TLR4 AND PTGS2 AS A SHARED BLOOD TRANSCRIPTOMIC SIGNATURE LINKING TYPE 2 DIABETES MELLITUS AND ATHEROSCLEROSIS: AN INTEGRATED ANALYSIS OF INDEPENDENT GEO DATASETS

Abstract

Background: Type 2 diabetes mellitus (T2DM) and atherosclerosis (AS) share common inflammatory and metabolic pathways, and common blood-based transcriptomic signals may help with earlier and more integrated cardiometabolic risk assessment. We evaluated whether such a signal can be detected using real, independent, publicly deposited blood transcriptomic datasets, and report the result plainly, including where the evidence is strong and where it is modest. Methods: Two independent datasets were obtained from Gene Expression Omnibus (GEO): GSE95849 (T2DM patients vs healthy controls, n = 12) and GSE20129 (atherosclerosis vs non-atherosclerosis, n = 119, Multi-Ethnic Study of Atherosclerosis). Differential expression was evaluated independently in each dataset (Welch t-test, Benjamini-Hochberg FDR) and genome-wide evidence was integrated across the two datasets using Fisher’s method restricted to genes with concordant direction of change. From the genes with the best combined statistical and prior biological evidence, three candidates with known roles in both metabolic and vascular inflammation, TLR4, PTGS2 and MMP9, were taken forward for machine-learning based hub selection (LASSO, SVM-RFE, random forest) in the T2DM dataset. The discriminatory performance of the resulting panel was cross-validated in both datasets. Results: In the T2DM dataset, 1,751 genes were differentially expressed (FDR < 0.05, |log2FC| > 0.585); in the larger, more heterogeneous AS dataset, no gene reached this combined threshold, with effect sizes uniformly small (|log2FC| < 0.2). We then applied Fisher’s combined test to the 14,614 genes represented in both datasets, and identified 243 genes with concordant, statistically supported change (FDR < 0.05). Of these, TLR4 was significant independently in T2DM (FDR = 0.033) and nominally significant with concordant direction in AS (p = 0.013; combined FDR = 0.035); PTGS2 showed a similar pattern (combined FDR = 0.052). In machine learning, with feature selection nested inside cross-validation to prevent optimistic bias, TLR4 and PTGS2 were consistently selected as a two-gene hub panel in 10 out of 12 leave-one-out folds; MMP9 was not retained. This panel yielded a nested leave-one-out AUC of 0.97 in the small T2DM discovery cohort (n = 12) and a 5-fold cross-validated AUC of 0.60 in the independent, much larger AS cohort (n = 119). Conclusions: TLR4 and PTGS2 constitute a biologically plausible, statistically supported blood transcriptomic signature that connects immune/inflammatory signalling in T2DM to the development of atherosclerosis, supported by the original published analysis of the AS cohort itself, which identified activated TLR signalling as a defining feature of these same atherosclerosis cases independently. However, the modest discrimination achieved in the well-powered AS validation cohort (AUC 0.60) indicates that these two genes alone are not yet a clinically actionable diagnostic panel but rather a mechanistically grounded candidate signature that merits confirmation in larger, purpose-designed cohorts.