Virtual Transcriptomics

Abstract

Objective: Therapeutic advancements in atherosclerotic cardiovascular disease have improved prevention of ischemic stroke and myocardial infarction, but diagnostic methods for atherosclerotic plaque phenotyping to aid individualized therapy are lacking. In this feasibility study, we aimed to elucidate plaque biology by decoding the molecular phenotype of plaques through analysis of computed-tomography angiography images, making a predictive model for plaque biology referred to as virtual transcriptomics. Approach and Results: We employed machine intelligence using paired computed-tomography angiography and transcriptomics from carotid endarterectomies of 40 patients undergoing stroke-preventive surgery for carotid stenosis. Computed tomography angiographies were analyzed with novel software for accurate characterization of plaque morphology and plaque transcriptomes obtained from microarrays, followed by mathematical modeling for prediction of molecular signatures. Four hundred fourteen coding and noncoding RNAs were robustly predicted using supervised models to estimate gene expression based on plaque morphology. Examples of predicted transcripts included ion transporters, cytokine receptors, and a number of microRNAs whereas pathway analyses demonstrated enrichment of several biological processes relevant for the pathophysiology of atherosclerosis and plaque instability. Finally, the ability of the models to predict plaque gene expression was demonstrated using computed tomography angiographies from 4 sequestered patients and comparisons with transcriptomes of corresponding lesions. Conclusions: The results of this pilot study show that atherosclerotic plaque phenotyping by image analysis of conventional computed-tomography angiography can elucidate the molecular signature of atherosclerotic lesions in a multiscale setting. The study holds promise for optimized personalized therapy in the prevention of myocardial infarction and ischemic stroke, which warrants further investigations in larger cohorts.