Coronary artery disease manifests as an age-dependent molecular phenotype in circulating extracellular vesicles

Background

Omics-based biomarkers for coronary artery disease (CAD) have shown limited performance, raising uncertainty as to whether this reflects the absence of a biological signal or inadequate modelling of baseline physiology.

Purpose

To determine whether CAD-associated extracellular vesicle (EV) molecular signatures are intrinsically age-dependent and therefore obscured in conventional biomarker models.

Methods

We analysed liquid-phase FTIR spectra of serum-derived EVs from 575 participants undergoing coronary CT angiography. Spectra (1,022 wavenumbers) were modelled using elastic net and random forest approaches. We compared unadjusted models with covariate-aware models that incorporated age, sex, and cardiometabolic risk factors and tested covariate×spectral interactions. Performance was assessed using 200-fold bootstrap resampling.

Results

Spectra-only models performed at chance level (AUC 0.53, 95% CI 0.47–0.60), suggesting minimal independent diagnostic signal. Clinical covariates exerted widespread influence on EV spectra (sex, n=167 bands; hypertension, n=112; age, n=50), with extensive covariate×CAD interactions. Incorporating interaction terms markedly improved discrimination (AUC 0.77, 95% CI 0.72–0.82; ΔAUC +0.24, p<0.001), with dominant contributions from age-dependent effects.

Key signals localised to amide II (~1547 cm⁻¹), carbohydrate/phosphate (~1020–1068 cm⁻¹), and lipid (~1477 cm⁻¹) regions, consistent with protein structural changes, metabolic remodelling, and lipid perturbation. These patterns indicate that CAD-associated EV signatures are not independent signals but represent age-dependent molecular shifts amplified in disease.

Conclusion

Extracellular vesicle molecular signatures of CAD are not absent, but age-dependent and therefore obscured in naïve models. CAD appears to manifest as an acceleration of baseline molecular ageing processes in circulating vesicles. These findings provide a mechanistic explanation for the limited performance of omics biomarkers and support age-aware modelling as a prerequisite for clinically meaningful biomarker discovery.

Figure 1. Visualisation of the most influential covariate × CAD interactions selected by the informed random forest model.