Introduction

Aortic dissection (AD) is characterized by a tear in the inner aortic layer or bleeding of the aortic wall, eventually leading to aortic wall rupture; due to its abrupt onset and poor prognosis, it is regarded as one of the most dangerous cardiovascular emergencies1. According to the Stanford classification, AD is classified into types A and type B based on whether the ascending aorta is involved. The incidence of AD ranges from 3/100,000 to 5/100,000 per year; however, it might have been underestimated as many patients have died outside the hospital due to delayed presentation2. AD mainly manifests as sudden onset of severe pain or tearing sensation in the anterior chest or the upper back, great difference in blood pressure between upper extremities, as well as neurological and ischemic symptoms; most patients also present with unspecific symptoms. At present, the pathogenesis of AD has not been fully elucidated and it is believed that this condition is the result of a combination of genetic, anatomical, hemodynamic, molecular biological and immunological factors3. Extracellular matrix (ECM) degradation in the aortic media and oxidative stress have become the focus of studies on AD.

The pathogenesis of AD is associated with the degradation of ECM in the aortic media. Matrix metalloproteinases (MMPs) are a family of endopeptidases secreted by a variety of cells such as vascular endothelial cells and smooth muscle cells, and can destroy the connective tissue of the arterial wall by degrading a variety of components in the ECM and basement membrane4. Among them, MMP-2 and MMP-9 are closely related to the onset and development of AD, as they promote the formation of aortic dissection by degrading elastin and collagen, leading to massive disintegration of the aortic media, structural damage, and apoptosis of smooth muscle cells5,6. Oxidative stress also plays an important role in the development of AD. In oxidative stress, reactive oxygen species (ROS) can inhibit the binding of Fibulin-5 to elastin, leading to abnormal structure and dysfunction of elastic fiber7, Fibulin-5 is a key protein in the regulation of elastic fiber synthesis, and its defect is the morphological basis for the development of AD8. Hypoxia also promotes the up-regulation of MMP-2 and MMP-9 expression through the HIF-1a pathway, which is involved in the formation of aortic dissection9. At present, the majority of studies on AD are still focusing on clinical diagnosis and treatment; however, researchers have started to look at the role of molecular mechanism in the early diagnosis, optimization of treatment and improvement of prognosis for AD.

Anoikis is a special mode of programmed cell death that occurs in normal anchorage-dependent cells after detachment from the ECM. It is mainly mediated by extrinsic and intrinsic pathways10. In the extrinsic pathway, corresponding receptor proteins are activated by TNF-α, FasL, and TRAIL and further bind to these cell factors via the associated death domain, resulting in the caspase cascade11; in the intrinsic pathway, the Bcl-2 protein family activates the caspase cascade through mitochondria, which promotes apoptosis12. At present, the research on anoikis mainly focuses on the field of oncology, such as the association between anoikis resistance and the metastasis of lung cancer13 and the role of anoikis resistance in the progression and metastasis of melanoma14. In recent years, the identification of diagnostic markers of AD using bioinformatics techniques and the relationship between extracellular matrix degradation, oxidative stress, smooth muscle cell phenotypic transformation and the pathogenesis of AD have become hotspots in studies of this condition. In the present study, we analyzed the anoikis-related genes (ARGs) of AD using bioinformatics methods, in order to inform further research and clinical interventions.

Materials & methods

Data acquisition

All data used in the present study were obtained from GEO Data Sets (https://www.ncbi.nlm.nih.gov/geo/). Data with the type of Expression profiling by array were screened out to find relevant cohorts containing cases of aortic dissection and normal controls, which were cohorts GSE52093 ,GSE190635 and GSE153434. The probe IDs of the matrix was converted to gene symbols according to the annotation files of GPL10558 ,GPL570 and GPL20795, respectively, and the three cohorts were integrated into one cohort for subsequent analysis.

Acquisition of differential ARGs

We performed differential analysis of ARGs expression between cases of AD and healthy controls using the limma package (Rversion 4.1.5) to obtain significant differential ARGs, and performed differential intergene correlation analysis using the corrplot package (Rversion 4.1.8) to generate the heat map of the correlation.

Enrichment analysis

We performed Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis of the differential ARGs using the clusterProfiler package (Rversion 4.1.5). Based on this up-to-date R package, we conducted Gene Set Enrichment Analysis (GSEA) of the subsequently selected TP53 and TUBB3 genes, and Gene Set Variance analysis (GSVA) enrichment analysis of the two genes using the GSVA package (Rversion 4.1.5). Functional pathway enrichment analysis was also performed for the differential ARGs using the Metascape database (metascape.org/gp/index.html).

Machine learning

We applied LASSO regression and Support Vector Machine Recursive Feature Elimination (SVM-RFE) in parallel to establish a machine learning model for the screening of ARGs. By identifying the intersection of the signature genes selected by both LASSO regression and SVM-RFE, we obtained the ARGs with the highest diagnostic value for aortic dissection. These genes are regarded as potential biomarkers for this medical condition.

Receiver operating characteristic analysis

The pROC package (Rversion 4.1.5) was used to perform receiver operating characteristic analysis of selected signature genes and plot receiver operating characteristic curve, and validation filtering of aortic dissection signature genes was performed with the criterion of Area Under Curve (AUC) being greater than 0.5.

Results

Identification of differentially expressed genes (DEGs) in aortic dissection

Using cohorts GSE52093 and GSE190635, which contained both cases of AD and healthy controls, we found 82 statistically significant DEGs related to anoikis, including 56 upregulated and 26 downregulated genes (Fig. 1A). The top five upregulated and downregulated genes are shown, and the correlation of all differential genes was presented in Fig. 1B.

Fig. 1
figure 1

Identification of differentially expressed genes in aortic dissection.(A) Heatmap of expression of ARGs. ***p <0.001, **p <0.01, *p <0.05. (B) Heatmap of co-expression among ARGs.Abbreviation: Anoikis anchorage-dependent cell death, ARGs anoikis-related genes.

Full size image

Enrichment pathway analysis of ARGs

To further explore the mechanism underpinning the impact of anoikis on AD, we performed GO enrichment analysis, KEGG enrichment analysis, and metascape. The GO analysis showed that the impact of ARGs was mainly associated with changes in oxygen content levels and with collagen-containing extracellular matrix and plasma membrane signaling receptor complex (Fig. 2A-C). The KEGG enrichment analysis revealed that ARGs were also significantly associated with PI3K-Akt signaling pathway, focal adhesion, lipids and atherosclerosis, as well as apoptosis (Fig. 2D, E). The enrichment analysis of differential genes using the Metascape online database showed that these genes were mainly enriched in processes such as apoptosis signaling pathways, regulation of cellular response to hypoxia, and regulation of cellular response to stress (Fig. 2F-H). This suggests that anoikis might be involved in the development and progression of AD through oxidative stress and changes in extracellular matrix components.

Fig. 2
figure 2

Enrichment analysis of ARGs. (A) Color barplot of GO enrichment analysis. (B) Barplot of GO enrichment analysis. (C) Bubble chart of GO enrichment analysis. (D) Barplot of KEGG enrichment analysis. (E) Bubble chart of KEGG enrichment analysis. (F) Network diagram of enrichment analysis by cluster. (G) Network diagram of enrichment analysis by P value. (H) Color barplot of enrichment analysis. (KEGG pathway data used in (D) and (E) are sourced from the KEGG database: www.kegg.jp/kegg/kegg1.html

Full size image

Identification of hub genes using machine learning algorithm

Two validated machine learning algorithms (LASSO, SVM-RFE) were applied to identify hub genes from ARGs associated with AD. In both methods, we used gene expression values as features and the binary disease status (AD vs. normal) as labels. SVM-RFE is widely used for classification and regression analysis; its model has nonlinear discriminant characteristics, which allows for the comparison of results obtained from models involving different numbers of variables and the selection of the best combination of variables. The SVM-RFE algorithm iteratively eliminated features based on their contribution to the classification accuracy, and a total of 19 DEGs were screened out (Fig. 3A, B). LASSO algorithm is a regression analysis commonly used to improve prediction accuracy; it belongs to the family of linear regression models, and performs feature selection by shrinking coefficients of less important features to zero. Using the default ten-fold cross-validation to optimize the penalty parameter, three genes were identified using LASSO regression (Fig. 3C, D) analysis. These genes showed stable non-zero coefficients across different penalty values, indicating their strong predictive power. Ultimately, three overlapping (via intersection) genes, i.e., GRSF1, TP53 and TUBB3, were identified in a screening model based on the two machine learning algorithms (Fig. 3E), demonstrating their robust selection by both linear (LASSO) and nonlinear (SVM-RFE) methods, suggesting that the three genes were of high diagnostic value for AD.

Fig. 3
figure 3

Identification of hub genes using machine learning algorithms. (A) Cross-validation plots of lasso regression. (B) Graph of lasso regression. (C) Graph of SVM-RFE cross-validation error. (D) Graph of SVM-RFE cross-validation accuracy. (E) Venn plot of lasso and SVM-RFE analyses.

Full size image

Validation of hub genes

The ROC curve analysis of our machine learning model in the training dataset (GSE52093, GSE190635) demonstrated excellent diagnostic capability (Fig. 4A). Additionally, GRSF1, TP53, and TUBB3 showed strong discriminative power between AD cases and healthy controls, with AUCs of 0.919, 0.906, and 0.889, respectively (Fig. 4B). To validate these findings, we assessed both the machine learning model and the three genes’ diagnostic performance in validation datasets (GSE153434). The machine learning model maintained robust performance in the validation sets (Fig. 4C), and the ROC analysis for individual genes (Fig. 4D-F) confirmed their strong diagnostic value, with AUCs approaching 1. Furthermore, differential expression analysis revealed significant differences in GRSF1, TP53, and TUBB3 expression levels between normal and AD tissues in the validation datasets (Fig. 4G-I), further substantiating our findings from the training cohort.

Fig. 4
figure 4

Validation of hub genes. (A) ROC curve of the machine learning model. (B) ROC curve of the hub genes. (C) ROC curve of the machine learning model in the validation set. (D – F) ROC curve of GSRF1, TP53, TUBB3 in the validation set. (G - I) Boxplot of differential expression for GSRF1, TP53, TUBB3 in the validation set.

Full size image

Potential signaling pathway of hub genes

The protein-protein interaction (PPI) network analysis was performed on key genes related to anoikis in AD (Fig. 5A - B). To further investigate the functional implications of our identified genes, we conducted gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA) on TP53 and TUBB3, while acknowledging that a comprehensive functional analysis of GRSF1 remains to be conducted in future studies. The result of single-gene GSEA for TP53 showed that TP53 was associated with processes such as cell chromosome segregation and sister chromatid segregation, while the GSVA analysis showed no significantly regulated gene set pathways (Fig. 5C,D). The result of single-gene GSEA for TUBB3 showed that this gene was associated with metaphase and anaphase, and GSVA analysis revealed 8 positively regulated gene set pathways and 6 negatively regulated gene set pathways (Fig. 5E,F). These findings provide insights into the potential mechanisms of TP53 and TUBB3 in the context of anoikis and AD, while further studies are needed to fully elucidate the role of GRSF1.

Fig. 5
figure 5

Potential signaling pathway of hub genes. (A) PPI plot of TP53. (B) PPI plot of TUBB3. (C) GSEA plot of TP53. (D) GSVA plot of TP53. (E) GSEA plot of TUBB3. (F) GSVA plot of TUBB3.

Full size image

Discussion

AD is a life-threatening cardiovascular disease. Data from a recently updated study have shown that the mortality rate among non-surgical patients who present with type A acute aortic dissection was 0.5% per hour within the first 48 h and 5.8% during the following 48 h15. However, the pathogenesis of AD is still unclear at present, which may delay the diagnosis of AD; thus, it is necessary to find indicators to ensure quick and accurate diagnosis of AD. Anoikis is a form of apoptosis induced by the detachment of anchorage-dependent cells from adjacent cells and/or surrounding matrix, which leads to their incompatibility with the microenvironment16,17. Anoikis resistance has been extensively studied in the field of oncology due to the ability of invasive cancer cells to survive away from the primary site18, while degradation of ECM components, oxidative stress, and apoptosis of smooth muscle cells are closely related to the functions of ARGs in the pathogenesis of AD, therefore, we speculated that certain ARGs might be early diagnostic indicators for AD.

First, we used two cohorts from the GEO database to conduct differential analysis of ARGs expression between cases of AD and healthy controls and obtained significant genes; the findings suggested that most of the genes were up-regulated and only a small number of them were down-regulated in the cases of AD. We then performed GO, KEGG, and Metascape enrichment analyses of DEGs associated with anoikis, which indicated that the functions of ARGs were mainly enriched in cellular response to oxygen levels, regulation of cellular stress response, components of the extracellular matrix, and lipids and atherosclerosis, which might be mechanisms underpinning the association between anoikis and AD.

Studies have shown that intermittent hypoxia (IH) in obstructive sleep apnea syndrome (OSAS) is associated with the pathogenesis of AD, and IH-induced reactive oxygen species (ROS) and hypoxia-inducible transcription factor-1 (HIF-1) can contribute to the harmful consequences of cardiovascular disease19. Experiments have demonstrated that IH can promote the progression of AD through the ROS-HIF-1α-MMP pathway20. Furthermore, macrophages can mediate the HIF1α-ADAM17 pathway through metabolic reprogramming, which promotes the pro-inflammatory response and destruction of elastic fibers, ultimately leading to aggravation of AD21. A recent study suggested that the proportion of M1 macrophages, lipid metabolism indicators such as low-density lipoprotein, high-density lipoprotein, and apolipoprotein, and inflammatory factors such as TNF-α, IL-1β, IL-6, and IL-10 were significantly associated with the development of AD22. Lipid metabolism disorders can eventually lead to thickening and sclerosis of the arterial wall, and rupture of atherosclerotic plaques can lead to intimal tears, leading to the formation of AD23. The aortic intima is composed of elastic fibers and vascular smooth muscle cells (VSMCs), which bind to collagen fibers, proteoglycans, glycosaminoglycans, and various adhesion proteins to form ECM and play a crucial role in the elasticity and tensile strength of the aorta24. Studies have shown that a variety of metabolic abnormalities can eventually result in the development of AD via ECM degradation25,26. Thus, the oxidative stress of cells, changes in components of the extracellular matrix, and lipids and atherosclerosis may play an important role in AD, further supporting our hypothesis that anoikis might lead to the development of AD.

With the use of two machine learning algorithms (LASSO, SVM-RFE), we finally identified three most relevant genes: GRSF1, TP53, and TUBB3, and verified the function of these genes using ROC curve analysis, PPI network, GSEA, and GSVA enrichment analyses. Next, we validated our findings in a new validation set, further confirming the diagnostic value of TP53 and TUBB3. Additionally, we analyzed their functions using PPI network, GSEA, and GSVA enrichment analyses.

Tumor suppressor p53 (TP53) is an important tumor suppressor gene that plays an important regulatory role in apoptosis induction, DNA damage and abnormal proliferation27. Studies have shown that p53 can mediate the protection of acute lung injury by inhibiting the iron death Nrf2/HIF-1/TF signaling pathway28. TP53-induced glycolysis and apoptosis regulator (TIGAR) can reduce the level of fructose-2,6-bisphosphate in cells, leading to glycolysis inhibition and reduction of intracellular ROS, thereby inhibiting microglial pyroptosis and protecting newborns from hypoxic-ischemic brain injury29. The above studies have demonstrated that TP53 can mediate iron death, pyroptosis, and apoptosis of cells through oxidative stress pathways. βIII-tubulin (TUBB3) is a major microtubule (MT) protein; it is a cytoskeletal protein involved in many cell pathophysiological processes, such as maintenance of shape, intracellular transport, mitosis, carcinogenesis, and chemoresistance30, and has been found to inhibit apoptosis and promote the growth and invasion of gallbladder cancer cells through the Akt/mTOR signaling pathway31.

The above experimental studies have indicated that the three genes identified in the present study can promote cell death in different forms through a variety of pathways, with GRSF1 being the most significant one, which is closely related to the pathogenesis of AD. VSMCs play a critical role in the pathogenesis of AD, and their phenotypic transformation promotes pro-inflammatory responses and MMP production, eventually leading to degradation of extracellular matrix and weakening of the aortic wall32. A growing body of evidence suggests that programmed cell death pathways, including apoptosis, necroptosis, pyroptosis, and iron death, play a key role in VSMC loss33; however, there has been few in-depth studies on the anoikis of VSMCs. Based on our current bioinformatics analysis, differentially expressed ARGs have been demonstrated in both cases of AD and healthy controls, and various programmed cell death pathways might jointly promote the development of AD, which might provide new ideas and insights for the diagnosis of AD.

Some limitations to the present study also need to be mentioned. First, this study was conducted primarily using bioinformatic methods and laboratory-based experiments are needed to verify these findings. To offset the above limitation, this study performed rigorous data analysis with reference to a large body of literature to ensure the reliability of results. Secondly, anoikis is a relatively new concept; although research on anoikis in the field of oncology has made significant progress, the specific mechanism of anoikis in the development of AD needs further investigation. To this end, the present study used bioinformatic methods to identify potential biomarkers for AD, which provided a strong basis for further experiments.

Conclusion

Through comprehensive bioinformatic analysis, three ARGs (GRSF1, TP53, and TUBB3) were initially identified, with TP53 and TUBB3 being further functionally validated, and the possible mechanism underpinning the association between anoikis and AD was explored using GO, KEGG, and Metascape analyses. Machine-based learning models and functional validation suggested that TP53 and TUBB3 contributed to the development of AD through the anoikis pathway, while further studies are needed to fully elucidate the role of GRSF1. Hopefully, our analyses helped elucidate the potential mechanism of the pathophysiology of AD and identify new diagnostic indicators.