Genetic vulnerability to Crohn’s disease reveals a spatially resolved epithelial restitution program

Abstract

Effective tissue repair requires coordinated intercellular communication to sense damage, remodel the tissue, and restore function. Here, we dissect the healing response in the intestinal mucosa by mapping intercellular communication at single-cell resolution and integrating with spatial transcriptomics. We demonstrated that a risk variant for Crohn’s disease, hepatocyte growth factor activator (HGFAC) Arg509His (R509H), disrupted a damage-sensing pathway connecting the coagulation cascade to growth factors that drive the differentiation of wound-associated epithelial (WAE) cells and production of a localized retinoic acid (RA) gradient to promote fibroblast-mediated tissue remodeling. Specifically, we showed that HGFAC R509H was activated by thrombin protease activity but exhibited impaired proteolytic activation of the growth factor macrophage stimulating protein (MSP). In Hgfac R509H mice, reduced MSP activation in response to wounding of the colon resulted in impaired WAE cell induction and delayed healing. Through integration of single cell transcriptomics and spatial transcriptomics, we demonstrated that WAE cells generated RA in a spatially restricted region of the wound site and that mucosal fibroblasts responded to this signal by producing extracellular matrix and growth factors. We further dissected this WAE cell-fibroblast signaling circuit in vitro using a genetically tractable organoid coculture model. Collectively, these studies exploited a genetic perturbation associated with human disease to disrupt a fundamental biological process and then reconstructed a spatially resolved mechanistic model of tissue healing.

One sentence summary:

The HGFAC-MSP-RON axis activates wound-associated epithelial cells that promote intestinal healing via a spatially restricted retinoic acid gradient.

Introduction

Wound healing is orchestrated by a spatial and temporal network of intercellular communication between epithelial, stromal, and immune compartments (1–5). This homeostatic program coordinates hemostasis, inflammation, epithelial restitution, and tissue remodeling (1, 3, 6–8). Upon tissue damage, platelets aggregate at sites of damaged vasculature and the coagulation cascade is elicited to produce fibrin clots, which collectively control bleeding and restore hemostasis. Subsequently, in the inflammatory phase, neutrophils, monocytes, and lymphocytes are recruited to the wound site. There, they respond adaptively to contextual cues, with either an antimicrobial response to infectious insult or homeostatic response to sterile trauma. Cytokines and chemokines, including fibroblast growth factors (FGFs) and transforming growth factor beta (TGF-β), facilitate tissue repair mediated by fibroblasts, which assist in restoration of tissue function (9, 10). At barrier surfaces, including the intestine, lung, and skin, epithelial cells also play a critical role in tissue healing by undergoing a dynamic transition into wound-associated epithelial (WAE) cells that cover the wound bed and restore barrier integrity in response to inflammation (6, 11). Accordingly, WAE cells depolarize and disassemble intercellular junctional complexes to proliferate and migrate across the wound site. Despite the recognized importance of WAE cells in epithelial restitution, it remains unclear how the WAE cell transition occurs and how these cells communicate with stromal and immune cells to coordinate healing responses in barrier tissues.

The coagulation cascade is exquisitely attuned to sense tissue damage, leading to the formation of fibrin clots and activating ancillary healing response pathways. Components of this system circulate systemically as zymogens that are activated locally at sites of tissue damage through a protease cascade that exponentially amplifies the initial stimulus. HGFAC controls a specialized pathway for growth factor activation that bifurcates from the coagulation cascade (12, 13). HGFAC is a serine protease synthesized in hepatocytes and secreted into the circulation as proteolytically inactive proHGFAC (12, 13). Local tissue damage initiates the coagulation cascade (14, 15), and activated thrombin cleaves proHGFAC to generate the active HGFAC protease (16, 17). HGFAC proteolytically activates proforms of hepatocyte growth factor (HGF) and MSP, two key cellular growth factors (13, 18) that signal through tyrosine kinase receptors, MET and Récepteur d’Origine Nantais (RON, also known as macrophage stimulating 1 receptor (MST1R)) respectively, to induce proliferation, cell migration, and extracellular matrix (ECM) adherence (19–23). Collectively, the HGFAC system functions as a locally triggered link between tissue damage and growth factor activation. However, it remains unclear how HGFAC-dependent growth factors mechanistically regulate intercellular communication during tissue healing and restitution.

Here, we address how wound healing is coordinated by spatially dissecting responses at single cell resolution after perturbation through a disease-relevant genetic variant. We demonstrate that HGFAC R509H—a risk variant for Crohn’s disease (CD)—preferentially impairs proMSP processing, and Hgfac R509H mice show delayed intestinal tissue healing. Leveraging this genetic model to disrupt healing pathways, we showed that the HGFAC-MSP-RON axis elicited the WAE cell functional response to injury. Integrating single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics data, we defined the WAE cell differentiation program and demonstrated that WAE cells instruct fibroblasts to promote tissue healing through ECM remodeling.

Results

HGFAC R509H missense variant exhibits reduced catalytic activity and preferentially impairs proMSP processing.

Genetic studies in inflammatory bowel disease (IBD) and CD have highlighted multiple vulnerabilities in the HGFAC-MSP-RON axis (Fig.1A, Table 1) (24, 25). In particular, HGFAC R509H was recently identified as a risk variant for CD (p=6.92×10⁻¹⁵, odds ratio=1.17) (24). Because the R509H mutation is located in the HGFAC catalytic protease domain near the thrombin cleavage site (Fig.1B), we sought to determine if R509H hinders activation of proHGFAC by thrombin or if it impairs the catalytic activity of HGFAC toward its substrates. We produced recombinant human proHGFAC R509 and R509H to assay thrombin cleavage (fig.S1, A and B) and determined that both variants were equally effectively cleaved by thrombin (fig.S1, C and D). We repeated the assay with increasing concentrations of thrombin and confirmed that both proHGFAC R509 and R509H were cleaved with similar efficiencies (fig.S1E), suggesting that proHGFAC R509H cleavage by thrombin is not impaired. Next, we addressed whether activated HGFAC R509H impacts proteolytic activity towards its substrate growth factors. ProMSP cleavage was significantly impaired by R509H (fig.1C, fig.S1, F and G; p=0.0094 at 2hr in G). Although proHGF cleavage was also impaired by R509H, the deficit was milder in comparison to proMSP cleavage (fig.S1, H to J). We confirmed this finding by assaying proMSP processing activity in serum from healthy human donors genotyped as homozygous for HGFAC R509 or R509H. Consistent with our in vitro assays, proMSP processing was delayed in HGFAC R509H serum (fig.S1, K and L). We conclude that HGFAC R509H substitution does not affect proteolytic cleavage by thrombin, rather it reduces catalytic activity and preferentially limits proMSP processing.

Figure 1. HGFAC R509H missense variant impairs proMSP processing and leads to delayed wound healing in vivo.

(A) Schematic diagram of the HGFAC system and coagulation cascade. Tissue damage initiates the coagulation cascade, which activates the HGFAC system via thrombin-mediated cleavage of proHGFAC. HGFAC activates key growth factors MSP and HGF. Asterisks indicate gene associations with IBD listed in Table 1. (B) Domain structure of the HGFAC protein (51). R509H is located in the catalytic protease domain of the HGFAC protein. SP: Signal peptide, FNI & II: Fibronectin type I & II. Red triangle indicates the thrombin cleavage site, black triangle indicates the kallikrein cleavage site. (C) ProMSP processing by HGFAC R509 and HGFAC R509H. Percentage of proMSP processing activity was measured at indicated time points and pooled across four independent experiments; *p<0.05, **p<0.01 (unpaired student’s t-test). (D) (upper panel) Experimental scheme of colonic endoscopic-guided wound model. Using a miniature video endoscope and biopsy forceps, wounds were created in the dorsal side of the distal colon at day 0 and imaged by endoscopy on days 1 and 3. (lower panels) Representative endoscopic images. (left) Forceps grasping tissue to make a wound. (right) A fresh wound. (E) Representative endoscopic images of Hgfac WT, R509H, and KO wounds on day 1 (left panels) and day 3 (right panels); the yellow dotted line indicates an unhealed area. (F) Quantification of wound closure in colonic mucosal wounds on day 3 relative to day 1 (n=9–10 wounds per genotype; n=6 mice per genotype in two independent experiments). ****p<0.0001, ns: not significant (one-way ANOVA with Tukey’s multiple comparison test). (G) Serum proMSP processing at baseline and day 1 post wounding across three genotypes (n=10–11 mice per condition). *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns: not significant (one-way ANOVA with Tukey’s multiple comparison test). (H) (upper panel) Schematic diagram detailing endoscopic procedure and MSP enema treatment. (lower panel) Delayed wound healing is improved by MSP treatment in Hgfac WT, R509H, and KO mice (n=5–8 wounds per condition, n=3 mice per condition). Wound closure was analyzed on days 1 and 3 and expressed as a percentage of the size of the day 1 wound area. **p<0.01, ****p<0.0001, ns: not significant (one-way ANOVA with Tukey’s multiple comparisons). Data represents mean +/− SEM.

Table 1.

HGFAC system-related IBD risk genes and variants.

Gene	Conseq.	rsid.	Chr.	Position	A0	A1	p-value	OR	Citation
HGFAC	p.Arg509His	rs16844401	4	3,447,925	G	A	6.9e-15	1.17	Sazonovs et al. 2021 (24)
MST1	p.Arg651Ter	rs142690032	3	49,684,379	G	A	1e-6	1.27	Huang et al. 2017 (25)
	p.Arg689Cys	rs3197999	3	49,684,099	G	A	1.4e-44	1.18	Huang et al. 2017 (25)
MST1R	p.Arg322Gln	rs2230593	3	49,902,645	C	T	1.4e-44	1.18	Huang et al. 2017 (25)
SPINT1	p.Ala649Thr	rs142240169	15	40,856,790	G	A	8.8e-6	0.77	Sazonovs et al. 2021 (24)
HGF	n.d.
MET	n.d.

Chr.=chromosome, OR=odds ratio, n.d.=not determined

Hgfac R509H mice show delayed wound healing and insufficient MSP activation.

To characterize how the HGFAC R509H variant mechanistically contributes to mucosal homeostasis and healing, we generated knock-in and knock-out mouse models using CRISPR/Cas9 technology. In mice, R507 corresponds to R509 in the human protein, thus we introduced the R507H mutation into the endogenous mouse Hgfac locus to model the human allele and refer to these mice as Hgfac R509H (Fig.S2, A to D). Hgfac KO mice have a 13bp deletion in exon12, generating a null allele and no detectable HGFAC protein (Fig.S2, A and B, D and E). We collected serum from Hgfac WT, R509H, and KO mice and performed the proMSP processing activity assay. Corroborating our findings in humans (Fig.S1, K and L), Hgfac R509H mice exhibited defects in proMSP processing in serum samples (Fig.S2, F and G). Serum proHGF processing was comparable between Hgfac WT and R509H mice (Fig.S2, H and I). Previous studies demonstrated that recovery from acute chemically induced colitis is delayed in Hgfac KO mice (26); however, the precise mechanistic functions of HGFAC and the R509H mutation in mucosal healing remain unclear. To determine the impact of R509H and impaired proMSP processing in tissue healing, we employed an endoscopy-guided wound healing model (11) (Fig.1D). Hgfac R509H and KO mice exhibited significantly delayed wound healing compared to WT mice (Fig.1, E and F; p<0.0001). Cleavage of systemic serum proMSP after local intestinal wounding was increased in WT mice compared to baseline, but this increase was severely limited in Hgfac KO and partially limited in R509H mice (Fig.1G). Serum HGF activation was comparable among the three genotypes (fig. S3). These data suggest proMSP cleavage is more sensitive to loss of HGFAC function in R509H and KO mice during mucosal injury. To determine if MSP supplementation is sufficient to promote wound healing, we employed the endoscopy-guided wound healing model in Hgfac WT, R509H and KO mice and treated them with PBS or recombinant mouse MSP by enema on days 1 and 2 after wounding (Fig.1H). MSP-treated mice showed significantly improved wound closure compared to PBS-treated mice (Fig.1H, p<0.0001). Local MSP administration rescued the delayed wound healing in Hgfac R509H and KO mice to equivalent degrees as PBS-treated WT mice. Together, these results indicated that impaired serum MSP activation in both Hgfac R509H and KO mice during injury leads to delayed wound healing, but MSP supplementation is sufficient to promote tissue healing.

Single-cell transcriptional profiling identifies cell lineages and dynamic functional programs associated with wound healing in mice.

Having demonstrated that the IBD risk allele of HGFAC disrupts a critical tissue healing program, we exploited this model as a clinically meaningful perturbation of mucosal homeostasis to further dissect the cellular and molecular mechanisms of healing. We turned again to the endoscopy-guided wound model and collected day 2 wounds or intact tissues from Hgfac WT and KO mice. Samples were subsequently analyzed by scRNA-seq using the 10x Genomics platform (Fig.2A). We obtained 60,871 individual cell transcriptomes after quality control and doublet removal across wounds (3 WT, 2 KO) and intact tissues (3 WT, 3 KO). Unsupervised clustering of the data identified 31 cellular subpopulations and corresponding marker genes defining epithelial, immune, and stromal subsets (Fig.2B, Fig.S4, A and B). Within the epithelial lineage, we identified stem cells, transit amplifying (TA) cells, absorptive enterocytes, and secretory cell lineages (Fig.2, B and C, Fig.S4, A and B). To identify cell types and states associated with healing, we compared epithelial cell composition between wounds and intact tissues (Fig.2C) and uncovered a unique cell population enriched in wounds, hereafter referred to as WAE cells. Additionally, we found proliferating mucosa-associated fibroblasts (MAFs) enriched amongst stromal cells (Fig.2D), and monocyte-derived dendritic cells (mo DCs) and neutrophils enriched amongst immune cells (Fig.2E) in wound samples. Dynamic expression of ligand-receptor pairs across epithelial, immune, and stromal clusters suggests cell-cell communication between different cell types in the tissue and inducible communication during wound healing (Fig.2F, fig.S5). T lymphocytes and CD103⁺ DCs showed specific expression of X-C motif chemokine ligand 1 (Xcl1) and X-C motif chemokine receptor 1 (Xcr1) (Fig.S5), a chemotactic circuit for intestinal immune homeostasis (27). Higher expression of oncostatin M (Osm) in monocytes and oncostatin M receptor (Osmr)/interleukin 6 cytokine family signal transducer (Il-6st) in stromal cells suggests a key interaction between innate immunity and the stroma during wound healing (Fig.2F, Fig.S5) (28). Furthermore, interleukin 17a (Il-17a) expression in T cells and interleukin 17 receptor C (Il-17rc) expression in epithelial cells suggests crucial immune-epithelial dialog during wound healing (Fig.S5) (8). No significant differential expression of these genes was observed between KO and WT samples (lowest p_adj=0.23 for interleukin 1 receptor antagonist (Il-1rn), with log(fold change)=0.59 in WT versus KO).

Figure 2. A single-cell transcriptome atlas of colonic wounds.

(A) Schematic diagram of the experimental design for colonic endoscopic-guided wound scRNA-seq. Hgfac WT mice (n=3 for intact tissue, n=3 for wounds) and Hgfac KO mice (n=3 for intact tissue, n=2 for wounds) were used. To minimize variation, three wounds or three intact tissues per mouse were collected by a 2mm skin biopsy punch tool and pooled as a single sample. (B) UMAP visualization of all epithelial, stromal, myeloid, and lymphoid cell clusters colored by cluster identity (n=6 and 5 for intact tissues and wounds, respectively). (C-E) Fraction of each epithelial (C), stromal (D), and immune (E) cell subpopulation per condition. Subpopulations that were significantly different in wounds (compared to intact tissues) are marked with arrows (up/down represents enriched/depleted in wound). Statistical significance was determined using a hierarchical Bayesian model with false discovery rate<0.05 (Methods). (F) Dot plot showing expression of cytokines or chemokines (left) and their corresponding receptors (right) across five cell types differentially abundant between wounds and intact tissues. Ligand-receptor pairs are linked by lines. Interactions between Osm and Osmr/Il-6st are highlighted in bold.

Extensive cellular crosstalk during wound healing suggests that WAE cells may perform unique functions during this process. We compared the transcriptional profiles of WAE cells with epithelial cell types annotated in our dataset, identifying 356 differentially expressed genes (DEGs) including clusterin (Clu), annexin A1 (Anxa1), aldehyde dehydrogenease 1 family member A3 (Aldh1a3), keratin 14 (Krt14), and claudin4 (Cldn4) (Fig.3A, Table S1; p_adj<0.05, average log₂(fold change)>0.5). Performing KEGG enrichment analysis of these genes, we found that pathways regulating actin cytoskeleton, focal adhesion, phosphoinositide 3-kinase (PI3K)-Akt signaling, and mitogen-activated protein kinase (MAPK) signaling were significantly enriched (Fig.3B; Fisher’s exact test p_adj<0.05). These findings are consistent with the notion that WAE cells proliferate and migrate to cover the wound bed in response to injury (11). To define the spatial location of WAE cells relative to wound sites, we performed immunofluorescence imaging by high-resolution confocal microscopy. Several specific markers selected from DEGs enriched in WAE cells (Krt14, Cldn4, Aldh1a3, Clu) were expressed specifically in epithelial cells covering the wound bed (Fig.3C, Fig.S6A) but rarely in cells in unwounded tissue (Fig.S6B). To identify the origin of WAE cells, we performed an RNA velocity analysis and found that WAE cells likely differentiated from TA cells and immature enterocytes (Fig.3D). To determine whether the WAE cell gene signature is present in human intestinal diseases, we analyzed previously generated scRNA-seq data comprising over 700,000 profiles from the terminal ileum and colon of 71 CD patients and healthy donors (29). We analyzed colonic samples from 18 CD patients and 16 healthy controls (total 97,788 epithelial cells) and identified an epithelial cell subpopulation with elevated expression of WAE cell markers (Fig.S6C, cluster 19; Materials and Methods). This subpopulation was almost exclusively found in samples from patients with CD (except for seven cells from healthy controls), and the average expression of WAE cell markers was significantly higher in inflamed relative to non-inflamed regions (Wilcoxon test p<10⁻¹⁵). These results suggest that WAE cells appear in response to damage or inflammation and share similar gene expression programs between human and mouse. Overall, we identified a dynamic transcriptional program associated with WAE cell responses to injury and mapped this program at single-cell resolution.

Figure 3. Characterization of murine wound-associated epithelial (WAE) cells.

(A) Top 50 differentially expressed genes (DEGs, two-sided Wilcoxon test) in WAE cells vs other epithelial cells (ranked by absolute average log₂(fold change), p_adj<0.05 for all genes) obtained from all day 2 samples. (B) KEGG functional enrichment analysis (one-sided Fisher’s exact test) of DEGs in WAE cells and other epithelial cell types obtained from all day 2 samples. The top 20 curated pathways are shown. (C) Immunohistochemistry (Krt14, Cldn4, Aldh1a3) and RNAscope (Clu) of WAE cell (green) and epithelial (EpCAM or ß-catenin, red) markers in WT day 2 wounds. W.B. indicates wound bed. Right images are magnifications of left images. Scale bars: 100 μm (left) and 50 μm (right). Data are representative of two independent experiments. Images were acquired with identical parameters as fig. S6B. (D) RNA velocity estimates of WT WAE cells projected on the UMAP plot. (E) Volcano plot depicting DEGs (two-sided Wilcoxon test) in WT WAE cells compared to KO WAE cells (p_adj<0.05, enriched in WT: average log₂(fold change)>0.5, enriched in KO: average log₂(fold change)<−0.5). Genes overlapping with WAE cell markers are labeled (p_adj<0.05, average log₂(fold change)>0).

To define the role of the HGFAC-MSP-RON axis in WAE cell function, we first mapped expression of the MSP receptor RON (Mst1r). RON was highly expressed in proliferative cell populations, including TA cells and immature enterocytes and, along with Met, was most prevalent in WAE cells (Fig.S6D). We then compared WAE cell transcriptomes between Hgfac WT and KO wounds, identifying 51 genes enriched in WT cells (Fig.3E) and thus a subset of WAE cell genes influenced by the HGFAC system. Notable enriched genes were related to RA metabolism (Aldh1a3, dehydrogenase/reductase 9 (Dhrs9), cytoskeletal dynamics (tropomyosin 3 (Tpm3), thymosin beta 10 (Tmsb10), actin related protein 2/3 complex subunit 1b (Arpc1b), filamin b (Flnb), myristoylated alanine-rich c-kinase substrate-like protein 1(Marksl1)), and cell adhesion and migration (carcinoembryonic antigen-related cell adhesion molecule 1 (Ceacam1), integrin beta-1(Itgb1), placenta expressed transcript 1 (Plet1)). These data were consistent with delayed restitution observed in Hgfac KO wounds.

MSP stimulates WAE cell gene expression in vitro to promote epithelial restitution.

Having implicated HGFAC and MSP in WAE cell function, we sought to precisely define the MSP-induced transcriptional program. To mimic WAE cell differentiation from immature enterocytes observed in our trajectory analysis, we isolated and cultured mouse colonic spheroids with 50% L-WRN CM (conditioned medium produced from mouse L cells secreting Wnt3a, R-spondin-3, and Noggin) for 2 days, then with 0% L-WRN CM to induce differentiation. We treated spheroids with MSP or PBS and collected samples at 0, 2, 4, and 6hr for bulk RNA-seq (Fig.4, A and B). MSP treatment of colonic spheroids induced a gene expression pattern similar to the WAE cell signature derived from scRNA-seq, with significant overlap (Fisher’s exact test, p=6×10⁻¹⁹) of 58 DEGs and increased protein expression of Aldh1a3, laminin subunit gamma 2 (Lamc2), and Cldn4 (Fig.4, C and D). To determine whether MSP can induce expression of WAE cell-specific genes regardless of differentiation state, we compared spheroids (stem cell-enriched organoids) with differentiated spheroids treated with MSP or PBS (Fig.S6E). MSP upregulated WAE cell-specific genes in both conditions but to a stronger degree in stem cell-enriched organoids (Fig.S6F). We then stimulated stem cell-enriched organoids with MSP for 24hr and analyzed epithelial marker expression. MSP treatment upregulated enterocyte marker intestinal alkaline phosphatase (Alpi) and downregulated stem cell marker leucine rich repeat containing G protein-coupled receptor 5 (Lgr5) but did not affect goblet cell marker mucin 2 (Muc2) (Fig.S6G), suggesting that MSP may encourage stem cell-to-enterocyte differentiation and supporting our finding that WAE cells differentiate from TA cells or immature enterocytes.

Figure 4. MSP can induce WAE cell transcriptional changes to promote epithelial restitution in vitro.

(A) (left) Schematic diagram of the murine colonic spheroid bulk RNA-seq experiment. Day 2 colonic spheroids were cultured in 0%L-WRN CM with or without recombinant mouse MSP (100 ng/ml) for 0, 2, 4, or 6hrs. Three biological replicates were included. (right) Volcano plot showing DEGs (two-sided negative binomial test) in MSP-treated relative to untreated colonic spheroids at 4 hrs. DEGs upregulated by MSP treatment that overlap with WAE cell markers are labeled (p_adj<0.05, average log₂(fold change)>0.5 for both). (B) Heat map showing WAE cell gene signature of colonic spheroids treated with or without MSP at indicated time points. (C) Venn diagram showing 58 of 207 WAE cell marker genes (p_adj<0.05, average log₂(fold change)>0.5) in scRNA-seq data that overlap with colonic spheroid bulk RNA-seq data at 4hr with MSP treatment (p_adj <0.05, average log₂(fold change)>0.5; union of all DEGs at three time points). (D) Immunoblot analysis of day 2 colonic spheroids cultured in 50% or 0% L-WRN CM treated with or without recombinant mouse MSP (100 ng/ml) for 24hr. (top) Representative blot image for Aldh1a3, Cldn4, Lamc2, and Gapdh. Three biological replicates were included. (bottom) Quantitative analysis of blot images (n=6 per condition). Expression was first normalized to Gapdh, then calculated relative to the PBS/50% L-WRN-treated condition. *p<0.05, ***p<0.001, ****p<0.0001 (two-way ANOVA with Sidak’s multiple comparison test). (E) Estimated proportion of WAE cells based on bulk RNA-seq data in colonic spheroids treated with or without MSP at indicated time points. (F-G) Cell migration assay. Colonic spheroids established from C57BL/6J mice were used. (F) Representative digital phase-contrast images of colonic monolayer. Yellow dotted lines show the margin of migrated cells at 42hr. Scale bars: 1mm. (G) Quantification of cell migration at 42 hr. Data are representative of three independent experiments. **p<0.01 (unpaired student’s t-test). Data represent mean ± SEM.

Next, we used transcriptional signatures for each cell type from our scRNA-seq of wounds as a reference for cellular deconvolution of the bulk RNA-seq of spheroids with CIBERSORT (30). These analyses showed induction of WAE cell genes but not of markers for other epithelial cell types in MSP-treated spheroids (Fig.4E, Fig.S6H). Finally, we examined whether the MSP-induced gene expression changes in organoids corresponded to functional healing responses that can be modeled in vitro. We developed a wound healing assay in primary colonic epithelial monolayers and demonstrated that wound closure and epithelial migration were significantly enhanced by MSP treatment (Fig.4, F and G; p=0.0030), consistent with our in vivo observation that MSP enema accelerated wound closure. Altogether, our results indicated that MSP signaling promotes WAE cell-associated wound healing.

WAE cells produce a retinoic acid gradient to coordinate fibroblast responses during wound healing.

Our efforts to characterize the function of WAE cells in response to tissue injury revealed a gene expression program that promotes epithelial restitution and wound closure. This WAE cell program is elicited and executed in a spatially distinct manner that requires extensive coordination with neighboring cells. Thus, we leveraged spatial transcriptomics to define WAE cell-mediated healing responses in situ. We repeated the endoscopy-guided wound assay and harvested three wounds and three intact tissues each from Hgfac WT and KO mice for spatial transcriptomics using the Visium platform (10x genomics) (Fig.5A left panel, Fig.S7, A and B). We then used our scRNA-seq data from wounds to locate individual cell types in the tissue transcriptomics using Tangram (31), finding broad cell lineages including epithelial, stromal, and immune cells on wounds versus intact tissue (Fig.S7C). In intact tissue, the three lineages stratified as expected, with the epithelial layer localized towards the luminal surface, stromal cells beneath that, and the signal for immune cells dim and scattered (Fig.S7C, bottom panels). This stratification was also observed in wounds, with enrichment in immune cell signature and upregulation of cytokines interleukin 1 beta (Il-1b), interleukin 11 (Il-11), chemokine ligand 9 (Ccl9), and Tgfb2 in the wound bed (Fig.S7, D and E). To determine where WAE cells are located relative to wounds, we mapped the WAE cell signature from scRNA-seq onto Visium coordinates and demonstrated that WAE cells were strictly localized at the surface of the wound sites (Fig. 5A, right panel). This was not observed in intact tissues (Fig.S8A).

Figure 5. Spatial transcriptomics reveal co-localization of WAE cells and proliferating MAFs in wounds in mice.

(A) (left) H&E images of day 2 wounds from 3 individual WT mice used for spatial transcriptomic analysis. Yellow squares show wound beds (W.B.). (right) Tangram mapping of WAE cells on the Visium spatial data. Color gradient depicts the percentage of mapped WAE cells. Number of nuclei per spot is indicated. The distance between the centers of each circle is 100 μm. (B) Tangram mapping of WAE cells and proliferating MAFs (pMAFs) on the Visium spatial data. Color gradient depicts the percentage of mapped WAE cells (yellow) and proliferating MAFs (purple). WAE cell spots are shown as bold squares. Spatial plot showing WAE cells and proliferating MAFs are overlapping or located in adjacent spaces. The distance between the centers of each square is 100 μm. (C) Distance (μm) to nearest WAE cell spot across all cell types. Center lines are the medians, box limits are the upper and lower quartiles, and whiskers are 1.5 times the interquartile range. Number of samples included is indicated in parentheses.

We next sought to integrate spatial localization of WAE cells with their functional responses to wounding. Our gene expression analysis between Hgfac WT and KO wounds identified components downstream of the HGFAC system that may coordinate wound healing responses via intercellular crosstalk: Aldh1a3 converts retinal into RA, and Dhrs9 catalyzes the interconversion of retinal and retinol. These genes were upregulated in WAE cells (Fig.S8B), and this response was impaired in Hgfac KO mice (Fig.3E). We confirmed via immunohistochemistry that Aldh1a3 is markedly decreased in Hgfac KO wounds (Fig.S8C), suggesting that the HGFAC system regulates Aldh1a3 expression. Given the role of RA as a morphogen, we hypothesized that WAE cells may utilize this pathway to spatiotemporally coordinate intercellular communication and promote wound healing in the mucosa. Indeed, Aldh1a3 was localized strictly in WAE cells at wound sites (Fig.3C, Fig.S8C). We used our scRNA-seq data to determine that cycling endothelial cells and proliferating MAFs express RA receptors (Fig.S8D) and potentially respond to WAE cell-derived signals. Moreover, proliferating MAFs and WAE cells mapped to overlapping or adjacent locations within the wound bed (Fig.5, B and C) but not intact tissue (Fig.S8E). Proliferating MAFs are Il-11- expressing cycling cells enriched in wounds that also express genes involved in ECM remodeling (lysyl oxidase (Lox), lysl oxidase like 3 (Loxl3), hyaluronan synthase 2 (Has2), EGF-containing fibulin-like extracellular matrix protein 2 (Efemp2), and collagen type VIII alpha 1 chain (Col8a1)), regeneration and tissue repair (Il-11, cellular communication network factor 4 (Ccn4), Fgf2, Fgf7, Tgfb2) (Fig.2D, Fig.S9, A to C), suggesting a central role in tissue healing.

These findings led us to hypothesize that WAE cells generate an RA gradient from wound sites to coordinate healing with fibroblasts. To test this, we treated mouse fibroblasts in vitro with RA and performed bulk RNA-seq to define an inducible RA transcriptional signature (Fig.S10, A and B) that included changes in retinol metabolism, ECM dynamics, and growth factor signaling (Fig.S10, C to E). We confirmed these changes by qPCR, demonstrating that RA-treated fibroblasts upregulated genes highly expressed in proliferating MAFs in vivo, including genes in the RA (dehydrogenase/reductase 3 (Dhrs3), cellular retinoic acid binding protein 2 (Crabp2), retinoic acid receptor beta (Rarb), and retinol binding protein 1 (Rbp1) and tissue healing (Fgf2, Lox, Loxl3, Col8a1) pathways (Fig.S10F). The top 30 RA-induced DEGs were particularly enriched in proliferating MAFs compared to other cell types in our scRNAseq data (Fig.S10G). Mapping the RA transcriptional signature onto spatial transcriptomic data revealed localized expression close to WAE cells within the wound bed that was absent in intact tissue (Fig.6A, Fig.S11A). Immunofluorescence imaging demonstrated that podoplanin⁺ MAFs accumulated in the wound bed adjacent to WAE cells (Fig.6B, Fig.S11B). CD45⁺ immune cells also accumulated in the wound bed, but the majority were localized within the interstitial space, with limited numbers at wound edges (Fig.6C, Fig.S11C). To determine which MAF subpopulations were spatially proximal to WAE cells, we performed RNAscope in situ hybridization. Consistent with our Visium analysis, proliferating MAFs (Il-11⁺ and/or Lox⁺) were located adjacent to WAE cells at wound edges (Fig.6D, Fig.S11D). Next, we performed spatial transcriptomic analysis of the RA transcriptional signature in Hgfac KO wounds (Fig.S11E) and evaluated if the WAE cell-derived RA gradient or MAF localization is impacted. MAF localization was not affected (Fig.S11F, p<0.0617) but the RA gradient appeared disrupted in Hgfac KO wounds as seen by reduced expression of RA-responsive genes (Fig.6, E and F). For validation, we performed RNAscope in situ hybridization on Hgfac KO wounds. Proliferating MAFs at wound edges showed specific expression of the RA-responsive gene Rbp1, suggesting that these cells may have received an RA stimulus from WAE cells (Fig.6, G and H, Fig.S11G). Rbp1 expression in immune cells was significantly lower than in proliferating MAFs in both Hgfac WT and KO mice (Fig.6I, p<0.0001), and Rbp1 expression in proliferating MAFs was reduced in Hgfac KO compared to WT mice (Fig.6, G to I, Fig.S11G). These data suggest that proliferating MAFs receive RA signals from WAE cells, and that RA production may be disrupted in Hgfac KO WAE cells. To determine the impact of the RA gradient on wound healing, we utilized an RA receptor inhibitor (BMS493) in vivo. Mice treated with BMS493 showed delayed wound healing compared to control-treated mice (Fig.S12A). WAE cells and MAFs were both present in wounds of BMS493-treated mice, although the wound structure was disorganized (Fig.S12B).

Figure 6. WAE cells produce retinoic acid (RA) locally to coordinate fibroblasts during wound healing.

(A) RA response signature visualized on the spatial transcriptomic data of wounds from 3 individual mice. Color gradients in each square depict RA transcriptional response (blue to yellow) and percentage of mapped proliferating MAFs (pMAFs, purple). RA transcriptional response was derived from the average expression of the top 30 DEGs 16hrs after RA treatment of fibroblast culture (fig. S10). Bold squares show the location of WAE cells mapped by Tangram (shown in Fig. 5B). Density plots depict the distribution of spots with high RA response score (greater than the upper quartile) in the wound beds (approximately three spots away from the wound), highlighting the concentrated area of RA-induced transcripts near WAE cells. The distance between the centers of each square is 100 μm. (B) Immunohistochemistry of WAE cells and proliferating MAFs. (left) WAE cells (Krt14) and MAFs (Pdpn) with DAPI staining. Scale bar: 100 μm. (right) Magnified image of yellow dotted square in left image. Scale bar: 50 μm. (C) Immunohistochemistry of MAFs (Pdpn) and immune cells (CD45). Scale bar: 100 μm. (D) RNAscope showing spatial distribution of WAE cells (probed with Clu) and proliferating MAFs (probed with Il-11/Lox) in WT day 2 wound. (left) Lower-magnification image. Scale bar: 100μm. (right) Magnified view of yellow dotted square in left image. Scale bar: 50 μm. Images in B-D are from serial sections; W.B. indicates wound bed. (E) Expression of RA transcriptional response genes (defined in A) with high gradient effect (negative correlation with distance to nearest WAE cell spot on the Visium slide). Hgfac WT and KO day 2 wound samples were analyzed. Expression of these genes was higher near WAE cells in WT compared to KO samples. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns: not significant. (F) Bar plot showing the gradient effect (negated Spearman correlation between expression and distance to nearest WAE cell spot; Spearman correlation<0, FDR<0.05) of DEGs induced by RA in fibroblast culture (log(fold change) >1). (G-H) RNAscope showing spatial distribution of Rbp1, a canonical RA-responsive gene, in relation to WAE cells, pMAFs, and immune cells in Hgfac WT and KO day 2 wounds. (G) Lower-magnification images. Scale bars: 100 μm. (H) Magnified views of yellow dotted squares in left images. In addition to Rbp1, Clu was used as a probe for WAE cells, Il-11/Lox were used for proliferating MAFs, and Ptprc was used for immune cells. Scale bars: 50 μm. (I) Semi-quantitative analysis of Rbp1 expression in Hgfac WT and KO day 2 wounds. Two fields of view per wound and two mice per genotype were used. **p<0.01, ***p<0.001, ****p<0.0001, ns: not significant (one-way ANOVA with Tukey’s multiple comparisons). Data represents mean +/− SEM.

Our results suggest that WAE cells may communicate with fibroblasts via RA to organize and promote wound healing. To more rigorously test this hypothesis, we first evaluated whether WAE cells produce RA and if Aldh1a3 is required. We generated Aldh1a3-deficient colonic spheroids by CRISPR knock-out and differentiated them into WAE cell monolayers with MSP (Fig.S13, A to C). After addition of the RA precursor retinol to the upper chamber of the transwell culture, we collected media from the lower chamber to measure RA concentration. Retinol-treated WAE cells produced more RA than control-treated cells as expected, and RA production was significantly reduced in Aldh1a3-deficient compared to WT cells (Fig.7A, p=0.00311), suggesting that Aldh1a3 is critical for WAE cells to produce RA. To reconstruct the spatial microenvironment in which WAE cells and fibroblasts interact, we developed a co-culture system and employed confocal microscopy to visualize fibroblasts on the bottom surface of a transwell apposed to a WAE cell monolayer on the top (Fig.7B). After treatment of WT WAE cells in the upper chamber with retinol, co-cultured fibroblasts upregulated RA-inducible genes (Dhrs3, Fgf2, Col8a1) associated with tissue remodeling (Fig.7C). This response was impaired in fibroblasts co-cultured with Aldh1a3-deficient WAE cells, indicating that Aldh1a3 may be necessary for WAE cells to derived RA and induce the observed changes in fibroblasts. Altogether, our data support a model in which tissue damage elicits a pathway through the HGFAC-MSP-RON axis to induce WAE cell differentiation and suggest that a localized RA gradient controlled by WAE cells may coordinate fibroblast-mediated tissue healing.

Figure 7. Ald1a3 is involved in generation of WAE cell-derived RA to promote fibroblast-mediated tissue repair in vitro.

(A) RA production from WAE cell monolayers. RA concentration was measured by ELISA (n=3–4 per condition). *p<0.05, ns: not significant (two-way ANOVA with Sidak’s multiple comparison test). Two independent experiments showed similar results. (B) (top left) Schematic diagram of WAE cell-fibroblast co-culture system, with a WAE cell monolayer on the top surface of a transwell and fibroblasts on the bottom surface. (top right) Representative z-stack image of WAE cell-fibroblast co-culture. WAE cells were stained with wheat germ agglutinin (WGA). Fibroblasts were stained with Vimentin. Nuclei are visualized with DAPI. Scale bar: 20 μm. (bottom) Representative confocal images of the WAE cell monolayer and fibroblasts. Scale bars: 20 μm. (C) qPCR analysis of RA-responsive genes in fibroblasts co-cultured with WT or Aldh1a3-deficient WAE cell monolayers. Fibroblasts were collected after 1 μM retinol (ROL) or DMSO treatment for 24 hr. Representative qPCR data (n=4 per condition). Three independent experiments showed similar results. *p<0.05, **p<0.01, ****p<0.0001, ns: not significant (two-way ANOVA with Sidak’s multiple comparison test). Fig. 7B was created in part using BioRender.com.

Discussion

Tissue homeostasis relies critically on wound healing. We demonstrate that HGFAC links the coagulation pathway with the growth factor MSP to promote a WAE cell-mediated response to injury, and that the IBD risk variant HGFAC R509H exhibits impaired proteolysis of proMSP and inefficient production of active MSP. In a murine model, we demonstrated that the mutation impaired MSP activation in response to injury and delayed wound healing. Our in vivo findings also support that MSP supplementation could potentially provide benefits to IBD patients with HGFAC R509H. These observations implicate the HGFAC-MSP-RON axis in WAE cell differentiation and function, and human genetics offer additional insights into the importance of this pathway in mucosal healing. HGFAC R509H was identified in an exome sequencing study as a risk factor for CD (24), and prior genome-wide association studies in IBD implicated the HGFAC locus (32) and a region on 3p21–22 that encompasses several genes including MST1 (encoding MSP) and MST1R (encoding its receptor RON) (33, 34). Two independently associated missense variants in MST1 [R703C (rs3197999) and R651X (rs142690032)] were identified as IBD risk variants (Table1) (33–35) and exist in linkage disequilibrium with three MST1R variants [R322Q (rs2230593), Q523R (rs2230590), and G1335R (rs1062633)]. MST1 represents an attractive candidate causal gene based on the independent low frequency R651X variant that has fewer variants in linkage disequilibrium and no other strong coding proxies. Moreover, IBD patients bearing heterozygous (C/T) and homozygous (T/T) MST1 R703C showed 27% and 50% lower serum MSP concentrations compared to carriers of the common variant (19). Additionally, MSP and RON expression were significantly reduced in colonic tissue from patients with ulcerative colitis (p<0.05) (36). To the best of our knowledge, no IBD risk variants have been identified in HGF and MET (Table1).

Despite the recognized importance of WAE cells in wound healing, WAE cell function and regulation are incompletely understood due to their paucity in tissue and technical challenges associated with molecular profiling of rare cell types. To deeply characterize WAE cell programs in space and time, we used an endoscopy-guided model to precisely monitor wounds, and advancements in scRNA-seq and spatial transcriptomics allowed us to perform detailed mechanistic assessments of wound healing. We demonstrate the importance of HGFAC and downstream MSP in efficient WAE cell differentiation and function both in vitro and in vivo. Amongst the functional programs induced in WAE cells, we identified genes related to actin cytoskeleton rearrangement, focal adhesions, cell proliferation, and survival that likely facilitate WAE cell migration and wound closure. Importantly, several WAE cell marker genes overlap with gene signatures from WAE cell-like subsets in other mouse injury models and IBD patient samples (37, 38), suggesting that they share a common epithelial regeneration program in response to injury. We additionally provide evidence that WAE cells communicate with fibroblasts through RA production.

RA performs key morphogenic functions. Although RA improves wound healing when locally administration to the skin (39) and can stimulate fibroblasts (40–44), mechanistic questions remain regarding its cellular sources, inducing stimuli, and cellular targets at wound sites. Immune cells are important effectors of RA signaling (45–47), although less is known about the role of RA within the larger intercellular network that governs wound healing. Under homeostatic conditions, vitamin A from the diet is absorbed by intestinal epithelial cells, converted into RA via Aldh1a1, and secreted into the small intestinal tissue (47). CD103+ DCs synthesize RA from retinol via Aldh1a2 to modulate B and T cell functions, such as Treg differentiation and Th17 cell development (45, 47–50). Here, we demonstrated that WAE cells are a major source of RA in wounded intestinal mucosa and that RA may be a key mediator of the fibroblast response in tissue repair and regeneration.

Our study has several limitations. We demonstrated involvement of the HGFAC-MSP-RON axis in WAE cell differentiation and function; however, detailed transcription factor networks governing these processes remain to be investigated in future studies. We additionally examined the impact of WAE cells specifically on stromal cells in the context of wound healing, but impacts on other cell types, such as immune cells, are also important to consider.

Overall, we leveraged a CD risk variant in HGFAC that is associated with impaired mucosal healing to uncover a coordinated wound healing response mediated by WAE cells and fibroblasts. Integrative spatial profiling methods enabled characterization of an intercellular communication network whereby WAE cells are involved not only in epithelial restitution but also in orchestrating fibroblast activity to promote tissue healing.

Materials and Methods

Study design

We aimed to identify the molecular mechanisms by which Hgfac R509H increases the risk of CD and to study the wound healing process using this genetic perturbation. Serum was obtained from healthy donors carrying either HGFAC R509 or R509H. Mechanistic and functional studies utilized inbred mouse strains, C57BL/6 mice purchased from the Jackson Laboratory, primary colonic epithelial spheroids, a mouse fibroblast cell line (NIH3T3), and a human embryonic kidney cell line (HEK293T). For the in vivo model study, age- and sex-matched mice were randomly assigned to experimental groups. Data collection and analysis were performed in a blinded manner whenever possible. Data analysis was not blinded for experiments grouped by genotypes due to the experimental design, but unbiased quantification was applied to obtain results. For the endoscopic-guided wound model, investigators were blinded to group allocation. A power calculation was not performed to determine the sample size; instead, sample sizes were estimated based upon our experience and previous studies in the field. Sample sizes are noted in the figure legends. All relevant data were included in this study, and all attempts at replication were successful. Further details are described in each figure legend. Primary data for panels with n<20 are reported in Data File S1.

Statistical analysis

Unless otherwise stated, a significant difference was considered p<0.05. Data are presented as mean ± SEM. Comparisons between two groups were performed by using a two-tailed unpaired t-test or two-way ANOVA with Sidak’s multiple comparison test. Comparisons between three groups were performed by one way-ANOVA with Tukey’s multiple comparison test. All statistical analysis was performed in GraphPad Prism 8.2 (GraphPad Software). The p values are presented as follows: *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. P values were corrected for multiple testing using the Benjamini-Hochberg method.

Data and materials availability

All data associated with this study are in the paper or supplementary materials. Sequencing data have been deposited in Genome Expression Omnibus (GEO) under accession number GSE235743 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE235743). Scripts for reproducing the analysis can be accessed via Zenodo (DOI 10.5281/zenodo.8011746).