gene regulatory network. This network was built based on 12 key COPD diagnostic biomarkers (CD300C, GNG13, HNRNPAB, ITGB2, LDB2, LTC4S, MCOLN3, NAIP, NOV, PDAP1, S100B, and SMAD7), as illustrated in Fig 6. The resulting network comprised 58 nodes and 82 edges, combining 12 seed genes and 46 TFs. Notably, ITGB2 was regulated by nine TFs, while HNRNPAB is controlled by five. Notably, the diamond-shaped transcription factors FOXC1 and NFIC regulated five key genes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Transcription factor-gene regulatory network in COPD.

The circular nodes represent key COPD genes, while the adjacent nodes indicated the transcription factors that regulate these genes.

https://doi.org/10.1371/journal.pone.0349338.g006

3.6. Immune infiltration analysis

We further applied the CIBERSORT algorithm to estimate immune cell infiltration between the COPD and normal groups. The proportion of 22 kinds of immune cells in samples of COPD disease group and control group was shown in Fig 7A. Compared to the control group, the infiltration of Plasma cells, Monocytes, Macrophages M0, Macrophages M1, activated Mast cells, and Neutrophils significantly increased in the COPD group. In contrast, T cells CD8, activated NK cells and Eosinophils showed a significant decrease (Fig 7B).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Immune Cell Infiltration Profiles and Correlation with Key COPD Biomarkers.

(A) Relative proportions of 22 immune cell types in COPD patients and control subjects, as predicted by the CIBERSORT algorithm. (B) Boxplots showing the expression of 12 key COPD biomarkers (CD300C, GNG13, HNRNPAB, ITGB2, LDB2, LTC4S, MCOLN3, NAIP, NOV, PDAP1, S100B, SMAD7) across immune cell types. (C) Heatmap depicting correlations between the key biomarkers and various immune cell types, revealing significant associations between gene expression and immune cell infiltration (correlation coefficients shown).

https://doi.org/10.1371/journal.pone.0349338.g007

To further investigate the relationship between gene expression and immune cell infiltration in the COPD microenvironment, we conducted a correlation analysis between multiple genes and immune cell types. The heatmap in Fig 7C illustrates the correlations between various genes, such as CD300C, GNG13, and HNRNPAB, and immune cells including neutrophils, eosinophils, M1 and M2 macrophages, activated NK cells, and memory B cells. Notably, HNRNPAB exhibited a strong negative correlation with plasma cells (correlation coefficient < 0.4) and a positive correlation with naive B cells. Additionally, ITGB2 showed significant negative correlations with T cells CD4 memory resting, NK cells resting, and Mast cells resting, while demonstrating positive correlations with T cells gamma delta and Macrophages M0.

3.7. Causal relationship between IGTB2 and HNRNPAB in COPD

We employed Mendelian randomization (MR) to infer causal relationships between 12 COPD biomarkers and COPD risk. Initially, SNPs associated with the proteome of these biomarkers were extracted and aligned with SNPs linked to COPD outcomes. Using the IVW method, we selected results with the highest statistical power (P < 0.05) while ensuring consistency in OR directionality. Among these, IGTB2 and HNRNPAB met the criteria, with four SNPs identified for IGTB2 and six for HNRNPAB (Fig 8A). Notably, a genetic predisposition for higher expression of ITGB2 and HNRNPAB significantly increased COPD risk (OR > 1, P < 0.05, Fig 8B). A sensitivity analysis was performed to ensure stability of the MR Results, which confirmed no heterogeneity or horizontal pleiotropy in this study.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Mendelian randomization method to identify the causative COPD genes.

(A) Scatter plot of COPD pathogenic genes (HNRNPAB and ITGB2) and COPD. (B) Inverse variance weighting (IVW) was used as the primary method to access the two-way causal relationship between COPD pathogenic genes (HNRNPAB and ITGB2) and COPD.

https://doi.org/10.1371/journal.pone.0349338.g008

3.8. GSVA pathway analysis of IGTB2 and HNRNPAB

To further explore the potential biological functions of ITGB2, we performed GSVA combined with KEGG pathway enrichment analysis, which quantifies the activity of biological pathways by aggregating the expression levels of genes within each pathway.

Through this integrated analysis, we evaluated the association between ITGB2 gene expression and the activity of various biological pathways. Fig 9A highlighted the significant pathways associated with ITGB2, including immune signaling and metabolic pathways (all p < 0.05). Notably, ITGB2 expression was upregulated in the B Cell Receptor Signaling Pathway, Fc epsilon RI Signaling Pathway, and Fc Gamma Receptor Mediated Phagocytosis, indicating its potential role in modulating B cell function and antibody-dependent cellular cytotoxicity. Additionally, ITGB2 expression was associated with significant changes in the activity of multiple metabolic pathways, such as the biosynthesis of unsaturated fatty acids, PPAR signaling, ether lipid metabolism, other glycan degradation, tryptophan metabolism, and sphingolipid metabolism (with positive t-values)(Table 3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. The pathway information in the GSVA analysis along with their corresponding t-values and p-values.

https://doi.org/10.1371/journal.pone.0349338.t003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 9. Gene Set Variation Analysis.

(A) Gene Set Variation Analysis pathway of HNRNPAB. (B) Gene Set Variation Analysis pathway of ITGB2.

https://doi.org/10.1371/journal.pone.0349338.g009

Similarly, for the HNRNPAB gene, we conducted the same GSVA combined with KEGG pathway enrichment analysis to investigate its association with pathway activity. GSVA-based KEGG enrichment analysis revealed distinct differences in pathway activity associated with HNRNPAB expression across immune and metabolic pathways. Fig 9B demonstrated significant immune-related upregulation in pathways such as apoptosis, p53 signaling, Cytosolic DNA Sensing Pathway, and Cytokine-Cytokine Receptor Interaction. Conversely, HNRNPAB was downregulated in pathways like Wnt signaling, ABC transporters, Basal Cell Carcinoma, Melanogenesis, and Hedgehog signaling. In terms of metabolism, HNRNPAB showed upregulation in the sphingolipid metabolism, arginine and proline metabolism, galactose metabolism, and amino sugar and nucleotide sugar metabolism pathways (positive t-values), while displaying a significant downregulation in the propanoate metabolism pathway (all p < 0.05) (Table 3).

3.9. Cell experiments verified that ITGB2 and HNRNPAB are key genes for COPD

The expression of ITGB2 and HNRNPAB mRNA was detected in mouse lung alveolar epithelial cells (MLE-12) from the COPD model. Compared with the blank group, both ITGB2 and HNRNPAB expression significantly increased in the COPD group (P < 0.01)(Fig 10A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 10. The expression of HNRNPAB and ITGB2 in the cell model in the Control group and the COPD group.

(A) Quantitative real-time PCR analysis of HNRNPAB and ITGB2 mRNA expression levels in control and compound-treated groups. Data are presented as mean ± SEM; *p < 0.05, **p < 0.01 vs. Control. (B) Immunohistochemical (IHC) staining showed ITGB2 and HNRNPAB in tissue sections from control and COPD groups. (C) Immunofluorescence (IF) staining showed subcellular localization of HNRNPAB and ITGB2 in control and compound-treated cells. Nuclei were counterstained with DAPI. Merged images illustrate colocalization.

https://doi.org/10.1371/journal.pone.0349338.g010

Compared with the control group, the IHC results of COPD showed that the average optical density of ITGB2 and the expression of HNRPAB protein were both decreased. At 40x magnification, three random fields of view were selected, and the average optical density values for relevant protein expression were statistically analyzed. Compared with the control group, ITGB2 expression decreased in the COPD group (P < 0.01), showing a statistically significant difference. The mean optical density value of HNRPA in the COPD group decreased significantly (P < 0.01)(Fig 10B). It should be noted that lung tissue sections contain a mixture of cell types, including epithelial cells, immune cells, and stromal cells, whereas MLE-12 cells represent a pure alveolar epithelial cell line. Therefore, the protein signals in tissue-based analyses reflect the combined contribution of multiple cell populations and may differ from those observed in cultured epithelial cells alone.

Compared with the control group, the mean fluorescence intensity value of ITGB2 protein positive expression decreased in the COPD group (P < 0.05). Compared with the control group, the mean fluorescence intensity value of HNRPAB protein positive expression decreased in the COPD group (P < 0.05)(Fig 10C). Notably, the mRNA levels of ITGB2 and HNRNPAB were elevated, whereas their protein levels detected by IHC and IF were decreased in the COPD model. This discrepancy may reflect post-transcriptional regulation or differences between the epithelial cell line and mixed lung tissue, and is further addressed in the Discussion.

3.10. Animal studies have confirmed that ITGB2 and HNRNPAB are key genes in COPD

To confirm successful establishment of the COPD mouse model, we assessed pulmonary function (Table 3) and performed histopathological examination (Fig 11A). Compared with controls, cigarette smoke‑exposed mice exhibited typical obstructive ventilatory dysfunction, with significantly increased respiratory frequency (F) and decreased tidal volume (TV, P < 0.001). Prolonged inspiratory and expiratory times (Te, P < 0.05) together with reduced peak inspiratory flow (PIF, P < 0.01) indicated airway obstruction and increased resistance. Minute ventilation (MV) increased compensatorily, but effective alveolar ventilation did not improve. H&E staining (Fig 11B) revealed marked alveolar destruction and enlargement in model mice, confirming emphysematous changes. Collectively, these results demonstrate that chronic cigarette smoke exposure successfully established a COPD mouse model with obstructive ventilatory dysfunction and emphysema. To investigate the expression changes of ITGB2 and HNRPAB in the COPD mouse model, we performed double immunofluorescence staining to detect the protein levels and co-localization of these two molecules in lung tissues from both groups (Fig 11C). In the control group, lung tissues exhibited high expression levels of both ITGB2 and HNRPAB, with a substantial proportion of double-positive cells. In contrast, the COPD group showed significantly reduced expression levels of ITGB2 and HNRPAB, accompanied by a marked decrease in the percentage of double-positive cells. Quantitative analysis (Table 4) further revealed that the mean optical density of ITGB2 decreased from 0.051 in the control group to 0.032 in the model group, with positivity rates declining from 34.00% to 25.02%. Similarly, the mean optical density of HNRPAB dropped from 0.057 to 0.025, with positivity rates decreasing from 61.87% to 34.74%. These results demonstrate that chronic cigarette smoke exposure significantly downregulates both ITGB2 and HNRPAB protein expression and reduces their co-expression in lung tissues of COPD mice, suggesting that these molecules may be involved in the immunoinflammatory regulatory mechanisms associated with COPD pathogenesis.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Pulmonary function data.

https://doi.org/10.1371/journal.pone.0349338.t004

Download:

• PNG
  larger image
• TIFF
  original image

Fig 11. The expression levels of HNRNPAB and ITGB2 in the animal model in the control group and the chronic obstructive pulmonary disease group.

(A)The result of H&E. (B) IHC staining showed ITGB2 and HNRNPAB in tissue sections from control and COPD groups. (C) IF staining revealed the subcellular localization of HNRNPAB and ITGB2 in the animal samples of the control group and the COPD group.

https://doi.org/10.1371/journal.pone.0349338.g011