PSRC Main Site  |  Past & Future Meetings
Plastic Surgery Research Council

Back to 2020 Abstracts


Identifying The Myeloid Subpopulation Responsible For Tissue Fibrosis Across Organ Systems Via Machine Learning Parameterization And Predictive Transcriptomics
David M. Stepien, MD, PhD1, Simone Marini, PhD1, Charles Hwang, BS1, Chase A. Pagani, BA1, Michael Sorkin, MD1, Noelle D. Visser, MS1, Amanda K. Huber, PhD1, Kaetlin Vasquez, MS1, Jun Li, PhD1, Sarah Hatsell, PhD2, Aris Economides, PhD2, Shamik Mascharak, PhD3, Michael T. Longaker, MD3, Benjamin Levi, MD1.
1University of Michigan, Ann Arbor, MI, USA, 2Regeneron Pharmaceuticals, Tarrytown, NY, USA, 3Stanford University, Stanford, CA, USA.

Purpose: Fibrosis secondary to ischemia and musculoskeletal polytrauma is a clinical problem common in plastic surgery with limited effective therapeutics currently available. Until recently, high throughput analyses of fibrosis have been untenable. We developed an automated parameterization of fibrosis histology via machine learning (ML) methods capable of distinguishing healthy from fibrotic muscle, correlated with transcriptomic profiles of infiltrating macrophages following musculoskeletal polytrauma. Furthermore, we established a transcriptomic analysis pipeline of multiple muscle and organ fibrosis models using single cell RNA sequencing (scRNAseq) to allow identification of a “fibrotic” macrophage. We hypothesize that inflammatory and fibrotic myeloid cell phenotypes are similar across musculoskeletal and visceral organ systems and that proteomic and transcriptional profiles can be used to predict homeostasis, injury and fibrogenesis. Methods: Ischemia/reperfusion of the mouse hindlimb with concomitant cardiotoxin injection into the tibialis anterior (IR/CTX) was induced in mice with genetic (LysmCre;Tgfb1fl/fl) and pharmacologic (TGFB1-Fc ligand trap) Tgfb1 deletion. H&E, Masson Trichrome, and picrosirius red histology was performed at 1 week post injury (n=3-4/group). Automated quantification of picrosirius micrographs were performed with a novel image processing pipeline resulting in a panel of 13 relevant fiber and branchpoint parameters and t-SNE dimensional reduction. Separately, we performed 10X scRNAseq on IR/CTX injured tibialis anterior muscle at homeostasis and 3 days post-injury. For comparison studies, IR/CTX macrophages were isolated and aggregated with additional macrophages from distinct murine data sets for examination of gene expression profiles: wild-type muscle 1-2 days following gluteus muscle crush (data gifted by Regeneron), lung tissue macrophages 14 days after asbestos (GSE127803), and 0, 3, 6 days after LPS (GSE120000) exposure. Results: IR/CTX injury with inhibition of TGF-β1 signaling recapitulated the appearance of uninjured muscle (Fig A) with ML delineating distinct, self-aggregating clusters between WT injury vs. all other groups. While injured and naive muscles were robustly separated, conditional Tgf-β1 deletion and systemic sequestration were observed as intermediate populations (Fig B). Unsupervised clustering in IR/CTX TA muscles resulted in 11 (uninjured) and 8 (IR CTX) clusters. Examining for overexpression of gene families in TGFβ1 signaling, we observed stark contrast between uninjured and IR/CTX macrophages (Fig C) suggesting an important role in fibrosis. In order to determine phenotypic resemblance of macrophages between muscular polytrauma vs. crush vs. inhalation injury, we visualized the combination of IR/CTX, muscle crush, and lung macrophages, demonstrating overlap (Fig D). Furthermore, we observed high levels of Tgfb1 expression in Adgre+ Csf3r- macrophages of muscle crush and lung injury, suggesting a shared myeloid phenotype associated with spatially distinct niches of fibrosis (Fig E). Conclusions: ML methods can aid in high-throughput analysis of muscle fibrosis histology. Given transcriptomic similarities between macrophages from IR, muscle crush, and inhalation induced fibrosis, it is possible that these cells may exhibit a conserved mechanism across organ systems and models, elucidating important parallels that could be amenable to ML and identification of important common therapeutic targets .


Back to 2020 Abstracts