Exploring Endocrine Signaling between Adipose Tissues and the Heart using Transfer Learning and Induced Pluripotent Stem Cell Models

Abstract

Background: Because of the amount of evidence pointed toward obesity as a co-morbidity of heart disease, our goal in this study is to use a data-driven method to discover potential cross talk signals between adipose and cardiac tissues, which may reveal a physiological reason for this association. For this study, we hypothesized that adipose-cardiac crosstalk signaling can be found by unexpected co-expression between adipose tissue genes encoding secreted proteins and cardiac tissue genes and vice versa. By discovering new crosstalk signals between these two tissues, we can find new associations between heart disease and obesity. Methods: Cross-tissue correlations of gene expression have been used to identify candidate endocrine signals that mediate tissue crosstalk, which are otherwise diffcult to detect in circulation. However, the identification of significant cross-tissue gene co-expression is hindered by immense multiple-testing burden. Our goal here is to abstract gene expression patterns of tissues into eigengenes and then test their correlation with secreted protein coding genes across tissues in a pairwise fashion. We first used the NIH Common Fund Genotype-Tissue Expression data set (GTEx v8) which contains RNA seq gene expression data from matched tissues of over 17,000 transcriptomes, to learn the latent structure of gene expression across tissues. To do so, we used PLIER (pathway level information extractor), a matrix factorization algorithm to extract the latent targets as eigengenes that map to specific biological pathways in the target tissue (e.g., pathways in the heart in the case of adipose to heart crosstalk). The learned gene expression structure is then transferred to a smaller data set with matching donor adipose and heart transcript expression data, from which a correlation matrix is constructed to find significant relationships between source tissue crosstalk signals and target tissue eigengenes and nominate novel crosstalk signals. Results: Adipose to heart tissue cross talk correlations using PLIER derived latent eigengene targets revealed many adipokines that had strong correlations with pathways found in the heart tissue through similar gene expression. Among these strong correlations were eigengene targets that were wrapped in metabolism pathways particularly those of fatty acid, phospholipid, and protein metabolism. Adipokines that showed high significant and positive correlation with cardiac latent variable expression included Adiponectin (R = 0.32, p = 4.7e−13), Glutathione Peroxidase 3 (R = 0.34, p = 5e−15) and Galectin 9 (R = 0.37, p < 2.2e−16) among others. Based on these results, we nominated new potential relationships between adipose and cardiac tissue. Work is currently under way to validate these candidate crosstalk signals using human induced pluripotent stem cell (iPSC)-derived cardiomyocytes. Future Directions In conclusion, we have found strong evidence in silico to suggest a relationship between known secreted proteins from adipose tissue (adipokine) and the expression of important pathways (such as fatty acid and protein metabolism) in the heart. Based on the findings of these correlations between adipokines and latent eigengene pathways within the heart, our future directions will involved the understanding of potential downstream functional consequences mediated by the adipose secreted proteins in iPSC derived models. This project was supported in part by NIH award R03-OD032666 and the University of Colorado Translational Research Scholar Program to E.L. This is the full abstract presented at the American Physiology Summit 2024 meeting and is only available in HTML format. There are no additional versions or additional content available for this abstract. Physiology was not involved in the peer review process.