Bidirectional CRISPR screens decode a GLIS3-dependent fibrotic cell circuit

Abstract

The stromal cell compartment plays a central role in maintaining tissue homeostasis by coordinating with the immune system throughout the inception, amplification, and resolution of inflammation¹. Chronic inflammation can impede the phased regulation of tissue restitution, resulting in the scarring complication of fibrosis. In inflammatory bowel disease (IBD), stromal fibroblasts have been implicated in treatment-refractory disease and fibrosis^2,3; however, their mechanisms of activation remain undefined. Through integrative single-cell and spatial profiling of intestinal tissues from IBD patients, we uncovered a pathological cell nexus centered on inflammation-associated fibroblasts (IAFs). These IAFs were induced by pro-inflammatory macrophages (FCN1⁺, IL1B⁺) and, in turn, produced the pro-fibrotic cytokine IL-11. Mechanistic dissection of the IAF activation program was achieved through genome-wide CRISPR knockout and activation screens, identifying the transcription factor GLIS3 as a key regulator of a gene-regulatory network governing expression of inflammatory and fibrotic genes. We further demonstrate that the magnitude of the GLIS3 gene expression program in intestinal biopsies stratifies patients with ulcerative colitis (UC) by disease severity, and fibroblast-specific deletion of Glis3 in mice alleviates pathological features of chronic colitis. Taken together, our findings unveil a critical immune–stromal cell circuit that functions as a central node in the inflammation–fibrosis cycle.

Chronic inflammation overstimulates fibrogenesis, culminating in fibrosis—a condition accounting for 45% of disease-related deaths with limited treatment options¹. Single-cell profiling has uncovered functionally distinct and location-specific fibroblasts central to driving fibrosis⁴. Apart from pan-tissue PI16⁺ or COL15A1⁺ fibroblasts, inflammation-associated signals instigate fibroblast states that impair the resolution of injury-induced inflammation, deposit excessive fibrotic collagen, and alter tissue mechanics^5–7. In several human inflammatory diseases, CXCL10⁺CCL19⁺ and SPARC⁺COL3A1⁺ fibroblasts expand in immune- or vasculature-associated tissue niches, respectively, where they promote pathology⁸. Targeting these processes is challenging due to a limited understanding of the molecular basis of disease-associated fibroblasts and the fact that existing immunosuppressives block pro-inflammatory mediators, which are not fibroblast-specific.

We previously reported IAFs expand in IBD, and express an inflammatory and fibrogenic gene signature associated with resistance to anti-tumor necrosis factor (TNF) therapy^2,3. This signature was enriched prior to treatment resistance, implying that IAFs promote disease progression despite medical intervention⁹. RNA-sequencing has also implicated oncostatin M (OSM) signaling and neutrophil recruitment in therapy resistance; however, the IAF-specific determinants and their relation to other resistance markers are not fully understood^10,11. Identifying the underlying cellular and molecular wiring of fibroinflammatory processes is critical to developing new treatments, as general immunosuppressives do not improve long-term fibrotic outcomes and biologics can have adverse effects or lose efficacy¹². As fibroblasts have emerged as central drivers of tissue remodeling after inflammation, we aimed to decipher the inter- and intra-cellular wiring of IAFs in a disease marked by fibrosis development.

A single-cell and spatial atlas of IBD

To systematically decipher the shared and distinct cellular and molecular drivers of Crohn’s disease (CD) and UC, the principal clinical subtypes of IBD, we integrated single-cell RNA-sequencing (scRNA-seq) data from the small and large intestine with publicly available IBD datasets^2,9,11,13, constructing a single-cell IBD atlas comprising 29 non-IBD control, 29 UC, and 57 CD patient samples (Fig. 1a, Extended Data Fig. 1a). To contextualize cellular-spatial relationships and their associations with distinct histopathological features, we applied Xenium-based single-cell spatial profiling and mapped these cells in the intestinal tissue. We included pathologist-annotated tissue resections from the small or large intestine of 4 non-IBD control (normal cuff of colon adjacent to diverticulitis), 3 ileal and 3 colonic CD, and 6 colonic UC patient samples (Supplementary Data 1). These atlases profile over four million intestinal cells across epithelial, immune, stromal, and fibroblast compartments and provide a comprehensive framework to dissect the etiology of IBD (Fig. 1a, Extended Data Fig. 1b).

Figure 1: An integrated single-cell and spatial atlas reveals inducible, inflammation-associated fibroblasts in pathological cellular niches.

(a) Schematic of the integrated IBD atlas workflow that utilizes scRNA-seq or Xenium to profile non-IBD, CD, and UC patients. IBD: inflammatory bowel disease; CD: Crohn’s disease; UC: ulcerative colitis.

(b) Proportion changes of fibroblasts stratified across disease. Box plots represent the quartiles with medians as the center, and whiskers the 10–90% range. Statistical analysis was performed using scCODA (Bayesian Dirichlet-multinomial model) with SMC as reference (FDR < 20%) (Methods). Number of samples for each category: non-IBD=29; CD inflamed=28; CD non-inflamed=54; UC inflamed=25; UC non-inflamed=22.

(c) Pseudobulk expression heatmap of scaled average IAF-specific genes (Wilcoxon signed rank test [two-sided], p < 0.05; log fold change > 3; expression in > 25% of IAFs, < 10% of non-IAFs). IAF CSGs: inflammation-associated fibroblast cell-specific genes.

(d) Pseudobulk scaled expression heatmap of IAF genes involved in extracellular matrix (ECM) deposition/organization or cytokine/chemokine production.

(e) Dot plot showing effect size (β) and absolute log₂ fold change (|LogFC|) for niche enrichment across CD and UC compared to non-IBD. Blue indicates enrichment in non-IBD and red in CD and UC. Analysis performed using scCODA with N3 reference niche. FDR < 20% for niche change in abundance.

(f) Heatmap of statistically enriched cell type proportion abundance across niches. Chi-squared test with p < 0.05 was set as the significance threshold.

(g) Left: visualization of cellular niches projected onto a Xenium-profiled UC patient tissue section. Right: distribution of IAFs on the same tissue section, showing dense distribution in niche N1.

(h) H&E section of UC tissue from (g) showing annotated anatomical and pathological tissue regions. Images are representative of the sample cohort. n=16 patients.

(i) Heatmap depicting the enriched niches within anatomical and pathological tissue domains.

Across IBD, we observed significant changes in cell-type abundance within defined intestinal compartments (Extended Data Fig. 1c). Within the fibroblast compartment, most cells did not change in abundance in disease, with the exception of reparative ADAMDEC1⁺ fibroblasts¹⁴, which decreased in inflamed CD and UC, while a subset of IL11-expressing IAFs expanded in inflamed CD and UC (Fig. 1b). IL-11 is a constituent of the IL-6 family, a group of cytokines that dictate the balance between tissue repair and fibrosis, with IL-11 emerging as a pro-fibrotic cytokine¹⁵. Alongside IL11, IAFs specifically expressed genes associated with fibrosis and impaired tissue functionality, including CD82, PRRX1, and CHI3L1^16–18 (Fig. 1c). IAFs also upregulated extracellular matrix (ECM) remodeling genes, including COL1A1 and COL6A1, as well as ABL2, which drives cytoskeletal rearrangement^19,20 (Fig. 1c,d). Furthermore, IAFs expressed neutrophil-recruiting chemokines (CXCL3, CXCL5, CXCL8), indicating that they may direct immune cell activity during disease^9,11 (Fig. 1d). Pathway analysis of the IAF transcriptome revealed enrichment of ECM organization alongside inflammatory gene expression (Extended Data Fig. 1d). We previously reported that multiple IAF genes are associated with a signature or refractory response to anti-TNF treatment^2,21 (Extended Data Fig. 1e).

Given the increased prevalence of IAFs in inflamed UC or CD patients, we hypothesized their expansion reflects concurrent intestinal tissue remodeling. We defined IAF-anchored multicellular niches by implementing cell neighborhood analysis, systematically quantifying cellular neighbors within a 30-μm radius to capture contact- and secreted-ligand–receptor interactions²². We identified 19 distinct cellular niches, each comprising 5,298 to 778,470 cells summed across all samples (Extended Data Fig. 2a). The niches corresponded to distinct anatomical layers and several were significantly changed in abundance in disease (Extended Data Fig. 2b,c). To account for anatomical variation, we examined niche–cell type composition separately in colonic and ileal samples. While niche compositions remained broadly consistent between anatomical sites, several epithelial cell niches (N5, N11, N12) displayed greater compositional variation (Extended Data Fig. 2d). In CD and UC, mucosal epithelial niches (N7, N16, N5, N11, N12) were depleted; conversely, lymphocyte-enriched (N2 and N15) and myeloid-enriched (N1, N9, N14) niches were significantly expanded (Fig. 1e,f). IAFs were statistically enriched in niches N1 and N14 in both colon and ileum, which comprised stromal and immune cells, with the strongest shared cell type enrichment being FCN1⁺IL1B⁺ macrophages (activated macrophages) that resemble CD68⁺ infiltrated mucosal macrophages in IBD²³ (Fig. 1f, Extended Data Fig. 2e). These monocyte-derived macrophages were characterized by canonical activation pathways and specifically expressed pro-inflammatory cytokines (IL1B, TNF, OSM) and innate immune sensors (TLR2, NLRP3), indicating that N1 and N14 mark inflamed regions influenced by stromal–immune interactions (Extended Data Fig. 2e,f). Histological characterization suggested that N1 and N14 were highly prevalent in tissues in advanced stages of fibrosis and ulceration (Extended Data Fig. 2g–i). We then refined our analyses to map the specific histopathological regions associated with these niches, observing that N1 and N14 were localized to mucosal regions with active or chronic colitis (Fig. 1g–i, Extended Data Fig. 2j). Collectively, our results reveal that IAFs localize to niches associated with active disease, potentially integrating signals from the tissue microenvironment to enact fibroinflammatory programs associated with disease.

An IL-11 cell circuit governs fibrosis

To define the spatial roles of IAFs, we generated conditional Il11 knockout mice alongside a reporter line, enabling elucidation of IAF function and distribution in a colitis model with stromal-driven fibrosis. We flanked Il11 exons 2-4 with LoxP sites (Il11^f/f) and bred this line onto tamoxifen-inducible cre/ERT2, enabling temporal deletion of Il11 (Il11^f/f;Cre) (Extended Data Fig. 3a,b). Concurrently, we engineered a novel Il11 reporter mouse by inserting T2A-mNeonGreen (mNG) before the Il11 stop codon (Il11^mNG), leveraging the selective expression of IL11 in IAFs (Extended Data Fig. 3c).

We modeled chronic intestinal inflammation with features of stromal-driven fibrosis using the chronic dextran sodium sulfate (DSS) regimen of three 7-day DSS cycles, each followed by a recovery phase. Il11^f/f and Il11^f/f;Cre mice were injected with tamoxifen after each DSS cycle to ablate Il11 expression in Il11^f/f;Cre mice. To assess whether IL-11 promotes fibrotic collagen deposition, we performed Masson’s trichrome staining on DSS- or water-treated colon Swiss-rolls, and quantified collagen-positive area as a percentage of total tissue area. Chronic DSS exposure increased collagen deposition in Il11^f/f mice relative to water-treated controls, but this was reduced in DSS-treated Il11^f/f;Cre mice (Fig. 2a). Correspondingly, tissue hydroxyproline, a proxy for total collagen content, was elevated in DSS-treated Il11^f/f mice but reduced in DSS-treated ll11^f/f;Cre mice (Fig. 2b). Expression of pro-fibrotic collagen genes (Col1a1, Col5a1, Col5a2, Col6a1) followed the same pattern of reduction in DSS-treated ll11^f/f;Cre compared to Il11^f/f mice (Fig. 2c), indicating that IL-11 promotes injury- and inflammation-induced fibrosis. Both genotypes exhibited shared trends in weight loss following DSS administration and histopathological scores of tissue inflammation, indicative of similar extents of colitis development in the absence of Il11, yet DSS-treated Il11^f/f;Cre mice exhibited reduced inflammation-driven tissue remodeling (Fig. 2d, Extended Data Fig. 3d,e). These findings extend previous reports showing exacerbated colitis after acute intestinal injury of Il11^−/− mice and spontaneous or aggravated colitis in transgenic or recombinant IL-11-treated mice^24–26, thereby highlighting a key role for endogenous Il11 in promoting fibrotic remodeling during chronic colitis.

Figure 2: An IL-11 cell circuit governs fibrosis.

(a) Masson’s trichrome-stained Il11^f/f and Il11^f/f;Cre colons (8–18 weeks) treated with water or chronic DSS (left). Total collagen percentage from three pooled experiments (right). Il11^f/f-water, n=10; Il11^f/f;Cre-water, n=14; Il11^f/f-DSS, n=17; Il11^f/f;Cre-DSS, n=10 mice.

(b) Colonic hydroxyproline normalized to total protein from tissues from (a). (c) qPCR quantification of collagens normalized to Eef2 from tissues from (a).

(d) Colon length measurements from (a).

(e) Percentage of IL-11^mNG cells across lineages after indicated treatments. Water-treated, n=2; DSS-treated, n=3 mice.

(f) Masson’s trichrome- (left) and immunofluorescence- (right) stained Il11^mNG tissues after DSS. Images are representative of 3 independent experiments.

(g) Schematic of PDGFRA+ fibroblast isolation from acute and chronic DSS-treated Il11^mNG mice (8–14 weeks) (left). Dot plot mapping human fibroblast gene signatures across mouse fibroblasts (right).

(h) Pseudobulk expression heatmap depicting scaled average expression of Il11 and mNeonGreen from acute and chronic DSS.

(i) Spatial niche-aware probability of intercellular communication. Edge thickness or node size depicts communication strength. Significant signals received by IAFs (left) and sent from activated macrophages (right).

(j) Immunofluorescence of chronic DSS-treated colons from Il11^mNG mice depicting proximal macrophage (CD68, red) and IL-11^mNG fibroblast (green) localization. Arrows indicate signal adjacency. Images are representative of 3 independent experiments.

(k) Spatial projection of IAFs and activated macrophages in non-IBD and CD tissues.

(l) Dot plot of IAF IL11 expression as a function of proximity to activated macrophages.

(m) Secreted IL-11 measured from co-cultures of polarized primary human monocyte-derived macrophages, removed of agonists, with colonic fibroblasts for 24 hours. Fibroblasts only, n=4; fibroblasts+macrophages, n=2; fibroblasts+polarized macrophages, n=3 cell lines.

Mice were co-housed and DSS followed the same regimen: acute (2.0%, 7 days), chronic (2.0%, 42 days). Unless otherwise stated, statistics are by a two-way ANOVA with Tukey’s multiple comparison test on distinct biological replicates and error bars are mean ± s.e.m. ns, not significant.

We next utilized Il11^mNG mice to map the spatial distribution of Il11 expression within the colon and interrogate its associated histopathological features. Following water or acute (one 7-day cycle) or chronic DSS treatment, we stained dissociated colonic cells for lineage-specific markers. mNG fluorescence was undetectable under water treatment and in non-fibroblast lineages, emerging exclusively in DSS-treated PDGFRA⁺ fibroblasts, with higher intensities observed after chronic DSS (Fig. 2e). Immunofluorescence detected IL-11^mNG fibroblast clusters only after DSS-treatment, localizing to regions of epithelial damage as in acute intestinal injury^25,27. Importantly, IL-11^mNG fibroblasts accumulated in areas of pronounced collagen deposition and inflammation after repeated cycles of injury-repair, modeling progressive stromal-driven tissue remodeling (Fig. 2f). These findings reveal an interplay with infiltrating immune cells that may influence IAF development or function, including collagen deposition.

To decipher the relevance of Il11^mNG mouse fibroblasts to human IAFs, we utilized scRNA-seq to profile PDGFRA⁺ fibroblasts recovered after acute and chronic DSS, identifying ten distinct fibroblast clusters (Extended Data Fig. 3f,g). Cross-species comparison revealed mouse IAFs (mIAFs) had the strongest enrichment for human IAF and anti-TNF resistance genes²¹ (Fig. 2g, Extended Data Fig. 3h,i). Consistent with increased IL-11^mNG expression after chronic DSS, mIAFs increased in relative abundance after chronic versus acute DSS (Extended Data Fig. 3j). mIAFs specifically co-expressed Il11 and the mNG reporter gene, confirming the specificity of mNG for IL-11 and mIAFs (Fig. 2h). Collectively, we identify cross-species conservation of IAFs in fibroinflammatory intestinal disease.

Given the presence of IAFs in inflamed and fibrotic tissue regions, we investigated the ligand-receptor interactions guiding their development (Methods). Amongst all the immune cell types sending signals to IAFs, activated macrophages relayed the strongest signals; meanwhile, IAFs were their top recipients (Fig. 2i). In chronic DSS-treated mouse colons, CD68 co-staining confirmed close proximity between macrophages and IL-11^mNG fibroblasts, including direct adjacency of mNG and CD68 fluorescence (Fig. 2j). Whereas Il11 was previously reported proximal to LysM-expressing cells after acute intestinal injury, our data resolve their spatial association in chronic disease^25,27. Importantly, within human CD and UC but not non-IBD patients, IAFs and activated macrophages co-occupied niches of active and chronic colitis, with IAF IL11 expression correlating with activated macrophage proximity (Fig. 2k,l). Our results suggest an activated macrophage–IAF circuit in which spatial proximity promotes IAF development.

We next co-cultured primary colonic fibroblasts with monocyte-derived macrophages preconditioned for either inflammation resolution (IL-4 or TGF-β) or pro-inflammatory activation (IFN-γ, LPS/IgG complexes, or adenosine). After removal of the stimuli, only pro-inflammatory macrophages induced fibroblast IL-11 secretion, whereas unstimulated or inflammation resolution macrophages had no effect (Fig. 2m). Macrophages challenged with an array of microbial- or damage-associated ligands triggered IL-11 secretion in co-cultured fibroblasts, whereas fibroblasts alone were unresponsive (Extended Data Fig. 3k). This secretion was driven by de novo transcription, as IL11 mRNA increased in co-cultured fibroblasts (Extended Data Fig. 3l). Collectively, our results indicate that activated macrophages are sufficient to induce IL11-expressing IAFs, and spatial co-association of these cells in vivo suggests they coordinate a critical inflammation-driven intercellular circuit in disease.

TGF-β and IL-1β drive IAFs

We implemented a computational approach to define the macrophage-derived signaling and gene regulatory networks driving IAF activation (Extended Data Fig. 4a–d, Methods). We inferred putative active IAF transcription factors from expression and activity scores, then predicted their converging upstream ligands using NicheNet’s ligand-receptor model²⁸, restricted to expressed receptors. This approach yielded 35 candidate IAF agonists and screening primary colon fibroblasts with the candidate agonists revealed IL-11 secretion exclusively by TGF-β or IL-1β/α (Extended Data Fig. 4e). Consistent with these results, activated macrophages highly expressed IL1B as well as TGFB1, while IAFs expressed their cognate receptors (Extended Data Fig. 4f,g).

To ascertain the contributions of TGF-β and IL-1β to IAF activation, we utilized knockouts of these ligands or their receptors in co-cultures between fibroblasts and TLR2/6-activated macrophages, since activated macrophages highly expressed TLR2, and TLR2/6-activated macrophages were robust activators of fibroblasts (Extended Data Fig. 2e, 3k). Upon TGF-β or IL-1β receptor knockout in fibroblasts, we observed impaired secretion of IL-11 and reduced expression of all measured IAF-specific genes (Extended Data Fig. 5a–c). Conversely, upon TGFB1 or IL1B1 knockout in macrophages, IL-11 secretion was reduced in fibroblasts, demonstrating dual pathway involvement (Extended Data Fig. 5a). Notably, macrophage IL1R1 was also required for fibroblast activation, implicating autocrine feedback (Extended Data Fig. 5a). Beyond TLR2/6, diverse immune ligands induced macrophage TGF-β and IL-1β secretion, indicating these cytokines are produced during co-culture with fibroblasts (Extended Data Fig. 5d). Stimulating fibroblasts with both cytokines led to synergistic rather than additive IL-11 production compared to individual treatments, consistent with elevated IL-6 family cytokines correlating with maladaptive tissue repair, and in alignment with the synergizing effect of this dual cytokine stimulation regimen^29,30 (Extended Data Fig. 5e, Methods).

To demonstrate that TGF-β and IL-1β are required to drive IL-11 production in vivo, we neutralized these cytokines with monoclonal antibodies (mAbs) in chronic DSS-treated Il11^mNG mice. Neutralization of TGF-β, IL-1β, or both decreased the number of IL-11^mNG fibroblasts and Il11 mRNA in the colon (Extended Data Fig. 5f,g). Cytokine blockade showed a subtle trend in alleviating fibrosis and did not ameliorate overall inflammation (Extended Data Fig. 5h–l). The failure to rescue disease severity likely reflects disruption of broader TGF-β- and IL-1β-dependent signaling programs involved in immune homeostasis and inflammation beyond IAFs, as receptors for both cytokines are broadly expressed in other cell types^31,32 (Extended Data Fig. 4g). Importantly, our findings instead identify TGF-β and IL-1β as crucial for driving the IAF state characterized by elevated IL11 and fibroinflammatory genes.

CRISPR screens reveal IAF determinants

Given the synergy between TGF-β and IL-1β in fibroblast activation and the need to identify fibroblast-specific targets, we investigated the signaling pathways underpinning this effect using parallel genome-wide CRISPR knockout and activation screens anchored on IL11 as a readout. We introduced mNG at the endogenous terminus of IL11 in human fibroblasts (IL11^mNG), which increased in fluorescence upon dual-cytokine stimulation (Extended Data Fig. 6a). We then stably expressed Cas9 or dCas9-VP64 for gene knockout or activation, respectively. After TGF-β and IL-1β stimulation, we sorted the top and bottom 15% IL11^mNG expressers and sequenced gRNAs enriched in IL-11 hyper- or hypo-producers, respectively (Fig. 3a). Integrating the loss- and gain-of-expression screens uncovered 61 shared hits, including TGF-β and IL-1β signaling components (TGFBR1, TGFBR2, SMAD3, IRAK2, MAPK1, RELA), known immunomodulatory genes (LAMTOR, HIC1, NFKBIZ, ARNT2), and IL11 itself (Fig. 3a–c, Supplementary Data 2). Importantly, several of these genes were selectively expressed in IAFs and upregulated during inflammation, underscoring their relevance to IAF-driven disease (Fig. 3d, Extended Data Fig. 6b).

Figure 3: Genome-wide CRISPR screens discover novel IAF determinants.

(a) Volcano plots of enriched hits based on fold change (FC) enrichment and p values. CRISPRko, CRISPR knockout; CRISPRa, CRISPR activation. One-sided hypergeometric test. Multiple comparisons adjusted by FDR.

(b) Pathway diagram of enriched hits from CRISPRko (red), CRISPRa (blue), or both screens (green) in known pathways.

(c) Scatter plot of CRISPRko and CRISPRa screens’ shared hits. Statistically significant hits (p < 0.05) are boxed and in black, and selected hits labeled. One-sided hypergeometric test. Multiple comparisons adjusted by FDR.

(d) Heatmap depicting log fold change (FC) expression of CRISPR screen hits across human fibroblasts. Hit selection filtered by differential expression in IAFs (Wilcoxon test, adjusted p < 0.01 [two-sided]; > 1% expression in IAFs).

(e) scRNA-seq of primary human fibroblasts stimulated with TGF-β and IL-1β (10 ng/mL). UMAP of timepoint clusters (left) and IL11 expression (center). High IL11 expression in subclusters 6 and 10 (right).

(f) Top-ranked mean correlation values between high IL11-expressing subclusters 6 and 10 from (e) and shared CRISPR screen hits from (c) reveal top enrichment of GLIS3, with p = 1.6x10⁻¹³ in cluster 6, and p = 3.9x10⁻⁹ in cluster 10.

(g) Relative percentage of IL-11^mNG median fluorescence intensity (MFI) in GLIS3-perturbed immortalized Cas9- or dCas9-VP64 fibroblasts compared to controls after TGF-β and IL-1β stimulation (24 hours). Control, n=4; CRISPRko/a, n=2 cell lines.

(h) Secreted IL-11 from distinct biological replicates measured after co-culture of primary human colonic GLIS3 CRISPRko (left) or GLIS3 CRISPRa (right) fibroblasts with TLR2/6-activated monocyte-derived macrophages. Error bars are the mean ± s.e.m. ns, not significant. Two-way ANOVA with Tukey’s multiple-comparisons test. n=3 cell lines.

To complement these screens, we performed time-resolved scRNA-seq of fibroblasts stimulated with TGF-β and IL-1β, unveiling IL11 expression within 6 hours of stimulation that persisted at 24 hours (Fig. 3e). We observed pronounced IL11 expression in stimulated fibroblast clusters 6 and 10, suggestive of molecular determinants within these clusters that enhanced its transcription (Fig. 3e). Correlation analysis of our bidirectional CRISPR screen hits with highly expressed genes in clusters 6 and 10 revealed GLIS3 as the top-ranked gene (p=1.6x10⁻¹³ in cluster 6, and p=3.9x10⁻⁹ in cluster 10) (Fig. 3f). GLIS3, a GLI-related protein regulating pancreatic β-cell development and thyroid gland function³³, was distinctly expressed in both IAFs and mIAFs during inflammation (Fig. 3d, Extended Data Fig. 6b,c). Functional studies confirmed the role of GLIS3 in IL-11 production, as its CRISPR knockout (CRISPRko) impaired IL-11^mNG fluorescence, whereas its CRISPR activation (CRISPRa) heightened it (Fig. 3g). Furthermore, during co-culture with TLR2/6-activated macrophages, IL-11 secretion was attenuated in GLIS3 CRISPRko and elevated in CRISPRa fibroblasts (Fig. 3h).

To assess the role of GLIS3 in this macrophage–fibroblast circuit, we endogenously tagged GLIS3 (GLIS3^3XFLAG), fluorescently labelled fibroblasts and macrophages with different dyes, and co-cultured them with or without TLR2/6 stimulation. Activated, but not resting, macrophages increased nuclear GLIS3, which formed puncta-like structures (Extended Data Fig. 6d). This increase originated from transcriptional upregulation (Extended Data Fig. 6e) and was potentiated and sustained in fibroblasts by dual TGF-β and IL-1β stimulation in an additive manner, mirroring the enhanced IL-11 production observed after dual stimulation (Extended Data Fig. 6f). Importantly, in vivo neutralization of TGF-β, IL-1β, or both decreased Glis3 expression in the colon of chronic DSS-treated mice (Extended Data Fig. 6g). These results implicate nuclear GLIS3 in integrating the combinatorial effects of macrophage-derived TGF-β and IL-1β, thereby directing IL11 expression in IAFs.

GLIS3 controls the IAF gene program

Having identified GLIS3 as a central regulator of IL11 expression in IAFs, we sought to determine its broader transcriptional program through RNA-sequencing of GLIS3 CRISPRko and CRISPRa fibroblasts stimulated with TGF-β and IL-1β. We identified >150 genes with reduced expression in GLIS3 CRISPRko yet enhanced expression in CRISPRa fibroblasts, identifying them as GLIS3 effector genes (Fig. 4a). In addition to GLIS3 and IL11, we observed enrichment of other IAF-specific genes implicated in fibroblast-driven pathologies, including LIF, which sustains fibroblast activation through positive feedback signaling³⁴, and FAP, an IAF gene linked to intestinal strictures in CD³⁵ (Fig. 4a). GLIS3 also regulated the IAF gene MMP2, a matrix remodeling enzyme that enables monocyte trafficking into damaged tissue, likely fueling further macrophage–fibroblast crosstalk^18,36, and PTGFR and SERPINE1, both of which regulate epithelial regeneration and contribute to mucosal damage during colitis^37,38 (Fig. 4a). Importantly, GLIS3 modulated the expression of fibrotic collagens COL6A1 and COL6A3, which were elevated in steady-state GLIS3 CRISPRa compared to WT fibroblasts (Fig. 4a, Extended Data Fig. 7a).

Figure 4: GLIS3 controls the IAF gene program.

(a) Top: Venn diagram of downregulated genes in GLIS3 CRISPRko or upregulated in GLIS3 CRISPRa fibroblasts after TGF-β and IL-1β stimulation (10 ng/mL, 24 hours). Significant genes with Benjamini & Hochberg adjusted p < 0.05 (Wald test, two-sided) were intersected to derive effector genes. Bottom: heatmap of average log fold change (FC) expression relative to controls. Displayed are intersecting effector genes; asterisks mark ChIP-seq peaks. n=3 per condition.

(b) Schematic of ChIP-seq in IL11^mNG fibroblasts with GLIS3^3XFLAG knock-in after TGF-β and IL-1β stimulation (10 ng/mL, 24 hours) (left). Pie chart depicts GLIS3^3XFLAG IP peak distribution (right).

(c) Gene ontology analysis of GLIS3^3XFLAG peaks at 24- versus 0-hours. Benjamini-Hochberg adjusted p values from hypergeometric test (one-sided). See Methods for more details.

(d) ChIP-seq tracks upstream of the first exon of IL11 in IgG and GLIS3^3XFLAG IP samples, summed across all replicate samples.

(e) Predicted TF binding motifs and their associated TFs enriched in all GLIS3^3XFLAG peaks.

(f) ChIP-qPCR schematic of FOSL1 and TEAD1 IP in control or GLIS3 CRISPRko IL11^mNG fibroblasts after TGF-β and IL-1β stimulation (10 ng/mL, 24 hours) (left). Heatmaps depict Z-score fold enrichment for FOSL1 (center) or TEAD1 (right) targets across replicates. n=4 cell lines per condition.

(g) Top: schematic of GLIS3 signature derivation and PROTECT cohort analysis. Bottom: CIBERSORT-estimated IAF and macrophage proportions across PROTECT samples stratified by Mayo score. Box plots represent the quartiles with medians as the center, and whiskers represent 1.5* inter-quartile range. Grey lines indicate mean GLIS3 ssGSEA score ± s.e.m. **p< 0.001 (one-sided) from ordinal probit regression of Mayo scores with ssGSEA and cell proportions (n=226). See Methods for more details.

(h) Heatmap of scaled average expression of refined GLIS3 signature across control and UC patients stratified by the combined Mayo score.

To define direct GLIS3 target genes, we performed chromatin immunoprecipitation sequencing (ChIP-seq) in activated GLIS3^3XFLAG fibroblasts, identifying 1,291 GLIS3-bound peaks, of which 15.2% mapped within −1kb to +100bp of transcription start sites (TSS) and 40.4% at intragenic regulatory regions (Fig. 4b). Of these peaks, 30.3% increased in enrichment after TGF-β and IL-1β stimulation, with gene ontology analysis implicating GLIS3 in the regulation of ECM organization and inflammatory response genes (Fig. 4c). These genes consisted of GLIS3 effector genes (p<0.001, Chi-squared test), including IL11, where GLIS3 bound upstream of its TSS (Fig. 4d, Extended Data Fig. 7b). To identify additional transcriptional regulators that may co-reside in GLIS3 peaks, we performed motif analysis and identified DNA motifs of FOSL1 (FRA1), a component of the AP-1 complex linked to IL-11 production in other disease models^39–43, and TEAD1/TEAD3, downstream effectors of YAP/TAZ signaling (Fig. 4e). Notably, YAP signaling inhibition has beneficial outcomes across various organ-specific models of fibrosis^44,45. To determine whether GLIS3 regulates the binding of FOSL1 and TEAD1 to target genes shared with GLIS3, we performed ChIP-qPCR of immunoprecipitated FOSL1 and TEAD1 from WT or GLIS3 CRISPRko fibroblasts stimulated with TGF-β and IL-1β (Fig. 4f). The absence of GLIS3 impaired both FOSL1 and TEAD1 binding to their targets, except for IL11, whose expression was dependent on GLIS3 or dual knockdown of TEAD1/TEAD3 in response to TGF-β and IL-1β (Fig. 4f, Extended Data Fig. 7c–f). These data indicate GLIS3-dependent IL11 induction in IAFs, but additional regulatory contributions in other cellular contexts are plausible. Collectively, these results demonstrate that GLIS3 determines the IAF state, including IL11 expression, and a transcriptional gene program involved in tissue remodeling and inflammation.

Having defined the GLIS3 transcriptional program in IAFs in vitro, we derived a GLIS3 signature score to quantify IAF activity in vivo. We integrated data from RNA-seq of GLIS3-perturbed fibroblasts, activated GLIS3-bound ChIP-seq peaks, and IAF-specific transcripts to identify a core GLIS3 signature (Fig. 4g). We projected this signature onto RNA-seq profiles of colonic biopsies from over 200 treatment-naïve pediatric UC patients from the PROTECT cohort⁴⁶ (Fig. 4g). We observed a disease severity-dependent increase in the enrichment score of the GLIS3 signature (Fig. 4g) and further refined this signature to a subset of 50 genes, each of which was independently predictive of disease severity, including GLIS3 and IL11 (Fig. 4h, Methods). Deconvolution of bulk transcriptomes to infer cellular composition in the PROTECT cohort revealed a direct relationship between heightened disease severity and increased frequency of both IAFs and activated macrophages, correlating with the GLIS3 signature (Fig. 4g).

GLIS3 governs intestinal fibrosis in vivo

To investigate the role of GLIS3 in driving intestinal pathology after chronic DSS treatment, we developed conditional Glis3 knockout mice by flanking exon 3 with LoxP sites (Glis3^f/f) and crossing them to fibroblast-specific Pdgfra-Cre, enabling conditional knockout of Glis3 (Glis3^f/f;Cre) (Extended Data Fig. 8a). Masson’s Trichrome staining revealed a reduction in the percentage of total colonic collagen in Glis3^f/f;Cre compared to Glis3^f/f mice treated with chronic DSS, mirrored by reduced tissue hydroxyproline and pro-fibrotic collagen gene expression (Fig. 5a–c). Following chronic DSS treatment, Glis3^f/f;Cre mice also showed lower histopathological scores of tissue inflammation, reduced weight loss, and longer colon lengths compared to Glis3^f/f mice (Fig. 5d,e, Extended Data Fig. 8b). These results indicate a central role of Glis3-expressing fibroblasts in mediating collagen deposition and driving fibroinflammatory signaling in vivo.

Figure 5. GLIS3 is required for IAF induction and aberrant collagen deposition during colitis.

(a) Masson’s trichrome stained Glis3^f/f and Glis3^f/f;Cre mouse colons (5–18 weeks) treated with water or chronic DSS (left). Total collagen percentage quantification across distinct biological replicates from four pooled experiments (right). Glis3^f/f-water, n=18; Glis3^f/f;Cre-water, n=21; Glis3^f/f-DSS, n=20; Glis3^f/f;Cre-DSS, n=15 mice.

(b) Quantification of colonic hydroxyproline normalized to total protein from tissue lysates from (a).

(c) qPCR quantification of collagens normalized to Eef2 from tissue lysates from (a).

(d) Colon length measurements from (a).

(e) Histopathological scoring (see Methods) on H&E-stained tissues from (a).

(f) Top: Xenium-based spatial profiling schematic of water- or chronic DSS-treated Glis3^f/f and Glis3^f/f;Cre mice. Bottom: distribution of cell type proportions across water- or chronic DSS-treated Glis3^f/f and Glis3^f/f;Cre mice. n=3 mice per condition.

(g) Spatial projection of mIAFs and activated macrophages on colonic Swiss-rolls of water- or chronic DSS-treated Glis3^f/f and Glis3^f/f;Cre mice.

(h) Spatial projection of the GLIS3 signature module score on colonic Swiss-rolls of water- or chronic DSS-treated Glis3^f/f and Glis3^f/f;Cre mice.

(i) Dot plot showing Il11, Glis3, and the GLIS3 signature expression on each replicate of the Xenium spatial sequencing cohort. n=3 mice per condition.

(j) Dot plot showing highlighted pro-inflammatory gene expression from activated macrophages and neutrophils in water- or chronic DSS-treated Glis3^f/f and Glis3^f/f;Cre mice. n=3 mice per condition.

All mice were co-housed and models of chronic DSS followed the same regimen (2.0%, 42 days). Images are representative of the sample cohort. Unless otherwise stated, statistics are by a two-way ANOVA with Tukey’s multiple comparison test on distinct biological replicates and error bars are mean ± s.e.m. ns, not significant.

To further define intra- and inter-cellular processes by which fibroblast GLIS3 mediates fibrotic tissue remodeling, we utilized Xenium single-cell spatial profiling of whole colon tissue from water- and DSS-treated Glis3^f/f and Glis3^f/f;Cre mice (Fig. 5f, Extended Data Fig. 8c). We observed no consistent differences between water-treated Glis3^f/f and Glis3^f/f;Cre mice; however, DSS-treated Glis3^f/f;Cre colons revealed a reduction of mIAFs and co-localized activated macrophages relative to those of Glis3^f/f (Fig. 5f,g). The GLIS3 signature, including Il11, was reduced in Glis3^f/f;Cre mice (Fig. 5h,i). In addition to mIAFs and activated macrophages, we observed a reduction of other immune cells in DSS-treated Glis3^f/f;Cre mice, particularly neutrophils (Fig. 5f). Gene expression profiling of these two myeloid subsets showed diminished levels of fibroinflammatory mediators, including Il1b, Osm, and Tnf (Fig. 5j). Collectively, we discover that mIAFs dictate disease outcomes not only in a cell-intrinsic manner by expressing fibrotic genes, but also by orchestrating intercellular interactions with pro-inflammatory myeloid cells in the inflamed intestine, positioning these fibroblasts as central effectors of a collapse in tissue homeostasis.

Discussion

The relationship between tissue inflammation and homeostasis remains incompletely understood yet has important implications for treating chronic inflammatory diseases associated with fibrosis. Despite the successes of targeted biologic therapies in moderate to severe IBD patients, only 20-30% achieve remission, and 13-46% of those patients will lose responsiveness, resulting in complications like recurrent fibrotic strictures that necessitate surgical intervention^47,48. Since maladaptive tissue remodeling leads to fibrosis and stems from aberrant stromal cell activation, we sought to decipher the cellular and molecular architecture of IAFs and delineate their roles in chronic intestinal inflammation. In both CD and UC patients, IAFs and activated macrophages engage a reciprocal cellular circuit initiated by macrophage-derived TGF-β and IL-1β at regions of tissue with active or chronic colitis. Genome-wide screens centered on this macrophage-induced fibroblast state switch revealed GLIS3 as a central transcriptional regulator. Importantly, fibroblast-specific Glis3 knockout mice were protected against intestinal inflammation, excessive collagen deposition, and pro-inflammatory gene expression.

Our bidirectional CRISPR screens identified GLIS3 as an essential IAF regulator. While other transcription factors such as SPI1, ETS1, TBX3, NR4A2, and TWIST1^13,49–53 have been implicated in driving disease-associated fibroblast states, we observed a stronger enrichment of GLIS3 in our functional dissection of IAFs. SPI1 and TBX3 promote myofibroblast ECM deposition, while ETS1 promotes its degradation, and its loss causes increased thickening of the submucosa and muscularis propria in DSS-induced colitis^13,49,50. Additionally, TWIST1 was shown to be expressed in a subset of CD patient fibroblasts with high FAP expression, and Col1a1-driven Twist1 knockout mice displayed less severe colitis^51,52. Our scRNA-seq data indicated that Fap and Twist1 expression mark distinct fibroblast subsets in mice—submucosal mGrem2⁺, mRspo3⁺ fibroblasts and muscularis propria-localized mLy6h⁺ fibroblasts, respectively—with Glis3 not being expressed in either of these subtypes. Notably, we observed specificity of GLIS3 in IL11-expressing IAFs, which was revealed through mechanistic elucidation of a two-signal activation mechanism whereby TGF-β and IL-1β converge to activate GLIS3. Thus, our work positions GLIS3 as a central integrator of tissue repair processes and inflammation, inducing a disease-enriched subset of intestinal fibroblasts to express fibrosis-associated cytokines, collagen, and genes associated with therapy resistance. These findings shed new light on immune-stromal cell interactions and mechanisms linking inflammation to fibrosis, raising the possibility that the IAF cell state may be conserved across tissues and diseases. Given the clinical burden of fibrotic complications in chronic inflammatory diseases and the fact that therapeutic interventions primarily act through immune suppression, the prospect of targeted anti-fibrotic therapy represents a promising opportunity.

Methods

Generating the integrated IBD single-cell RNA-seq atlas

Previously published scRNA-seq datasets on Crohn’s disease (CD) and ulcerative colitis (UC) were used for integrative analysis^2,9,11,13. Smillie et al.² was generated with colonic tissue biopsies obtained from 12 non-IBD controls and 18 UC patients (from both inflamed regions and non-inflamed regions). Library preparation for single-cell profiling was performed by fractionating the tissues into epithelial and lamina propria fractions before downstream processing. The Kong et al.¹³ dataset was generated from 13 non-IBD controls and 46 CD patients, which consisted of biopsies obtained from inflamed regions of 17 subjects and non-inflamed regions of 43 diseased patients. Library preparation for single-cell profiling was performed in a mixed manner, where some samples were separated into epithelial and lamina propria fractions and some were not separated. Biopsies were obtained from three segments of the GI tract – small bowel, terminal ileum, and colon. The Friedrich et al.¹¹ dataset was generated by fractionation and sorting of EPCAM⁻ and CD45⁻ cells of colonic biopsies obtained from 4 non-IBD controls and 11 UC patients, comprising 7 inflamed region biopsies and 4 non-inflamed region biopsies. The Martin et al.⁹ dataset consisted of separated lamina propria fractions of paired involved and non-involved region biopsies from 11 CD patients. Altogether, the combined dataset consisted of 1,143,316 cells from 115 patients before performing downstream quality control filtering. Data analysis was performed using the scanpy implementation⁵⁴.

Single-cell RNA-seq analysis and cell type identification

Gene expression normalization was performed on the combined scRNA-seq dataset to account for differences in sequencing depth across cells. Unique molecular identifier (UMI) counts were normalized by the total UMI count per cell, and the total count of each cell was set to 10,000 transcripts per cell. After natural logarithm conversion and scaling of the gene expression matrix, the top 2,000 highly variable genes were selected for a first round of dimensionality reduction. Thereafter, batch correction was performed on each individual patient sample using harmony⁵⁵, followed by neighborhood clustering and uniform manifold approximation and projection (UMAP) embedding of the single cells⁵⁶. Based on the expression of known markers of epithelial (KRT8, EPCAM), stromal (PDGFRA, PECAM1, ACTA2, S100B, RGS5), and immune (CD79A, MZB1, CD3D, TRAC, C1QA, TPSAB) populations, the clusters were then sub-divided into these three major compartments for subsequent rounds of clustering and analysis.

For compartment-specific analyses, gene expression normalization was performed by excluding genes with higher UMI counts (> 5% of the total UMI count per cell) to minimize the contribution of highly expressed genes to the normalization.

In the stromal compartment dataset, only genes expressed in > 5 cells were considered for further integrative analysis. Dimensionality reduction and batch correction was performed by adjusting for the following covariates: 10x Genomics Single Cell Gene Expression Solution chemistry (V1, V2 or V3), patient, study, and tissue site (small bowel, ileum, or colon). The top 45 adjusted principal components were considered for neighborhood clustering using the Leiden algorithm⁵⁷ and visualized using UMAP embedding. Post-hoc analysis was performed on identified clusters to remove poor-quality cells; i.e., clusters with low UMI counts, high mitochondrial gene fraction, or expressing lineage markers of non-stromal cells were removed as doublet cells. The Wilcoxon rank sum test was performed to define the markers specific to individual clusters and annotate each cluster.

Analyses for the epithelial and immune compartments followed a similar workflow as described above, except that additional rounds of iterative clustering were performed after removal of doublet and low-quality cells.

Generation of a Xenium-based spatial transcriptomics dataset

Human sample collection:

16 patients diagnosed with UC, CD, or diverticulitis (DC) who were recruited into the Prospective Registry in IBD Study at MGH (PRISM) study at Massachusetts General Hospital (MGH) participated in this study. Informed consent was obtained from all patients in accordance with the IRB approved protocol 2004P001067. The study protocol complied with all relevant ethical regulations. DC samples were pathologist-confirmed histologically normal. Excess tissues from clinically warranted surgical resections were collected for research purposes. The IRB approved secondary use protocol 2020P001262 allowed for the use of these tissues in research at the Broad Institute of MIT and Harvard.

Human colon sample processing:

Resected human colon samples were fixed in 10% formalin overnight and stored in 70% ethanol. Tissue samples were trimmed and oriented in a tissue embedding cassette (Thomas Scientific, 230274-000B) such that all of the histological features of the colon were preserved. Paraffin embedding was done at the Koch Institute at Massachusetts Institute of Technology (MIT) Histology Facility to generate formalin-fixed paraffin-embedded (FFPE) blocks and stored at 4 °C. For hematoxylin and eosin (H&E) staining and pathology assessment of the tissues, FFPE tissue blocks were sectioned at 5 μm thickness on a microtome (Leica HistoCore AUTOCUTl) and placed on glass slides (VWR, Superfrost plus). Slides were baked at 42 °C for 3 hours before being stored in a desiccator overnight. Glass slides containing tissue sections were baked at 60 °C for 30 minutes, then H&E stained using an automatic H&E stainer at the MIT histology facility. Tissue slides were imaged with a 10x objective using a Zeiss Axio light microscope with a Metafer slide-scanning platform. During microtome sectioning, 40 μm tissue scrolls were collected to assess tissue RNA quality with DV200 scores. These sections were placed in 1.5 mL microcentrifuge tubes and stored at −80 °C until analysis. RNA was extracted using the Qiagen RNeasy FFPE Kit (Qiagen, #73504), and DV200 scores were assessed with an RNA pico chip on a 2100 Bioanalyzer (Agilent).

Sample preparation for Xenium profiling:

Human colon FFPE blocks were sectioned at 5 μm thickness on a microtome (Leica HistoCore AUTOCUTl) and placed on Xenium glass slides (10x Genomics) following slide equilibration at room temperature for ≥30 minutes. Slides were baked at 42 °C for three hours and thereafter stored in a desiccator overnight. Tissues were baked at 60 °C for two hours, deparaffinized in xylene (Millipore Sigma, 214736-4L), ethanol rehydrated, and decrosslinked according to the manufacturer’s protocol (10x Genomics, demonstrated protocol CG000580 Rev C). Protocol CG000582, Rev F (10x Genomics) was used for probe hybridization, ligation, and amplification for eight of the samples, whereas user guide CG000749 Rev B (10x Genomics) with additional cell segmentation staining was used for the other eight human samples. A customized 480-plex gene probe panel (Probe ID: HC3GPZ) was designed and used for all human samples. The probes were hybridized onto the tissue for 19 hours at 50 °C. When used, the slides were incubated with cell segmentation markers for 17 hours and 30 minutes at 4 °C. Buffers were prepared for the Xenium Analyzer instrument (software version 3.1.0.0) according to manufacturer protocol for the Xenium v1 workflow (10x Genomics, user guide CG000584 Rev G). Selected regions of interest covering the full tissue of each sample were imaged, and spatially-localized transcriptional data were collected.

Mouse colon tissue blocks embedded in optimal cutting temperature (OCT, TissueTek) were cryosectioned at 10 μm thickness in a cryostat (Leica, CM1950) at −20 °C. Tissue sections were placed on Xenium glass slides (10x Genomics) and stored in a slide mailer at −80 °C for 4 days. Tissues were fixed in 4% paraformaldehyde (Thermo Fisher Scientific, BP531-25) and permeabilized according to the manufacturer’s protocol (10x Genomics, demonstrated protocol CG000581 Rev D). Tissue slides were further processed following the 10x Genomics user guide (CG000749 Rev B). A customized 480-plex gene probe panel (ID:JUR2CE) was used and hybridized for 21 hours at 50 °C. Tissues were stained with cell segmentation markers for 17 hours and 30 minutes at 4 °C. Buffers for the Xenium Analyzer instrument (software version 3.2.0.1) were prepared following the Xenium v1 workflow (10x Genomics, user guide CG000584 Rev G). Slides were imaged, and transcriptional data was collected by the Xenium Analyzer instrument.

Post-Xenium tissue slides were incubated for 10 minutes with 10 mM Sodium hydrosulfite (Sigma-Aldrich) to remove the tissue quencher (10x Genomics, demonstrated protocol CG000613 Rev B) and H&E stained using the automatic H&E stainer at the MIT Histology Facility. Glass slides were imaged on a Zeiss Axio microscope with a Metafer slide-scanning platform using a 10x objective.

Analysis of human spatial transcriptomics dataset

Samples were processed in two batches, with and without the cell segmentation staining kit. For tissues processed without the cell segmentation staining kit, we used the nucleus segmentation mask to restrict the counting of transcripts per cell located within its nucleus. In total, a gene expression matrix of 4,477,548 cells X 480 genes was obtained from 16 human tissues, and analyzed using a similar workflow as described for the single-cell RNA-seq dataset (see Supplementary Data 3 for panel gene list). The analysis was implemented in Scanpy. Briefly, quality control was performed to remove poor-quality cells with total RNA counts <10. Count-normalization and log transformation was followed by regression of the effect of total RNA per cell and scaling. Principal component analysis was conducted to generate the top 50 pcs. Batch correction was then performed using the harmony package, considering the following variables as covariates: Patient + Batch. After inspection of the elbow plot, the top 30 eigenvectors were used to construct a k-nearest neighbor (k-NN) graph with n_neighbors=15. Subsequently, the leiden clustering algorithm was applied (resolution = 0.5) to cluster the spots and the resulting clusters were visualized by the uniform manifold approximation and projection (UMAP) embedding algorithm. Based on the expression of known markers of epithelial (KRT8, EPCAM), fibroblast (PDGFRA, C1S), stromal (PECAM1, ACTA2, DES, S100B, RGS5, CD9, ANO1), lymphocyte populations (CD79A, MZB1, CD3D, TRAC), and myeloid cells (C1QB, CD83, AIF1, TPSAB, CD36, FCGR3B), the clusters were then sub-divided into five major compartments for subsequent iterative rounds of clustering and analysis. In contrast to the scRNAseq analysis workflow, we further sub-divided the stromal cells into fibroblast and non-fibroblast compartments, and likewise the immune cells into lymphocytes and myeloid cells for resolving the cell type heterogeneity.

Compartment-specific analysis and clustering was implemented using a similar workflow as described above by setting the leiden algorithm parameter resolution = 1. Clustering was followed by post-hoc analysis of identified clusters to remove poor-quality cells, i.e., clusters with low total RNA counts or expressing lineage markers from other compartments. The Wilcoxon rank sum test was performed to define the markers specific to individual clusters and annotate them. Altogether, we identified 56 unique cell types across the different compartments in the spatial atlas.

Harmonizing single-cell and spatial atlases

The single-cell RNA-seq IBD atlas (scIBD) revealed 51 distinct cell types, whereas the spatial IBD atlas (spIBD) revealed 56 distinct cell types. Broadly, most of the cell types identified in the scIBD were also identified in the spIBD, which also recovered additional cell types that are known to be located deeper in the muscularis propria, i.e. interstitial cells of Cajal (ICC), neurons, muscle cells located in the inner circular (SMC-MPI) and outer longitudinal layer (SMC-MPO) of the muscularis proporia. Because the scRNAseq was based on biopsies, these cell types are not captured. Likewise, monocytes, follicular B cells and IL6-hi fibroblasts were only identified in the spIBD. On the other hand, colonic plasma B cells, naive CD8 T cells, and M cells (microfold) were only captured in the scIBD.

Moreover, we also observed some cell types/states that were captured in the scIBD but could not be distinguished in the spIBD. For instance, we recovered the Th1 and Th17 polarized states of differentiated CD4 T cells in the scIBD, but the heterogeneity between these cells could not be resolved in the spIBD, and hence were annotated as the same cluster CD4 Th1/Th17. Naive and memory B cells were recovered in the scIBD, but the heterogeneity could not be distinguished in spIBD. Likewise, capillaries and arteries were distinct in the scIBD, but grouped into the same cluster in spIBD. WNT5B+, NPY-hi and PTGIS-hi cells were identified in scIBD, but could not be distinguished in spIBD. Conversely, SPP1+ macrophages were identified in the spIBD, while C1Q-hi and MHCII-hi macrophages could not be resolved.

For the ligand–receptor analysis (described below), we focused on the common set of cell types between the two atlases and wherever possible used the closest cell type as a proxy when the cell types could not be identified in the spIBD.

Cellular niche analysis based on spatial transcriptomics

We evaluated the composition of cell types in the proximal neighborhoods of each cell to identify the cellular niches, i.e. microenvironments shared across multiple samples. For every cell in our spatial atlas, we counted the number of distinct cell types present within a radius of 30 μm, thereby generating a frequency matrix of 3,382,726 microenvironments by 56 cell types. The median number of total cells sampled within the 30 μm radius was 14 cells, ranging from 1 to 113. Next, we implemented a k-means clustering approach using scikit-learn (v.0.22) package and evaluated the stability of clusters across a range of k (5 - 35) using the silhouette score. We obtained the highest score at k = 19, suggesting the existence of 19 distinct niches (N0-N18) in our dataset.

Niche and cell type enrichment analysis

Chi-squared test of independence was performed to evaluate the enrichment of cell types in a given niche. A contingency table summarizing the count distribution of each cell type across niches in the overall spatial atlas was generated, and the Chi-squared statistic (χ²) was computed using observed and expected frequencies assuming independence between variables. Degrees of freedom (df) were calculated as: df=(N−1)×(C−1), where N is the number of distinct niches, and C is the number of cell types. p-values for cell type enrichment in each niche were obtained using standardized residuals and corrected for multiple testing using the Bonferroni method. p-values <0.05 were considered statistically significant.

Cell type composition analysis in single-cell and spatial atlases

To compare the compositional differences between disease and healthy patients, we used the scCODA toolbox⁵⁸, which provides a Bayesian implementation of the Dirichlet-multinomial model. For the scRNA-seq atlas, we first aggregated cell count data for each library and visualized the relative abundance of each cell type across the different conditions. Given that Martin et al. predominantly profiled immune cells and Freidrich et al. predominantly profiled stromal cells, we restricted our analyses to Kong et al. for CD and Smillie et al. for UC. Within these two studies different tissue enriched fractions, i.e. epithelial sorted, lamina propria, immune sorted or the whole tissue were used to prepare the libraries, hence we performed the analyses at sample library level. On the study-specific cell count matrices, scCODA was implemented by adding a pseudocount of 0.001 and considering the following covariates: patientID + enriched_fraction + disease status, using the SMC as a reference cell type because its abundance did not change between the conditions. The analysis was performed separately for UC and CD patients, using the respective control subjects as the reference, and cell types were considered to be significantly changing in abundance at false discovery rate (FDR) < 20%. Disease-specific effect sizes (β) and log fold change in abundances of only cell types with credible effects were visualized.

For the spatial atlas, we performed the analyses at patient level. Specifically, we first visualized the relative abundance of each niche across the different conditions, and identified niche N3 as the reference niche that did not change between conditions. scCODA was then implemented by adding a pseudocount of 0.001 using the non-IBD controls as reference. Niches were considered to be significantly changing in abundance at false discovery rate (FDR) < 20%. Disease specific effect sizes (β) and log fold change in abundances of only niches with credible effects were visualized.

Differential expression analysis on scRNA-seq atlas

Pseudobulk expression for each gene was first computed for each cell type in every patient and the analysis was performed using the decoupler and pydeseq2 packages. Differential expression was evaluated using the respective controls for UC-inflamed or CD-inflamed subjects, including 10x chemistry as a covariate in the model. For UC-inflamed, genes with FDR-adjusted p-value <0.05 were considered statistically significant, and for CD-inflamed, genes with FDR-adjusted p-value <0.1 were considered statistically significant because we observed differential regulation of fewer genes in the CD-inflamed patients.

Inference of cell communication network based on spatial constraints

Intercellular crosstalk analysis was performed using the cellphonedb (v5.0) package using the DEG analysis method. The scRNA-seq IBD atlas was used as the reference and analysis was performed separately for UC and CD patients. First, differentially upregulated genes passing the significance threshold in inflamed tissues compared to control tissues, described in the Differential expression analysis on scRNA-seq atlas section, were used as the input for each cell type. Since IAFs were predominantly observed in inflamed tissues, the differential expression analysis did not yield significantly upregulated genes in IAFs because of low numbers in controls. Hence, to obtain the IAF upregulated genes, differential expression at pseudobulk level was performed between IAFs and non-IAF fibroblast populations and appended to the input table. The threshold parameter in cellphonedb was modified to include genes with >5% expression in a given cell type for scoring the interactions.

For inferring spatially constrained ligand-receptor communication scores, the list mapping each niche to significantly enriched cell types, described in the Niche and cell type enrichment analysis section, was used as input. Because some cell types identified in the scRNA-seq IBD atlas could not be identified in the spatial IBD atlas, as described in the Harmonizing single-cell and spatial atlases section, we used the niche membership of their closest cell type as a proxy. Specifically, CD4 Th1 and Th17 cells in the scRNA-seq atlas were assumed to be members of the same niches as CD4 Th1/Th17 cells in the spatial atlas, WNT5B+ NPY-hi and WNT5B+ PTGIS-hi were considered the same as WNT5B+, capillaries and arteries were assumed to the same as capillaries/arterioles, naive and memory B cells were assumed to be the same as naive B cells in the spatial atlas, and C1Q-hi macrophages were assumed to localize to the same niches as LYVE1-hi macrophages.

The total sum of significant ligand–receptor interaction scores were counted between a given sender–receiver pair of cell types and used to derive the adjacency matrix for network analysis. The igraph package was used to plot the directed and weighted network graph on the adjacency matrix, with edge thickness representing the sum total interaction score. Stromal compartment cells were removed from the network graph to highlight the communication with IAFs and non-stromal cells.

Histopathology annotations

H&E-stained sections of each tissue were used to manually identify structures pertaining to various anatomical and pathological regions. Using the Xenium Explorer (10x Genomics), the H&E images were aligned to the slide, and the regions were selected manually and cell barcodes were extracted. We ascertained the following anatomical regions: the normal mucosa, submucosa, and muscularis propria, and the following pathological regions: mucosa with predominantly active colitis, mucosa with predominantly chronic colitis, fibromuscular hyperplasia, lymphoid aggregates, and granuloma. Each tissue was also qualitatively graded for stages of fibrosis, ulceration, and severity of inflammation.

Identification of putative driver ligands of IAF gene program

First, we identified transcription factors (TFs) enriched in IAFs based on their expression levels. Cluster specific enrichment analysis in the stromal compartment atlas was performed using the Wilcoxon rank sum test to identify genes enriched in IAFs. The list was then subset to the list of known TFs to yield 23 putative mediators. In parallel, we also implemented DoRothEA to infer the activity of known TFs^59,60. Then we identified the enriched TFs in IAFs using the Wilcoxon test by considering all non-IAF stromal cells as the background. We selected TFs with a mean activity difference of > 0.75 and p-value < 0.01, yielding a list of 34 putative TFs. In total, the final list consisted of 55 unique TFs putatively orchestrating the IAF program.

Next, we leveraged the ligand–target regulatory potential model available from NicheNet to identify potential ligands whose signaling converges onto the enriched TFs. Regulatory potential scores are derived by incorporating multiple data sources that cover ligand–receptor relationships, signal transduction, and gene regulation events and then applying network propagation methods to quantify the signal flowing from a ligand through its receptors onto the signaling proteins, transcriptional regulators, and finally the target genes. For each TF in our list, we selected the top 10 ligands based on the strongest regulatory potential scores derived from the model. Only the ligands that appeared ≥ 5 times in the top 10 list were selected for further analysis. Subsequently, we restricted the list of ligands to those whose receptors were expressed in > 5% of IAF cells. Ultimately this pipeline yielded 13 putative ligands, which were further expanded upon based on ligand family membership.

Analysis of mouse spatial transcriptomics dataset

Tissues were processed with the cell segmentation staining kit, thus counting of transcripts was not restricted to the nucleus. In total, a dataset of 2,792,913 cells X 480 genes was generated from 12 mouse large intestines, representing three biological replicates from the following conditions: Glis3^f/f;Cre treated with DSS, Glis3^f/f;Cre treated with water, Glis3^f/f treated with DSS, and Glis3^f/f treated with water. Data analysis was performed using a similar workflow as described above (see Supplementary Data 3 for panel gene list used for defining cell types).

Briefly, cells with total RNA counts <10 were removed, followed by normalization and log transformation, regression of total RNA, and scaling. Top 40 principal components (PCs) were used for the k-NN graph construction, and leiden clustering was applied (resolution = 0.5) and visualized by UMAP. Subsequent iterative rounds of clustering were performed by grouping the clusters into four major compartments: epithelial, fibroblast, stromal, and immune, including lymphocyte and myeloid cells. The following markers were broadly used: epithelial (Krt8, Epcam), fibroblast (Pdgfra, Dpt), stromal (Pecam1, Acta2, Des, S100b, Rgs5, Snap25, Ano1), lymphocyte (Cd79a, Mzb1, Cd3d, Trac, Gata3, Xcl1), and myeloid (Tyrobp, Aif1, Ccr2, Fgr, C1qa, Cd68, Cd83, Trem2, Lyve1, Clec9a, Cd209a, Fpr1, Cxcr2).

Compartment-specific analysis and clustering was implemented using a similar workflow as described above by setting the leiden algorithm parameter resolution = 1, except for the immune compartment where resolution = 1.5 was used. Post-hoc inspection of clusters led to removal of poor-quality cells. Finally, for cell type annotation, we performed a Wilcoxon rank sum test to define the markers specific to each cluster, inspected the spatial distribution of each cluster, and inspected the expression of cell type specific markers listed in Supplementary Data 3. Ultimately, our approach led to the identification of 55 unique cell types exhibiting distinct expression signatures and spatial distribution in the mouse large intestine.

Mice

Animal studies complied with all relevant ethical regulations. All mice utilized in this study were co-housed in specific pathogen-free (SPF) facilities at Massachusetts General Hospital. For all experiments, 5- to 20-week-old male or female sex-matched mice were used. Mouse strains included C57BL/6J Il11^mNG mice, Il11^f/f mice (C57BL/6JGpt-Il11^em1Cflox/Gpt; Gempharmatech #T051922), Glis3^f/f mice (C57BL/6JGpt-Glis3^em1Cflox/Gpt; GemPharmatech, #T010949), R26-CreERT2 mice (B6.129-Gt(ROSA)26Sor^{tm1(cre/ERT2)Tyj}/J; Jax, #008463), and PDGFRa-CRE mice (C57BL/6-Tg(Pdgfra-cre)1Clc/J, Jax, #013148). Statistical calculations were not performed to determine sample size. Genotype-matched mice were randomly assigned to each treatment group, and blinding was not performed. Mice were housed with a 12 hour dark, 12 hour light cycle and provided food and water ad libitum. Mice were housed at an ambient temperature of 18–24 °C and 30–70% relative humidity. All animal procedures were conducted in accordance with protocol 2003N000158 approved by the Massachusetts General Hospital Institutional Animal Care and Use Committee (IACUC), and animals were cared for according to the requirements of the National Research Council’s Guide for the Care and Use of Laboratory Animals.

Generation of Il11^mNG mice

Mouse Il11 was targeted with an in vitro-verified gRNA (>90% cleavage efficiency) near the stop codon of exon five (gRNA-targeting sequence: TAAAGACTCGACTGTGACTC). Double-stranded DNA (dsDNA) was synthesized containing a T2A-mNeonGreen sequence flanked by approximately 400 nucleotide long homology arms of the Il11 gene upstream and downstream of the stop codon. A single-stranded DNA (ssDNA) sequence of this donor template was synthesized with the Guide-it Long ssDNA Production System following the manufacturer’s protocol (Takara, #632644). Verified gRNA, ssDNA, and Cas9 with NLS (PNA Bio, #CP02) complex were microinjected into zygotes by the Harvard Transgenic Mouse Core. Resulting Il11^mNG mice were backcrossed on the C57BL/6J background for three generations, followed by heterozygous crosses to obtain homozygous reporter mice. The license to utilize mNeonGreen was acquired from Allele Biotechnology.

Generation of Il11^f/f;Cre and Glis3^f/f;Cre mice

Il11^f/f mice (C57BL/6JGpt-Il11^em1Cflox/Gpt; Gempharmatech #T051922) and Glis3^f/f mice (C57BL/6JGpt-Glis3^em1Cflox/Gpt; GemPharmatech, #T010949) were generated by sperm reconstitution at the UMass Chan Transgenic Animal Modeling Core on behalf of Gempharmatech. Heterozygous pups were crossed together to generate homozygous conditional knockout mice. Homozygous Il11^f/f mice were then bred with R26-CreERT2 mice (B6.129-Gt(ROSA)26Sor^{tm1(cre/ERT2)Tyj}/J; Jax, #008463) to generate homozygous Il11^f/f, hemizygous Cre mice. Glis3^f/f mice were bred with hemizygous PDGFRa-CRE mice (C57BL/6-Tg(Pdgfra-cre)1Clc/J, Jax, #013148) to generate homozygous Glis3^f/f, hemizygous PDGFRa-CRE mice.

Acute and chronic DSS colitis

DSS experiments were performed as previously described¹⁴. Briefly, mice were given 2.0% (weight/volume) DSS (Thermo Fisher Scientific, #J14489-22) dissolved in water for 7 days for the acute model of DSS. In the chronic model, DSS administration was followed by a water recovery phase lasting for another 7 days. This cycle was repeated 2 more times, with mice weights being recorded daily. Upon cessation of each model at day 7 of the acute DSS model, day 42 for Il11 and Glis3 mouse experiments, or day 35 for cytokine neutralization experiments, mice were euthanized, and colons were obtained for phenotyping. For Il11 mice, chronic DSS experiments represent three pooled independent experiments with the following cohort replicate per experiment: Il11^f/f - water (n = 3, 4, 11); Il11^f/f;Cre - water (n = 0, 0, 14); Il11^f/f - DSS (n = 4, 3 10); Il11^f/f;Cre - DSS (n = 3, 4, 3). For Glis3 mice, chronic DSS experiments represent four pooled independent experiments with the following cohort replicate per experiment: Glis3^f/f - water (n = 4, 4, 4, 6); Glis3^f/f;Cre - water (n = 0, 0, 0, 21); Glis3^f/f - DSS (n = 4, 7, 3, 6); Glis3^f/f;Cre - DSS (n = 3, 4, 3, 5). For cytokine neutralization, chronic DSS experiments represent two pooled independent experiments with the following cohort replicate per experiment (n = 3, 4). Data pooling was performed by aligning animal data by timepoint (e.g., % weight change by day of DSS/water treatment), with all downstream analyses restricted to end-point measurements.

Tamoxifen administration

Il11^f/f and 11^f/f;Cre mice were administered DSS following the chronic DSS model. After cessation of each cycle of DSS administration, mice were intraperitoneally (IP) injected with tamoxifen (Sigma, #T5648) dissolved in corn oil (Sigma, #C8267) (200 μl of 20 mg/ml). Tamoxifen was administered for three consecutive days, each at a different site of the mouse abdomen for each cycle of water treatment. Weights were monitored daily.

Cytokine neutralization

Il11^mNG mice were administered DSS following the chronic DSS model. At the start of each cycle of DSS administration, mice were IP injected with monoclonal antibodies (mAbs) directed against IgG (BioXCell, #BP0083), anti-TGF-β (BioXCell, #BP0057), anti-IL-1β (Invivogen, #mil1b-mab9-1T), or the combination of anti-TGF-β and anti-IL-1β. Antibodies were diluted in PBS and IP injected at a different site of the mouse abdomen on the first and third day of each cycle of DSS administration (100 μl, 100 μg/mouse). Weights were monitored daily.

Mouse colon cryosectioning

Water- and DSS-treated mice were euthanized with CO₂ in accordance with IACUC protocol. Colons were removed, flushed and rinsed in PBS (Sigma, #D8537-500ml), then cut longitudinally to expose the lumen. Colons were rolled from the distal to the proximal end with forceps and then placed in cryo-blocks (Tissue-Tek, #25608-916) containing OCT (Tissue-Tek, #25608-930) on dry ice. Frozen colon Swiss rolls were sectioned in a Leica CM1950 cryostat at −20°C and at 10 μm slices and placed onto glass slides (Fisherbrand 22-230-900) for imaging. 50 μm slices were collected for RNA extraction, and 100 μm slices were collected for protein extraction.

Mouse colon staining

Immunofluorescence:

Fresh-frozen tissue sections were fixed in 4% PFA (Electron Microscopy Services, #15710-S) diluted in PBS (v/v) for ten minutes at room temperature, followed by three washes in PBS. Tissues were then permeabilized in 0.2% Triton X-100 in PBS (v/v) for ten minutes at room temperature, followed by three washes in PBS. Tissues were blocked in 4% BSA (LGC Clinical Diagnostics #1900-0016) in PBS (w/v) for 20 minutes at room temperature, then incubated with diluted primary antibodies in 4% BSA in PBS for one hour at room temperature (1:400 for anti-mNeonGreen [Proteintech, #nfms], 1:400 for anti-CD68 [Cell Signaling Technology, #97778S]). Tissues were then washed three times in PBS, then incubated with secondary antibodies in 4% BSA in PBS for one hour at room temperature (1:500 AF-488 conjugated anti-mouse antibody [Proteintech, #sms1AF488-1], 1:1000 AF-594 conjugated anti-rabbit [Thermo Fisher Scientific, #A-21207]). Tissues were washed three times in PBS, rinsed in deionized water, mounted with ProLong^™ Diamond Antifade Mountant with DAPI (Thermofisher, #P36962) and sealed with a cover glass (Corning, #2975-245) for 24 hours. Images were captured on a Nikon Ti2-E inverted microscope equipped with a CSU-W1 spinning disc confocal.

QuPath software v0.6.0 was used to quantify the percentage of IL-11^mNG fibroblasts in whole-colon Swiss-rolls of Il11^mNG mice subjected to chronic DSS. Channel min/max and gamma values were set at equal measurements across the sample cohort. Cell detection was performed based on DAPI-stained nuclei. IL-11^mNG cells were then defined by setting a single measurement classifier on objects filtered on all DAPI-stained cells. A channel filter was then set for “Cell: 488 nm mean”. An “Above Threshold” was then set which defined IL-11^mNG+ cells based on the background signal distribution of 488 nm mean cell measurements in water-treated colon samples.

Masson’s Trichrome:

Staining was performed following the Masson’s Trichrome Stain Kit protocol (Vitroview, #VB-3016). Briefly, frozen tissue was acclimated to room temperature, fixed in 10% formalin for 30 minutes, and rinsed in distilled water three times. Tissues were then incubated in Bouin’s fluid for one hour at 60°C. Tissues were washed in running tap water for five minutes, then incubated with Weigert’s A + B solution mixture (1:1 mixture) for seven minutes. Tissues were then rinsed with running tap water for five minutes, followed by incubation with Biebrich scarlet/acid fuschin stain for five minutes. Tissues were rinsed in distilled water, then incubated with phosphomolybdic-phosphotungstic acid solution for a total of ten minutes. Without rinsing, aninline blue solution was added for 15 minutes, followed by three washes in distilled water. Tissues were briefly rinsed in acetic acid then dehydrated with 90%, then 100% ethanol for two minutes each. The tissues were cleared in xylene (Sigma, #214736-1L) for five minutes, three times, and finally mounted with Cytoseal 60 (EMS, #18007) and sealed with a cover glass. Images were captured on a Nikon Ti2-E inverted microscope for detailed magnification images or Leica APERIO VERSA scanning system for quantification.

For quantification of colonic collagen with Masson’s trichrome stained colon samples, FIJI imaging quantification software v1.54p was used on images captured on the Leica APERIO VERSA scanning system. Images were increased in brightness and contrast by setMinAndMax (−12, 228). Masson’s trichrome stain deconvolution was then performed using the “Colour Deconvolution 2” plugin, setting the vectors to User Values defined by Colour[1](R1: 0.412010759, G1: 0.849633569, B1: 0.329195889), Colour[2](R2:0.587269268, G2:0.67744923, B2: 0.442919122), Colour[3](R3: 0.804103811, G3: 0.584262903, B3:0.109790353). Only the blue collagen channel was used; red (muscle) and black (nuclei) signals were excluded from analyses. Collagen-positive pixel areas were identified by setting the signal threshold at 140 units to obtain the percentage of collagen-positive pixels relative to the total image area. To determine the tissue area, images were converted to 8-bit format, the signal threshold was set at 180 units, and the area was then measured. The final collagen percentage was obtained by dividing the calculated collagen-positive pixel area of the tissue by the total tissue area of the image.

H&E staining:

Staining was performed following the Hematoxylin and Eosin Stain Kit protocol (Vitroview, #VB-3000). Briefly, frozen tissue slides were fixed in cold 80% methanol (Sigma, #179337-1L) at 4°C for five minutes. The slides were then placed in room temperature PBS for ten minutes, followed by a rinse in running deionized water for two minutes. Next, slides were placed in Mayer’s Hematoxylin Solution for five minutes and then rinsed in running tap water for five minutes. Slides were then placed in 95% ethanol for one minute and then Eosin solution for one minute and thirty seconds. Slides were then dehydrated by placing them in two changes of 95% ethanol for two minutes each, and two changes of 100% ethanol for two minutes each. The tissues were cleared in xylene (Sigma, #214736-1L) for five minutes, three times, and finally mounted with Cytoseal 60 (EMS, #18007) and sealed with a cover glass. Images were captured on a Leica APERIO VERSA scanning system.

Histopathological quantification of tissue inflammation

Blinded histopathological assessment on H&E-stained tissues was carried out by a trained pathologist (A.R.S.). H&E-stained slides were graded by a standardized score of tissue injury and inflammation as described previously⁶¹. The total histological score ranged from 0 to 10 and was derived from the sum of four scoring criteria: mucosal ulceration, epithelial hyperplasia, lamina propria mononuclear infiltration, and lamina propria neutrophil infiltration. Mucosal ulceration scoring criteria were either normal (0), surface epithelial inflammation (1), erosions (2), or ulcerations (3). Epithelial hyperplasia scoring criteria were either normal (0), mild (1), moderate (2), or pseudopolyps (3). Lamina propria mononuclear infiltration scoring criteria were either normal (0), slightly increased (1), or markedly increased (2). Lamina propria neutrophil infiltration were either normal (0), slightly increased (1), or markedly increased (2).

Hydroxyproline determination

100 μm slices of OCT-embedded frozen colonic tissues were obtained by cryosectioning and flash frozen in liquid nitrogen. Tissues were crushed using manual dissociation on dry ice and incubated in Pierce^™ RIPA buffer (ThermoFisher, #89900) for twenty minutes. Following a freeze-thaw cycle, hydroxyproline extraction was performed following the manufacturer’s protocol with a few modifications (Sigma, #MAK463-1KT). 50 μL of lysates were incubated with 50 μL of 10N NaOH for two hours at 100 °C at constant agitation. Samples were then allowed to cool to room temperature, where they were neutralized with the addition of 50 μL of 10N HCl. Samples were then diluted 1:1 with 150 μL of water. Samples were then centrifuged for five minutes at 14,000 × g at room temperature to remove any particulates, and lysates were transferred to new tubes. A hydroxyproline standard (ranging from 0 - 50 μg/mL) was then prepared and 20 μL of each standard and tissue lysate were transferred to individual wells on a 96-well plate (Corning, #29444-018). A reaction mixture containing 8 μL of Reagent A and 90 μL of Oxidation Buffer was prepared for all samples, and 90 μL of this mixture was added to each well, followed by ten minutes of incubation at room temperature. Then, 90 μL of Reagent B was added to all wells and mixed by pipetting up and down until turbidity dissipated. The plate was incubated for ninety minutes at 37 °C in an incubator, whereupon the optical density (OD) of each well was measured at 560 nm on a BioTek Synergy H4 Hybrid Multi-Mode Microplate Reader. To quantify the amount of hydroxyproline, the OD values of the standards were plotted against their concentrations. Each sample’s OD measurement was then divided by slope of the standard curve to yield a hydroxyproline concentration in μg/mL. Values were normalized to protein concentrations determined by measurements from a bicinchoninic acid assay (BCA).

Protein concentration determination

Total protein concentrations were determined by the Pierce^™ BCA protein assay kit (ThermoFisher, #23225) on prepared tissue lysates with a few modifications. A standard curve with Pierce^™ bovine serum albumin standard (ThermoFisher, #23209) was prepared ranging from 0 - 10 μg/μL on individual wells on a 96-well plate. 1 μL of each sample was also added to individual wells on a 96-well plate. Next, a reaction mixture was prepared by adding 50 parts of BCA reagent A with 1 part of BCA reagent B. 200 μL of BCA reaction mixture was added to each well of the 96-well plate, and the plate was then incubated for thirty minutes at 37 °C in an incubator. The OD of each well was measured at 562 nm on a BioTek Synergy H4 Hybrid Multi-Mode Microplate Reader. Each sample’s OD measurement was then divided by slope of the standard curve to yield a protein concentration in μg/μL.

Mouse colon dissociation

Dissociation of mouse colons were performed following a modified protocol⁶². Following acute or chronic DSS treatment, mice were euthanized and colons placed in ice-cold HBSS (ThermoFisher, #14170112). Colons were cut longitudinally and cut into small pieces. Tissue was then placed into 15 mL of epithelial cell solution (10mM HEPES [ThermoFisher, #15630080], 0.5M EDTA [ThermoFisher, #AM9260G], 100U/ml Pen/Strep, 2% FBS, 100 μg/ml DNase I [Roche, #10104159001] in HBSS). Samples were incubated at 37°C for 15 minutes and manually shaken every five minutes. After incubation, samples were placed on ice for ten minutes. Tissue pieces were then transferred to a new tube with BSS and vigorously shaken for 40 seconds. The supernatant was collected by filtering through a 100 μm cell strainer(Corning, #352360) into a new tube and pieces of tissue placed in the original tube. Shaking and filtering were performed an additional two times in order to dissociate and deplete this epithelial fraction. The remaining tissue was placed in HBSS, and washed with gentle shaking for 30 seconds each, three times. With a razor blade, the tissue was cut into smaller pieces and transferred to a new tube containing 10 mL of lamina propria solution (100U/ml Pen/Strep, 100 μg/ml LiberaseTM [Sigma, #5401127001], 100 μg/mL DNase I in RPMI). Tissues were incubated 37°C for 20 minutes. Manual pipetting with a 5ml pipette was used to break apart the tissue. The tissue was incubated 37 °C for an additional ten minutes. Manual pipetting with a 1ml pipette was used to break apart the tissue further until dissociation. Samples were transferred to ice and 1ml of FBS and 80μl of EDTA were added to stop the digestion reaction. Tissue samples were pipetted again with a 1ml pipette and filtered with a 40μm cell strainer strainer (Corning, #352340). HBSS was added and cells spun down at 400g for ten minutes.

Cell surface staining of colonic cells

Isolated colonic cells were washed in PBS and re-suspended in PBS with 2% FBS. Cells were then FC blocked (Biolegend, #101320) on ice for ten minutes at 1μg/100ml buffer. Without removal of the FC block, a master-mix of fluorophore-conjugated antibodies diluted 1:200 each in PBS with 2% FBS was added to the cells for 30 minutes on ice (APC anti-mouse CD45 [Biolegend #157605], PE/Dazzle^™ 594 anti-mouse TER-119, [Biolegend, #116243], Brilliant Violet 421^™ anti-mouse PDGFRA [Biolegend, #135923], PE anti-mouse PECAM1 [Biolegend, #102407], Brilliant Violet 711^™ anti-mouse EPCAM1[Biolegend, #118233], and LIVE/DEAD^™ Fixable Near IR (780) viability stain [ThermoFisher, #L34992]). Cells were spun at 500g for three minutes and washed with PBS. This spin and wash was repeated an additional time. Cells were re-suspended in live cell sorting buffer (PBS with 5% FBS, 25mM HEPES, 1mM EDTA) and sorted on a Sony SH800 cell sorter. Single-antibody stains for each fluorescent channel using compensation beads (BeckmanCoulter, #B22804) were used to set up compensation.

Cell culture

Primary colon fibroblasts (CRL-1459), and immortalized fibroblasts (CRL-4001) were obtained from the American Type Culture Collection (ATCC). Lenti-X 293T cells for lentivirus production were obtained from Takara (#632180). Primary monocyte-derived macrophages were from isolated from healthy donor blood (unpurified buffy coats, 25-50mL; ages within 18-65 years, unknown distribution of male and female) were collected at Research Blood Components, LCC, MA, USA after obtaining a signed consent form. Standard testing for blood-borne pathogens was performed. THP-1-Cas9-expressing cells were a gift from the Genomics Platform of the Broad Institute of MIT and Harvard. Commercially-purchased cell lines were authenticated by ATCC with short tandem repeat identification. Authentication of Lenti-X 293T cells was not provided by Takara. Donor or donated cells were not authenticated. All cell lines tested negative for mycoplasma except donor-derived cells, which were not tested.

Fibroblasts and Lenti-X 293T cells were cultured in Dulbecco’s modified Eagle’s medium (Thermo Fisher Scientific, #10569044) supplemented with 10% (v/v) heat-inactivated fetal bovine serum (FBS) (MilliporeSigma, F2442), 1X penicillin-streptomycin (Corning, #30-002-CI). Myeloid cells were cultured in RPMI-1640 medium (Gibco, #22400-089) supplemented with 10% FBS, 1X penicillin–streptomycin. All cells were cultured at 37 °C, atmospheric O₂ and 5% CO₂.

Human monocyte purification and macrophage differentiation

Human monocytes from unpurified buffy coats were isolated with RosetteSep^™ Human Monocyte Enrichment Cocktail (STEMCELL, #15028) following the manufacturer’s protocol, then enriched with SepMate tubes and Ficoll-mediated separation. Isolated cells were washed with PBS before plating in RPMI-1640 containing recombinant M-CSF (PeproTech, #300-25) for seven days on non-treated petri dish plates (Falcon #351029). For polarization to different macrophage subsets, on day six following monocyte isolation, attached macrophages were washed with PBS and incubated with macrophage polarizing ligands for 24 hours (see Supplementary Data 4 for concentrations and catalog numbers). For differentiation of THP-1-Cas9 monocytes into adherent macrophages, cells were stimulated with 300 ng/ml phorbol myristate acetate (PMA) (Invivogen, #tlrl-pma) for three hours at 37°C on non-treated petri dishes. Cells were then washed in PBS and re-seeded onto tissue culture treated plates for two days in RPMI-1640.

Cell stimulation

Fibroblasts were first seeded onto a multi-well tissue culture plate. Following adherence at 75-90% confluency overnight, cells were washed with PBS, and appropriate ligand diluted in complete DMEM was added for the indicated time point (Supplementary Data 4). Following stimulation, plates were spun down, and media supernatant was collected for analysis, or cells were washed with PBS and harvested into TRIzol reagent (Thermo Fisher Scientific, #15596026) for RNA extraction.

Calculation of additive or synergistic IL-11 production

A mathematical formulation published by Sanford et al.⁶³ was used to determine whether IL-11 was produced additively or synergistically with dual TGF-β and IL-1β stimulation, as follows:

The amount of IL-11 produced in response to single cytokine stimulations or at baseline was set as these variables.

Expression of IL-11 at baseline $={\mathrm{X}}_{\text{baseline}}$

Expression of IL-11 after TGF-β = ${\mathrm{X}}_{\text{baseline}}+\mathrm{\Delta }\mathrm{T}\mathrm{G}\mathrm{F}-\beta$

Expression of IL-11 after IL-1β = ${\mathrm{X}}_{\text{baseline}}+\mathrm{\Delta }\mathrm{I}\mathrm{L}–1\beta$

If expression of IL-11 after TGF-β & IL-1 β is additive

$={\text{X}}_{\text{baseline}}+\mathrm{\Delta }\mathrm{T}\mathrm{G}\mathrm{F}–\beta +\mathrm{\Delta }\mathrm{I}\mathrm{L}–1\beta$

$=388.8$

If expression of IL-11 after TGF-β & IL-1β is synergistic

$={\text{X}}_{\text{baselin}}\mathrm{*}\text{fold change}\Delta \mathrm{TGF}\beta \mathrm{*}\text{fold change}\Delta IL1\beta$

$={\text{X}}_{\text{baseline}}\mathrm{*}\left(\left({\text{X}}_{\text{baseline}}+\mathrm{\Delta }TGF–\beta \right)/{\text{X}}_{\text{baseline}}\right)\mathrm{*}\left(\left({\text{X}}_{\text{baseline}}+\mathrm{\Delta }IL–1\beta \right)/{\mathrm{X}}_{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{e}\mathrm{l}\mathrm{i}\mathrm{n}\mathrm{e}})\right.$

$={\mathrm{X}}_{\text{baseline}}+\mathrm{\Delta }\mathrm{T}\mathrm{G}\mathrm{F}–\beta +\mathrm{\Delta }\mathrm{I}\mathrm{L}–1\beta +(\mathrm{\Delta }\mathrm{T}\mathrm{G}\mathrm{F}–\beta \mathrm{*}\mathrm{\Delta }\mathrm{I}\mathrm{L}–1\beta )/{\mathrm{X}}_{\text{baseline}}$

$=1123$

RNA isolation and quantitative RT–PCR (qPCR)

In vitro cell cultures:

RNA from in vitro cell cultures was isolated as previously described⁶⁴. RNA was extracted from cells in TRIzol reagent following the manufacturer’s protocol with chloroform and isopropanol precipitation, followed by a wash in 75% ethanol (Thermo Fisher Scientific).

Tissue samples:

RNA from in vivo tissue samples were isolated with the Aurum^™ Total RNA Mini Kit (Biorad, #7326820) following the manufacturer’s instructions. 50 μm slices of OCT-embedded frozen colonic tissues were obtained by cryosectioning and flash frozen in liquid nitrogen. Tissues were crushed using manual dissociation on dry ice and then lysed in 700 μL of lysis solution, pipetting up and down to lyse the tissue. The lysate was centrifuged for three minutes at room temperature at 14,000 x g and then transferred to a new tube. 700 μL of 60% ethanol was then added to the lysate, and the mixture was transferred to an RNA binding column. The column was centrifuged for sixty seconds at room temperature at 14,000 x g, whereupon the filtrate was discarded. 700 μL of low stringency wash solution was added to the column, and samples were then centrifuged for thirty seconds at room temperature at 14,000 x g to remove the solution. A 80 μL mixture of DNase I was added to the column and allowed to digest at room temperature for twenty-five minutes. 700 μL of high stringency wash solution was added to the column, and samples were then centrifuged for thirty seconds at room temperature at 14,000 x g to remove the solution. 700 μL of low stringency wash solution was added to the column, and samples were then centrifuged for thirty seconds at room temperature at 14,000 x g to remove the solution. The column was then centrifuged for two minutes at room temperature at 14,000 x g to remove any remaining liquids. RNA was eluted by adding 40 μL RNA elution solution and centrifuged for two minutes at room temperature at 14,000 x g to collect the RNA.

Equal amounts of RNA were used to synthesize cDNA with the iScript cDNA synthesis kit (Bio-Rad Laboratories, #1708891). iTaq Universal SYBR Green Supermix (Bio-Rad Laboratories 1725124) was used for qRT-PCR on the C1000 Touch Thermal Cycler (Bio-Rad Laboratories). Relative gene expression was calculated using the 2^−ΔΔCT method⁶⁵, with HPRT as the reference housekeeping gene for human samples or Eef2 for mouse samples. See Supplementary Data 5 for primer sequences. Heatmaps of gene expression Z-scores were generated in Microsoft Excel v16.

Secreted protein quantification

Secreted cytokines were measured by spinning down cells and debris of cell culture media. IL-11, IL-1β, and TGF-β were measured with a cytometric bead array assay (BD Biosciences, #560228, #558279, #560429) following the manufacturer’s protocol. Samples were measured for median fluorescence intensity (MFI) of the cytokine of interest using the CytoFLEX S cytometer (Beckman Coulter). MFI readings for each cytokine were plotted against a standard curve to obtain final cytokine concentrations. Data was analyzed using FlowJo^™ v10.8 Software (BD Life Sciences) software⁶⁶.

Generation of CRISPR knockin fibroblasts

IL11^mNG:

Human IL11 was targeted with an in vitro-verified gRNA (>90% cleavage efficiency) near the stop codon of exon five (gRNA-targeting sequence: TGAAGACTCGGCTGTGACCC). Double-stranded DNA (dsDNA) was synthesized containing a T2A-mNeonGreen sequence flanked by approximately 400 nucleotide-long homology arms of the IL11 gene upstream and downstream of the stop codon. Single-stranded DNA (ssDNA) was then synthesized with the Guide-it Long ssDNA Production System following the manufacturer’s protocol. ssDNA was electroporated along with Cas9 with nuclear localization signal (PNA Bio, CP02) and validated IL11 gRNA using the Lonza 4D-Nucleofector^™ X Unit kit (Lonza, #AAF-1003X) with the SE Cell Line 4D-Nucleofector^™ X Kit S (Lonza, #V4XC-1032). Cells were expanded, and single-cell clones were sorted for mNeonGreen expression following fibroblast activation. Positive clones were then stably infected with lentivirus encoding for Cas9 (pXPR_101) or dCas9-VP64 (pXPR_109) (plasmids were obtained from Genetic Perturbation Platform of Broad Institute of MIT and Harvard) and activity for CRISPRko or CRISPRa verified with gRNAs. Cells were passaged under blastidicin S HCl (Thermo Fisher Scientific #A1113903) to maintain Cas9 expression.

GLIS3^3XFLAG:

For 3X FLAG-tagged GLIS3 fibroblasts, a 3X FLAG sequence flanked by short homology arms to the GLIS3 insertion site was synthesized and electroporated (as described above for IL11^mNG) into IL11^mNG-Cas9 fibroblasts along with a gRNA targeting GLIS3 (gRNA-targeting sequence: CCAAGAGAGCTTTTAGCCTT). Cells were single-cell cloned, and 3X FLAG knockin cells were selected and verified for FLAG expression with anti-FLAG antibody (MilliporeSigma #F1804-200UG) in an image-based assay (see Immunofluorescence staining).

CRISPRko/CRISPRa/siRNA cell line generation

Lentiviral gRNA cloning:

Oligos targeting a gene of interest were designed based on CRISPick of the GPP platform of the Broad Institute of MIT and Harvard. See Supplementary Data 5 for sequences. Oligos were annealed and phosphorylated using T4 ligation buffer (New England Biolabs, #M0202L) and T4 PNK (New England Biolabs, #M0201L). Esp3i-digested (Thermo Fisher Scientific, #FD0454) pXPR_003 or pXPR_502 for CRISPRko or CRISPRa, respectively, were then ligated with annealed oligos using quick ligase (New England Biolabs, #M2200L). Plasmids were transformed into One Shot^™ Stbl3^™ Chemically Competent E. coli cells (Thermo Fisher Scientific, #C737303), single colonies were selected, and DNA was extracted using a spin miniprep kit (Qiagen, #27106). Correct integration of gRNA sequence was verified by Sanger sequencing.

Lentivirus production:

Lenti-X 293 cells were seeded and transfected with a combination of psPAX2 (Addgene #12260), VSV-G (Addgene #8454), and CRISPRko or CRISPRa vectors in OptiMEM I reduced serum media (Thermofisher #31985088) and complexed to Lipofectamine 2000 (Thermo Fisher Scientific, #11668019). After 16 hours of incubation of cells with the multi-plasmid complex, media was replenished, and cells were allowed to grow for 48 hours. Media was then harvested and spun to remove debris, and the supernatant containing lentivirus was frozen at −80°C. IL11^mNG Cas9 or dCas9-VP64 fibroblasts were stably infected with lentivirus-containing gRNAs targeting the gene of interest. Cells were then selected using puromycin (Thermo Fisher Scientific, #A11138-03) and blastidicin.

Ribonucleoprotein KO:

KO cells were generated with gRNA-Cas9 electroporation, following the manufacturer’s recommendations (Integrated DNA Technologies (IDT)). Briefly, a synthesized gRNA for the gene of interest was obtained through IDT and re-suspended in water. The RNP complex was formed by incubating 120 pmol of gRNA with PBS and 104 pmol of Alt-R Cas9 enzyme (IDT, #1081058) at room temperature for 20 minutes. Cells were electroporated with the resultant RNP complex using the Lonza 4D-Nucleofector^™ X Unit kit with the SE Cell Line 4D-Nucleofector^™ X Kit S.

siRNA:

Pre-designed siRNA oligomers were purchased from Sigma-Aldrich and re-suspended in nuclease-free water at 20μM (scrambled control [Sigma, #SIC001], FRA1 [Sigma, #SASI_Hs01_00191186], TEAD1 [Sigma, #SASI_Hs01_00031302], TEAD3 [Sigma, #SASI_Hs01_00090158]). Seeded fibroblasts were transfected with 150pmol siRNA complexed with Lipofectamine RNAiMAX (Thermo Fisher, #13778030) in Opti-MEM media (Thermo Fisher, #31985062). 24 hours later, cells were re-seeded into 12-well plates and plated in triplicate knockdown per condition. 24 hours later, cells were washed with PBS and replenished with fresh media with or without the addition of 10 ng/ml of human TGF-β and IL-1β for 24 hours. Cells were washed in PBS and resuspended in TRIzol reagent for RNA isolation.

Fibroblast and macrophage co-culture

On day six after M-CSF-treatment of primary human monocytes, differentiated macrophages were polarized with agonists for 24 hours (for concentrations, see Supplementary Data 4). Following removal of agonist and several washes in PBS, fibroblasts were seeded with the macrophages for 24 hours. For experiments utilizing THP-1-macrophages, following two days of rest after PMA stimulation, fibroblasts were seeded onto the differentiated THP-1-macrophages, and agonist was added to the media.

For qPCR-based measurements of fibroblasts in co-culture, fibroblasts were seeded at the bottom of 24-well plates, with macrophages separated through transwell (Corning, #3462). Following co-culture, macrophages were removed and RNA isolated from fibroblasts as described above.

Immunofluorescence staining

Monoculture of fibroblasts:

Cells were seeded on a coverglass (Celltreat, #229173) in a 12-well tissue-culture dish, and after completion of experiment, cells were fixed in 2% PFA followed by three washes in PBS for five minutes each. Cells were permeabilized with 0.2% Triton X-100, then washed in PBS for five minutes each. Cells were blocked with 4% BSA-PBS and then incubated with 1:500 FLAG antibody (MilliporeSigma, #F1804-200UG) or 1:500 Collagen 6 antibody (Thermo Fisher Scientific, #PA5-106556) in 4% BSA-PBS for one hour at room temperature. Cells were washed with PBS and incubated with a 1:1000 dilution of AF-647 conjugated anti-mouse antibody (Thermo Fisher Scientific, #A-21235) or a 1:1000 dilution of AF-594 anti-rabbit antibody (Thermo Fisher Scientific, #A-21207) in 4% BSA-PBS for one hour. Cells were washed three times in PBS, rinsed in distilled water, and mounted with ProLong^™ Diamond Antifade Mountant with DAPI onto a glass slide for 24 hours. Images were captured on a Nikon Ti2-E inverted microscope equipped with a CSU-W1 spinning disc confocal.

Co-cultures of fibroblasts and macrophages:

Fibroblasts and THP-1-macrophages were pre-stained with fluorescent labeling dyes prior to co-culture. Briefly, differentiated macrophages were labeled with CellTracker Orange CMRA Dye at 10μM (Thermo Fisher Scientific, #C34551) for 30 min at 37°C in serum-free RPMI-1640 media. Cells were then washed in PBS and seeded onto a coverglass in a 12-well dish in complete RPMI-1640 overnight. The following day, adherent fibroblasts were labeled with CellTracker Blue CMAC Dye at 10μM (Thermo Fisher Scientific, #C2110) for 30 min at 37 °C in serum-free RPMI-1640 media. Cells were then washed, trypsinized, and seeded onto the labeled macrophages along with the macrophage-activating ligand FSL-1 (1 ng/ml) (Invivogen, #tlrl-fsl) overnight. Immunofluorescent staining was performed as above (Monoculture of fibroblasts). Images were captured on a Nikon Ti2-E inverted microscope equipped with a CSU-W1 spinning disc confocal. Cells were then washed with PBS and imaged again.

FIJI imaging quantification software v1.54p was used to quantify the fluorescence intensity of AF647-labeled nuclear GLIS3^3XFLAG or AF594-labeled collagen 6. The median fluorescence intensity of AF647-labeled nuclear GLIS3^3XFLAG was quantified in each individual fibroblast nucleus.

Statistics and Reproducibility

Statistical analyses were performed using GraphPad Prism v10. One-way ANOVA was used in comparisons between three or more groups. Two-way ANOVA was used to compare factorial groups. For Tukey’s, Dunnet’s, and Dunn’s multiple comparison tests, the family-wise error rate was set at level 0.05 (95% confidence interval). Multiplicity-adjusted p-values < 0.05 were considered significant. All experiments were repeated at least twice in successful, independent experiments.

CRISPR screens

Lentivirus stocks of the Brunello CRISPRko and Calabrese CRISPRa libraries were obtained from the Genomics Platform of the Broad Institute of MIT and Harvard. Viral titers were calculated using a puromycin selection assay in which a target multiplicity of infection (MOI) of 0.3 was selected in order to target each individual fibroblast with one gRNA. IL11^mNG-Cas9 or IL11^mNG-Cas9-VP64 fibroblasts were infected with the determined titers of the CRISPRko or CRISPRa libraries, respectively, at 500x representation of each library for 24 hours in DMEM supplemented with 10 μg/ml polybrene (MilliporeSigma, #TR-1003-G). Fibroblasts were washed in PBS the following day, then incubated with 1 μg/mL puromycin and 3μg/mL blastidicin in DMEM for three days. Following selection, fibroblasts were re-seeded and expanded under selection. Expanded fibroblasts were then re-seeded in DMEM with no selection for 24 hours, and following attachment, fibroblasts were washed with PBS and stimulated with a dual combination of TGF-β and IL-1β (10 ng/ml each) in DMEM for 24 hours. After stimulation, fibroblasts were detached with trypsin, washed in PBS, and re-suspended in live cell sorting buffer (PBS with 5% FBS, 25mM HEPES, 1mM EDTA) and the top and bottom 15% of IL11^mNG fibroblasts were sorted on a Sony SH800 cell sorter (see Extended Data Fig. 6a for gating strategy). Collected fibroblasts were then spun down and frozen for subsequent DNA isolation and sequencing.

CRISPR gDNA isolation and sequencing

Fibroblast DNA was isolated with the QIAamp DNA Micro Kit (Qiagen, #56304) following the manufacturer’s instructions. Isolated DNA was then eluted in water and PCR was performed as previously described with P5 primer and a unique P7 primer for each sample⁶⁷. SPRI-cleanup was performed with AMPure XP beads (Beckman Coulter, #A63881) and concentrations quantified with Qubit dsDNA HS assay (Thermo Fisher Scientific, #Q32854). After running each sample on a 2% E-Gel (Thermo Fisher Scientific, #G402022), the dominant bands were excised, and 4nM of pooled DNA was sequenced with the Illumina Nextseq 500 instrument for sequencing (NextSeq 500/550 High Output v2 kit ,75 cycles) following parameters: single end, index 8, read 51.

CRISPR screen analyses

PoolQ 2.2.0 (https://portals.broadinstitute.org/gpp/public/software/poolq) was used to map raw data to gRNA reads to obtain the raw count data. EdgeR was used to identify differentially abundant (enriched) guides in sorted samples⁶⁸. STARS v1.3 (https://portals.broadinstitute.org/gpp/public/software/stars) was used to rank the targeted genes from enrichment scores of multiple guides. Genes were selected if the perturbation was greater or equal to 3 and the fold change was greater than or equal to 2 with p < 0.05.

Bulk RNA-seq of stimulated fibroblasts

IL11^mNG-Cas9 and IL11^mNG-Cas9-VP64 fibroblasts were infected with non-targeting lentiviral gRNA or GLIS3 gRNA to make control, GLIS3 CRISPRko, or CRISPRa fibroblasts. Fibroblasts were expanded under 1 μg/mL puromycin and 3 μg/mL blasticidin selection to select for a bulk genetically perturbed population. GLIS3-perturbed fibroblasts were then seeded in DMEM. The next day, cells were washed with PBS and new media with or without dual TGF-β and IL-1β (10 ng/ml each) was added at 0, 6, 12, and 24 hour time points in a reverse time course stimulation. After 24 hours, fibroblasts were washed with PBS, RNA was extracted with TRIzol, and each sample’s total RNA was quantified.

Purification of mRNA was performed with the Dynabeads mRNA direct purification kit (Thermo Fisher Scientific, #61012) with 40 ng of total RNA per sample. RNA-seq libraries were generated with previously described preparations⁶⁹ and sequenced with a NextSeq 500/550 high output kit V2.5 (Illumina 20024906).

Analysis of bulk RNA-seq of stimulated fibroblasts

Raw FASTQ reads were processed using FastQC⁷⁰, Trim Galore! (https://github.com/FelixKrueger/TrimGalore) and SortMeRNA⁷¹ to remove low-quality reads, adapters, and ribosomal RNA. Sequencing data was aligned to the human reference genome (GRCh38) using STAR and gene abundance was quantified using Salmon⁷². Differentially expressed genes were identified by using the DESeq2 R package⁷³. Genes with p-value < 0.05 were considered as differentially expressed.

scRNA-seq of stimulated fibroblasts and isolated colonic mouse fibroblasts from DSS treatment

scRNA-seq of TGF-β/IL-1β stimulated fibroblasts:

Fibroblasts were stimulated as for bulk RNA-seq; however, upon stimulation completion, at least 2,000 live single cells per sample were resuspended in PBS with 0.05% BSA and submitted for 10x Genomics single cell sequencing 3’ transcriptome V3 with Cell Suspension with HiSeq X up to 300 cycles.

scRNA-seq of sorted PDGFRA+ fibroblasts:

20,000 live single cells resuspended in PBS with 0.4% BSA were submitted for 10x Genomics single cell sequencing Chromium Next GEM Chip G (v3.1). Libraries were sequenced across 1.2 lanes of an Illumina NovaSeqX flow cell (10B reads, up to 300 cycles).

Analysis of scRNA-seq of stimulated fibroblasts

Gene expression matrices for each individual sample were obtained by aligning the FASTQ sequence reads to the reference hg38 human transcriptome using CellRanger v3.1.0 software (10x Genomics). Data analysis was performed using the scanpy package. Further, the count matrices of cell X UMI barcodes were filtered to exclude poor-quality cells, i.e. cells expressing < 200 detected genes, > 10000 UMIs, or a mitochondrial gene fraction > 10%. Count matrices of each individual sample were merged and read-depth normalized using the standard logTP10K normalization procedure, i.e. number of transcripts per 10,000 transcripts in a cell. The top 2,000 variable genes were then identified, and the normalized expression matrix was centered and regressed for effects of cell cycle scores (S phase and G2M phase) + UMI count per cell. Dimensionality reduction was then performed on the top 60 principal components (PCs) selected based on inspection of the elbow plot, followed by the construction of a k-nearest neighbor (k-NN) graph and the application of the leiden clustering algorithm. The resulting clusters were visualized in the UMAP embedding space.

Analysis of scRNA-seq of isolated colonic mouse fibroblasts from DSS treatment

Gene expression matrices for 6 individual mice, 3 acute DSS- and 3 chronic DSS-treated, were obtained by aligning the FASTQ sequence reads to the reference mm10 mouse transcriptome using CellRanger v7.0 software (10x Genomics). However, one chronic DSS-treated sample failed the sequencing quality control metrics demonstrating poor quality and was excluded. Data analysis was performed using the scanpy package and followed the similar workflow as described in the Single-cell RNA-seq analysis and cell type identification section. Altogether, we identified 10 unique fibroblast cell types.

Analysis of mouse spatial transcriptomics dataset

In total, a gene expression matrix of 1,540,301 cells X 480 genes was obtained from colons of 3 wild-type DSS-treated and 3 Glis3^f/f;Cre DSS-treated mice, and analyzed downstream using a similar workflow as described for the Analysis of human spatial transcriptomics dataset section. Altogether, we identified 41 unique cell types across the different compartments.

GLIS3^3XFLAG ChIP-seq

ChIP-seq was performed according to the protocol of Tian et al.⁷⁴ with a few modifications. Following treatment of GLIS3^3XFLAG fibroblasts with media alone or TGF-β and IL-1β (10 ng/ml each) for 24 hours, adherent fibroblasts were washed with PBS three times, then fixed with 0.25M disuccinimidyl glutarate (DSG) (Thermo Fisher Scientific, #20593). Fibroblasts were washed three times in PBS, then fixed with 1% formaldehyde (MilliporeSigma, #1040021000). Following three more washes, fibroblasts were scraped into PBS, spun down into pellets, and resuspended in SDS lysis buffer supplemented with protease inhibitor cocktail (Sigma-Aldrich, #P8340-1ML). Lysates were frozen for at least one hour at −80 °C, followed by thawing and sonication in a microTube-500 AFA fiber screw cap (Covaris, #520185) on a Covaris LE220Rsc Focused Ultrasonicator-110586. Sonication was performed at 7 °C, peak incident power 450, duty factor 30%, 200 cycles per burst, treatment time 100 seconds. Sonication was repeated four times. Sonicated lysates were spun down and quantified for total protein and DNA utilizing a BCA assay (Thermo Fisher Scientific, #23225) and measuring absorbance at 260 nM (A260). Equal lysate concentrations were then brought up to 900 μl in low ionic strength chip dilution supplemented with 1mM DTT (Thermo Fisher Scientific, #R0861), and 4 μl of either IgG control (Thermo Fisher Scientific, #14-4714-82) or FLAG antibody (MilliporeSigma, #F1804-200UG) was added to the appropriate sample and rotated overnight at 4 °C. Following incubation, Pierce^™ Protein G Magnetic Beads (Thermo Fisher #88847) were added and incubated for another 4 hours at 4 °C. Beads were separated from the solution and washed twice in low ionic strength ChIP dilution buffer, once with high salt buffer, once with LiCl, and finally twice with TE buffer following previously published methods⁷⁴.

To elute from the beads, elution buffer was added for 15 minutes, beads separated by magnet, and elution repeated once more. Pooled eluates were then de-crosslinked overnight at constant rotation at 65 °C by adding decrosslinking buffer. Following incubation, 50 μg/ml RNase A (Qiagen #19101) was added for one hour at 65 °C, then 0.4 mg/ml proteinase K (Thermo Fisher Scientific, #AM2548) was added to digest proteins for four hours at 65 °C. DNA was then extracted with phenol:chloroform:isoamyl alcohol (25:24:1, v/v) (Thermo Fisher Scientific, #15593031) extraction, and the DNA pellet was concentrated in a 2X volume of 100% ethanol, 1 μl of glycoblue coprecipitant (ThermoFisher #AM9516), and 0.1X volume of sodium acetate (Thermo Fisher Scientific, #AM9740). DNA pellets were spun down, washed in 75% ethanol, and resuspended in water. Samples then underwent library preparation for ChIP-seq using the KAPA HyperPrep Kit (Roche, #07962363001), following the manufacturer’s protocol.

Library yield was quantified using the dsDNA-specific assay Qubit (Thermo Fisher Scientific, #Q33231). The size distribution of libraries was determined by Agilent 4200 TapeStation analysis (Agilent, #G2991BA). Barcoded sample libraries at equimolar concentrations were pooled and paired-end sequenced using the Illumina NextSeq 2000 Sequencing System with the NextSeq 2000 P3 Reagents Kit (200 cycles) (Illumina #20040560), following the manufacturer’s instructions.

GLIS3^3XFLAG ChIP-seq analysis

ChIP-seq data were processed using the nf-core/chipseq 2.0.0 workflow⁷⁵. Raw sequencing reads were preprocessed using FastQC and Trim Galore! to remove low-quality reads and adapters. The remaining reads were then aligned to a reference genome (GRCh38)⁷⁶ using BWA. Picard’s MarkDuplicates (https://broadinstitute.github.io/picard/), SAMtools⁷⁷, and BAMTools⁷⁸ were used post-alignment for filtering and removing unmapped, multi-mapped, PCR duplicate, and mismatched reads. BEDTools⁷⁹ and bedGraphToBigWig⁸⁰ were used to create normalized bigwig files which were visualized using Integrated Genomics Viewer (IGV)⁸¹. The retained high-quality alignment results were used to call narrow peaks using Model-based Analysis of ChIP-seq version 2 (MACS2)^82,83 against an IgG control with a q-value < 0.05. Consensus peaks set across all FLAG antibody samples were created using BEDTools. featureCounts⁸⁴ was used for counting the consensus peaks in each sample. HOMER was used for annotating peaks relative to gene features and performing motif enrichment analysis.

Functional enrichment analysis was performed using the ClusterProfiler package⁸⁵. Consensus annotated peaks enriched in 24 hr samples were selected based on mean log2 fold change ≥0.15 relative to 0 hr samples. Enriched peaks mapped to the nearest peaks were used for the enrichment analysis considering the Gene Ontology Biological Processes (BP) database⁸⁶. Pathways with Benjamini-Hochberg adjusted p-value < 0.05 were considered for further analysis⁸⁷.

ChIP-qPCR

Following treatment of fibroblasts with media alone or TGF-β and IL-1β (10 ng/ml each) for 24 hours, ChIP isolation was performed as described above (GLIS3^3XFLAG ChIP Seq), but using 4μg each of ChIP-grade IgG control (Thermo Fisher Scientific, #14-4714-82), FLAG antibody (MilliporeSigma, #F1804-200UG), FOSL1 antibody (Cell Signaling, #5841S), or TEAD1 antibody (Active Motif, #61643). Once the DNA pellets were resuspended in water, the samples underwent qPCR with iTaq Universal SYBR Green Supermix for the indicated target genes (see Supplementary Data 5 for primers). Data was normalized using the fold enrichment method where IP signals are divided by the IgG signals, representing the IP signal as the fold increase in signal relative to the background. Heatmaps of gene expression Z-scores were generated in Microsoft Excel v16.

GLIS3 signature and PROTECT cohort analysis

Ordered probit regression, implemented using the polr function in the MASS R package^88,89, was used to assess the association between GLIS3 signature gene expression and the clinical Mayo scores for each patient. Mayo scores were considered as the response ordinal variable, with control samples as the lowest category. The model was fit using single sample Gene Set Enrichment ssGSEA scores and each gene’s expression independently as predictors. Bonferroni adjusted p-values < 0.05 were considered statistically significant.

Derivation of gene set enrichment scores

ssGSEA scores were calculated from transcriptome profiles for each subject using the ssGSEA module (v10.1.0) implemented in GenePattern⁹⁰.

Deconvolution of bulk RNA-seq profiles

RNA-seq profiles of the PROTECT cohort were downloaded from GSE109142. The TPM data were deconvoluted to estimate the fraction of each cell type with the CIBERSORT algorithm implemented in the R package RNAMagnet⁹¹. To define the cell type specific markers, the Wilcoxon rank sum test was performed on the Integrated IBD Atlas. Only genes with p-value < 1x10⁻⁸ and average log-scale fold change ≥ 0.75 compared to the background, and that were expressed in < 20% of cells in the background population, were considered further. Pseudobulk expression matrix of selected markers at cell type population level was used as input for CIBERSORT.

Human participants and ethical statement

The IRB of Mass General Brigham reviewed and approved of the protocol, including a waiver of informed consent from the 16 patients recruited into the PRISM cohort at MGH. The study protocol complied with all relevant ethics guidelines and regulations. Excess tissues from clinically warranted surgical resections of patients with diverticulitis, CD, and UC were collected for research purposes. All samples were deidentified.

Extended Data

Extended Data Figure 1: An integrated single-cell and spatial atlas reveals inducible, inflammation-associated fibroblasts.

(a) UMAP of epithelial, immune, and stromal cell compartments from profiling IBD patients with scRNA-seq.

(b) UMAP of epithelial, lymphoid, myeloid, fibroblast, and stromal cell compartments from profiling IBD patients with Xenium-based spatial transcriptomics.

(c) Dot plot showing effect size (β) and absolute log₂ fold change (logFC) for cell type enrichment across CD and UC compared to non-IBD controls. Blue indicates enrichment in non-IBD and red indicates enrichment in CD and UC. Analysis was done using scCODA with the SMC cells as the reference. Only cell types that passed the significance threshold (FDR < 20%) were considered changed in abundance. See Methods for more details.

(d) Pathway enrichment analysis of IAF-specific genes. Benjamini-Hochberg adjusted p values from a hypergeometric test (one-sided) for enriched pathways were plotted. See Methods for details.

(e) Pseudobulk scaled expression heatmap of anti-TNF resistance signature²¹ plotted across all fibroblast subtypes in the integrated IBD atlas.

Extended Data Figure 2: An integrated single-cell and spatial atlas reveals that inflammation-associated fibroblasts reside in pathological cellular niches.

(a) Sum of cell counts of each niche across all patients profiled in the spatial atlas.

(b) Barplot depicting relative proportion of each niche along the gastrointestinal tract.

(c) Barplot depicting relative proportion of each niche across the 16 patients profiled.

(d) Heatmap depicting the proportion of cell type abundance across ileal (n=3) or colonic (n=13) niches. Only statistically enriched cell types are plotted (Chi-squared two-sided test, Bonferroni adjusted p < 0.05).

(e) Expression dot plot of selected immune genes from activated macrophages.

(f) Pathway enrichment analysis of activated macrophage-specific genes. Benjamini-Hochberg adjusted p values from a hypergeometric test (one-sided) for enriched pathways were plotted.

(g) Barplot depicting relative proportion of each niche across qualitative grades of tissue ulceration.

(h) Barplot depicting relative proportion of each niche across qualitative grades of tissue fibrosis.

(i) Barplot depicting relative proportion of each niche across qualitative grades of tissue inflammation.

(j) Heatmap of cell type proportions across distinct anatomical and histopathological tissue domains as defined by a pathologist on all human patient tissues profiled. Only statistically enriched cell types are plotted (Chi-squared two-sided test, Bonferroni adjusted p < 0.05).

Extended Data Figure 3: Pro-fibrotic IAFs secrete IL-11 in response to activated macrophages.

(a) Schematic of Il11^f/f mice generated by flanking exons 2-4 with LoxP for Cre-mediated excision.

(b) qPCR quantification of Il11 from tissue lysates from Fig. 2a, normalized to Eef2.

(c) Schematic of mNeonGreen knock-in at the Il11 terminus with homology-directed repair.

(d) Percentage of starting weight of mice from Fig. 2a (see Methods for treatment). Filled lines represent s.e.m. Two-way repeated measures ANOVA.

(e) Histopathological scoring (see Methods) on H&E-stained tissues from Fig. 2a.

(f) Gating strategy to quantify IL-11^mNG after water or DSS treatment. Plots are representative of the sample cohort. Related to Fig. 2e, g.

(g) UMAP of PDGFRA+ fibroblasts after acute or chronic DSS.

(h) Pseudobulk scaled expression heatmap of human IAF genes across PDGFRA+ fibroblasts after acute or chronic DSS.

(i) Pseudobulk scaled expression heatmap of human anti-TNF resistance genes across PDGFRA+ fibroblasts after acute or chronic DSS.

(j) Proportion changes of PDGFRA+ fibroblasts across DSS models. Box plots represent the quartiles with medians as the center, and whiskers the 10-90% range. Number of samples for each category: Acute DSS=3; Chronic DSS=2.

(k) Secreted IL-11 measured from co-cultures of ligand-activated monocyte-derived macrophages (see Methods for concentrations) and primary human colon fibroblasts. n=3 cell lines per condition.

(l) qPCR quantification of IL11 from fibroblasts co-cultured with TLR2/6-activated macrophages over time, normalized to HPRT. Filled lines represent s.e.m. n=3, except at 12 hours where n=2 cell lines.

All mice were co-housed. Unless otherwise stated, statistics are by a two-way ANOVA with Tukey’s multiple comparison test on distinct biological replicates and error bars are mean ± s.e.m. ns, not significant.

Extended Data Figure 4: TGF-β and IL-1β drive IL-11 production.

(a) Schematic of IAF activation inference. See Methods for more details. TF, transcription factor.

(b) Left: heatmap of scaled cell type-averaged IAF TF activity (one-sided Wilcoxon, p < 0.01; mean activity difference > 0.75). Center: heatmap of pseudobulk IAF TF expression (Wilcoxon, p < 0.01). Right: ligand frequency ( >5 among top 10 NicheNet predicted regulators).

(c) Dot plot of ligand/receptors driving activated macrophage and fibroblast communication.

(d) Ingenuity upstream regulator analysis of IAF genes (agonists: Z-score > 0).

(e) Secreted IL-11 measured from primary human colonic fibroblasts stimulated with agonists from (a) (10 ng/mL, 24 hours). n=3 per condition for all except TSLP where n=2 cell lines. One-way ANOVA with Dunnet’s multiple comparison test on distinct biological replicates and error bars are the mean ± s.e.m. ns, not significant.

(f) Dot plot of RNA of macrophage or fibroblast markers and TGF-β and IL-1β ligands/receptors across myeloid and fibroblast subtypes.

(g) UMAP of myeloid/lymphoid compartments alongside IL1B or TGFB1 expression.

Extended Data Figure 5: Inflammatory macrophages activate fibroblasts through both TGF-β and IL-1β.

(a) Secreted IL-11 after co-culture of fibroblast knockouts for TGFB1- and IL1B-related ligands or receptors with TLR2/6-activated macrophages (top) or knockouts in activated macrophage knockouts with fibroblasts (bottom). n=3 cell lines per condition.

(b) Z-score heatmap of IAF genes in TGFBR1/2, IL1R1 CRISPRko fibroblasts co-cultured with TLR2/6-activated macrophages, normalized to HPRT. n=3 cell lines per condition.

(c) qPCR of gene knockouts normalized to HPRT. Macrophages: IL-4/IL-13 (10 ng/mL); FSL-1 (10 ng/mL) + ATP (5 mM). Fibroblasts: media. n=2 cell lines per condition.

(d) Secreted TGF-β and IL-1β after co-culture of primary human macrophages and fibroblasts (24 hours) (Methods). n=3 cell lines per condition.

(e) Secreted IL-11 from primary colonic fibroblasts stimulated with TGF-β and/or IL-1β (10 ng/mL, 24 hours). Dashed lines: additive or synergistic response (Methods). One-way ANOVA with Tukey’s multiple-comparisons test. n=3 cell lines per condition.

(f) Left: immunofluorescence of Il11^mNG colons after intraperitoneal injection with IgG control or dual anti-TGF-β and anti-IL-1β antibodies (100 μL in PBS, 100 μg/mouse). Right: IL-11^mNG cell percentage in all DAPI-imaged cells from two pooled independent experiments. Mice (co-housed, 13-20 weeks) were treated with chronic DSS (2.0%, 35 days). n=7 mice per condition.

(g) qPCR quantification of Il11 from lysates from (f) normalized to Eef2. Kruskal-Wallis test with Dunn’s multiple-comparison test.

(h) Total colonic collagen percentage quantification from (f).

(i) Quantification of colonic hydroxyproline normalized to total protein from lysates from (f).

(j) Percent starting weight of mice from (f). Filled lines represent s.e.m. Linear mixed-effects analysis with Dunnett’s multiple comparison test.

(k) Histopathological scoring (Methods) of H&E-stained tissues from (f).

(l) Colon length measurements from (f).

Unless otherwise stated, statistics are by a one-way ANOVA with Dunnet’s multiple comparison test on distinct biological replicates and error bars are the mean ± s.e.m. ns, not significant.

Extended Data Figure 6: Genome-wide CRISPR screens discover novel IAF determinants.

(a) Gating strategy to quantify IL11^mNG after TGF-β and IL-1β stimulation (10 ng/mL, 24 hours).

(b) Dot plot of IL-11 determinants increased in expression during inflammation in CD and UC (Wilcoxon signed rank two-sided test, Benjamini-hochberg adjusted p < 0.05).

(c) Pseudobulk scaled average expression heatmap of Il11, mNeonGreen, and Glis3 from acute and chronic DSS-treated mice.

(d) Left: immunofluorescence of dual-color IL11^mNG fibroblast-Thp-1 macrophage co-cultures with or without TLR2/6 activation. Right: nuclear GLIS3 MFI quantification. Lines represent the median. Two-tailed Mann–Whitney U test. Steady-state, n=156; TLR2/6 stimulation, n=186 individual cells.

(e) qPCR measurement of GLIS3 from primary human colonic fibroblasts co-cultured with TLR2/6-activated monocyte-derived macrophages, normalized to HPRT. Filled lines represent s.e.m. n=3 cell lines.

(f) Nuclear GLIS3 quantification in IL11^mNG fibroblasts with knock-in of GLIS3^3XFLAG stimulated with TGF-β and/or IL-1β (10 ng/mL). Filled lines represent s.e.m. One-way ANOVA with Dunnett’s multiple comparison test. n=4 wells of median values from 6,908-12,459 fibroblasts.

(g) qPCR measurement of Glis3 from the sample cohort in Extended Data Fig. 5f, normalized to Eef2. One-way ANOVA with Dunnett’s multiple comparison test.

Statistics are on distinct biological replicates and error bars are the mean ± s.e.m. ns, not significant.

Extended Data Figure 7: GLIS3 increases in the nucleus to regulate transcriptional control of the IAF gene program.

(a) Left: immunofluorescence of control or GLIS3 CRISPRa fibroblasts at steady state, stained for COL6 and DAPI (nuclei). Right: COL6 MFI quantification. Lines represent the median. Two-tailed Mann–Whitney U test. Control, n=244; GLIS3 CRISPRa, n=198 individual cells.

(b) ChIP-qPCR of IL11 DNA in GLIS3^3XFLAG knock-in fibroblasts stimulated with TGF-β and IL-1β (10 ng/mL, 24 hours). Unpaired Student’s t-test (two-sided). n=5 cell lines per condition.

(c) qPCR measurement of IL11 in IL11^mNG fibroblasts treated with scrambled control or FRA1 siRNA stimulated with TGF-β and IL-1β (10 ng/mL, 24 hours), normalized to HPRT. Two-way ANOVA with Tukey’s multiple-comparisons test. n=3 cell lines per condition.

(d) Z-score heatmap for relative fold change in gene expression for FOSL1 targets after stimulation of control or FRA1 knockdown fibroblasts with TGF-β and IL-1β (10 ng/mL, 24 hours) against control non-stimulated fibroblasts, normalized to HPRT. n=3 cell lines per condition.

(e) qPCR of IL11 in IL11^mNG fibroblasts treated with scrambled control orTEAD1, TEAD3, or dual TEAD1 and TEAD3 siRNA stimulated for TGF-β and IL-1β (10 ng/mL, 24 hours), normalized to HPRT. Two-way ANOVA with Tukey’s multiple-comparisons test. n=3 cell lines per condition.

(f) Z-score heatmap for relative fold change in gene expression for TEAD1 targets after stimulation of control or TEAD1, TEAD3, or dual TEAD1 and TEAD3 knockdown fibroblasts with TGF-β and IL-1β (10 ng/mL, 24 hours) against control non-stimulated fibroblasts, normalized to HPRT. n=3 cell lines per condition.

Statistics are on distinct biological replicates and error bars are the mean ± s.e.m. ns, not significant.

Extended Data Figure 8: GLIS3 is required for IAF induction and aberrant collagen deposition during colitis.

(a) Schematic of Glis3^f/f mice generated by flanking exon 3 with LoxP for Cre-mediated excision.

(b) Percent of starting weight of Glis3^f/f and Glis3^f/f;Cre mice from Fig. 5a. Filled lines represent s.e.m. Two-way repeated measures ANOVA.

(c) UMAP of epithelial, immune, fibroblast, and stromal compartments in Xenium-based spatial profiling of water- and chronic DSS-treated Glis3^f/f and Glis3^f/f;Cre mice.

Data availability

Raw count matrices of the single-cell RNA-sequencing data used in this study were downloaded from various repositories. Martin et al. 2019 is available at NCBI Gene Expression Omnibus (GSE134809). Smillie et al. 2019 and Kong et al. 2023 are available at the Broad Single Cell Portal (SCP259 and SCP1884, respectively). Friedrich et al. 2021 was downloaded from ImmPort (SDY1765). Processed anndata objects of single cell RNAseq IBD atlas are available at the Broad Single Cell Portal (SCP2927).

Single-cell RNA-sequencing data of stimulated fibroblasts profiled at various time points are available in the NCBI Gene Expression Omnibus (GSE250516). Raw single-cell RNA-sequencing data for PDGFRA+ fibroblasts from the mouse large intestine are available in the NCBI Gene Expression Omnibus (GSE288481). Processed anndata object of PDGFRA+ fibroblasts is available at the Broad Single Cell Portal (SCP3384). Bulk RNA-sequencing data generated in this study are available in the NCBI Gene Expression Omnibus (GSE250515). ChIP-seq data generated during this study are available in the NCBI Gene Expression Omnibus (GSE250514). CRISPR screen data generated during this study are available in Supplementary Data 2. Publicly available RNA-sequencing data for the PROTECT cohort was downloaded from the NCBI Gene Expression Omnibus (GSE109142). Anndata objects of Xenium based spatial transcriptomics profiling is available in the Broad Single Cell Portal (SCP2927 for human intestinal tissue, SCP3384 for mouse intestinal tissue). Raw Hematoxylin and eosin (H&E) stained images post-spatial profiling are available on Zenodo at https://doi.org/10.5281/zenodo.17518435⁹².

All unique biological materials generated in this study are available upon request.

Code availability

No new software code was generated in this study. All code used for data analysis in this study is available upon request from the authors.