Integrative multi-cohort bioinformatics and machine learning analysis reveals dysregulation of pro-inflammatory and pro-resolving lipid mediator pathway genes in atherosclerosis
摘要
Atherosclerosis is increasingly recognized as a disease of failed inflammation resolution rather than merely excessive inflammation. Specialized pro-resolving mediators (SPMs) such as resolvin E1 (RvE1) and their balance with pro-inflammatory leukotriene B4 (LTB4) play critical roles in this process. However, the transcriptomic landscape of the genes governing this pro-inflammatory/pro-resolving balance—including SPM synthesis enzymes, lipid mediator receptors, and competing pro-inflammatory pathways—remains incompletely characterized in human atherosclerotic tissue using integrative multi-cohort approaches with experimental validation at the level of circulating lipid mediators. We performed integrative bioinformatics analysis of 50 curated genes spanning the pro-inflammatory and pro-resolving lipid mediator pathways (48–49 of which were successfully mapped on each microarray platform) across three independent transcriptomic cohorts totaling 197 arterial tissue samples: GSE100927 (69 atherosclerotic vs. 35 control peripheral arteries; discovery cohort), GSE43292 (32 carotid atheroma vs. 32 paired macroscopically intact tissue; validation cohort-1), and GSE28829 (16 advanced vs. 13 early atherosclerotic carotid plaques; validation cohort-2). Differential expression analysis was performed with Benjamini-Hochberg FDR correction. Three machine learning algorithms—LASSO regression, Random Forest, and Support Vector Machine-Recursive Feature Elimination (SVM-RFE)—were applied for consensus hub gene identification. Classification performance was evaluated using nested cross-validation and independent cohort validation. Additionally, serum concentrations of pathway end-products (LTB4, RvE1, and EPA) were measured by enzyme-linked immunosorbent assay (ELISA) in an independent clinical cohort of 67 subjects (40 coronary artery disease patients vs. 27 healthy controls). Total study population: 264 samples across four independent cohorts. In the discovery cohort (GSE100927), 34 of 48 (70.8%) analyzed genes were significantly dysregulated (FDR < 0.05, |log2FC| > 0.3). The LTB4 biosynthesis cascade was coordinately upregulated: ALOX5 (log2FC = + 0.77, FDR = 2.25 × 10−13), ALOX5AP (log2FC = + 0.86, FDR = 1.09 × 10−7), and LTA4H (log2FC = + 0.33, FDR = 8.54 × 10−6). SPM receptors CMKLR1/ChemR23 (log2FC = + 0.38, FDR = 4.41 × 10−7) and GPR18/DRV2 (log2FC = + 0.49, FDR = 5.38 × 10⁻¹⁴) were paradoxically upregulated, suggesting potentially compensatory, but possibly insufficient, resolution signaling. These findings were independently confirmed: GSE43292 showed 26/48 significant genes and GSE28829 showed 24/49 significant genes. Five genes (CD36, CD68, HMOX1, ALOX5AP, ALOX5) demonstrated consistent upregulation across all three cohorts. Three ML algorithms identified 8 consensus hub genes (PLA2G2A, CD36, TNF, GPR18, PTGER4, ELOVL5, HMOX1, CD68) achieving AUC = 0.89 (95% CI: 0.82–0.94, sensitivity = 0.85, specificity = 0.82) in the independent GSE28829 validation cohort using nested cross-validation. ELISA validation was concordant with the transcriptomic predictions: serum LTB4 was 4.19-fold elevated (P = 6.92 × 10−31) and the RvE1/LTB4 ratio was significantly reduced (P = 1.15 × 10−8) in patients, consistent with upregulation of the LTB4 biosynthesis pathway at the mRNA level. This multi-cohort, multi-level study provides integrated transcriptomic and biochemical evidence indicating that dysregulation of the pro-inflammatory/pro-resolving lipid mediator axis in atherosclerosis is consistent with impaired inflammation resolution, showing coordinated upregulation of LTB4 synthesis genes alongside potentially compensatory, but possibly insufficient, SPM receptor upregulation, validated across 264 samples in four independent cohorts spanning three measurement platforms and two analytical modalities.