Single cell sequencing data identify distinct B cell and fibroblast populations in stricturing Crohn's disease

Single cell RNA sequencing of human full thickness Crohn's disease (CD) small bowel resection specimens was used to identify potential therapeutic targets for stricturing (S) CD. Using an unbiased approach, 16 cell lineages were assigned within 14,539 sequenced cells from patient‐matched SCD and non‐stricturing (NSCD) preparations. SCD and NSCD contained identical cell types. Amongst immune cells, B cells and plasma cells were selectively increased in SCD samples. B cell subsets suggested formation of tertiary lymphoid tissue in SCD and compared with NSCD there was an increase in IgG, and a decrease in IgA plasma cells, consistent with their potential role in CD fibrosis. Two Lumican‐positive fibroblast subtypes were identified and subclassified based on expression of selectively enriched genes as fibroblast clusters (C) 12 and C9. Cells within these clusters expressed the profibrotic genes Decorin (C12) and JUN (C9). C9 cells expressed ACTA2; ECM genes COL4A1, COL4A2, COL15A1, COL6A3, COL18A1 and ADAMDEC1; LAMB1 and GREM1. GO and KEGG Biological terms showed extracellular matrix and stricture organization associated with C12 and C9, and regulation of WNT pathway genes with C9. Trajectory and differential gene analysis of C12 and C9 identified four sub‐clusters. Intra sub‐cluster gene analysis detected 13 co‐regulated gene modules that aligned along predicted pseudotime trajectories. CXCL14 and ADAMDEC1 were key markers in module 1. Our findings support further investigation of fibroblast heterogeneity and interactions with local and circulating immune cells at earlier time points in fibrosis progression. Breaking these interactions by targeting one or other population may improve therapeutic management for SCD.


| INTRODUC TI ON
Chronic inflammation is a significant factor in driving intestinal fibrosis and reciprocal interactions between fibroblasts and inflammatory immune cells are key to the pathogenesis of fibrosis. 1ricture formation occurs in 30-50% of Crohn's disease (CD) patients 2,3 and stricturing CD (SCD) is linked with high levels of morbidity and healthcare utilization. 4,5Fibrosis is a transmural process and current pharmacological treatments, including biologic and small molecule drugs, neither reduce fibrosis nor the requirement for stricture resection, 6 and stricture recurrence is common. 7brosis is characterized by increased deposition of extracellular matrix (ECM) proteins such as Collagen-I with accompanying thickening of both submucosa and smooth muscle cell layers. 8[14] Here, we report the use of human full thickness CD resection specimens and single cell RNA sequencing (scRNA-seq) to gain information on the cell types in the SCD and non-stricturing (NSCD) regions of the CD small bowel.Note: All cases included in our study were "sporadic" Crohn's disease (CD) with no cases of very early onset CD.This is evidenced by the fact that the disease duration at the time of surgery for the 2 paediatric diagnoses was 16 and 30 years.Several patients had a history of penetrating disease (Montreal B3), but resection specimens used in this analysis were harvested by a GI pathologist to ensure that they derived from fibrotic structures (SCD) or non strictured (NSCD) segments away from the site of any penetration.

F I G U R E 1 Legend on next page
Sequence reads were aligned with cellranger (v3.1.0;10x genomics).Imported single cell gene counts were analysed using R package Seurat (v4).A data set was prepared per sample and each was normalized (function "NormalizeData") followed by the identification of variable features (function "FindVariableFeatures" using parameters "vst"method and nfeatures = 2000).The fraction of mitochondrial genes was calculated for quality assessment (function "PercentageFeatureSet"). Data sets were integrated by selecting integration features and anchors and then using the function "IntegrateData."Integration data were then scaled (function "ScaleData") and batch effects regressed out by incorporating patient ID.Principle component analysis (functions "RunPCA" and "FindNeighbors") was used for graph based clustering at a resolution of 0.3 and Uniform Manifold Approximation (UMAP) and t-SNE dimensionality reduction was then computed (functions "RunUMAP"and "RunTSNE").Batch correction was performed using Harmony by integrating the patient ID variable. 15Fibroblast and B clusters were identified by visualizing cell markers (e.g., Lumican (LUM) and CD83 respectively) before being organized as independent sub-sets and analysed further.Monocle3 was used to subcluster and analyse pseudotime trajectories of B cell and fibroblast lineages. 16[19] Cross validation analysis of fibroblast cells (CD data set to the fibroblast atlas) used custom R scripts and marker gene identification (Seurat FindMarkers function) from both data sets.Tables of overlapping marker genes were assembled with the respective fold change metrics.Data imported into Cytoscape and StringDB networks were prepared using Omics visualizer app. 20

| Stricture-specific cell types in stricturing Crohn's disease
We performed scRNA-seq of cells isolated from full-thickness tissue surgically resected from the strictured ileum of CD patients (n = 4) and non-strictured CD controls (NSCD n = 4); 3/4 were patient matched (Table 1).No prior selection was applied (e.g., FACS) to avoid cell type bias on sequencing.Instead, all cells were sequenced (total of 14,539 cells), integrated and batch corrected. 15UMAP visualized the complete integrated data set.This identified 24 clusters which were assigned to 16 definable cell lineages based on cell markers obtained from the published literature (Figure 1). 20,21NSCD and SCD preparations contained the same cell types (Figure 2; Table 2) and the same proportions of endothelial and fibroblasts, indicating that phenotype did not impact cell release (Table 2; Figure S1A,B).

| Stricturing Crohn's disease is associated with selective increases in B cells and IgG plasma cells
Fibroblast function under inflammatory conditions is shaped by interaction with immune cells. 1 Amongst immune cell types, B cells and plasma cells were most markedly increased in frequency in SCD compared to NSCD (Figure 2; Table 2).As well as differences in absolute numbers of plasma cells, there were differences in the distribution of antibody subclasses produced by these cells.
The proportion of IgG subclasses was increased in SCD, whereas IgA2, was increased in NSCD (Figure 3A).CXCR4, which is reported to facilitate plasma cell homing to inflamed tissue was expressed by 62% of plasma cells in SCD versus 44% in NSCD (Figure 3B). 22ditional sub-clustering of B cells identified a population of naïve B cells (Figure 3C, cluster 1), expressing IGHD and IGHM; a cluster expressing genes consistent with a memory B cell phenotype (Figure 3C, cluster 2); and a small subset of B cells expressing germinal centre markers including BCL6 and AICDA (Figure 3C, cluster 3), the latter of which were almost exclusively found in SCD but not NSCD.

| Identification of two fibroblast subtypes
LUM is an ECM-secreted proteoglycan known to regulate collagen fibrogenesis and two LUM+ve fibroblast subtypes were identified from our scRNA-seq data (Figure 1A,B).These were further subclassified as two clusters (fibroblast cluster 12 (C12) and fibroblast cluster 9 (C9)) based on expression of selectively enriched genes (fold change >1.5 and adjusted p ≤ 0.001; Figure 4; Table S1).Ranking these genes according to the numbers of cells in which they were expressed demonstrated that the highest proportion of cells within the two clusters expressed the profibrotic genes Decorin (DCN) for F I G U R E 1 Cellular profile of stricturing Crohn's disease (SCD).ScRNA-seq was performed on cells from full-thickness tissue surgically resected specimens from stricturing (SCD; n = 4) and non-stricturing (NSCD; n = 4) ileum of CD patients (n = 5).Three patients provided both SCD and NSCD specimens.No prior cell selection was applied (e.g., FACs sorting) and all cells (n = 14,539) were sequenced.(A) Uniform Manifold Approximation (UMAP) identified 24 clusters assigned to 16 definable cell lineages (Figure S1A).Undifferentiated cells (Undiff1/2) contained markers of undifferentiated cells as defined by protein atlas webtool, 20 and expressed TOP2A and MKI67 unlike any other cell type.Unknown (U) had no identifiable gene marker linked to any cell type.(B) Bubble plot shows levels of expression of key markers for different cell types identified from published data sets.
C12 and JUN for C9 (Table S1).Based on prior studies we considered the expression of key genes (collagen genes or potential markers of fibrosis) involved in the fibrotic process (Table S1).[25]

| Extracellular matrix, stricture organization and regulation of WNT pathway genes are associated with fibroblast clusters
The GO and KEGG Biological terms associated with the fibroblast cluster C12 and C9 gene sets included ECM and stricture organization (both clusters; GO terms), and positive and negative regulation of WNT pathway genes (C9; GO terms) and the IL-17 signalling pathway (C9 KEGG terms) (Figure S2A-H).The fibroblast subtypes corresponded to those identified previously by scRNA-seq in ulcerative colitis (UC) and other fibrotic diseases. 23,26C9 fibroblast markers (Table S2) were validated independently by comparing with PI16 fibroblasts in the perturbed human fibroblast atlas (fibroXplorer database 26 ; n = 3604 fibroblasts; Figure 5A,B).Single cell trajectory analysis was performed on all fibroblasts to identify possible differentiation states 27 and from this, four sub-clusters were identified (Figure 5C).Differential gene analysis identified gene markers for each sub-cluster and intra sub-cluster gene analysis identified co-regulated gene modules that aligned along predicted pseudotime trajectories (Figure 5D,E).The modules represent groups of cells with gene expression profiles that are sufficiently unique to allow their separation into individual modules with a unique identify.Some genes may be included in more than one module.The gene members of each of the 13 gene modules are shown in Figure S3.The top two markers of module 1 were CXCL14 and ADAMDEC1 (Figure S3), previously identified to be markers in a distinct fibroblast population in UC and other fibrotic diseases, 23 indicating that sub-cluster 1 are the same ADAMDEC1-like fibroblasts.

| DISCUSS ION
We have, for the first time, generated sequencing data from all the cell types isolated from resected full-thickness SCD and NSCD intestinal specimens from stricturing CD patients.Access to well phenotyped, fresh surgical specimens, rather than the more readily accessible mucosal biopsies, allowed sampling from the deeper layers of the bowel that contribute to CD fibrosis.This underlines the clinical importance of our dataset.In contrast to studies in other organ systems that have used enrichment, we determined natural groupings from an atlas of 14,539 sequenced cells.This identified An Imbalance in the gut flora has been reported to possibly promote the mucosa of the terminal ileum to increased IgA production and thereby expand the proportion of IgA-coated bacteria in Diarrhoea-Predominant Irritable Bowel Syndrome. 28Activating IgA class switching may serve to regulate inflammation and subsequent pathologies, 28 and it remains to be seen whether this includes intestinal fibrosis.
Several studies have recently identified a role for B cells in the pathogenesis of UC, 22,29,30  interactions with Fc receptors and the consequent promotion of inflammatory Th17 responses. 30In UC, IgG plasma cells infiltrate the inflamed intestine via CXCR4, 22 and this chemokine receptor was expressed by many plasma cells in our CD study.IgG plasma cells expressing CXCR4 were enriched in SCD compared with NSCD samples.Reclustering of B cells revealed naïve, memory and germinal centre B cell populations, the latter of which were almost exclusively found in SCD.These finding are consistent with the formation of organized tertiary lymphoid tissue (TLT) in SCD. 31 TLT formation is a feature of CD, 32 but its causal contribution to the inflammatory process or to fibrosis is currently poorly understood.In other inflammatory contexts, B cells have been shown to interact directly with fibroblasts to influence the fibrotic process in inflammation, 33 raising the possibility that a similar process contributes to fibrosis in CD. contribute to CD pathogenesis, or conversely represent a protective response.As our analysis used full thickness intestinal tissues it is likely that we captured such structures and cells that might indicate TLS were identified.However, we cannot comment on the spatial location of different cell types because of the dissociation process necessary for scRNA-seq.To resolve help these issues, work could be undertaken in the future using single cell spatial transcriptomics.
Whilst our data might suggest a link with fibrosis the cellular involvement and mechanisms remain unclear.
Lumican is a member of the class II small leucine-rich proteoglycan superfamily known to be involved in fibroblast function and organization of collagen fibrils in the ECM. 34In this context, LUM is critical for the pathogenesis of several fibrotic diseases, including lung, heart, cornea and liver, [35][36][37][38][39] and may promote early fibrotic responses to injury. 35LUM is also known to be involved in inflammation, 40 and immune cell recruitment after injury. 41Lumican secretion from TNFα stimulated fibroblasts can promote differentiation of monocytes to fibroblasts. 42 have identified two LUM+ve fibroblast subtypes, which were further subclassified as two clusters, C12 and C9.4][45] In addition, a number of genes were identified that are involved in the fibrotic process.Cells in the C9 cluster express ACTA2, the marker of myofibroblasts.Increased production of ECM proteins, such as collagen, is a key feature in the mechanism for development of fibrotic strictures and is part of the of wound healing response to inflammation.Accordingly, we identified COL4A1, COL4A2, COL15A1, COL6A3, COL18A1 expression in C9 fibroblasts.
These cells also expressed ADAMDEC1, previously identified to be a marker in a distinct population in UC and other fibrotic diseases. 23MB1 and GREM1 were also found to be expressed.Interestingly, LAMB1 and COL4A2 are potential markers of fibrosis progression in liver and GREM1, a highly conserved member of the TGFβ superfamily, is involved in fibrosis across multiple organ systems. 24,25It has been suggested that Lumican-inhibiting drugs could be used to modulate fibrosis. 42In this context, our finding of two LUM+ve fibroblast subtypes and the expression of profibrotic genes by C12 and C9 cluster cells may support the identification of suitable drug targets.
Interestingly, positive and negative regulation of WNT pathway genes were found to be associated with the C9 fibroblast cluster in SCD.Fibrosis has been associated with WNT pathway activation in a number of organs such as the lung, liver, kidney, and skin, 46 and in the development of aggressive complications in inflammatory bowel disease (IBD) patients, notably IBD-associated colorectal cancer, 47 and formation of intestinal fistulae. 48Recently we identified the importance of cross-talk between WNT and TGFβ signalling pathways in CD intestinal fibrosis. 49The results presented here for C9 cluster fibroblasts are in accordance with this prior study and support further the potential use of smallmolecule inhibitors of WNT signalling in treating intestinal fibrosis in CD by targeting particular fibroblast clusters identified by scRNA-seq.
Although IL-17 is considered to play a role in fibrosis in some organs, its role in intestinal fibrosis in human CD is unclear.were analysed using Monocle3. 16(D) Identification of the most significant gene marker for each cluster using differential gene analysis.(E) Co-regulated gene modules identified from predicted pseudotime trajectories using Monocle3. 16Each module contains multiple genes that have similar expression patterns between cell neighbours (i.e.intracluster gene analysis).Gene members of each gene module are listed in Figure S3.

2 | ME THODS 2 . 1 | 2 . 2 |
Patient recruitment and single cell isolation from surgically resected tissue Appropriate local Ethics Committee approvals (London-City Road and Hampstead Research Ethics Committee; 15/LO/2127) and informed consent were obtained prior to patient recruitment.Human full thickness CD resection specimens from the CD small bowel were washed with Hanks' Balanced Salt solution (HBSS) supplemented with 0.01% Dithiothreitol and then HBSS-EDTA (1 mM) for 10 min per wash under agitation at 37°C.Finely dissected full thickness tissue was incubated with 20 mL of Dulbecco's modified Eagle's medium (DMEM) (Gibco) supplemented with collagenase type 1A (1 mg/ mL, c2674-500 mg, Sigma) and DNase I (10 units/mL) for 45-60 min under gentle agitation, then filtered through a 100 μm cell strainer and the cell pellet washed twice in PBS.For scRNA-seq cells were re-suspended in 1 mL FBS supplemented with 10% DMSO and cryostored in line with the recommended 10x genomic protocol for Fresh Frozen Human Peripheral Blood Mononuclear Cells.Single cell RNA sequencing SCD and NSCD tissue was collected and processed independently (n = 8 in total; SCD, 4 and NSCD 4 samples) as above.Before library preparation, cell suspensions were thawed, washed (PBS BSA 0.04%), filtered and cell viability assessed; a viability cut-off of >70% immune cells at earlier time points in fibrosis progression.Breaking these interactions by targeting one or other population may improve therapeutic management for SCD.K E Y W O R D S B cells, Crohn's disease, fibroblasts, fibrosis, stricturing TA B L E 1 Clinical characteristics of Crohn's disease cohort.
fibroblasts and immune cell populations in the gut without any potential bias associated with pre-selection.The availability of human IBD datasets, such as fibroXplorer, 26 supported independent validation.The numbers of plasma and B cells were increased in SCD tissue to a much greater degree than other immune cell populations.The differential role of IgA1 and IgA2 in the intestine is poorly understood.Cells secreting IgA1 usually predominate in the small intestine and those secreting IgA2 are increased in the colon.IgA2 is more resistant to proteases and thus may persist longer, with an increased effect on the microbiota in the gut lumen when secreted.The altered pattern of expression of IgA1 and 2 in strictured versus non-strictured tissue might reflect a change in the production of factors controlling isotype switching perhaps as the result of altered microbial status in the strictures versus normal gut tissue.

F I G U R E 3
The observed increases in immune cells, both proportionally and in number, in the strictured compared to non-strictured tissue is probably not surprising given the role of inflammation in the fibrotic process.However, more interesting are the selective changes observed such as the discriminating increase in B cells and the changes observed in the composition of this B cell population.Collectively these observations suggest that B cells and/or IgG may play a significant role in the fibrotic process associated with development and maintenance of strictured regions.The large increase in B cells (and plasma cells) may partially mask smaller increases in populations of other cell types throughout the gut that could contribute potentially to the fibrotic process.Furthermore, we know that recurrent stricture development is common after surgical resection.7 Tertiary lymphoid structures (TLS) are a known histological feature of CD.Nevertheless, it remains to be determined if they Analysis of B cells and plasma cells in stricturing (SCD) and nonstricturing (NSCD) small bowel tissue.(A) Proportions of antibody subclasses expressed by plasma cells in SCD and NSCD tissues.(B) Expression of CXCR4 in plasma cells.(C) Sub-clustering of B cells identified three distinct clusters consistent with naïve (1), memory (2) and germinal centre (3) subsets.

50 A
putative role for IL-17 signalling in CD fibrosis is supported by the finding that an anti-IL-17 antibody can diminish colitis and fibrosis in the TNBS murine model through downregulation of collagen expression and pro-fibrogenic cytokines such as IL-1β, TGF-β1, and TNFα,51 or possibly by reducing EMT.52However, data from two clinical trials did not identify any notable efficacy from blocking IL-17, which may relate to the importance of the cytokine in intestinal homeostasis including mucosal protection.50Whether blocking of IL-17 or a family member has a direct anti-fibrotic effect or is related to an ability to reduce inflammation remains to be established.In conclusion, our study provides an unbiased reference dataset for SCD versus NCD, which is controlled internally because most of these tissues are matched from the same individual.This enables the internal relationships of different cell types to be compared in a single time window.Our data indicate clearly that therapeutic intervention must come much earlier in CD progression than the timepoint described here to break the apparent interaction between immune cells and fibroblasts.In this context, it F I G U R E 4 Identification of the C12 and C9 fibroblast subtypes in stricturing and non-stricturing small bowel tissue.UMAP of fibroblasts identifying cluster 12 (all C12, blue symbols) and cluster 9 (all C9, red symbols) from the integrated stricturing (SCD; triangles) and non-stricturing (NSCD, dots) scRNA data set.Insert shows a comparison of the numbers of LUM+ve C12 and C9 clusters in resected small bowel from both NSCD and SCD tissue.The percentage of each cluster in either the NSCD or SCD data set is given in parenthesis.The fold change in C12 and C9 fibroblasts in NSCD and SCD is shown as SCD/ NSCD.would be interesting to consider whether increased immune cells residency in the mucosa and/ or assay of circulating antibodies/ immune cells in blood could indicate fibrotic progression at a much earlier time point.AUTH O R CO NTR I B UTI O N S David T. Humphreys: Data curation (equal); formal analysis (equal); investigation (equal); methodology (equal); resources (equal); writing -review and editing (equal).Amy Lewis: Conceptualization F I G U R E 5 Workflow and stringdb network assembly, trajectory and differential gene analysis for fibroblast populations and clusters.(A) Schematic of scRNA-seq workflow, incorporating stringdb network assembly for C9-like fibroblasts and target validation workflow.(B) Stringdb network for C9-like SCD fibroblasts.(C) Four clusters resolved from differentiation trajectories of the two fibroblast populations Stricturing (SCD) and (B) non-stricturing (NSCD) small bowel tissue comprise the same cell types but proportions differ markedly for B cells and plasma cells.A UMAP of NSCD and SCD preparations indicate they are comprised of the same cell types.B cells and plasma cells showed the most marked increased in frequency in SCD compared to NSCD (see Table2).
COL15A1, COL6A3, COL18A1; a marker of a distinct fibroblast population in fibrosis, ADAMDEC1; and LAMB 1 and GREM1, both reported to be involved fibrosis (C9 compared to C12; p-value adjusted < 9.0E-05; Table B cells and plasma cells showed the most marked increased in frequency in stricturing Crohn's disease (SCD) compared to non-strictured (NSCD).
but their contribution to CD remains poorly understood.There was a shift towards plasma cells expressing IgG subclasses, and away from IgA, in SCD.IgG drives inflammation in UC by activating innate immune cells viaTA B L E 2Note: Numbers and frequency (%) of cells identified from NSCD and SCD small bowel tissue.