Background: Aldosterone-producing adenomas (APAs) are a common cause of primary aldosteronism that can lead to cardiovascular complications if left untreated. Machine learning-based bioinformatics approaches have emerged as powerful tools for identifying potential disease markers, gaining widespread recognition in biomedical research. We aimed to use machine learning to discover novel biomarkers of APAs to identify new pathophysiological mechanisms.
Methods: We applied 2 machine learning algorithms to published RNA sequencing data to identify APA feature genes. Validation was performed using APA tissue samples, spatial transcriptomics, pig adrenal glands, and in vitro assays in a human adrenocortical cell line.
Results: Machine learning identified ATP2A3 as a key feature gene in APA, and its upregulation in APAs compared with the adjacent cortex was confirmed by spatial transcriptomics. In human adrenocortical cells, angiotensin II treatment increased ATP2A3 gene expression 9.15-fold. Silencing ATP2A3 decreased basal CYP11B2 expression and aldosterone secretion by 3.51-fold and 1.46-fold, respectively, and by 1.77-fold and 1.94-fold under angiotensin II stimulation. Dietary sodium restriction in pigs significantly increased ATP2A3 mRNA and protein levels. Spatial transcriptomics showed that APA cells exhibited higher ATP2A3 gene expression compared with all other adrenal cell types. The suppressive effect of ATP2A3 silencing on CYP11B2 expression was further enhanced by Ca2+ inhibitors.
Conclusions: The ATP2A3 gene is highly expressed in APA and is a key regulator of CYP11B2 expression and aldosterone production.
Keywords: adenoma; adrenal glands; aldosterone; hyperaldosteronism; sodium.