BS18 Development and validation of deconvolution methods for bulk RNA-seq data analysis in human myocardium and skeletal muscle using cibersortx

Single-cell RNA-seq is a vital tool for deciphering patterns of gene expression within the multitude of cell lineages comprising any tissue. However, price and technical complexities limit the use of this technology. Deconvolution of bulk RNA-seq data from whole tissues offers an alternative approach to derive some of these insights and gain valuable contextual information from existing data. Cardiovascular diseases and their risk factors alter myocardial and skeletal muscle biology. We aimed to develop and validate approaches to deconvolute bulk RNAseq data from human right atrium (RA), left ventricle (LV) and skeletal muscle biopsies, allowing inference of the cellular composition of these samples. Methods We used single-cell RNA-seq data from the Heart Cell Atlas project to construct signature matrices encompassing all major cell lineages using the CIBERSORTx deconvolution package. Randomly selecting 200 cells from each cell lineage, we generated a signature matrix for each tissue, including fibroblasts, endothelial cells, lymphoid cells, myeloid cells, neuronal cells, pericytes, smooth muscle cells, and myocytes (defined as atrial cardiomyocytes and ventricular cardiomyocytes in the RA and LV matrices, respectively). Cardiac signature matrices also included adipocytes and neuronal cells, while the skeletal muscle signature matrix also incorporated satellite cells. These matrices can be applied in CIBERSORTx by any scientist wishing to infer cell lineage proportions in their bulk RNAseq data. To assess the accuracy of these signature matrices, we generated 100 synthetic RA, LV and skeletal muscle datasets with varying proportions of each cell lineage and compared their known composition with the composition predicted by CIBERSORTx. To illustrate their value in deconvolution, we applied our signature matrices to publicly available bulk RNA-seq data from the Genotype Tissue Expression (GTEx) project to explore the association of age and sex with cellular composition of these tissues. Results Linear regression analyses revealed excellent agreement between known and predicted composition of synthetic tissues, with R2>0.95 for all cell lineages, and good calibration. Deconvolution of skeletal muscle data from GTEx revealed, amongst other differences, a lower proportion of satellite cells and a higher proportion of fibroblasts with age, both of which have been demonstrated previously using alternative methods, corroborating the accuracy of our approach. Wide-ranging age- and/or sex- differences were noted in all tissues, which warrant further assessment using complementary approaches. Conclusions Our validated CIBERSORTx signature matrices represent user-friendly tools for elucidating how disease processes influence cell lineage composition of human myocardium and skeletal muscle. These allow additional insights to be gained from existing and new bulk RNAseq data, which may help to define important cell lineages for further characterisation using complementary approaches. Conflict of Interest None