| dc.description.abstract | Circulating cell-free RNA (cfRNA) holds significant potential for the minimally invasive assessment and monitoring of atherosclerosis, a disease characterised by plaque formation in the arterial walls. While plasma cfRNA provides a snapshot of real-time gene expression, its lack of cellular context makes it challenging to discern whether observed changes are due to shifts in cell type proportions or to intrinsic transcriptomic alterations. This distinction is particularly important in atherosclerosis, where disease progression and outcomes are heavily influenced by various cell types involved. Deep learning-based cellular deconvolution of bulk RNA sequencing data from cfRNA, using single-cell RNA-sequencing data as a reference, facilitates the prediction of the cellular origins underlying the measured cfRNA. Leveraging the Tabula Sapiens, a multi-organ single-cell RNA atlas, this study validated its suitability as a reference for cfRNA deconvolution, demonstrating its ability to yield accurate deconvolution predictions for both solid organs and plaque samples. Plaque deconvolution identified macrophage content as a predictor of major adverse cardiovascular events (MACE) (HR [95% CI] = 1.27 [1.07, 1.52], P = 0.01). Cellular deconvolution analysis of cfRNA revealed a protective association between hepatocyte- and monocyte-derived cfRNA with atherosclerosis severity (β [95% CI] = -0.30 [-0.56, -0.06], P = 0.02) and MACE risk (HR [95% CI] = 0.69 [0.51, 0.95], P = 0.02), respectively. Despite these findings, the accuracy of predicted cell type identities decreased with increasing tissue heterogeneity in the bulk RNA-sequencing data. This was particularly evident in the deconvolution of cfRNA, where its inherently complex nature posed significant challenges to precise cell type identification. To overcome the current limitations of circulating cfRNA deconvolution, further research is required, which could unlock its clinical potential and ultimately improve patient outcomes. | |
