Abstract
This study identified novel biomarkers for endothelial dysfunction (ED). These findings may provide both a potential biomarker and therapeutic target for the prevention and treatment of atherosclerosis and coronary artery disease (CAD).
Introduction
Cardiovascular disease (CVD) remains a leading cause of global mortality, often rooted in endothelial dysfunction (ED)—the breakdown of the inner lining of blood vessels. A major driver of this breakdown is Endothelial-to-Mesenchymal Transition (EndMT), a process where endothelial cells lose their identity and transform into fibroblast-like cells that promote scarring and plaque buildup (atherosclerosis). While metabolic stressors like high blood sugar (hyperglycemia) and high cholesterol (hyperlipidemia) are known triggers, the specific molecular "hubs" and the role of long non-coding RNAs (lncRNAs) in regulating this transition have remained unclear. This research seeks to identify reliable biomarkers for early diagnosis and potential therapeutic targets to halt this transition.
Methodology
The study employed a multi-stage workflow combining "wet lab" biology with "dry lab" computational modeling. Human Umbilical Vein Endothelial Cells (HUVECs) were treated with pooled serum from patients with high glucose (GLU), high LDL cholesterol, or high triglycerides (TG) to mimic the complex chemical environment of a patient's blood. Whole-transcriptome RNA sequencing was then performed to identify differentially expressed mRNAs and lncRNAs.
To identify the most significant genes, a "triple-threat" machine learning approach was used, integrating Random Forest (RF) for ranking, SVM-RFE for feature elimination, and LASSO regression to prevent overfitting. The findings were validated using external public datasets (GSE43292, GSE28829, GSE163154), single-cell RNA sequencing (scRNA-seq) of human plaques (GSE253903), and qRT-PCR on blood samples from 100 individuals (50 CAD patients and 50 healthy controls).
Results
The analysis revealed that while high sugar and high fat trigger different initial gene responses, they converge on the same EndMT pathways, specifically involving inflammation and lipid metabolism. A total of 306–364 mRNAs and 471–562 lncRNAs were identified as significantly altered under different metabolic stress conditions.
The machine learning models consistently identified a four-gene signature—CD36, ISG15, HSPB2, and IRS2—as the most reliable indicators of atherosclerosis. This signature achieved an Area Under the Curve (AUC) of 0.997 in the training set and maintained strong performance (AUCs of 0.808 to 0.958) in external validation sets. Furthermore, specific lncRNAs such as LINC02381, VIM-AS1, and ELF-AS1 were found to be significantly upregulated in CAD patients, acting as central nodes in the regulatory network pushing cells toward EndMT. Single-cell data confirmed that IRS2 and ISG15 are specifically altered within the endothelial cells of actual human plaques.
Discussion and Conclusion
This study provides a robust framework for identifying biomarkers of metabolic-induced endothelial dysfunction. The identification of the four-gene signature and key regulatory lncRNAs offers promising targets for early diagnosis and therapeutic intervention. However, limitations include the relatively small clinical cohort and the use of 2D cell cultures, which do not fully replicate the 3D mechanical stresses of living vessels. Future work should focus on larger multi-center trials and mechanistic studies involving gene knockdown or overexpression in animal models to confirm the therapeutic potential of these biomarkers.
Summary updated