Interpretable machine learning coupled to spatial transcriptomics unveils mechanisms of macrophage-driven fibroblast activation in ischemic cardiomyopathy

Myocardial infarction (MI) often leads to ischemic cardiomyopathy, which is characterized by extensive cardiac remodeling and pathological fibrosis accompanied by inflammatory cell accumulation. Although inflammatory responses elicited by cardiac macrophages are instrumental in post-MI cardiac remodeling, macrophage microniche-mediated fibroblast activation in MI are not understood. Analyses of the spatial transcriptomics data of the hearts of patients with ischemic cardiomyopathy and a history of MI using a novel workflow combining Significant Latent Factor Interaction Discovery (SLIDE), which is an interpretable machine learning approach recently developed by us, regulatory network inference, and in-silico perturbations unveiled unique context-specific cellular programs and corresponding transcription factors driving these programs (that would have been missed by traditional analyses) in macrophages, and resting and activated cardiac fibroblasts. More nuanced analyses to examine the microniches comprising these cells in failed hearts uncovered additional cellular programs reflective of altered paracrine signaling among these cells. Silencing of niche-specific key genes and TFs from these cellular programs in both mouse and human macrophages altered the expression of pro-fibrotic genes. Furthermore, the secretomes from these macrophages suppressed myofibroblast differentiation. Finally, macrophage-specific in vivo silencing of Tvp23b, Tdrd6, and B3galnt1 and the transcription factors Mafk and Maz, which are differentially expressed in macrophage/activated fibroblast niches, using a novel lipidoid nanoparticle approach in mice with MI significantly improved cardiac function and suppressed fibrosis. Our study uncovers novel macrophage niche-mediated fibroblast activation mechanisms and provides a new generalizable framework, coupling interpretable machine learning, regulatory network inference, in-silico perturbations, and in vitro and in vivo testing.