Single Cell Analysis Reveals Cell Type-Specific Senescence Signatures in the Foreign Body Response to Synthetic Materials

Introduction: The foreign body response to biomaterials (FBR) is a significant barrier to the success of regenerative medicine therapies and can lead to medical device failure. We recently demonstrated that senescent stromal cells (SnCs)—non-proliferative cells with a secretory phenotype—accumulate and drive excess fibrosis during the FBR [1]. However, SnCs are known to play diverse roles in wound healing and fibrosis, and our lack of understanding of in vivo SnC phenotypes hinders our ability to target pathological SnCs in the FBR. Furthermore, SnCs are difficult to identify using a single marker, making it challenging to study the phenotype and function of these cells in biomaterial-mediated fibrosis. Transfer learning algorithms enable the identification of cell populations in single cell RNA-sequencing (scRNA-seq) datasets using gene signatures obtained from data of known cell types. In this study, we apply a transfer learning approach, ProjectR [2], to a bulk RNA-seq-derived in vivo signature of senescent cells (SenSig [3]) to define the SnC phenotypes in the FBR and identify SnC-specific therapeutic targets.

Materials and

Methods: To conduct scRNA-seq of the FBR, we isolated cells from the fibrotic capsules of synthetic implants in mice and humans. Mouse tissues were obtained from polycaprolactone (PCL) particle implants placed into volumetric muscle loss (VML) injury defects for 6 weeks—the timepoint in which senescent cells and fibrotic tissue peak around the particles [1]. Human samples were obtained from de-identified fibrotic capsules from patients undergoing breast implant replacement surgeries under an approved JHU IRB exemption. Cells from mouse and human tissues were isolated, enriched for CD45+ immune cells, and sequenced using 10x Chromium. Data were processed and analyzed using a standardized pipeline in Seurat. We applied transfer learning of SenSig to the datasets via ProjectR (SenSig-ProjectR) to identify cells that were concordant with the senescence signature. Cells with a projection weight >0 and adjusted p-value < 0.01 were considered senescent. To characterize SnC phenotypes, we conducted differential gene expression analysis between SnCs and non-SnCs on non-cycling cell clusters with >5% SnCs. Differentially expressed genes were further annotated based on biological processes known to be involved in senescence or fibrosis. For comparisons between mouse and human SnCs, human gene names were converted to their mouse ortholog names using the Ensembl BioMart database. Genes without human-to-mouse ortholog conversions available were excluded from the species comparison analysis.

Results, Conclusions, and Discussions: Using SenSig-ProjectR, we identified multiple stromal cell clusters with >5% SnCs in the mouse and human FBR (Figure 1A-B).

SnCs in the mouse FBR demonstrated some differences across phenotypes. All SnC subtypes upregulated senescence-associated cell stress and metabolism genes, and select ECM genes associated with wound healing and fibrosis. Myofibroblast SnCs upregulated genes associated with protein translation, collagen crosslinking, and cytoskeletal contraction, suggesting a pro-fibrotic, secretory phenotype. By contrast, activated fibroblast SnCs upregulated genes associated with wound healing, including MMPs, Hif1a, and select inflammatory factors; cartilage-like fibroblast SnCs upregulated proteoglycans; and pericyte/SMC SnCs upregulated vascular regulation and inflammatory genes. These results suggest that FBR SnCs are phenotypically diverse, and that targeting specific subtypes may be necessary to avoid off-target effects.

Human FBR SnCs displayed certain similarities to mouse SnCs, but we observed some key differences (Figure 1B-C). The myo/cartilage-like and progenitor fibroblast SnCs expressed similar genes to the mouse myofibroblast and activated fibroblast SnCs, respectively; however, both human SnC subtypes upregulated cartilage-specific genes that were not observed in mouse SnCs. Likewise, human pericyte/SMC SnCs upregulated fewer inflammatory factors compared to mouse SnCs, and expressed greater numbers of lipid metabolism genes. These results suggest that while SnCs between the mouse and human FBR are similar, species- and implant-specific factors may influence SnC phenotypes.

As myo/cartilage-like fibroblast SnC phenotypes appeared the most fibrotic, we screened for cell surface markers specific to this population to enable therapeutic targeting. We found that LRRC15 was upregulated in these SnC subpopulations in mouse and human, and confirmed protein expression of these genes in p16+ cells—a surrogate marker for SnCs—in the human breast implant capsules (Figure 1D).

In conclusion, using scRNA-seq and SenSig-ProjectR, we identified one perivascular and multiple fibroblast SnC subtypes in the FBR. While some similarities were observed between SnC subtypes, SnCs expressed diverse gene profiles associated with wound healing, fibrosis, and vascular regulation. We identified LRRC15 as a candidate cell surface marker for the most pathological, fibrotic SnC subtype in the mouse and human FBR. These data can inform SnC-targeted therapeutic development to mitigate pathological fibrosis and improve outcomes of biomaterial-based regenerative therapies.

Acknowledgements (Optional):

References: [1] PMID:32295900; [2] PMID:31121116; [3] PMID:37079217