Oxidative stress-associated genes TPPP3 and VEGFA in COPD revealed by bulk and single-cell sequencing analysis

Abstract

Chronic obstructive pulmonary disease (COPD) is characterized by chronic airway inflammation and is closely linked to oxidative stress. This study aimed to identify and validate key oxidative stress-related genes and pathways involved in COPD using integrated bioinformatics and experimental approaches. Public COPD datasets were obtained from the Gene Expression Omnibus (GEO) database, and oxidative stress-related genes were retrieved from the GeneCards database. Differentially expressed genes (DEGs) were screened and analyzed for functional enrichment. Machine-learning algorithms, including Least Absolute Shrinkage and Selection Operator (LASSO) regression and Random Forest, were used to identify hub genes and evaluate diagnostic value by calculating the area under the receiver operating characteristic (ROC) curve (AUC). Single-cell RNA sequencing (scRNA-seq) data were analyzed to determine the distribution of hub genes across different cell types. Finally, a COPD combined oxidative stress cell model was established using human bronchial epithelial cells (BEAS-2B), and key gene expression was experimentally validated. We identified 76 overlapping genes associated with both COPD and oxidative stress, mainly enriched in necroptosis, JAK-STAT, MAPK, and related pathways. 12 hub genes were screened using machine-learning methods. Single-cell analysis showed that TPPP3 and VEGFA were predominantly expressed in epithelial cells. Experimental validation confirmed the bioinformatics predictions at the gene level. This study identified and validated 12 oxidative stress-related hub genes in COPD, highlighting TPPP3 and VEGFA as key genes enriched in epithelial cells and potentially involved in tissue remodeling. These findings not only provide insights for exploring new therapeutic strategies but may also serve as potential diagnostic biomarkers or candidate therapeutic targets for COPD.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37375-4.

Introduction

Chronic obstructive pulmonary disease (COPD) is an incompletely reversible lung disease characterized by chronic inflammation and airflow limitation¹. COPD is currently the third leading cause of death worldwide and is a global burden², but the primary mechanism of action is still unclear. Oxidative stress, defined as an imbalance between the production of reactive oxygen species (ROS) and the body’s ability to utilize antioxidants to counteract their harmful effects^3,4, has been increasingly recognized as a central contributor to COPD pathogenesis. Substantial evidence from mechanistic studies has established oxidative stress as a key driver in both disease initiation and progression^5–10, highlighting its critical role in COPD pathophysiology.

Cigarette smoking is a major cause of COPD, and induces oxidative stress in airway epithelium^11–13. As oxidative stress drives inflammatory response¹⁴ and COPD is fundamentally a chronic inflammatory disease, oxidative stress may be the key mechanistic link to COPD-associated inflammation^5,15.

The role of oxidative stress in COPD pathogenesis is well-established. However, conventional studies are often limited by inadequate sample sizes. To address this, we employed single-cell sequencing and machine-learning methods. These approaches facilitate the analysis of large cohorts through sophisticated computational integration^16,17. The diagnostic gold standard for COPD is lung function testing¹⁸. However, recent evidence and updated guidelines highlight that many high-risk patients with symptoms and CT-detected structural lung damage fall outside spirometry-defined COPD, and that severe disease or poor cooperation can limit spirometry’s reliability, underscoring the need for broader diagnostic criteria^19–21. Therefore, we obtained data from a large number of human samples in public databases and utilized multiple computational approaches to find relevant genes and pathways. To empirically prove the results, we added cell experiments to evidence the results. We hope our results effectively identify potential biomarkers for COPD diagnosis and provides valid targets for subsequent studies.

Materials and methods

Datasets preparation

The COPD datasets were obtained from the Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/) database, including GSE76925, GSE38974, and GSE173896. The mRNA expression profile dataset GSE76925 contained 151 smokers, with 111 patients with COPD and 40 controls, while the GSE38974 dataset had 59 smokers, including 50 COPD patients and 9 controls. To ensure balanced groups and manageable data volume, the control samples GSM5282543, GSM5282544, and GSM5282545 and COPD samples GSM5282537, GSM5282538, and GSM5282539 were randomly selected from the scRNA dataset GSE173896. The oxidative stress dataset was derived from GeneCards (https://www.genecards.org).

Identification of differentially expressed genes (DEGs)

The GSE76925 and GSE38974 datasets received data normalization and probe annotation using the R software (version 4.0.1) ‘limma’ (linear models for microarray data) and ‘GEOquery’ packages. Then, we combined datasets GSE76925 and GSE38974 to form a data cohort and corrected the batch effect among platforms using the Surrogate Variable Analysis (SVA) package in R.The efficacy of this correction was assessed and visualized by principal component analysis (PCA) before and after the procedure (Figure S2) DEGs were defined based on the screening criteria of adjusted p-value < 0.05 and |log fold change (FC)|> 0.5. The oxidative stress-related genes were gained from GeneCards (https://www.genecards.org/). Finally, we identified the intersection of oxidative genes and DEGs as differentially expressed oxidative-related genes. Venn diagrams were conducted by the ‘VennDiagram’ R package to visualize differentially expressed oxidative-related genes.

Functional enrichment analysis

Using the R packages ‘clusterProfiler’ and 'org.Hs.eg.db’ analyses of Kyoto Encyclopedia of Genes and Genomes (KEGG) (www.kegg.jp/kegg/kegg1.html ) and Gene Ontology (GO) were carried out, with a statistically significant difference defined as p < 0.05 ²². We conducted visualization at the online website (https://www.bioinformatics.com.cn).

Screening and validation of hub genes

For the identification of hub genes, we performed two machine-learning algorithms, including the Random Forest algorithm and LASSO based on differentially expressed oxidative-related genes. We calculated the average modeling error rate for all genes using the R package ‘random forest’. Then we constructed the random forest model and calculated the dimensional importance values using the Gini coefficient method²³. Then, the candidate hub genes were crossed using the ‘glmnet’ package of R software, and LASSO analysis was performed to screen the final hub genes²⁴.We ensured reproducibility through detailed reporting of hyperparameters and random seeds.The area under the receiver operating characteristic (ROC) curve (AUC) were used to assess the diagnostic performance of the hub genes.The AUCs of the identified hub genes, along with their 95% CIs, are presented in Supplementary Table 1.

Single-cell data analysis

We utilized the Seurat R package (version 3.0.2) and applied standard downstream processing to analyze the scRNA-seq data (https://github.com/satijalab/seurat). Data were filtered based on the following criteria for homogeneity, including the number of features (high threshold: 2500; low threshold: 200), proportion of hemoglobin (high threshold: 0.003), and proportion of mitochondrial RNA (high threshold: 0.01). The data were then normalized by applying the “LogNormalize” method. We conducted Principal Component Analysis (PCA) for unsupervised clustering and unbiased visualization of cell populations on 2D maps. After that, we chose t-Distributed Stochastic Neighbor Embedding (t-SNE) and uniform manifold approximation and projection (UMAP) as the non-linear dimensionality reduction methods. Subsequently, the “FindClusters” function was utilized to identify clusters. Subsequently, the ‘SingleR’ package (version 1.0.0) was used for marker-based cell type annotation.

Cell experiment

Cell culture and treatment

Human bronchial epithelial cells (BEAS-2B) were cultured using the Bronchial Epithelial Cell Medium (BEpiCM; ScienCell, California, USA) at 37 °C, 5% CO₂, and 95% air humidity. Normal control (NC) group: BEAS-2B without intervention; Cigarette smoke extract (CSE) group: BEAS-2B cells were treated with 2.5% CSE for 24 h.

Preparation of CSE

The study reference cigarette 3R4F (Louisville, KY, USA) was used in the experiments. A standard preparation of one cigarette incorporated into 5 mL of BEpiCM was prepared by using a vacuum pump at a constant rate (5 min per cigarette smoked). The CSE solution was subsequently filtered (0.22 μm; Merck Millipore, SLGS033SS) to remove insoluble particles and designated as 100% CSE solution. Standardization was performed for quality control by monitoring the 320 nm absorbance, and between 0.8 and 1 was considered as the 100% CSE solution.

Detection of inflammation factors and oxidative stress-related factors

ROS detection

Following a 24-h incubation with CSE, the medium was removed, and the cells were washed three times with phosphate buffered saline (PBS). The 2’,7'-dichlorodihydrofluorescein diacetate (DCFH-DA) probe from the Reactive Oxygen Detection Kit (Beyotime Biotechnology) was diluted 1000-fold in fresh medium. This working solution was then added to the petri dishes, and the cells were incubated for 20 min. Subsequently, the DCFH-DA-containing medium was removed and the cells were washed three times with PBS. Finally, the culture medium was added to observe the fluorescence intensity on a fluorescence microscope. The whole process was protected from light.

Detection of inflammatory factors and other oxidative stress-related factors

The levels of inflammatory factors and other oxidative stress-related factors were detected using commercial enzyme-linked immunosorbent assay (ELISA) kit for superoxide dismutase (SOD), catalase (CAT) (Nanjing Jiancheng Biological Engineering Research Institute, Nanjing, China), Interleukin-1β (IL-1β), Interleukin-6 (IL-6), Interleukin-8 (IL-8), and tumor necrosis factor-α (TNF-α) (BioTNT, Shanghai, China) in cell supernatants.

Quantitative real-time PCR (qRT-PCR)

We extracted RNA from BEAS-2B cells after CSE intervention for 24 h. Total RNA was reverse transcribed into complementary DNA (cDNA) per the manufacturer’s instructions (Takara Bio, Japan), and the A260/280 ratio was used to determine cDNA purity. The expression of TPPP3 and VEGFA was determined using quantitative polymerase chain reaction (qPCR) (primer sequences are presented in Table 1). The relative mRNA expression of target genes was normalized to the level of the GAPDH in each group.

Table 1.

qRT-PCR primer sequences.

Gene name	Forwards primers 5’–3'	Reverse primers 5’–3'
TPPP3	AAGTCTGCTCGGGTCATCAAC	GAGCCCGTGTATCTGCTGG
VEGFA	AGGGCAGAATCATCACGAAGT	AGGGTCTCGATTGGATGGCA

Statistical analysis

Data were analyzed and plotted using GraphPad Prism 8.0 software (San Diego, CA, USA). Data are presented as mean ± standard error of the mean (SEM). To compare normally distributed data, we used a t-test. The Kruskal–Wallis test was used to compare data that did not conform to normal distribution. A p-value < 0.05 was considered statistically significant.

Results

Identification of DEGs and screening hub genes

A total of 76 DEGs were screened, including 45 downregulated and 31 upregulated genes. The visualizations were performed through Venn diagrams (Fig. 1). Gene selection via Boruta algorithm (Random Forest classifier; Mean Decrease Gini (MDG) > 1 threshold) identified 26 features. LASSO logistic regression identified 32 candidate genes. Finally, 12 overlapping genes from both methods were defined as hub genes for downstream analysis (Fig. 2).

Fig. 1.

(A) The volcano plots show the differentially expressed genes between COPD and normal group; (B) Intersection of upregulated and (C) downregulated differentially expressed genes with oxidative stress genes.

Fig. 2.

Machine-learning approaches to discovering hub genes. (A) and (B) Visualization of LASSO; (C). and (D). Visualization of Random Forest; (E). Intersection of LASSO and Random Forest with 12 common genes.

Functional enrichment analysis

To further investigate the localization, function, composition, and associated pathways of the DEGs, we conducted GO (Fig. 3) and KEGG pathway (Fig. 4) analyses.

Fig. 3.

Bar plot showing significantly enriched Gene Ontology (GO) terms in Biological Process (BP), Cellular Component (CC), and Molecular Function (MF) categories. The y-axis (Count) indicates the number of genes associated with each term.

Fig. 4.

KEGG enrichment analysis. (www.kegg.jp/kegg/kegg1.html ).

KEGG pathway analysis revealed several key pathways implicated in COPD pathogenesis(Supplementary table2). For instance, the cytokine-cytokine receptor interaction pathway promotes chronic inflammation, immune dysfunction, and airway remodeling, which are central to COPD²⁵. The calcium signaling pathway performs a vital function in the inflammation and regulation of various cellular processes, such as muscle contraction, neurotransmitter release, gene expression and cell growth. In the context of COPD, disturbances in calcium signaling can contribute to the pathophysiology of the disease^26,27. Necroptosis, as a new focus of cell death research, is closely associated with oxidative stress and has been gradually increasing in COPD^28,29.The PI3K-AKT pathway, a classical pathway of oxidative stress, was also enriched in this study^30–32.

Identification of hub gene expression levels

We employed ROC curves to assess the diagnostic sensitivity and specificity of the 12 hub genes. The AUC for each gene was calculated, and all values > 0.6, demonstrating robust discriminatory power for COPD (Fig. 5).

Fig. 5.

ROC curves and AUC of 12 hub genes. (A) ACE; (B) BCHE; (C) CARS2; (D) EDN1; (E) G3BP1; (F) MAPT; (G). MRAP; (H). NPM1; (I). PTGIS; (J). SCN4B; (K). TPPP3; (L). VEGFA.

Single-cell data analysis

We completed the annotation of 9 major cell types: Endothelial cells, Epithelial cells, Monocytes, Macrophages, NK cells, B cells, T cells, Tissue stem cells, and Smooth muscle cells. At the single-cell level, TPPP3, PTGIS, NPM1, VEGFA, and EDN1 significantly differed in the COPD and control groups (Figure S1). The distribution of individual genes in different cells was also significantly different. PTGIS, ACE, and EDN1 were significantly distributed in endothelial cells, while TPPP3 and VEGFA were significantly distributed in epithelial cells (Fig. 6). We also examined alterations in intercellular co-expression and the ROS pathway between the two genes. The orange color represents gene expression levels, green indicates ROS pathway enrichment levels, and yellow denotes the degree of co-expression between the genes and the ROS pathway. This indicates that both genes show significant correlation with ROS in epithelial cells.

Fig. 6.

(A). The annotation of cell types between the control and COPD group; (B). Expression of 12 hub genes in different cells; (C). and (D). Expression of TPPP3, VEGFA, and ROS and co-determination with ROS in NC and COPD groups.

Indicators of oxidative stress and inflammation

Observing ROS under fluorescence microscopy, the fluorescence intensity of the CSE group was significantly stronger than that of the control group, indicating that the ROS content was upregulated. SOD and CAT were significantly downregulated in the CSE group, and together with the rise in ROS, an oxidative-antioxidative imbalance was seen. IL-1β, IL-6, IL-8, and TNF-α are significantly higher in the CSE group. The dysregulation of these inflammatory factors is consistent with COPD inflammatory characteristics. These inflammatory factors and oxidative-antioxidant factor imbalances are considered to be criteria for success in combined oxidative stress models of COPD^33,34.(Fig. 7).

Fig. 7.

(A). Expression of ROS under fluorescence microscopy; (B). Statistical plot of ROS for NC and CSE Groups; (C). (D). ELISA detection of indicators of oxidative stress; (E. F. G. H). ELISA detection of inflammation indicators; (I. J). Gene level expression of TPPP3 and VEGFA in NC and CSE groups.(n = 3).

Expression Levels of TPPP3 and VEGFA in BEAS-2B Cells.

As shown, TPPP3 and VEGFA were significantly upregulated in the CSE group and were statistically different.(Fig. 7).

Discussion

There have been numerous reports of oxidative stress in COPD³⁵. There are studies that reveal antioxidant diets and treatments are effective in improving COPD patients^36–38. Nevertheless, there is a lack of large-scale data analysis of COPD and oxidative stress to demonstrate a correlation. Current studies related to COPD and oxidative stress have focused on the Keap1-Nrf2 pathway, the PI3K-AKT pathway, and the MAPK pathway, etc.^39–42 This study introduces several novel methodological approaches that differentiate it from previous COPD-omics studies. While prior research typically relies on standard differential expression and clustering analyses of bulk RNA-seq data, our study integrates bulk RNA-seq and single-cell RNA-seq data with machine-learning techniques to identify oxidative stress-related hub genes in COPD. To our knowledge, this integrated pipeline has not been previously applied to investigate oxidative stress mechanisms in COPD. The results of the study showed that 12 hub genes were included, including SCN4B, MRAP, TPPP3, PTGIS, CARS2, BCHE, G3BP1, NPM1, ACE, MAPT, VEGFA, and EDN1. Among the 12 hub genes. We focused on two genes, TPPP3 and VEGFA, as they were associated with tissue regeneration and remodeling^43–46, which is vital in the pathophysiology and development of COPD. TPPP3, also known as tubulin polymerization-promoting protein family member 3, is involved in the regulation of microtubule dynamics and induces microtubule protein polymerization and microtubule bundling⁴⁷. A study has shown that TPPP3 knockdown inhibits the growth of tumor cells⁴⁸. Some studies have shown that TPPP3 can proliferate tissues and increase the occurrence of fibrosis^48,49. However, there is little published research on the association of TPPP3 with COPD. VEGFA is associated with angiogenesis, vascular permeability inflammation, and inflammation^50–52. VEGFA is an essential factor for angiogenesis and was also shown to enable cell proliferation and migration^53,54. And which is vascular proliferation is closely related to remodeling in pulmonary disease⁵⁵. One study showed that VEGFA was higher in all COPD groups than in the normal group ⁵⁶, which is consistent with our study. Moreover, angiogenesis was necessary for remodeling⁵⁷. Therefore, we hypothesized that the upregulation of VEGFA in our study was closely related to the formation of airway remodeling in COPD. TPPP3 and VEGFA were found to be predominantly distributed in epithelial cells³⁸. Airway remodeling and fibrosis occurred predominantly in epithelial cells^58,59. In the present research, TPPP3 and VEGFA were found to be significantly upregulated in COPD samples and our experimental validation confirmed the same result. There is currently no effective treatment to halt or reverse airway remodeling. Our findings identify the potential biomarker for this process, which could also serve as a novel candidate therapeutic target.

Cigarettes are the leading causative agent of COPD⁶⁰. Numerous literatures have reported that CSE-induced cells as a model of COPD^61–63. To simulate the COPD state, we detected important inflammatory factors of COPD and essential indicators of oxidative stress, including reactive oxygen species, SOD enzymes, IL-6, IL-1β, TNF-α, etc. to determine the effect of the model, which proved to be successful. We utilized the model to validate the bioinformatics results in the subsequent experiments and found that the cellular model supported our bioinformatics results. Our results can be confirmed by the validation of large samples and human-derived cells, providing instrumental value for subsequent clinical markers of COPD with basic research. However, limitations exist in our study. For example, although COPD is known to have distinct phenotypes, our research did not focus on investigating these specific COPD phenotypes. The experiments were conducted using only a single cigarette smoke extract (CSE) concentration and exposure duration (2.5% for 24 h). Furthermore, validation of TPPP3 and VEGFA was performed only at the mRNA level, without confirmation at the protein level—such as by Western blot, immunofluorescence, or ELISA—and no functional assays (e.g., proliferation, migration, fibrosis, or angiogenesis) were carried out. Additionally, the findings lack validation in animal models. Deeper investigations are also needed; for instance, examining the phenotypic effects of knocking out these hub genes in relevant COPD models would be valuable. Future studies will extend this work. First, the expression and functional role of TPPP3 and VEGFA in airway remodeling will be validated at the protein level (e.g., by Western blot/immunohistochemistry) and through functional assays (e.g., assessing proliferation, migration, or angiogenesis) in cellular and animal models of COPD. Second, the in vivo necessity of these genes will be examined by creating knockout models to assess their impact on disease-relevant phenotypes. Finally, to strengthen clinical relevance and translational potential, we will pursue tissue-level validation (e.g., using IHC/ISH on patient lung samples) and evaluate their presence as potential biomarkers in minimally invasive clinical specimens (e.g., plasma or sputum).

Conclusion

We utilized COPD datasets and oxidative stress-related genes from public databases and found the DEGs for functional enrichment analysis. We then derived 12 hub genes through various machine computational methods and dug deeper into the distribution of the hub genes at the single-cell level. Finally, we used CSE to establish a cellular COPD and oxidative stress model to validate the above bioinformatics results.

Abbreviations

COPD

Chronic obstructive pulmonary disease

ROS

Reactive oxygen species

DEGs

Differentially expressed genes

FC

Fold change

GO

Gene ontology

KEGG

Kyoto encyclopedia of genes and genomes

LASSO

Least absolute shrinkage and selection operator

ROC

Receiver operating characteristic

AUC

Area under the curve

scRNA-seq

Single-cell RNA sequencing

PCA

Principal component analysis

t-SNE

T-distributed stochastic neighbor embedding

UMAP

Uniform manifold approximation and projection

BP

Biological process

CC

Cellular component

MF

Molecular function

CSE

Cigarette smoke extract

PBS

Phosphate buffered saline

DCFH-DA

2',7'-Dichlorodihydrofluorescein diacetate

SOD

Superoxide dismutase

CAT

Catalase

IL-1β

Interleukin-1β

IL-6

Interleukin-6

TNF-α

Tumor necrosis factor-α

ELISA

Enzyme-linked immunosorbent assay

JAK-STAT

Janus kinase-signal transducer and activator of transcription

qRT-PCR

Quantitative real-time PCR

cDNA

Complementary DNA

qPCR

Quantitative polymerase chain reaction

SEM

Standard error of the mean

MDG

Mean decrease gini

SVA

Surrogate variable analysis

Author contributions

Wenglam Choi conceived of the presented idea and performed the computations and, experimentation. Yueren Wu assisted with calculations and article writing as well as code handling. Wenjing Chen and Yuting Shi assisted in laboratory operations and data processing. Weihang Luo and Xinyuan Wu assisted experimental operation. Zhenhui Ruan collected references and data. Jingcheng Dong provided guidance and article correction. All authors discussed the results and contributed to the final manuscript.

Funding

This study received financial support from the National Natural Science Foundation of China (No.81973631), The Science and Technology Commission of Shanghai Municipality (YDZX00003008), Basic research on kidney tonic formula for intervention in several airway inflammatory diseases (three projects)(2022QD056), 2023 Ground High Construction Project—First-class Integrative Medicine/Summit Discipline Construction—Cultivation of Provincial and Ministerial Key Laboratories of Integrative Medicine (DGF601013), and NDRC Equipment Reform Special Project-DK07 Basic Technology Platform for Scientific Research in the Disciplines of Integrative Medicine (DKF132001).

Data availability

The datasets and codes used or analyzed during the current study are available from the corresponding author or first author upon request.

Declarations

Competing interests

The authors declare no competing interests.