Identification of Anoikis-related potential biomarkers and therapeutic drugs in chronic thromboembolic pulmonary hypertension via bioinformatics analysis and in vitro experiment

Abstract

There is growing evidence that programmed cell death plays a significant role in the pathogenesis of chronic thromboembolic pulmonary hypertension (CTEPH). Anoikis is a newly discovered type of programmed death and has garnered great attention. However, the precise involvement of Anoikis in the progression of CTEPH remains poorly understood. The goal of this study was to identify Anoikis-related genes (ARGs) and explore potential therapeutic drugs for CTEPH. Differentially expressed genes were identified by limma and weighted gene co-expression network analysis (WGCNA) packages, and functional analyses were conducted based on the differentially expressed genes. Subsequently, a combination of protein-protein interaction (PPI), Least Absolute Shrinkage and Selection Operator (LASSO), and Support Vector Machine Recursive Feature Elimination (SVM-RFE) methodologies was employed to screen hub genes associated with CTEPH, which were further verified by dataset GSE188938, quantitative real-time polymerase chain reaction (qRT-PCR) and Western blot. CIBERSORT was utilized to evaluate the infiltration of immune cells and the relationship between infiltration-related immune cells and ARGs. Finally, targeted drug analysis and molecular docking were used to predict drugs targeting Anoikis process to treat CTEPH. Thirty-two differentially expressed genes related to Anoikis and CTEPH were screened through WGCNA analysis. Then, the key ARGs FASN, PLAUR, BCL2L1, HMOX1 and RHOB were screened by PPI, Lasso and SVM-RFE machine learning. Validation through dataset GSE188938, qRT-PCR, and Western blot analyses confirmed HMOX1 and PLAUR as powerful and promising biomarkers in CTEPH. In addition, CIBERSORT immunoinfiltration revealed that Mast_cells_activated and Neutrophils were involved in the pathological regulation of CTEPH. Correlation analysis indicated that HMOX1 was positively correlated with Neutrophils, while PLAUR was negatively correlated with Mast_cells_activated. Finally we used targeted drug analysis and molecular docking to identify that STANNSOPORFIN as a potential drug targeting HMOX1 for the treatment of CTEPH. HMOX1 and PLAUR emerge as potential biomarkers for CTEPH and may influence the development of CTEPH by regulating Anoikis. Mast_cells_activated and Neutrophils may be involved in Anoikis resistance in CTEPH patients, presenting novel insights into CTEPH therapeutic targets. STANNSOPORFIN is a potential agents targeting Anoikis process therapy for CTEPH.

Keywords: Anoikis; CTEPH; Diagnostic markers; Immune infiltration; Machine learning; Therapeutic drugs.

Fig. 1

Flowchart of the multistep screening and verification strategy on bioinformatics data and in vitro experiment. CTEPH chronic thromboembolic pulmonary hypertension, DEGs differentially expressed genes, WGCNA weighted gene co-expression network analysis, PPI network protein–protein interaction network, LASSO least absolute shrinkage and selection operator, SVM-RFE support vector machine recursive feature elimination, qRT-PCR quantitative reverse transcription polymerase chain reaction.

Fig. 2

Identification of differentially expressed genes in CTEPH by limma and WGCNA packages. (A) PCA diagram of the DEGs expression. Red represents gene up-regulation and blue represents gene down-regulation. (B) Scale independence diagram. (C) Average connectivity graph. The soft threshold is 22. (D) Hierarchical clustering tree of MRNA expression patterns of 18 samples. Sample aggregation, no outlier samples. (E) Correlation diagram between modules. (F) Dendrogram of all differentially expressed genes clustered based on different similarity measures. Turquoise, gray, blue and brown are the four modules with the highest degree of clustering. (G) Heat map of the characteristic genes of the module. (H) Scatterplot of correlation between modules and clinical traits. Gray (r = 0.78, p = 1.5e−61), blue (r = 0.63, p = 2.7e−17). (I) Intersection gene map of limma package and WGCNA package.

Fig. 3

Identification of differentially expressed Anoikis-related genes and functional enrichment analysis. (A) Venn diagram of DEGs and Anoikis to get 32 Anoikis-related genes. (B) The clustering heatmap of the expression pattern ofdifferentially expressed Anoikis-related genes. The abscissa represents samples, and the abscissa represents genes. Red for gene up-regulation and blue for gene down-regulation. (C) GO enrichment analysis. (D) KEGG enrichment analysis. (E) Immunologic Signatures enrichment analysis. (F) Reactome enrichment analysis. The Y axis represents the pathway name. The X-axis represents the ratio of the number of genes enriched to the target pathway to the total number of target genes. The color represents the adjusted p-values, the redder the color or the smaller the p value, the more significant the enrichment. The larger the bubble, the more genes present in each pathway.

Fig. 4

PPI network and machine learning. (A) PPI network. The number of lines between the two proteins represents the strength of the interaction. (B) 5 × CV Accuracy of SVM-RFE machine learning. (maximal accuracy = 0.86). The X axis is the number of features and the Y axis represents the accuracy of the curve change after 5 times cross-validation. 6–0.86 means that the accuracy rate of 6 features is 0.86. The closer the accuracy is to 1, the higher the accuracy. (C) 5 × CV Error of SVM-RFE machine learning. (minimal RMSE = 0.14). The X axis represents the number of features, and the Y axis represents the error rate of curve changes after 5 times cross-validation. 6–0.14 indicates that the error rate of 6 features is 0.14. The closer the accuracy is to 0, the lower the error rate. (D) Lasso machine learning. Lasso machine learning obtained 7 genes that contributed more to the model prediction through regression analysis and cross-validation. (E) Venn diagram of Lasso and SVM-RFE machine learning. PPI network protein–protein interaction network, LASSO least absolute shrinkage and selection operator, SVM-RFE support vector machine recursive feature elimination.

Fig. 5

Verification of PLAUR and HMOX1 by dataset GSE188938, qRT-PCR and Western blot. (A) Expression of HMOX1, PLAUR, RHOB, FASN and BCL2L1 in CTEPH and Control group of dataset GSE188938. (B,C) Validation of the expressions of potential diagnostic markers via qRT-PCR, n = 5. (D–F) Validation of the expressions of potential diagnostic markers via Western blot, n = 3. *p < 0.05; **p < 0.01. qRT-PCR quantitative reverse transcription polymerase chain reaction.

Fig. 6

Immune infiltration analysis. (A) The proportions of the 22 immune cells in each sample. The X axis represents each sample, and the Y axis represents the proportion of different immune cells. The 22 colors represent 22 types of immune cells. (B) The correlation heat map of 22 immune cells. (C) Box-type plot of immune cell infiltration differences.

Fig. 7

Correlation analysis between diagnostic markers and infiltration-related immune cells and construction of diagnostic markers-targeted drugs network. (A) Correlation analysis between PLAUR and infiltration-related immune cells. (B) Correlation analysis between HMOX1 and infiltration-related immune cells. (C) Scatter diagram indicating the correlation between PLAUR expression and Mast cells activated. The expression of PLAUR was negatively correlated with Mast_cells_activated (r =  − 0.502, p = 0.034). (D) Scatter diagram indicating the correlation between HMOX1 expression and Neutrophils. The expression of HMOX1 was positively correlated with Neutrophils (r = 0.538, p = 0.021). (E) PLAUR and HMOX1-targeted drugs network. Red represented diagnostic markers and light blue represented the drugs. (F) Molecular docking of diagnostic markers- targeted drugs. The binding energies of HMOX1-STANNSOPORFIN is − 2.51 kcal/mol.