Multi-omic analysis reveals retinoic acid molecular drivers for dermal fibrosis and regenerative repair in the skin

Abstract

Skin fibrosis is driven by fibroblast activation and excessive extracellular matrix deposition. To ascertain the fibroblast subpopulation(s) responsible for instigating fibrosis, we employed an established murine bleomycin skin fibrosis model. We characterized both the fibrotic and remodeling phases of dermal fibrosis, through a multi-omic approach. Using an unsupervised machine learning algorithm that quantifies 294 fiber features, we identified precise timepoints of fibrosis and regeneration. Single-cell transcriptomic and epigenomic sequencing then identified a Cyp26b1 expressing fibroblast subpopulation responsible for dermal fibrosis. The same fibroblast subtype was mapped to Visium spatial transcriptomic data. We further mapped the fibrotic subtypes to protein spatial data. To ascertain the functional impact of the fibroblast subpopulations, transplant delivery analysis showed their ability to drive skin fibrosis. Lastly, we identified a small molecular inhibitor of Cyp26b1 (Talarozole) to prevent and rescue dermal fibrosis. Conclusively, we establish an atlas of fibrotic and regenerative biological drivers of skin fibrosis.

Graphical Abstract

eTOC Blurb

Griffin et al., utilized the established murine bleomycin skin fibrosis model to identify the fibroblast subpopulation(s) and signaling pathways responsible for skin fibrosis. Using a multi-omic approach including single-cell transcriptomic, epigenomic sequencing, and spatial protein and transcriptomic analysis, a Cyp26b1 expressing fibroblast subpopulation was identified as contributing to dermal fibrosis progression.

Introduction

Wound healing in mammalian skin results in scar formation, leading to non-native tissue form and function.¹ Acute skin fibrosis can be triggered by a local dermal injury, including surgery, burn, chemical, or trauma injuries.² Alternatively, chronic skin fibroses may arise secondary to radiation treatment, systemic diseases including scleroderma (SSc), and graft-versus-host disease. In both acute and chronic forms, fibroblasts are the critical player in driving skin fibrosis, producing pathologic amounts of collageneous extracellular matrix through an orchestrated series of biological events.

Skin fibrosis is a dynamic process, with significant changes in the fibrotic extracellular matrix (ECM) composition and organization across time, encompassing both a pro-fibrotic phase in the short term and a long term remodeling phase.³ Prior studies data sets have focused on understanding the pro-fibrotic progression in acute wound healing.⁴ However, very little is known about progressive, chronic wound healing pathologies and the cell and ECM architecture during dermal fibrotic remodeling.

Several animal models have been utilized to mimic chronic skin fibrosis, with repetitive bleomycin injections being one of the most widely employed.⁵ Interestingly, the bleomycin-induced mouse model is able to progressively resolve fibrosis.⁶ Although many prior studies have focused on the fibrotic phase, the reparative phase is poorly understood. Comparative analysis of the fibrotic and resolution phases of fibrosis would therefore allow for the possible identification of fibroblast subpopulations specific to these divergent processes and deepen our understanding of both pathogenesis and treatment in chronic fibrotic diseases.

In this study, our overarching aim was to identify the specific fibroblast population(s) responsible for dermal fibrosis utilizing the bleomycin mouse model. First, we applied a machine-based fiber quantification algorithm to identify the specific stages of fibrosis and post-fibrotic resolution in the bleomycin mouse model. Next, we used transcriptomic and epigenetic single cell analysis to identify profibrotic and regenerative fibroblast subpopulations. We further explored timepoint-specific spatial transcriptomic data to identify putative locations of these fibroblast subpopulations during the fibrotic and resolving phases. Integration of these multi-omic datasets revealed cellular communications specific to dermal fibrotic progression, including a Cytochrome P450 26b1 (Cyp26b1) signaling pathway that was highly expressed upon fibrotic progression. We subsequently validated the functional role of Cyp26b1 signaling in fibrosis using a small molecule inhibitor with follow-up single-cell transcriptomic analysis. Collectively, these data provide biological insights into the fibrotic and regenerative drivers of bleomycin induced fibrosis, providing potential therapeutic targets to treat dermal fibrosis.

Results

Progression of fibrotic ECM can be defined based on dermal architecture composition

To determine the kinetics of the matrix architecture during progression of chronic skin fibrosis following bleomycin injections, we utilized a high dimensional fiber quantification algorithm previously employed for fibrotic tissue analysis (Figure 1A).^7,8 Bleomycin treated mice were injected daily for 10 days. Mice were then sacrificed at varying timepoints (0, 7, 12, 21, 28, 35) until Day 35 to ascertain the trajectory of the fibrotic and reparative phases (Figure 1B). Grossly, picrosirius red staining (PSR) showed reduced fiber branching and immature collagen fibers (green staining) at day 12 compared to days 21 and 28 (Figure 1C).

Figure 1: Extracellular matrix analysis reveals fibrotic and regenerative trajectories in bleomycin induced skin fibrosis.

(A) Schematic of automated matrix analysis pipeline composed of 294 ultrastructural features. (B) H&E analysis of different timepoints following bleomycin injections. (C) Temporal progression of dermal matrix architecture following bleomycin treatment. Top; UMAPs of Picrosirius Red Staining Algorithm. Bottom; Representative PicroSirius Red staining of each timepoint. Fibrotic dermis exhibited progressive matrix aberration until Postoperative day 12 (POD 12), followed by resolution to baseline by POD 35. (D) Pseudotime analysis of matrix analysis. Left; UMAP plot of timepoints. Right: UMAP of Pseudotime analysis of matrix analysis colored by Pseudotime. (E) Kinetics of matrix progression using picrosirius red staining, identifying POD 12 and 21 as snapshots of active fibrosis and early remodeling, respectively. Scale bar (B) 250 μm. (C) 40 μm. n = 6 per timepoint unless otherwise stated.

Each PSR image was then quantified to capture architectural variation. The resulting fiber feature matrix was reduced by Uniform Manifold Approximation and Projection (UMAP) to a two-dimensional manifold in order to characterize global variations in dermal architecture by timepoint (Figure 1C), then fitted to an unsupervised trajectory in DDRTree to quantitively assign pseudotime values to images based on degree of deviation from a root point representing baseline architecture (Figure 1D). PBS administration produced no significant deviation from baseline histological architecture by Day 35 (Figure S1A), while bleomycin induced progressive aberration of matrix architecture (i.e. increasing mean pseudotime by timepoint) up to Day 12, followed by full resolution to baseline (i.e. decreasing mean pseudotime by timepoint) by Day 35 (Figure 1C–E). Interestingly, the most significantly correlated parameters with low pseudotime, representing healthy dermal architecture, included measures of intra-image variation in stained matrix area, indicating disperse, patch-like ECM characteristic of normal skin (Figure 1C, S1B–C). High pseudotime, representing fibrotic progression, was associated with multiple measures of high fiber number, reflecting the confluent, dense ECM architecture of fibrotic skin (Figure 1C, S1B–C). Collectively, these data utilized a high-dimensional profile of dermal architecture to identify precise timepoints of fibrotic progression and post-fibrotic remodeling in the bleomycin-induced skin fibrosis model. Importantly, traditional metrics of fibrosis such as fibrotic gene expression (e.g. Col1a1, Col1a3, Acta2) and normalized tissue Collagen content supported these kinetics at representative timepoints (Figure S1D–E). Histological confirmation by H&E staining also demonstrated the expected phenotypic characteristics of skin fibrosis by timepoint, including increased dermal thickness, Collagen deposition, epidermal thinning, and adipocyte loss with progression of fibrosis (Figure 1B).

Fibroblast subtypes differ among fibrosis and remodeling trajectories

Based on our matrix architectural analysis, we first employed transcriptional analyses of Day 12 (peak of fibrotic progression), Day 21 (early post fibrotic matrix remodeling), and Day 28 (late post fibrotic matrix remodeling), profiling 11,370 cells isolated from harvested skin (Figure 1E, 2A–B). Cell types were identified using computational tools (i.e. SingleR, EnrichR) to evaluate differentially expressed genes and associated functional pathways. Upon analyzing all cells sequenced, we found that our dataset captured multiple characteristic dermal cell types known to play important roles in fibrosis and regeneration, including fibroblasts, endothelial cells, immune cells, smooth muscle cells, and epithelial cells, similar to those previously found in the setting of wound repair (Figure 2B, left).⁴ Across all timepints, we identified twelve fibroblast subtypes, with distinct transcriptional profiles (Figure 2B, right). Fibroblast subpopulations were identified through SingleR, with confirmatory expression of known fibroblast markers i.e. Col1a1, Col1a3, and were identified across all timepoints (Figure S1F). Interestingly, the fibroblast subpopulations dynamically changed across timepoints, with higher proportion of fibroblast cluster 0 at the fibrotic Day 12 (peak fibrosis) timepoint compared to the regenerative timepoint Day 21 (early remodelling) (Figure 2C). Similarly, cluster 1 had a higher proportion at fibrotic Day 12 compared to regenerative timepoint Day 21. In contrast, fibroblast cluster 3 was highly prevalent in the regenerative timepoint Day 21 compared to Day 12 (Figure 2C). In addition, cluster 2 increased during later stages of remodeling to a proportion very similar to that present at Day 0. Given the timepoint-specific enrichment of fibroblast clusters 0, 1, and 3, we directly compared the expression of several genes of interest to skin fibrosis. Interestingly, among the top 150 differentially expressed genes in cluster 0 were Hyaluron Synthase 1 (Has1) and Interleukin-6 (Il-6), both previously associated with cancer desmoplasia, indicative of a fibrotic phenotype (Figure S1G).^9,10 Alternatively, cluster 3 was associated with myelin basic protein (Mbp) and proteolipid protein (Plp1), associated with axonoal regeneration and indicative of a regenerative phenotype (Figure S1G).^11,12

Figure 2: Single cell analysis of fibrotic and remodeling skin reveals fibroblast subcluster specific to the fibrotic phase.

(A) Schematic of sequencing of baseline, fibrotic, and remodeling skin samples (B) Left; UMAP of overall cell phenotypes identified from single-cell transcriptomic profiling of skin. Right: UMAP of fibroblasts identified from single-cell transcriptomic profiling of skin. (C) Proportion analysis of fibroblast subtypes over time. (D) Gene analyses of top differentially expressed markers of all fibroblast clusters. Top; Fibroblast Cluster 0 Markers. Blue box shows expression of cluster defining markers. Bottom; Fibroblast Cluster 3 Markers. Purple box shows expression of cluster defining markers. (E) Gene ontology (GO) pathway and GO term analysis of cluster 0 (Top) and cluster 3 (Bottom). (F) Cell chat signaling of pathways enriched in Day 21 compared to Day 12 signaling including Apolipoprotein E (APOE) and Oncostatin (OSM) signaling (Top). Immunostaining of APOE and OSM in Day 21 skin (Bottom). Scale bar (F) 300 μm.

Interestingly, expression of Collagen Type 1 (Col1) and Collagen Type 3 (Col3) were also higher in clusters 0 and 1 compared to cluster 3 (Figure S1H). Furthermore, expression of Yes associated protein (Yap1), a mechanical pathway transcription factor known to cause multiorgan fibrosis^4,8,13,14, was highly expressed in cluster 0 compared to cluster 3 (Figure S1H). These data support the relevance of fibroblast clusters 0 and cluster 3 in fibrosis and regeneration, respectively.

We next sought to more broadly characterize gene expression programs characteristic of these fibroblast clusters. We performed Gene Ontology (GO) pathway enrichment analysis for cells from each cluster using top differentially expressed genes (Figure 2D–E). Cluster 0 (enriched in Day 12 skin) highly expressed genes related to extracellular matrix and adhesion, characterized by GO terms “TGF-beta signaling pathway” and “miRNA Targets in ECM and membrane receptors” (Figure 2E, top). In contrast, cluster 3 (higher in Day 21 skin) was enriched for GO terms related to metabolism, including “Urea Cycle and Metabolism of Amino Acids” (Figure 2E, bottom).

In order to further characterize these fibroblast subclusters of interest, we identified key genes whose expression was specific to each cluster. We found that Cyp26b1, Retinoic Acid Receptor Responder 2 (Rarres 2), and Adipocyte enhancer binding protein-1 (Aebp1) were highly expressed by cluster 0, while Fatty acid binding protein-5 (Fabp5), Cluster of differentiation 200 (Cd200), and Cd82 were highly expressed by cluster 3 (Figure 2D). These findings suggested that these genes and their corresponding proteins could potentially be leveraged to enable prospective isolation and/or specific manipulation of each fibroblast cell subpopulation. In order to verify that differential representation of the fibroblast cell subclusters in our scRNA-seq dataset was also reflected by differential representation of these cell subsets between experimental conditions in vivo, we performed immunofluorescent (IF) staining for these markers in histologic sections of Day 0, Day 12, and Day 21 along with co-staining for COL1 to specifically identify fibroblast cells (Figure S1I–J). Immunostaining showed enhanced expression of fibroblast 0 markers in Day 12 and fibroblast 3 markers in Days 0 and 21 (Figure S1I–J).

In order to further interrogate intercellular signaling and how cell-cell crosstalk involving these fibroblast cell clusters may differ under distinct experimental conditions, we applied the computational tool CellChat,¹⁵ which infers intercellular communication patterns based on scRNA-seq data (Figure S2). CellChat revealed intercellular communication axes/signaling pathways that were differentially active in Day 12 versus Day 21. Interestingly, overall there was greater number of communications at Day 21 relative to Day 12, suggesting reduced cell-cell interactions during fibrosis compared to remodelling (Figure S2A–C). Several signaling pathways were highly divergent between Day 21 and Day 12, especially between epithelial and fibroblast cells (Figure S2C–D). For instance, cell-crosstalk involving Apoliporotein E (Apoe) signaling was highly upregulated in Day 21 fibroblasts (Figure 2F). Similarily, Oncostatin (Osm) cell-crosstalk within fibroblasts and endothelial cells was highly enhanced in Day 21 fibroblasts (Figure 2F). Immunohistochemistry confirmed the upregulation of Apoe signaling in Day 21 fibroblasts and Osm in Day 21 fibroblasts, demonstrating the specific interactions at different timepoints (Figure 2F).

Supporting these results, computational pseudotime analysis similarly suggested an overall transcriptional trajectory that started at fibroblast cluster 3 and progressed toward cluster 2, via cluster 0 (Figure S1K). Collectively these data suggest that fibroblast cluster 0 may play an important role in the fibrotic progression of skin fibrosis.

Fibroblasts exhibit differences in chromatin accessibility during fibrosis and remodeling trajectories

To identify epigenetic variations across the progression of fibrosis, we applied scATAC-seq to over 22,855 nuclei at Days 0, 12, 21, and 28 (n=5 per timepoint), which were mapped to scRNA-seq by anchor label transfer using the Signac computational tool (Figure 2A and 3A). Interestingly, we were able to identify our RNA clusters in our scATAC-seq data, suggesting high correlation between our RNA and ATAC datasets (Figure 3A and S2E). The fibroblast clusters identified by scATAC-seq showed a high accesibility of fibroblast-related genes (e.g. Platelet derived growth factor-alpha (Pdgfra), Collagen Type 1 (Col1a1), Vimentin (Vim), Lumican (Lum), Zinc Finger E-Box Binding Homeobox 2 (Zeb2) (Figure 3B). Following transfer of our RNA data, 14 fibroblast ATAC defined clusters were identified, with distinct epigenetic profiles (Figure 3C and S2F), showing integration between RNA and ATAC datasets.

Figure 3: The fibroblast epigenomic landscape of skin reveals a transition to an open landscape in the remodeling phase compared to the fibrotic phase.

(A) UMAP of chromatin accessibility-based cell types identified via scATAC-seq. (B) Accessibility of characteristic fibroblast genes on scATAC-seq clusters. (C) UMAP of chromatin accessibility-based fibroblast clusters identified via scATAC-seq. (D) Coverage plots of top differentially accessible peaks with gene annotations for fibroblasts. (E) Gene ontology pathways (Top) and terms (Bottom) associated with fibroblasts at Day 0 (Left), Day 21 (Middle) and Day 28 (Right). (F) Highly enriched DNA motifs for fibroblasts comparing different timepoints. Red font indicates timepoint of enrichment. (G) Immunostaining of DNA motifs at Day 21 (top) and day 28 (bottom) identified through scATAC-seq. White arrows show positive staining. Scale bar (G) 150 μm.

Overall, a closed chromatin landscape was observed at Day 12 compared to Days 21, 28, and Day 0, with greater number of accessible peaks in Days 21, 28, and Day 0 compared to Day 12 (Figure S2G). Open chromatin peaks included “Twist Related Protein-1 (Twist1)”, “Tubulin Polymerization Promoting Family 10 Member 3 (Tppp3)”, and “Lysyl oxidase homolog 1 (Loxl1)” as shown by fibroblasts at Day 21 (Figure 3D). In addition, Day 28 demonstrated differentially accessible (DA) peaks associated with tissue patterning including “Homeobox B1 (Hoxb1)”, “Msh Homeobox 2 (Msx2),” and “Homeobox A (Hoxa)” (Figure 3D). DA peaks at Day 0 were involved in histone modification, adherens junction regulation, and fatty acid metabolism (Figure 3D). DA peaks at Day 0 were involved in histone modification, adherens junction regulation, and fatty acid metabolism (Figure 3D).

Next, we characterized global alterations to the fibroblast epigenome based on GO and DNA motif enrichment. GO analysis of Day 21 identified “extracellular matrix organization”, “cytoskeleton fiber organization”, and “tubulin binding” processes, further supporting active remodeling during this phase (Figure 3E). In contrast, Day 28, showed terms involved with hair follicle regeneration and transcription, suggesting a recovery back to a homeostatic state (Figure 3E).

Enhanced DNA motif analysis demonstrated motifs at Day 0 compared to Day 12 associated with Early growth response factors (Egr) including Egr1 and Egr3, known to regulate attachment and survival of normal skin (Figure 3F).¹⁶ Due to the closed chromatin state at Day 12, few differences were observed comparing Day 12 to Day 0. Comparing Day 21 to Day 12 showed enhancement of Specificity proteins (Sp) including Sp1, Sp3, and Sp9 (Figure 3F). Lastly, motif analysis associated with Day 28 relative to Day 12 included Kruppel-like factors (Klf) including Klf4 and Klf2 (Figure 3F). Interestingly, Sps and Kruppel-like factors have been shown to work in synergy to promote cell survival and stem cell maintenance with their dysregulation being involved in fibrosis and cancer.¹⁷ Immunostaining revealed a presence of SP1 at Day 21 and Klf4 in Day 28 (Figure 3G). Of note, when comparing DNA motif analysis of each individual timepoint to all timepoints, Day 12 was also associated with Jun and Fos pathways, highly established fibrotic pathways in skin (Figure S2H).¹⁸

Collectively, these terms suggested that fibroblasts exhibited distinct variations in chromatin accessibility at Day 21 and Day 12. Notably, the transition in GO terms from Day 12 to Day 21 suggested a switch from a closed to open landscape with upregulation of terms associated with cell proliferation, development, and tissue patterning consistent with a regenerative phenotype (Figure 3E–F, S2G–H).

To examine how transcription factors (i.e. DNA motifs) associated with each timepoint reflected differences in fibroblast phenotype, we extracted the frequency with which each gene in our scATAC-seq dataset was linked to each individual JUN, SP, and KLF motif of interest. These linkages were identified at the individual gene level for Fibroblast 0 (e.g. Cyp26b1, Rarres2, Aebp1) and Fibroblast 3 (e.g. Fabp5, Cd200, Cd82) markers previously identified in our scRNA-seq analysis (Figure S3A), then summated as signature scores per fibroblast subtype to assess the relative association of each motif with a Fibroblast 0 vs. 3 transcriptional phenotype (Figure S3B). Interestingly, the FOS::JUN and JUN::JUNB motifs enriched in Day 12 were highly associated with Fibroblast 0 genes (Figure S3B–C), which was largely driven by a high frequency of linkage to the Cyp26b1 gene. In contrast, the SP and KLF motifs enriched at Day 21 and 28, respectively, were all linked to a much greater degree with Fibroblast 3 signature genes, particularly for SP transcription factors (Figure S3B, S3D–E). Overall, this integrated analysis suggests that DNA motifs associated with fibrotic vs. regenerative timepoints were also linked to Fibroblast 0 (fibrotic) vs. Fibroblast 3 (regenerative) signature genes, respectively, which may infer some involvement of these transcription factors in promoting the corresponding fibroblast phenotypes.

Fibroblasts subpopulations during fibrosis and regeneration are common across established acute injury models and other organ systems

Additionally, we sought to examine connections between dermal fibroblast subpopulations identified in the present chronic fibrosis model and those established in prior acute injury models. Fibroblasts at Day 12 were distinctly enriched for established mechanofibrotic fibroblast genes in the context of acute injury to the skin, lungs, and other organs (e.g. En1, Tgfb2, Il6), all of which marked cluster 0 fibroblasts as well (Figure S3F).^8,19–21. A recently proposed progenitor-like marker for pro-fibrotic dermal fibroblasts (Cd201, i.e. Procr) was primarily expressed in unwounded skin in our study (Figure S3G) and also exclusively marked cluster 2 fibroblasts (Figure S3G).²² Traditional dermal lineage markers (e.g. Dlk1, Blimp1/Prdm1), on the other hand, did not appear to be variable by timepoint and were not clearly associated with the fibrotic or regenerative subpopulations identified in the present study (Figure S3H).²³ Interestingly, components of a regenerative signature recently uncovered by cross-species analysis of reindeer and fetal human fibroblasts (e.g.Crabp1, Tpm1) were highly associated with the early remodeling timepoint of Day 21, though they did not exclusively denote our clusters of interest (Figure S3I).¹⁹ Overall, comparison of acute vs. chronic injury-associated fibroblasts denoted strong parallels in the pro-fibrotic phenotype (e.g. Day 12, cluster 0), while identifying a smaller subset of shared markers between acute wound regeneration and early remodeling of chronic skin fibrosis (e.g. Day 21).

To examine potential connections between fibroblast subtypes in our skin model and other tissue contexts such as lungs, we integrated our scRNA-seq dataset with a prior reference dataset from bleomycin-induced lung fibrosis via anchor-based label transfer in Seurat.²⁴ Interestingly, pro-fibrotic skin fibroblasts such as cluster 0 appeared to map primarily to fibroblasts associated with a regenerative role in lungs, while regenerative skin fibroblasts such as cluster 3 did not have as clear of a correlate in lungs (Figure S3J–K). Further, individual lung fibrosis markers established in the reference dataset and prior literature (e.g. Csmd1, Serpine2, Spp1) were in fact expressed relatively more by regenerative skin fibroblasts (cluster 3) rather than pro-fibrotic skin fibroblasts (cluster 0) (Figure S3L).^24,25 Regenerative lung fibroblast markers (e.g. Cd248, Igfbp5, Pi16) were also conversely enriched in pro-fibrotic skin cluster (Figure S3M).²⁴ These contrasting associations, in addition to the established pro-regenerative role of Spp1 in foreign body response, may therefore suggest divergent biological mechanisms in skin vs. lungs.²⁶ Furthermore, fibroblasts in the bleomycin-induced lung dataset generally expressed low levels of our skin fibroblast markers of interest (e.g. Cyp26b1, Fabp5, Cd200, Cd82) or once again exhibited the reverse associations (e.g. Aebp1, Rarres2) (Figure S3N). Overall, our integrative analysis suggested highly tissue-specific mechanisms of fibrosis vs. resolution.

Fibroblasts Exhibit Distinct Cell Spatial Organization during fibrosis and remodeling trajectories

To further explore the fibrosis and remodeling phases, we applied the 10x Genomics Visium platform to analyze gene expression in the context of the spatial environment on Days 0, 12, 21, and 28 (Figure 4A). First, we performed anchor based transfer of our RNA-based defined populations onto our combined spatial data set, identifying 7 of the fibroblast transcriptionally defined subpopulations spatially (Figure 4B). As confirmation, the spatial location of known cell types was evaluated and found to disitribute in the skin as expected (Figure 4C). Echoing our scRNA-seq data, spatial-defined fibroblast clusters differentially expressed Cyp26b1 and Yap1 (Figure 4D).

Figure 4: Spatial analysis during fibrosis and repair in skin fibrosis reveals Cyp26b1 associated with fibrosis at a spatial level.

(A) Schematic analysis of Visium and PhenoCycler analysis on different timepoints. (B) UMAP of spatial RNAseq clusters identified by Visium analysis. (C) Visualization of spatially indexed cell type probabilities for various cell phenotypes. (D) Violin plot of Cytochrome P450 26B1 (Cyp26b1), Family 2 Cyp26b1 and Yes Associated Protein 1 (Yap1) expression spatially by fibroblast subclusters. (E) Spatial plot of fibroblast subtype 1 at Day 12 (Top) and Day 21 (bottom). (F) Gene ontology pathway of fibroblast cluster 1 (Top) and spatial plot of Serpin family A member 1 (Serphinh1) at Day 12. (G) Identification of cellular niches associated with each fibroblast subtype. (H) Spatial plot of Cyp26b1 expression at Day 12 and Day 21. (I) UMAP of all cell types identified through PhenoCycler analysis. (J) Ridge plot of Collagen Type 1(Col1), Yap1, Cyp26b1, Piezo1, Piezo2 and Focal Adhesion Kinase (Fak) expression modules analysis across fibroblast subtypes.

We next examined the spatial distribution of fibroblast subsets within each sample (Figure 4E). Through this analysis, we observed that spatial-defined fibroblast cluster 1 was enriched at Day 12 compared to Day 21 (Figure 4E). In addition, GO pathway analysis revealed upregulation of pathways associated with “macrophage differentiation”²⁷ and “metabolic pathway of fibroblasts”²⁸, suggesting a fibrotic inflammatory phenotype (Figure 4F, top). Interestingly, we observed high spatial gene expression of Serpinh1 at Day 12, suggesting correlation with skin fibrosis²⁹ (Figure 4F, bottom).

We further explored the cellular niche of the fibroblast subtypes by quantifying the frequency of specific cell phenotypes within their spatial neighborhood (Figure 4G). Interestingly, spatial fibroblast 1 occupied Spatial Niche A, characterized by high epithelial-fibroblast proximity, whereas spatial fibroblasts 2, 3, and 5 occupied Spatial Niche B, notable for greater epithelial-fibroblast as well as monocyte/macrophage-fibroblast proximity (Figure 4G).

To assess putative communication networks among cell types at Day 12 and Day 21, differential interaction maps were generated comparing the groups based on cell-cell spatial co-localization (Figure S4A). Day 12 showed high interaction between smooth muscle cells and fibroblasts, monocytes/macrophages, or epithelial cells compared to those at Day 0, which was characterized by more endothelial cell interactions (Figure S4A). When comparing Day 12 to Day 21, strong smooth muscle cell and monocyte/macrophage interactions were again observed in Day 12 (Figure S4A). Evaluating cell-cell interactions at Day 21 compared to Day 0 demonstrated differing cell-mediated communications including endothelial mediated communications (Figure S4A).

Cellchat utilizing the spatial VISIUM data, demonstrated high levels of cell-cell spatial communication between fibroblast 1 and 2 and other cell types at Day 12 compared to Day 21 (Figure S4B). We were also able to ascertain how cell-cell communications were instigated using CellChat with spatially weighted interaction analysis, showing that smooth muscle cells communicated via tyrosine-protein kinase KiT (C-kit), Hepatocyte growth factor (Hgf), Netrin, Nerve Growth Factor (Ngf), and Notch cell-signaling at Day 12 compared to Day 21 (Figure S4C, red box, left). Endothelial cells communicated at Day 21 compared to Day 12 through C type lectin receptor (Clec) and Nectin signaling (Figure S4C, red box, right). These data highlight the unique cell-cell communication patterns at Day 12 and Day 21 driving the fibrotic and remodeling phases respectively at a spatial level. Analyzing the fibroblast subclusters alone at a spatial level demonstrated high interactions of Fibroblast 1 and 3, the fibrotic subclusters, at Day 12 compared to Day 21 (Figure S4C). Interestingly, the cell-cell communication pathways observed at a sc-RNAseq level, were further shown to be upregulated at a spatial level at Day 12 compared to Day 21, including Perositin and Hepatocyte Growth Factor signalling (Figure S2C, S4D).

In addition, the fibroblast scRNA-seq markers of cluster 0 including Cyp26b1 were upregulated at a spatial level at Day 12 compared to Day 21 in the dermis where fibroblasts reside, showing the importance of these markers driving the fibrotic phase (Figure 4H). Collectively, these data suggested that similar RNA-defined fibroblast subclusters existed in situ at a spatial level and contributed to the fibrotic and regenerative activity at Day 12 and Day 21, respectively.

Spatial protein analysis reveals high expression of fibroblast driven retinoic acid signaling during the fibrotic phase of bleomycin induced skin fibrosis

While scRNA-seq is highly informative with regards to dermal fibrosis cell heterogeneity and transcriptomic subpopulations, to phenotype Day 12 and Day 21 in a spatially-informed fashion, we utilized the PhenoCycler platform. Using the CODEX platform allowed us to further integrate our spatial and scRNAseq findings, as we further investigated the spatial expression profile of CYP26B1 at a protein level. This is an assay in which a large panel of individual protein markers are sequentially labelled with, and iteratively imaged between, cyclic additions and washouts of dye-labelled oligonucleotide-conjugated antibodies (Figure 4A; markers shown in Table 1).^7,30 Protein staining patterns were used to project a manifold of cell-representative clusters (Figure 4I, Figure S4E). Across all treatment groups, 19 cell clusters were identified by unique PhenoCycler signatures, including five fibroblast clusters (Figure 4I).

Table 1:

List of protein markers with their associated barcodes for the CODEX experiment performed in this study.

CODEX MARKER	Barcode	Company	Antibody Product Number
PDGFRA	BX001	Abcam	Ab234965
CD31	BX002	Akoya	CD31-BX002
CD200	BX003	ThermoFisher Scientific	Ma170101
PIEZO1	BX005	ThermoFisher	PA5-72973
CORIN	BX006	ThermoFisher	711422
CALU	BX007	ThermoFisher	Pa5-109479
CJUN	BX010	Abcam	Ab247284
CD29	BX013	ThermoFisher	Ma5-17103
aSMA	BX014	Abcam	Ab5694
CD45	BX015	Akoya	CD45-BX015
CK15	BX016	Abcam	Epr1614Y
YAP	BX017	Abcam	Ab223126
PANCK	BX020	Biolegend	914204
SCA1	BX021	ThermoFisher	710952
ROBO2	BX022	Invitrogen	PA5-113491
ADIPOQ	BX023	ThermoFisher	PA184881
LEF1	BX024	Abcam	EPR2029Y
CK19	BX025	Novus Bio	EP1580Y
CD4	BX026	AKOYA	CD4-BX026
BLIMP1	BX027	ThermoFisher	Ma1-16874
CALD1	BX028	Abcam	Ab238782
CD8	BX029	Akoya	CD8-BX029
FAK	BX030	ThermoFisher	PA5-88093
CD20	BX032	Novus	NBP2-54591
PLIN1	BX033	Abcam	Ab3526
PDGFRb	BX034	ThermoFisher	PA5-96085
CD26	BX035	Abcam	EPR18215
COL 1	BX036	Abcam	Ab88147
DES	BX041	Abcam	Ab243931
DLK	BX042	Abcam	Ab119930
SOX9	BX043	Abcam	EPR14335-78
COL IV	BX045	Abcam	Ab6586
TAGLN	BX046	Abcam	Ab14106
Ki67	BX047	Akoya	Ki67-BX047
VIM	BX049	Abcam	Ab193555
MYH11	BX050	Abcam	Ab240983
CD68	BX052	Abcam	Ab237968
CYP26B1	BX054	ProteinTech	21555-1-AP
PIEZO2	BX055	ThermoFisher	PA5-72975

CD31: Cluster of Differentiation 31; CK19: Cytokeratin 19; CD4: Cluster of Differentiation 4; YAP: Yes-Associated Protein; CD68: Cluster of Differentiation 68; MGP: Matrix Gla Protein; CD45: Protein Tyrosine Phosphatase Receptor Type C; CD8: Cluster of Differentiation 8; CD20: Membrane Spanning 4-Domain A1; HLA-DR: Major Histocompatibility Complex Class II; CD26: Dipeptidylpeptiidase IV; PDGFRa: Platelet Derived Growth Factor Receptor Alpha; COLIV: Collagen Type IV; aSMA: Alpha Smooth Muscle Actin; Ki67: Marker of Proliferation Ki-67; VIM: Vimentin; PANCK: Pan-Cytokeratin; FAK: Focal Adhesion Kinase; MYH11: Myosin Heavy Chain 11; ADIPOQ: Adiponectin, C1Q and Collagen Domain Containing; PLIN1: Perilipin 1; CALD1: Caldesmon 1; TAGLN: Transgelin; PDGFRB: Platelet Dervied Growth Factor Receptor Beta; DES: Desmin; PIEZO1: Piezo Type Mechanosensitive Ion Channel Component 1; PIEZO2: Piezo Type Mechanosensitive Ion Channel Component 2; COL1: Collagen Type I; CYP26B1: Cytochrome P450 Family 26 Subfamily B Member 1, SCA1: SpinoCerebellar Ataxia Type 1, CD200: Cluster of Differentiation 200; LEF1: Lymphoid Enhancer Binding Factor 1, DLK1: Delta Like non-canonical Notch Ligand 1, SOX9: SRY-box transcription factor 9.

Overall, compared to Day 0, Day 12 and Day 21 had grossly apparent differences in spatial proteomic signatures. The proportion of five PhenoCycler defined fibroblast clusters varied across timpoints. As expression of Cyp26b1 was associated with fibrotic trajectory at Day 12 in the scRNA-seq and spatial RNA-seq data, we focused on determining whether the Cyp26b1 fibrotic marker showed similiar correlation at a protein level.

Focusing on the fibrotic trajectory, the gene signature of fibrotic clusters shown by scRNA-seq and Visium (CYP26B1⁺,YAP1⁺) was highly expressed in PhenoCycler defined fibroblast cluster 1 and cluster 3, compared to PhenoCycler defined fibroblast cluster 2, 4, and 5 (Figure 4J, top row and bottom left). Fibroblast cluster 3 also had high levels of COL1 compared to other clusters, indicative of a fibrotic phenotype (Figure 4J, top right row). Interesintgly, Fibroblast cluster 1 and 3 were also highly mechanically sensitive, expressing high levels of mechanotransduction proteins (YAP1⁺, FAK⁺, PIEZO1⁺, and PIEZO2⁺) suggesting a fibrotic phenotype^8,21 (Figure 4J, bottom row). Differential interaction maps were then used to visualize spatially-defined cell-cell interactions in different timepoints, particularly the communication of fibroblast cluster 3 (Figure S4F–G). Comparing cell communications at Day 12 to Day 0 or Day 21, Fibroblast cluster 3 showed high communication with immune cells including macrophages and CD8 T-cells (Figure S4F). These interactions suggest an inflammatory phenotype at Day 12, and reflect a fibrotic inflammatory state in skin fibrosis.⁴ In summary, PhenoCycler spatial protein expression analysis supported that fibrotic fibroblast subtype markers identified by scRNA-seq, ATAC-seq, and spatial RNA-seq (CYP26B1⁺YAP⁺) may be influential in driving fibrosis at Day 12 compared to Day 21.

In vivo interrogation of fibroblast subpopulations on fibrotic and remodeling trajectories

While transcriptomics and epigenetic analyses informed our understanding of molecular dynamics during fibrosis and remodeling, it is important to confirm these findings using functional assays. Given that our scRNA-seq analysis suggested the existence of fibrotic and remodeling clusters, we sought to determine whether delivery of cluster 0 and cluster 3 fibroblasts could alter the fibrotic and remodeling trajectories, respectively. Based on characteristic surface markers we had identified for these key fibroblast cell clusters (Figure 2D and S5A), we used fluorescence-activated cell sorting (FACS) to isolate CD248+CD55+ (Profibrotic cluster 0) and CD200+CD82+ (Proregenerative cluster 3) fibroblasts (defined as Lin-, CD45⁻CD31⁻CD326⁻ per a previously published lineage depletion strategy for isolating fibroblasts) from native mouse dorsal skin.³¹ Control fibroblasts, defined as Lin- were used as a control.

Following completion of ten bleomycin injections, injection of cluster 0 GFP⁺ fibroblasts or cluster 3 GFP⁺ fibroblasts was performed on Day 12 (Figure 5A). With injection of cluster 3 fibroblasts, histology of dorsal skin at Day 17 by H&E analysis revealed an earlier and a more profound regenerative phenotype usually observed later starting at Day 21 with appearance of hair follicles (Figure 5B–C). In contrast, injection of profibrotic cluster 0 fibroblasts following bleomycin administration showed a greater fibrotic phenotype at Day 17, more similar to that of Day 12, the time of most active fibrosis (Figure 5B–C).

Figure 5: CYP26B1 inhibition prevents and rescues skin fibrosis.

(A) Schematic of injection of pro-fibrotic and pro-regenerative fibroblasts following bleomycin treatment. (B) H&E analysis following injection of pro-fibrotic and pro-regenerative fibroblasts with (C) hair follicle quantitation. (D) Picrosirius red images of skin following pro-fibrotic and pro-regenerative fibroblasts. (E) RT-qPCR analysis of fibrotic markers [Collagen Type 1 (Col1), Collagen Type III (Col3), alpha-smooth muscle actin (α-SMA)] following treatment of bleomycin treated skin with profibrotic and proregenerative fibroblasts. (F) ELIZA Collagen type I Secretion following treatment of bleomycin treated skin with profibrotic and proregenerative fibroblasts. (G) Immunostaining of regenerative markers following treatment of bleomycin treated skin with profibrotic (Middle column), control (Right column) and proregenerative (Left column) fibroblasts with quantitation (Right). (H) Immunohistochemistry showing co-staining of CD248 (Green) (Right) and CD55 (Green) (Left) with CYP26B1(Red) at Day 12. White arrows indicate colocalization. Yellow dotted lines represent the dermis. (I) Schematic of concomitant prevention (Left) and rescue treatment (Right) experiments with CYP26B1 inhibition (Cyp26b1i). (J) H&E analysis and (K) hair follicle quantitation of concomitant prevention (Left) and rescue treatment (Right) experiments. (L) Schematic of sc-RNAseq analysis of CY26B1 inhibition of bleomycin treated skin. (M) UMAP of the fibroblast subclusters identified by sc-RNAseq (Left) and quantitation across the timepoints (Right). (N) Violin plots of cluster 0 fibrotic markers following Cyp26b1 inhibition. (O) Violin plots of fibrotic marker expression with and without Cyp26b1i. Scale bar (B) 250 μm, (D) 50 μm, (G) 200 μm, (H) 75 μm, (J) 250 μm. *p< 0.05. n = 3 per timepoint unless otherwise stated. Data shown as mean ± standard deviation (S.D.); statistical comparisons were made using one-way ANOVA with Bonferroni correction for multiple comparisons or unpaired T-test.

ECM analysis also demonstrated a similar dermal architecture of Day 17 following injection with pro-regenerative fibroblasts to Day 0 (Figure 5D). To validate our ECM algorithm data, we also analyzed ECM gene and protein expression. Collagen Type I, Collagen Type III, and α-SMA gene expression were significantly greater in skin treated with pro-fibrotic fibroblasts compared to control fibroblasts or proregenerative fibroblasts (Figure 5E). Furthermore, secretion of Collagen type I was significantly greater in skin treated with pro-fibrotic fibroblasts compared to control fibroblasts or proregenerative fibroblasts (*p< 0.05) (Figure 5F). Masson’s Trichome staining further revealed signifcantly enhanced collagen content following injection of profibrotic cluster 0 fibroblasts after bleomycin administration (Figure S5B).

Immunohistochemistry confirmed presence of proregenerative fibroblast markers CD200 and CD82 in skin with cluster 3 fibroblast administration, in concert with less fibrotic features (Figure 2D, 5A and 5G).

Retinoic Acid Metabolism is highly expressed in fibrotic fibroblast subpopulation in skin fibrosis

Given the fibrotic nature of CD248⁺CD55⁺ skin fibroblasts in vivo, we aimed to further characterize signaling pathways of cluster 0 fibroblasts. We observed high levels of Cyp26b1 signaling in cluster 0, our fibrotic subpopulation at Day 12 (Figure 2D). Retinoic acid (RA) is an important regulator of gene transcription through a specific set of nuclear receptors controlling transcriptional activation.³² Vitamin A (retinol) and its active metabolite, all-trans-retinoic acid (atRA), are critical signaling molecules in this pathway, modulating cell cycle and immune responses.³³ Clearance of atRA is mediated predominantly by cytochrome P450 family 26 enzymes.³⁴ The catabolic enzymes Cyp26a1 and Cyp26b1 have been studied in detail in the embryo, where they limit gradients of RA that guide patterns of gene expression critical for morphogenesis.³⁵

Endogenous atRA is critical in the maintenance of healthy skin epithelium and the immune system.³³ Interestingly, atRA has been found to play a role in the modulation of inflammatory response.³⁶ In acute mouse wounds, RA has been found to improve epithelization.³⁷ RA, synthetic vitamin A analogs, have been shown to have anti-inflammatory and anti-proliferative properties, inducing decreased Collagen Type 1 from SSc fibroblasts in culture.³⁸. Using immunohistochemistry, we demonstrated that CD248+CD55+ cells, were also highly positive for Cyp26b1 at Day 12 (Figure 5H). Our scRNA-seq and spatial transcriptomic data suggested that Cyp26b1+ fibroblasts were highly enriched at Day 12 (Figure 2D and 4H). In addition, scATAC-seq showed high levels of RA downstream targets including Crabp2 at Day 21 compared to day 12, suggesting enriched RA aids regeneration (Figure S2I). The kinetics of HOX gene expression during development is correlated to their response to retinoic acid, and interestingly, we observed high levels of HOX gene accessibility at Day 28 – including Hoxa9 and Hoxb9 – concomitant with RA downstream targets (Figure S2J).³⁹ Collectively, these data suggest that Cy26b1 signaling may be important among fibroblasts during the fibrotic trajectory.

Modulation of the Fibrotic Trajectory in vivo through Cyp26b1 inhibition

Talarozole is a potent inhibitor of Cyp26a1 and Cyp26b1 and has shown some success in preventing skin diseases in clinical trials.^40,41 As scRNA-seq fibrotic cluster 0 was associated with high levels of Cyp26b1 expression at Day 12 (Figure 2D), we hypothesized that inhibition of Cyp26b1 using Talarozole (Cyp26b1i), could prevent or mitigate bleomycin induced skin fibrosis.

First, to evaluate preventative treatment, we administered Cyp26b1i concomitant with bleomycin for 12 days (Figure 5I, left). Histology demonstrated presence of hair follicles and reduced dermal thickening at Day 12 following concomitant Cyp26b1i and bleomycin administration compared to bleomycin alone (Figure 5J–K). ECM analyses of Cyp26b1i treated skin showed a similar architecture to Day 0 when harvested at Day 12, the established peak of fibrosis with bleomycin injections alone (Figure S5C). Masson’s Trichrome staining also demonstrated significantly reduced collagen in concomitant Cyp26b1i treated skin (Figure S5D). Immunostaining showed greater presence of proregenerative fibroblast markers (i.e CD200) and reduced profibrotic fibroblast markers (i.e. CD248) in the Cyp26b1i treated skin compared to the bleomycin treated skin (Figure S5E). Lastly, secretion of Collagen was also significantly lower in the Cyp26b1i treated skin compared to bleomycin alone (Figure S5F).

Secondly, to evaluate rescue of fibrosis post-bleomycin treatment, we administered Cyp26b1i at Day 12 only following completion of bleomycin injections (Figure 5I, right). Histology by H&E demonstrated presence of hair follicles and reduced dermal thickening at Day 21 following post-bleomycin injection of Cyp26b1i compared to bleomycin treatment alone (Figure 5J–K). ECM analyses of post-bleomycin Cyp26b1i treated skin showed improved dermal architecture compared to bleomycin injection alone (Figure S5C). Masson’s Trichome staining further revealed reduced collagen content in the Cyp26b1i treated skin, when harvested at Day 21 (Figure S5D).

Immunostaining also showed greater expression of proregenerative fibroblast markers and reduced profibrotic fibroblast markers in the Cyp26b1i rescue treated skin compared to bleomycin alone (Figure S5E). In addition, secretion of Collagen was also significantly lower in the post-bleomycin Cyp26b1i treated skin compared to bleomycin alone when harvested at Day 21 (Figure S5F). To assess the effect of Cyp26b1i on retinoic acid metabolism, in vitro dorsal fibroblasts were cultured with bleomycin and Cyp26B1i. Enhanced RA levels after 48 hours were observed by ELISA compared to fibroblasts cultured with bleomycin alone (*p< 0.05) (Figure S5G, left). To further confirm these findings in our in vivo system, similar analysis was performed on harvested tissue. Again, significantly increased RA levels were seen with ELISA when skin was treated with bleomycin and Cyp26b1i relative to bleomycin alone (*p< 0.05) (Figure S5G, right). Collectively, these data suggest, that Cyp26b1i administered either concomitantly with or following bleomycin injection can mitigate bleomycin induced skin fibrosis.

CYP26B1 inhibition alters the transcriptional profile of fibrotic fibroblasts

As Cyp26b1 inhibition altered the histological trajectory of bleomycin induced fibrosis, to ascertain whether Cyp26b1 inhibition also causes a transcriptional profile change we performed scRNA-seq on bleomycin treated skin with or without concomitant Cyp26b1i (Figure 5L and S5H–I). We integrated the analysis of Day 0, Day 12 (bleomycin alone), and Day 12T (bleomycin with Cyp26b1i) cells to compare the differences with Cyp26b1i (Figure 5M). Upon analyzing all cells sequenced, we observed 8 fibroblast subclusters, which differed among the treatment groups (Figure 5M). The distribution of the fibroblast subclusters were more similar between Day 0 and Day 12T cells compared to Day 12 (Figure 5M, right). There was a higher proportion of fibroblast cluster 0 at the fibrotic Day 12 timepoint compared to Cyp26b1i treated skin (Day 12T) (Figure 5M, right). In contrast, fibroblast cluster 1 was highly prevalent in Cyp26b1i treated skin and more similar to that seen at Day 0 (Figure 5M, right). GO pathways demonstrated that cluster 0 was associated with “YAP1 ECM axis”, and “Collagen synthesis” terms, suggestive of fibrotic pathways, unlike Cluster 1, which was associated with “vascular endothelial growth factor (VEGF) signaling” and “integrin signaling” (Figure S5J). Furthermore, previous scRNA-seq defined markers of the fibroblast fibrotic cluster (Cyp26b1⁺, Rarres2⁺, Aebp1⁺) were more highly expressed in this scRNA-seq analysis of Day 12 compared to Day 12T skin (Figure 2D and 5N). Previous scRNA-seq markers of the fibroblast cluster responsible for fibrosis (Cd248, Cd55) (Figure S5A) were also highly also expressed in Day 12 compared to Day 12T within this dataset as well (Figure 5O). Collectively, these data suggest that Cyp26b1 inhibition may alter the fibrotic trajectory of bleomycin induced skin fibrosis towards greater regeneration by altering fibroblast subpopulation distribution and signaling.

Discussion

We report a multi-omic approach to bleomycin induced skin fibroses, which characterizes the transcriptomic/epigenomic and spatial organization of both fibrotic and remodeling phases. While the histological phenotype of human skin fibroses has been highly characterized, therapeutic strategies to overcome it are lacking. In our study, we have described a highly granular approach to ECM analysis following bleomycin, using an automated matrix architecture algorithm. We observed distinct features that correspond to the fibrotic phase including organization of the extracellular matrix. Previous reports have found that the progression of immature to mature Collagen fibers can occur following dermal injury and in advancement of dysplastic disorders.^42,43 Analyzing changes in ECM structure with high granularity has allowed for high resolution evaluation of the transition between fibrosis and regeneration.

Through scRNA-seq, we observed that fibroblast subpopulations were highly divergent at a transcriptomic level in the fibrotic (Day 12) compared to regenerative phases (Day 21). Tabib et al., echoed this finding in analyses of the transcriptome of healthy and SSc skin biopsies.⁴⁴. Of the twelve fibroblast subpopulations in our study, cluster 0 was implicated in fibrosis in our mouse model. The transcriptional signature of this fibroblast cluster was associated with two main features that correlated with its fibrotic phenotype.

First, cluster 0 was mechanically sensitive, with high expression of Yap1. Previous studies have found mechanically active fibroblasts to be associated with skin sclerosis. Knockdown of Yap and TAZ has been found to inhibit both inflammation and fibrotic response in skin and lung mouse models of chronic fibroses.⁴⁵ Ma et al., recently found that the Hippo pathway is crucial in SSc skin for both myofibroblasts and endothelial-to-mesenchymal transitioning cells (EndoMT).⁴⁶ Secondly, cluster 0 had a strong inflammatory signature with high expression of IL-6 and pathways associated with cytokine signaling. Inflammatory gene expression (IFN and HLA associated genes) has been highly reported in SSc scRNA-seq data sets.⁴⁷ More recently, in reindeer wounds, absence of inflammatory fibroblasts enabled regeneration.¹⁹ Taken together, these data suggest an inflammatory nature to fibroblast cluster 0, and that this fibroblast subpopulation may play a role in acute and chronic skin fibroses.

Interestingly, in our spatial transcriptomic dataset, we observed high expression of metabolic and inflammatory signaling at Day 12 compared to Day 21. Previous human scleroderma spatial data has shown an inflammatory CXCL9+ signature within fibroblasts in pansclerotic morphea patients compared to healthy skin.⁴⁸ In parallel, cell-cell interaction analysis of our spatial transcriptomic data showed high levels of smooth muscle cell driven communication at Day 12. Ma et al., recently identified strong SMA staining in the deep dermis of the skin in patients with SSc using the Visium platform.⁴⁶ Taken together, these data support the role of inflammatory, mechanosensitive fibroblasts during the fibrotic phase in skin fibrosis.

We also paired our transcriptomic analysis with epigenetic profiling. Interestingly, we observed a closed chromatin landscape at Day 12 compared to Day 21. Chromatin accessibility has recently been observed to be significantly reduced in diffuse scleroderma patients compared to healthy controls.⁴⁹ Interrogating differential motifs, we observed an upregulation of CJUN and EGR1 motifs at Day 12 compared to Day 21.^18,50,51 Several reports have shown JUN enhances fibrosis in skin injury and abdominal adhesions and thus these data further corroborate the role of c-Jun signaling in skin fibrosis.^18,52,53 Future studies should focus on understanding how Jun/Fos-mediated transcription may influence Cyp26b1 expression in the context of fibrosis and regeneration.

Our scRNA-seq analysis demonstrated the presence of fibrotic and remodeling clusters, marked by CD248+CD55+ (Profibrotic) and CD200+CD82+ (Proregenerative), respectively. Histology demonstrated the return of secondary elements and reduction of collagen following injection of progenerative fibroblasts at the peak of fibrosis when harvested at Day 17. Previous studies have suggested that injection of autologous fibroblasts may improve wound healing in large burns, diabetic ulcers, and skin wounds through the modulation of fibroblast-keratinocyte cross talk to stimulate ECM production, cell adhesion, and tissue remodelling.⁵⁴ In our study, injection of proregenerative fibroblasts may have provided optimal cell-cell cross-talk to overcome fibrosis and allow for regenerative healing through hair follicle induction.

The dominant fibroblast subpopulation (Fibroblast 0) observed during fibrosis expressed high levels of Cyp26b1, Rarres2, and Aebp1. Rarres2, an adipokine linked to inflammation has been implicated in several additional fibrotic conditions including human liver fibrosis,⁵⁵ renal fibrosis in human SSc patients^56,57, and metabolic syndrome linked with atherosclerotic cardiovascular disease.⁵⁸ Similarly, Aebp1, has been associated with human cardiac fibrosis⁵⁹ and liver fibrosis.^60,61 Shared expression of fibrotic markers across different organs suggests the presence of common fibrogenic mechanisms and potential for developing broad-spectrum anti-fibrotic therapies. However, future research is necessary to validate the therapeutic potential of these markers within each specific organ system.

As remodeling fibroblasts (Cluster 3) were present across all timepoints, we hypothesize that these cells are likely resident fibroblasts within the local wound niche, which become the dominant subpopulation following injury. Similarly, we have previously observed in pig skin wounds that a subpopulation of resident fibroblasts expands after wounding and promotes regenerative repair, suggesting that injury alters fibroblast dynamics. Interestingly, the fibrotic fibroblasts (Cluster 0) were not present at day 0, indicating that these cells become activated only after injury and contribute to fibrosis.⁶² Using single-cell RNA sequencing (scRNA-seq) trajectory analysis, we have also previously shown that specific fibroblast subpopulations can transition into activated fibroblasts in mouse skin wounds, being either absent or present at low abundance in unwounded skin.⁴ These findings suggest that skin injury activates resident fibroblasts, driving them into a more fibrotic state that contributes to skin fibrosis.

We identified transcriptional enrichment of Cyp26b1 expression at the peak of fibrosis. The beneficial effects of RA on tissue regeneration has been extensively studied, with several reports finding that RA has a protective effect on fibrosis, however others have challenged this notion.^63,64 In our study, scRNA-seq and Visium transcriptomic analyses suggested that high levels of enzymes responsible for RA breakdown, like Cyp26b1, may play a role in the fibrotic phase. Secondly, we observed enhancement of downstream targets of RA signaling enriched in the regenerative phase, including Crabp2. These findings led us to hypothesize that manipulation of Cyp26b1 signaling may alter the trajectories of skin fibrosis. To evaluate the proposed fibrotic RA pathway in the prevention and rescue of skin fibrosis, we employed a small molecule inhibitor, Talarozole, to inhibit Cyp26b1.⁶⁵ Global knockout of Cyp26a1 and Cyp26b1 are embryonic lethal, and postnatal knockout of Cyp26b1 has been reported to cause severe dermatitis and shortened lifespan, precluding utility of these mice to study bleomycin induced fibrosis.⁶⁶ Furthermore, we are not aware of a fibroblast specific knockout model. However, local inhibition of Cyp26b1 with Talarozole altered the fibrotic trajectory induced by bleomycin administration, as demonstrated by an ECM architecture more similar to native skin at time points where greater fibrosis was seen with bleomycin alone. Our scRNA-seq analysis highlighted the potential effectiveness of RA pathway modulation on shifting the distribution of fibroblasts to more closely mimic normal skin. Future work will identify downstream molecular consequences of Cyp26b1 inhibition on modulating fibroblast subpopulation/s activities. To date, Talarozole has shown to be well tolerated in early clinical trials of psoriasis and facial acne vulgaris.^67–69 However, these current data are from small early randomized control trials, warranting further investigation to understand the efficacy of Talarozole in fibrotic skin conditions. Collectively, our study thus highlights the promising role of RA signaling and Cyp26b1 inhibition as a potential therapeutic target for mitigating fibrosis.

Limitations

The mouse bleomycin induced skin model allowed us to study mechanisms underlying fibrosis and to ascertain timepoints and cell types involved in both fibrosis and regeneration. Although key aspects of fibrosis including collagen deposition occur following bleomycin, there is an exaggerated inflammatory response in mice, which can overestimate the effect of anti-inflammatory drugs.⁵ Thus, these mouse models do not fully recapitulate all aspects of skin fibrosis disease. Given these limitations, it is highly important to confirm our findings in human disease clinical specimens. Lastly, while our study focused on the role of Cyp26b1 signaling, several other fibrotic tissue neighborhoods were identified (CJUN+ regions) including spatial related fibrotic factors involved with metabolic signaling, which may be biologically relevant.

Resource Availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Michael T. Longaker (longaker@stanford.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

Single-cell RNA-sequencing data have been deposited in the Gene Expression Omnibus (GEO) (accession numbers GSE278069, GSE278169, and GSE277998) and are publicly available as of the date of publication. This paper does not report original code.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animals

Mouse ttrains used in this work are B6 (C57BL/6J, stock: 000664), 8–12 weeks of age, acquired from Jackson Laboratories. Mice were housed at the Stanford University Comparative Medicine Pavilion (CMP) in accordance with Stanford University guidelines, under an approved APLAC protocol (APLAC #21308 & #11048) and under the supervision of the Veterinary Service Center (VSC).

METHOD DETAILS

Bleomycin Fibrosis Model:

Induction:

Anesthesia was induced and maintained with 1–3% isoflurane at a flow rate of 2L/min. Adequate anesthesia was confirmed with the loss of hind-limb reflex to nociceptive stimuli. A portion of the dorsal skin was shaved, and depilatory cream (Nair^™) used to remove residual hair. Dorsal skin was sterilized with Betadine Surgical Scrub Veterinary (Aviro Health L.P.^™, Stamford, CT) followed by sterile alcohol prep pads (FisherScientific^™, Pittsburgh, PA).

Using an insulin syringe, 100 μL of bleomycin sulfate (Millipore Sigma, Cat: 1076308; 1 mg/mL in sterile phosphate buffered saline [PBS]) were injected intradermally in the center of the shaved region for 10 days. To ensure the injection site remained consistent across the treatment, 5 μL of 0.2% India ink solution (Fisher Scientific, Cat: 15433679) was injected intradermally 2 mm laterally and 2 mm longitudinally from the bleomycin-injected site. Mice were harvested at days 7, 12, 21, 28, or 35 following the first bleomycin or PBS control injection.

Transplant Experiments:

For the transplant experiments, 100,000 fibroblasts following flow cytometry sorting as below were injected intradermally into each site.

Inhibitor Experiments:

For the Cyp26b inhibitor experiments, Talarozole (10 uM)(MCE, ChemExpress) was intradermally injected into the same Bleomycin injected sites for 10 days as a single 30 ul dose for prevention experiments or as a single 30 ul dose at day 12 for rescue experiments.

Harvest:

Injected skin regions, with 1mm border were harvested with dissecting scissors and processed for downstream histology, transcriptomic, or proteomic analyses. Dissection followed fascial planes. Dorsal skin distal from the injected area was also collected as unwounded controls. Mice were euthanized by CO₂ necrosis and cervical dislocation. Harvested skin for Fluorescence-activated cell sorting (FACS) was mechanically digested using dissecting scissors to finely mince each specimen. Harvested skin for use in histology and immunofluorescent (IF) staining was placed in tissue embedding cassettes.

Harvesting cells for fluorescence-activated cell sorting (FACS):

Following CO2 euthanasia, bleomycin-injected skin was dissected and washed once in PBS. Tissue was then finely minced with sharp surgical scissors. Minced skin tissue was enzymatically digested using Collagenase Type II (ThermoFisher Scientific, Cat: 17101015) and Collagenase Type IV (ThermoFisher Scientific, Cat: 17104019) at a 1:1 ratio and concentration of 1500U/mL in DMEM (ThermoFisher Scientific Cat: 10569010) for 90 minutes at 37°C. During enzymatic digestion, samples continually agitated at 150rpm. Following digest, Enzyme activity was stopped using FACS buffer, and digested tissue strained through 70μm cell strainers. Cells were pelleted at 1500rpm for 5 minutes at 4C and the resuspended in 150μL of FACS buffer for primary antibody staining. For lineage negative (Lin-) FACS analysis, the following primary antibodies were used: Tie2 (ThermoFisher, Cat:13598782), CD45 (ThermoFisher Scientific, Cat:48045182 or ThermoFisher Scientific, Cat:13045181), CD31 (ThermoFisher Scientific, Cat:RM52280 or ThermoFisher Scientific 13031181), CD324 (ThermoFisher Scientific, Cat:13324982), CD326 (ThermoFisher Scientific, Cat:45592185), Ter119 (ThermoFisher Scientific, Cat:14592182). For injection experiments, fibroblasts were further co-stained, for regenerative markers CD200 (ThermoFisher Scientific, Cat MA5-17980) and CD82 (ThermoFisher Scientific, PA5-13228) or fibrotic markers CD248 (ThermoFisher Scientific, BS-2101R) and CD55 (Biolegend, 131803). Cells were stained with primary antibody for 30 minutes on ice. Following primary antibody staining, samples were washed with 500μL of FACS buffer, spun at 1500rpm for 5 minutes (4°C), and resuspended in 150μL of FACS buffer for secondary antibody staining using either Streptavidin-eFluor450 (ThermoFisher, Cat: 48431782) or Streptavidin-Alexa Fluor 647 (S21374). Cells were stained with secondary for 20 minutes on ice and then washed again with 500 μL of FACS buffer, and DAPI (4′,6-diamidino-2-phenylindole) (Bioligand, Cat: 422801) was added to label dead cells. A BD II FACS Aria machine was used for FACS sorting and analysis.

Histology and immunofluorescent staining:

Fixation:

Samples were fixed in 10% neutral buffered formalin (NBF; ThermoFisher Scientific^™, Waltham, MA) for 24 hours at 4°C. Tissues for the fluorescent transplant experiments were fixed in 4% paraformaldehyde (PFA) solution in PBS for 24 hours at 4°C.

Paraffin Sectioning:

Automated Tissue Processor (ThermoFisher Scientific^™, Waltham, MA) was used to dehydrate samples in a gradient of alcohols. Tissue was then embedded using ThermoFisher Histostar Tissue Embedding station. Paraffin blocks were trimmed as necessary and cut as 8 μm-thick sections. Paraffin ribbons were placed in a water bath at 40°C and mounted onto Superforst/Plus adhesive slides (ThermoFisher Scientific^™, Waltham, MA). Sections were baked at 37°C overnight.

Cryosectioning:

Fixe samples were placed in 30% sucrose dissolved in PBS at 4°C. After one week, samples were removed from sucrose and embedded at tissue blocks using Tissue Tek O.C.T. (Sakura finetek, Torrance, CA) over dry ice and 100% ethanol to achieve rapid freezing. Frozen blocks were mounted on a Thermo Scientific CryoStar NX70 cryostat, and 8 μm-thick sections were transferred to Superfrost/Plus adhesive slides (ThermoFisher Scientific^™, Waltham, MA).

Staining:

Hematoxylin and eosin (Cat: H-3502; Vector Laboratories, Burlingame, California), Masson’s Trichrome (ab150686; Abcam^®, Waltham, MA), Picro-sirius Red (ab150681; Abcam^®, Waltham, MA), and Oil Red O (Sigma-Aldrich^™, St. Louis, MO) stains with standard protocols were used.

Paraffin sections were hydrated prior to staining. They were placed in xylene for 20 minutes, followed by 10 minutes each of 100% ethanol (EtOH), 95% EtOH, 70% EtOH, 50% EtOH, and 30% EtOH. Slides were then submerged in running tap water for 10 minutes.

Cryosection samples were first dehydrated by submerging slides into PBS for 10 minutes, followed by 15 minutes each of 30% EtOH, 50% EtOH, 70% EtOH, 95% EtOH, and 100% EtOH.

Brightfield images were acquired with a Leica CTR4000 microscope.

Immunofluorescence staining:

For immunofluorescent staining, slides were washed twice in Tween 20 (Sigma-Aldrich^™, St. Louis, MO) followed by one wash in PBS. Slides were then blocked for 1 hour with Power Block (Biogenex^™, Fremont, CA) prior to addition of the following primary antibodies:

Abcam ab243839 (CD200), Invitrogen PA5-79006 (CD82), ThermoFisher Scientific 60170-1-IG (CD248), ProteinTech 18254-1-AP (APOE), Novus AF-495-SP (OSM), CYP26B1 (21555-1-AP), Proteintech 10216-1-AP (RARRES2) and ThermoFisher Scientific MA1-26771 (anti-collagen type I). Slides were then incubated for 2 h with Alexa Fluor 488, 594, or 647-conjugated anti-rabbit, anti-rat, or anti-mouse antibodies (Invitrogen, Waltham, MA). Finally, slides were mounted in Fluoromount-G mounting solution with DAPI (ThermoFisher Scientific^™, Waltham, MA). Fluorescent images were acquired with a LSM880 inverted confocal, Airyscan, AiryscanFAST, GaAsP detector upright confocal microscope.

Hematoxylin and eosin staining:

Slides were submerged into Hematoxylin for 10 minutes. Slides were then submerged into tap water for 5 minutes, then dipped into eosin 12 times. DI water baths were prepared, and slides were submerged until water was clear of eosin. Slide racks were then dipped into 70% EtOH 10 times, followed by 1 minute in 95% EtOH then 100% EtOH. Finally slides were dipped into Xylene 8 times until they were mounted on Superfrost/Plus adhesive slides (ThermoFisher Scientific^™, Waltham, MA) with Permount Mounting Medium (Electron Microscopy Sciences^™, Hatfield, PA).

Masson’s Trichrome Staining:

For Trichrome staining, Bouin’s solution was added to the samples for 60 minutes in a humidity chamber. To create humidity chambers, slide boxes were lined with wet paper towels. Next, slides were submerged into running tap water for 5 minutes. Working Weigert’s Iron Hematoxylin was then added to samples for 5 minutes, followed by submersion into running tap water for 5 minutes. Biebrich Scarlet /Acid Fuchsin Solution was then added to samples for 4 minutes. After slides were submerged in running tap water for 5 minutes, Phosphomolybdic/Phosphotungstic Acid was added to the sample for 45 minutes. Phosphomolybdic/Phosphotungstic Acid was removed from slides and Aniline Blue Solution was added for 4 minutes without a washing step. After Aniline Blue staining, slides were submerged into running tap water for 5 minutes. Correct staining was confirmed using Leica CTR400079 microscope. Finally, slides were dipped into 1% Glacial Acetic Acid Solution then running water 12 times each, followed by 15 dips each in 95% EtOH then 100% EtOH. Slides were submerged into Xylene 8 times prior to mounting with Permount.

Picrosirius Red Staining:

Dehydration/rehydration steps were not completed for Picrosirius red staining. Slides were first washed 3 times with PBS and then submerged into running tap water for 1 minute. Picrosirius red was added to slides for 60 minutes in a humidity chamber, previously described. After the completion of the 1-hour stain, slides were dipped into 2 different changes of 0.5% Glacial Acetic Acid 10 times (20 dips total), followed by 10 times into 2 different changes of 100% ETOH (20 dips total), and 8 times into Xylene. Slides were then mounted with Permount.

Picrosirius red stained histologic analysis:

Analysis of picrosirius red stained tissue sections took place using an image-processing algorithm. The algorithm profiles 294 ultrastructural features to provide a quantitative comparison of extracellular matrices. Each group (n = 3) was randomly imaged across 150–250 tiles at 40x.

Color deconvolution following previously described methods was performed to characterize each stain by absorbance in three RGB channels. Ortho-normal transformation was then used to determine each color’s contribution to the captured image. Red and green images were produced, representing mature and immature ECM fibers, and analyzed for individual features as detailed below. A Matlab script was used to achieve analysis, including noise reduction, preferential selection for smooth regions with low variance, and “skeletonization” of images to characterize the fiber networks. The algorithm allows for measurement of a collection of geometric features that characterize ECM morphology/architecture.^7,8

Ultrastructure quantitative analysis

Picrosirius red stained slides were imaged using polarization microscopy at 40x magnification. Each group (n = 3 biological replicates/ group) were imaged at random unique locations, 150–250 number of images per group), and then analyzed using an in-house image-processing algorithm as previously described.^7,8 Briefly, color deconvolution was performed to characterize collagen fibers by absorbance in red, green, and blue channels. Ortho-normal transformation followed to the contribution of each color to a given image. The red and green images produced represent mature and immature ECM fibers, respectively, and were analyzed for individual features. A Matlab-based script was applied to conduct downstream analysis, which included noise reduction, preferential selection for smooth regions with low variance, and “skeletonization” of images. The algorithm allows for measurement of a collection of geometric features that characterize ECM morphology/architecture.^7,8

Using quantified matrix values identified from the ultrastructure algorithm, all datapoints were run through the DDRTree algorithm, with a minimum spanning tree and minimum branch length of 15. Pseudotime values were then assigned to each ultrastructural datapoint based on their geodesic distance to the root point, representing baseline histological architecture. Pseudotime scores were averaged per animal to quantify tissue-level architecture across all technical replicates (tiles), the statistically summarized by mean and standard deviation at the group level.

Single-cell RNA-sequencing

Single cell isolation from mouse skin:

Using sharp surgical scissors, fibrotic skin harvested from the murine bleomycin-fibrosis model were mechanically digested and then enzymatically digested with Collagenase II (ThermoFisher Scientific, Cat: 17101015) and IV (ThermoFisher Scientific, Cat: 17104019) in DMEM-F12 (GIBCO^™, ThermoFisher Scientific). Samples were placed on an orbital shaker for 90 minutes at 150 RPM at 37°C. FACS buffer was added to quench the enzymatic digest. Samples were then passed through a 70μm cell strainer to remove large debris. Cells were centrifuged at 1500 RPM at 4°C, resuspended in 0.04% BSA in PBS, passed through a 40μm strainer, and counted. Cell suspensions were submitted to the Stanford Functional Genomics Facility for library preparation (10x Chromium Single Cell platform; Single Cell 3’ v3, USA) and sequencing.

Data processing, fastq generation, and read mapping:

Cell Ranger (10X Genomics, version 3.1)’s mkfastq was used to convert base calls to reads and align data against Cell Ranger’s mouse reference genome using the count function with SC3Pv3 chemistry. For quality control, maximum percent mitochondrial RNA was capped at 15%. A maximum of 7500 unique genes were used for downstream analysis.

Data normalization and cell subpopulation identification:

Unique molecular identifiers (UMIs) from each barcoded cell were normalized using a scale factor of 10,000 UMIs/cell. Using the R package Seurat (version 5), these data were natural log transformed. The first 20 principal components of normalized data were used for uniform manifold approximation and projection (UMAP) analysis. Batch correction was checked at the sample-level using Harmony.⁷⁰

CellChat receptor-ligand analysis:

The CellChat platform was applied to cell-cell interactions. CellChat’s Shiny App for its Cell-Cell Communication Atlas Explorer was applied to our scRNA-seq Seurat object in R. Cells were binned based on SingleR-defined cell types, and default parameters used to identify Secreted Signaling, ECM-receptor, and Cell-Cell Contact relationships.

Visium spatial transcriptomic analysis:

Fibrotic skin specimens were rapidly harvest and flash frozen in OCT. Using the Visium Tissue Optimization Slide and Reagent Kit, permeabilization time was optimized at a thickness at 10μm per section and 37 minutes for mouse tissue. Following cryo-sectioning at −20 degrees onto gene expression slides. The expression slide and reagent kit were used to produce sequencing libraries. The libraries were then sequenced using NextSeq (Illumina). Following demultiplication, raw FASTQ files and histology images were processed for each sample with the Space Ranger software for genome alignment. The raw spaceranger output files for each sample were then read into a Seurat class object in R using Seurat’s Load10x function. Data was normalized using SCT Transform with default parameters. To ascertain the integration of our scRNA-seq and Visium spatial analysis, we employed FindTransferAnchors function from Seurat, which allowed for the alignment of data using the two datasets. This cross-platform linkage is performed serially in an unconstrained and constrained fashion.

CODEX Spatial Analysis

To spatially phenotype mouse specimens, Co-Detection by Indexing (CODEX), an assay in which markers are labeled with oligonucleotide-conjugated antibodies and iteratively images between cyclic additions and washouts of dye-labeled oligonucleotides, was used.7 A custom CODEX panel was designed to assess wound cells within the tissue (see Table 1). In brief, primary antibodies were individually barcoded and validated using commercial supplier’s protocols. Mouse OCT blocks (n=3 per group) were sectioned at 8 μm thickness onto coverslips for CODEX antibody staining. Antigens were retrieved by standard citrate-EDTA processing prior to addition of CODEX antibodies. Using a CODEX-integrated Keyence BZ-X instrument (Akoya Biosciences) image acquisition was then performed. Using software from Akoya Biosciences, raw images were processed, and cell segmentation and rendering performed.

CODEX data were visualized using Akoya Biosciences Multiplex Analysis Viewer (MAV) in ImageJ. Resulting .fcs files were then concatenated in FlowJo and imported into Seurat and STvEA R packages for downstream analysis. Following debris removal, the processed UMAP was analyzed through Seurat. Analysis of the protein staining patterns was then used to assign cell types. The cell interactions were then inferred at k=20 nearest neighbors to quantify cell spatial interactions, and differential interaction maps were generated using ggraph scores.

Single Cell Assay for Transposase-Accessible Chromatin Sequencing

Nuclei isolation from mouse dermis

For nuclei isolation, 100mg of flash-frozen mouse dermis was finally minced with surgical scissors for 4 min. 0.1x lysis buffer was used for a 2-minute incubation, followed by quenching with wash buffer and centrifugation at 4°C at 500 rcf for 5 min. Following two wash and centrifugation steps, nuclei were resuspended in nuclei buffer (10x Genomics), counted, and then processed using 10x Genomics’ Chromium platform. During nuclei retrieval, samples were filtered using 40μm (Flowmi) and 20μm filters (PuriSelect).

Single-cell ATAC sequencing (scATAC-seq)

scATAC-seq was performed following 10x Genomics protocols. In brief, nuclei were isolated as described above from dermal mouse wounds. Mouse wounds from n =5 mice were pooled for each condition. Nuclei then underwent transposition, GEM generation and barcoding, post GEM incubation cleanup, library construction, and qualitative control on an Agilent Bioanalyzer High Sensitivity DNA chip (10x genomics protocol CG000496 Rev B). Libraries were pooled and cDNA libraries sequenced on the Illumina platform.

scATAC-seq Data Processing and Analysis

Transcriptomic and epigenomic data were acquired on the 10X Chromium platform using previously established protocols.⁴ 10x Genomics Cell Ranger tool cellranger-atac mkfastq was used to demultiplex raw base call (BCL) to fastq files. Using default parameters, cellranger-atac count was used to align sequencing files to the mouse genome (mm10). Signac (v. 1.10.0) was then used to conduct downstream analysis. Quality control for scATAC-seq included the following cutoffs: minimum 3000 peaks, maximum 30000 peaks, percent reads in peaks greater than 15%, nucleosome signal less than 4, and transcriptional start site enrichment score greater than 3.⁷¹ Following normalization, linear and non-linear dimensional reduction, nuclei were clustered using dimensions 2 to 30 and a gene activity matrix generated. Integration with scRNA-seq data was performed using Signac’s “Integrating with scRNA-seq data” vignette. Batch correction was checked at the sample-level for both scRNA-seq and scATAC-seq using Harmony.⁷⁰ Differentially expressed genes were calculated applying a log fold change threshold of 0.25. A minimum expression percentage of 25% using the default Wilcoxon Rank Sum test for scRNA-seq data in Seurat was also used. Differentially accessible pearks were calculated with a 0.25 log fold change threshold and minimum percentage of 10%. These were applied using the logistic regression framework for scATAC-seq data recommended in Signac.

Applying the default hypergeomtric test pipeline (*p < 0.05), enriched motifs were calculated and compared to background peaks matched for GC content. Using CreateMotifMatrix, motif-to-gene linkages were extracted from the scATAC-seq dataset. Genomic ranges were converted to annotated gene features using ClosestFeature, as previously described.⁷¹ Within the extracted matrix, the motif-to-gene associations were then calculated as a frequency with which a motif was associated to each gene of interest.

Retinoic acid enzyme-linked immunosorbent assay

To analyze effect of bleomycin with and without Cyp26b1 inhibitor treatment on retinoic acid levels, an ELISA (Cat#MBS706971) was performed on 100,000 in vitro cultured dorsal murine fibroblasts seeded into each well of a 6 well plate with either bleomycin (1μg/mL) alone (n=3) or bleomycin and Talarozole (10nM) (n=3). Media was collected after 48 hours and supernatant was analyzed for retinoic acid per manufacturer’s protocol with two technical replicates for each sample. In vivo retinoic acid levels were evaluated in mouse dorsal skin treated with either bleomycin (1mg/mL) alone (n=5 mice) or bleomycin and Talarozole (10μM) (n=5 mice) for 12 days before collection. 100mg of skin was digested with 5 IU/mL collagenase type II (Cat#17101015) and type IV (Cat#17104019) in DMEM F12 for 90 minutes at 37C. The digestion was then quenched with 10% FBS and centrifuged. Supernatent was collected and analyzed by ELISA for retinoic acid per manufacturer’s protocol with two technical replicates for each sample. A Tecan Magellan microplate reader was used to determine optical density and RA concentration was calculated by subtracting average optical density of blank wells and generating a four-parameter logistic (4-PL) curve.

QUANTIFICATION AND STATISTICAL ANALYSIS

All data are shown as mean ± standard deviation, unless otherwise specified. Statistical testing was performed in GraphPad Prism v9 unless otherwise specified. For two-group comparisons, unpaired t-tests were used. For multi-group comparisons, one-way ANOVAs were used with Bonferroni’s post hoc corrections to compare groups; p < 0.05 conferred statistical significance for all tests. * P < 0.05; ** P < 0.01; *** P < 0.001; **** P < 0.0001.

Star Methods

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-Mouse Monoclonal eFluor 450, CD45	Thermo Fisher Scientific	Cat# MCD4528
Anti-Mouse Monoclonal eFluor 450, Ter-119	Thermo Fisher Scientific	Cat# 48-5921-82;
Anti-Mouse Monoclonal eFluor 450, CD31	Biolegend	Cat# RM5228
Anti-Mouse Monoclonal eFluor 450, Tie-2	Thermo Fisher Scientific	Cat# 13-5987-82
Anti-Mouse Monoclonal eFluor 450, CD326	Thermo Fisher Scientific	Cat# 11-5791-82
Anti-Mouse Monoclonal eFluor 450, CD324	Thermo Fisher Scientific	Cat# 13-3249-82
Anti-Mouse Monoclonal eFlor FITC	Thermo Fisher Scientific	Cat# MA5-17980
Polyclonal CD82	Thermo Fisher Scientific	Cat# PA5-13228
Polyclonal TEM1/CD248	Thermo Fisher Scientific	Cat# BS-2101R
Anti-Mouse CD55 eFlor PE	Thermo Fisher Scientific	Cat# 131803
Anti-Rabbit Alexa Fluor 488	Thermo Fisher Scientific	Cat# A-11008
Anti-Rabbit Alexa Fluor 594	Thermo Fisher Scientific	Cat# A-11037
Anti-Rabbit Alexa Fluor 647	Thermo Fisher Scientific	Cat# A-21245
Rabbit Anti-CD200	Abcam	Cat# A-314662
Rabbit Anti-CD82	ThermoFisher Scientific	PA5-13228
Rabbit Anti-CD248	Bioss	BS-2101R
Rabbit Anti-APOE	Proteintech	18254-1-AP
Rabbit Anti-OSM	Novus	AF-495-SP
Rabbit Anti-CYP26B1	ThermoFisher Scientific	21555-1-AP
Rabbit Anti-RARRES2	Proteintech	10216-1-AP
Rabbit Anti-COLLAGEN1A1	ThermoFisher Scientific	PA5-29569
Chemicals, peptides, and recombinant proteins
Bleomycin	Millipore Sigma	Cat# 1076308
Talarozole	MedChemExpress	Cat# R115866
Fluoromount-G with DAPI	Thermo Fisher Scientific	Cat# 00-4959-52
Permount	Fisher Chemicals	Cat# SP15
Collagenase II/IV	ThermoScientifc	Cat# 17101015, 17104019
Biogenex Laboratories Power Block	Fisher Scientific	Cat# NC9495720
Trypsin antigen retrieval solution	Abcam	Cat# ab970
Ethanol	GoldShield	Cat# 64175
Xylene	Sigma	Cat# 534056
Triton-X 100	Sigma	Cat# X100
Phosphate Buffered Saline	Sigma	Cat# P5368
DAPI	BioLegend	Cat# 422801
Hematoxylin	Sigma	Cat# H3136
Eosin	Sigma	Cat# HT1101128
Picro Sirius Red stain kit	Abcam	Cat# ab150681
Mouse ELISA	MyBioSource	Cat# MBS706971
Critical commercial assays
10x Chromium Single Cell platform	10X Genomics	Single Cell 3’ v3
10x Chromium Single Nuclei platform	10X Genomics	Single Cell ATAC v2
Visium Tissue Optimization Slide and Reagent Kit	10X Genomics	Visium Tissue Optimization Slide and Reagent Kit
Agilent Bioanalyzer	Agilent	RRID:SCR_018043
Experimental models: Organisms/strains
C57/BL/6J	The Jackson Laboratory	Stock No. 000664
Deposited Data
Raw and analyzed data	This paper	GEO: GSE278069 GSE278169 GSE277889
Software and algorithms
ImageJ	National Institutes of Health	RRID:SCR_003070
MATLAB	MathWorks	RRID:SCR_001622
Rstudio	RStudio	RRID:SCR_000432
Other
Tegaderm dressings	3M	Cat# 1624W

Highlights.

1. We characterize fibrotic and remodeling phases of dermal fibrosis.

2. Timepoints of fibrosis and regeneration identified by fiber feature quantification.

3. Multiomic analysis identified a Cyp26b1 expressing profibrotic population.

4. CYP26B1 inhibitor Talarozole prevents and rescues bleomycin-induced dermal fibrosis.