The FOXP3⁺ Pro-inflammatory T cell: A Potential Therapeutic Target in Crohn’s Disease

Abstract

Background and aims:

The incidence of Crohn’s disease (CD) continues to increase worldwide. The contribution of CD4⁺ cell populations remains to be elucidated. Here, we aim to provide an in-depth transcriptional assessment of CD4⁺ T cells driving chronic inflammation in CD.

Methods:

We performed single cell RNA-sequencing in CD4+ T cells isolated from ileal biopsies of CD patients compared to healthy individuals. Cells underwent clustering analysis, followed by analysis of gene signalling networks. We overlapped our differentially expressed genes with publicly available microarray datasets and performed functional in vitro studies, including an in vitro suppression assay and organoid systems, to model gene expression changes observed in CD TREGs and to test predicted therapeutics.

Results:

We identified five distinct FOXP3⁺ regulatory T (Treg) subpopulations. Tregs isolated from healthy controls represent the origin of pseudotemporal development into inflammation-associated subtypes. These proinflammatory Tregs displayed a unique responsiveness to TNFα signaling with impaired suppressive activity in vitro and an elevated cytokine response in an organoid co-culture system. As predicted in silico, the histone deacetylase inhibitor Vorinostat normalized gene expression patterns, rescuing the suppressive function of FOXP3⁺ cells in vitro.

Conclusion:

We identified a novel, proinflammatory FOXP3⁺ T cell subpopulation in CD patients and developed a pipeline to specifically target these cells using the FDA-approved drug Vorinostat.

Lay Summary:

Crohn’s intestinal lesions are enriched in a unique population of FOXP3+ cells that exhibit a pro-inflammatory transcriptional signature consistent with strong TNFα signaling.

Introduction

The incidence of inflammatory bowel disease (IBD), comprising Crohn’s disease (CD) and ulcerative colitis, continues to increase worldwide¹. Though advances in the field have resulted in modern immunotherapies that continue to lessen disease burden, a significant number of patients remain unresponsive to or only partially responsive to clinically available therapies. Refractory disease results in significant morbidity to patients, highlighting the need for novel therapeutics.

Several studies have investigated CD-associated transcriptome-wide changes. Due to the analysis of whole-tissue samples, the contribution of individual cell populations remain largely unknown. To identify these cell types, as well as gene expression changes relevant to CD, recent investigations employed single cell RNA sequencing (scRNA-seq) approaches. Interestingly, Jaeger et al. detected an increased T helper 17 (Th17) over regulatory T cell (Treg)/T follicular helper cells (Tfh) ratio from single cell suspensions of lamina propria isolated from CD lesions². In general, CD4⁺ T cells develop in the thymus and naïve CD4⁺ T cells encounter polarizing cytokines in the periphery inducing the expression of specific transcription factors underlying the differentiation into specialized functional subsets³. The inflamed mucosa of IBD patients displays a substantial infiltration of Th17 cells⁴. In agreement with this finding, the levels of Th17-excreted cytokines were elevated in inflamed lesions, suggesting a proinflammatory function of these cells^5,6.

The X-chromosome-encoded forkhead domain protein 3 (FOXP3) transcription factor represents the master Treg lineage factor and is critical for Treg development, maintenance and function^7–10. We and others have additionally documented a predominant infiltration of CD4⁺FOXP3⁺ cells in the Crohn’s inflammatory lesion^11,12. FOXP3⁺ Treg cells are critical to the maintenance of immune self-tolerance and are key players in the regulation of inflammatory signals in many immune mediated disorders; yet emerging evidence suggests altered function within the inflammatory milieu. The notion of a proinflammatory Treg is supported by single cell sequencing technology suggesting distinct proinflammatory phenotypes resembling that of other effector T cells¹³. These previous scRNA-seq studies on CD focused on total lamina propria cellular populations limiting in-depth analysis of CD4⁺ subpopulations, such as the putative proinflammatory FOXP3⁺ cell. To overcome this issue and to explore transcriptome-wide changes in CD4⁺ cells in CD on a single-cell level, we purified CD4⁺ cells in human biopsies and performed scRNA-seq. In this study, we demonstrate a novel pro-inflammatory function of FOXP3⁺ T cells in CD which are responsive to Tumor Necrosis Factor alpha (TNFα) signaling. Importantly, the inhibition of histone deacetylases reversed the expression pattern of CD-associated inflammatory genes and rescued the suppressive function of these cells. These data have therapeutic import to patient selection for anti-TNF therapy and potentially to Treg adoptive cell transfer programs.

Methods

Isolation, cryopreservation, and preparation of single cell suspensions of TI biopsies

De-identified surgical specimens from inflamed terminal ileal mucosa were obtained from CD patients diagnosed by established clinical, radiologic, and histopathologic criteria (four males, three females). Tissues were immediately placed in ice-cold RPMI and processed within 2 h of resection. For comparison, biopsy specimens were obtained from age- and sex-matched healthy controls. Tissue resections were subjected to sampling using biopsy forceps along the mucosa of the specimens, collecting approximately 20 biopsies from each.

Biopsies were stored in HypoThermosol^® FRS (BioLife Solutions) on ice and transferred to CryoStor^® CS10 (StemCell) in 1.5 ml cryogenic vials. Samples were frozen at −80°C overnight and stored in liquid nitrogen until single-cell suspensions were prepared. Healthy controls consisted of patients evaluated for gastrointestinal symptoms with normal endoscopic evaluations. Clinical characteristics for CD patients are detailed in Supplementary Tables 1 and 2. Due to low cell numbers, the cells of patients 4, 6 and 7 were combined. Cell viability was similar in both healthy and Crohn’s populations. The study was approved by the Institutional Review Board (IRB) committee-approved protocol (13–000712) and written informed consent was received from participants before inclusion in this study.

For details on preparation of single cell suspensions, refer to Supplementary Methods.

Flow cytometry-based sorting of CD4⁺ cells

Following isolation, cells underwent cell-surface staining with AF488-CD4 (Biolegend, cat. no.: 300519). Cells were gated on live CD4⁺ cells utilizing high expression of AF488 (CD4⁺) and low expression of 7-AAD (Biolegend, cat. no.: 420404) and PO-PRO I (PO-PRO I, cat. no.: P3581) and subjected to sorting through a FACSAria digital cytometer equipped with FACSDiva v 8.0.1 software.

Single-cell RNA sequencing (scRNA-seq)

Whole live cells were washed twice in 1x PBS + 0.04% BSA and immediately submitted to the Core for single cell partitioning. Full methodology available in Supplemental Methods. The barcoded Gel Beads were thawed and the cDNA master mix was prepared according to the manufacture’s instruction for Chromium Next GEM Single Cell 3’ Library and Gel Bead Kit (10x Genomics, Pleasanton, CA). A per sample concentration of 500,000 cells/mL or better was required for the standard targeted cell recovery of approximately 5,000 cells. The cell suspension and master mix, thawed Gel Beads and partitioning oil were added to a Chromium Single Cell G chip. The filled chip was loaded into the Chromium Controller, where samples were processed and partitioned into uniquely labeled GEMs (Gel Beads-In-Emulsion). The GEMs were collected for reverse transcription, GEM dissolution, and cDNA clean-up. The resulting cDNA contained a pool of uniquely barcoded molecules. Standard Illumina sequencing primers and a 10X Genomics unique i7 sample index were added to each cDNA pool. Pools and libraries were measured using Qubit High Sensitivity assays and Agilent Bioanalyzer High Sensitivity chips. Libraries were sequenced at between 40,000 and 50,000 fragment reads per cell following Illumina’s standard protocol using the Illumina cBot and HiSeq 3000/4000 PE Cluster Kit. The flow cells were sequenced as 100 X 2 paired-end reads on an Illumina HiSeq 4000 HD using HiSeq 3000/4000 sequencing kit and HCS v3.4.0.38 collection software. Base-calling was performed using Illumina’s RTA version 2.7.7.

scRNA-seq data analysis

scRNA-seq data were aligned and quantified using the 10X Genomics Cell Ranger Software Suite (v6.1.1) against the human reference genome (hg38). Seurat package (v4.0)^28,29 was used to perform integrated analyses of single cells. Genes expressed in <3 cells and cells that expressed <200 genes and >20% mitochondria genes were excluded from downstream analysis. Data was normalized and scaled according to sequencing depth (raw counts per cell, nUMI) and condition. Additionally, a regression due to mitochondrial expression for each individual cell was performed. Each dataset was SCTransform-normalized and the top 3000 Highly Variable Genes (HVGs) across cells were selected. Datasets were integrated based on “anchors” identified between datasets before Principal Component Analysis (PCA) was performed for linear dimensional reduction. Shared Nearest Neighbor (SNN) Graph was constructed to identify clusters on the low-dimensional space (top 30 statistically significant principal components (PCs)). For PCA, we used the normalized and scaled single cell RNA-seq Seurat dataset. By separating by patient ID and cluster ID, we created a merged variable and took the average expression per these preselected conditions. We then used DESeq2 to perform estimation of size factors, estimation of dispersion and negative binominal Generalized Linear Model fitting as Wald statistics. We performed a variance stabilization before plotting these normalized counts according to the top 3,000 genes, selected by highest variance. Transcriptomic profiles were generated based on enriched marker genes in each cell subtype cluster and differentially expressed genes (DEGs) between two conditions were detected at the cluster level. Heatmaps were generated in Seurat (v4.1.0) using the top 10 DEGs based on the average log2 fold change after applying the FindAllMarkers function (min.pct=0.25, logfc.threshold=0.25). Tregs were further subclustered applying the FindNeighbors (dims 1:10) and FindClusters function (resolution=0.2), leading to identification of five distinct populations. Afterwards, the expression of the top 25 genes per cluster was calculated by applying the FindMarkers function (min. pct=0.25, idents cluster 1–5). For Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) calculations, the RunUMAP function (dims = 1:40, reduction = “pca”) was applied and both, DimPlot (seurat4) and plot_cells (monocle3, v.1.2.0) tools, were used for plotting.

PCA plots were created after normalizing and scaling the dataset. Then, we extracted condition- and cluster IDs, grouped them by a merged (condclust) variable and averaged the expression of each gene per this variable (n=16). We performed variance stabilization using the package DESeq2 (v1.30.1) and used the biplot function of the stats package (v4.0.3).

For pseudotime analyses, monocle3 (preprocessed with n=100 dims, reduced dimension, chosen order root node manually according to recommendations of monocle2) and monocle2 (v2.18.0, reduction method=DDRTree, max. components=2, heatmaps constructed with differential gene test according to “~sm.ns(Pseudotime)”, significance level qval<0.1, plotted with plot_pseudotime) were applied. For the signaling network, CellChat (1.1.3) has been utilized using the eight main clusters and aggregating a cell-cell communication network from significant signaling genes and interactions (threshold.p=0.05) according to netAnalysis_signalingRole after the centrality scores were calculated in the inferred intercellular communication network (“netP”, min.cells=10). The identification of senders, receivers (and others) in the intercellular communication network was computed by network centrality measures for each cluster. In these weighted-directed networks, out-degree and in-degree identified dominant senders and receivers. The mediators (gatekeeper) and influencers (controlling the communications) as additional parameters of centrality networks are also listed. In a separate analysis, genes upregulated in Tregs isolated from CD patients were compared to a publicly available microarray dataset in which human Tregs were treated with TNFα for 2 or 24 hours in vitro (GSE18893). Significance was calculated using Fisher’s exact test.

The drug target prediction was performed based on “A Single-cell Guided pipeline to Aid Repurposing of Drugs” (ASGARD, v1.0.0¹⁷). Briefly, DEGs (Crohns 1–5 vs. Control 1–6) were selected with limma (v3.46.0), mono-drug repurposing for each cell type followed for the “large-intestine” drug, gene and rankMatrix, calculating a mono drug score. Next, we created a final drug score (FDR<0.1) and a cluster-specific drug score (FDR<0.1) using FDA-approved drugs only. The results were exported as nodes and visualized in Cytoscape (v3.9.0³⁰). The dot charts were created with ggdotchart within the ggpubr (v0.4.0) package, after the FindAllMarkers function (min.pct=0.25, log2c.threshold=0.25) was applied and the top and bottom 10 genes selected for each cluster.

Gene set enrichment analysis (GSEA)

To compare genes upregulated in Tregs isolated from CD patients to gene expression patterns from Tregs isolated from mice with Treg-specific Hdac1 or Hdac2 deletion (GSE139480), we performed GSEA. We generated custom gene sets containing genes significantly upregulated (padj ≤0.05, (log2 FC ≥0.75) and carried out GSEA (GSEA software, Broad institute, UC San Diego) with default settings (1000 permutations for gene sets, Signal2Noise metric for ranking genes) using our own scRNA-seq dataset (CD vs. healthy).

T cell isolation and differentiation

Naïve CD4⁺ T cells as well as CD25⁺⁺ and CD25⁻⁻ cells were isolated from peripheral blood mononuclear cells (PBMCs) and differentiated using published protocols (see Supplementary Methods).

Cytokine and inhibitor treatments

Precise dosing and source of cytokine and inhibitor reagents are supplied in Supplementary Methods.

T cell suppression assay

CD25 cells were isolated from the same patients as CD4⁺ T cells (see Supplementary Methods). One day prior to the assay, CD25 cells were defrosted and cultured in cRPMI in the presence of CD28 on CD3-labeled plates at 37°C. 5,000 CD25⁻⁻ responders and Human T-activator CD3/CD28 Dynabeads^® (Life Technologies, cat. no.: 11131D) were added per well of 96-well round bottom plates and mixed with Tregs at noted dilutions. Proliferation was assessed after culturing cells at 37°C for 5 days and upon addition of 1 μCi tritiated thymidine for 18 h.

Patient-derived organoids and T-cell co-culture

Organoids were generated using published procedures ^31–34 (IRB 21–006244), see Supplementary Methods for details. Tregs from the following conditions were stimulated, counted, and pelleted: Tregs only, Tregs treated with TNFα, Tregs treated with TNFα and subsequently, vorinostat. Organoid cells and Tregs were combined at a ratio of 1:1 (10,000 epithelial cells and 10,000 Tregs per condition), replated together into a dome of Matrigel and overlaid with human colon medium containing 50 U/ml human IL2. The organoids reformed in the presence of Tregs. Media was changed every 3 days and the co-cultures were maintained in duplicates for 7 days. Whole-mount organoids were subjected to immunofluorescence analysis (see Supplementary Methods), RNA was extracted for qRT-PCR and the media was collected for cytokine measurement.

RNA isolation, cDNA synthesis, and quantitative real-time PCR (qRT-PCR)

Standard protocols were followed as per manufacturer’s recommendations. See Supplementary Methods for specific lot numbers and reagents.

Cytokine measurement

Media from organoid/T-cell co cultures were collected and stored at −20°C. Cytokine levels were determined using the Human Cytokine 71-Plex Discovery Assay (Eve Technologies).

Statistics

Statistical analyses were performed using a D’Agostino & Pearson test for normality. If passed, an unpaired t-test was performed. Otherwise, a Mann-Whitney test was performed (* p≤0.05, ** p≤0.01, *** p≤0.001). Raw count comparisons were calculated with a One-way ANOVA. To handle multiple comparisons in the scRNA-seq datasets, the MAST package (v1.22.0) and for heatmaps, the Wilcoxon Rank Sum test was used. Graphs were designed using GraphPad Prism 8.0.1 (GraphPad Software, Inc., San Diego, CA, USA), BioRender.com and R (4.0.3).

Results

scRNA-seq reveals cellular dynamics in Crohn’s disease patients

Ileal biopsies were extracted from CD patients, as well as age- and sex-matched healthy individuals. CD4⁺ cells were purified for subsequent single cell RNA-sequencing (scRNA-seq; Fig. 1A). The single cell transcriptomes of 11,520 CD4⁺ T cells and 8,686 CD4⁺ cells were isolated from healthy subjects and CD patients, respectively, and cells were assigned to distinct cell populations based on the expression of established marker genes (Fig. 1B, Extended Data Fig. 1A). Clustering analysis revealed the presence of eight cell types including Th17, Th1, regulatory T cells (Tregs), effector and central memory cells, follicular T helper cells, and small populations of CD4^low cells and cells which could not be assigned to a distinct cell type (“unknown”). Total cell numbers per population were determined in healthy and CD samples (Fig. 1C) as well as the cellular composition of each sample (Extended Data Fig. 1B, C, Suppl. Tables 3 and 4). Notably, the proportion of Tregs was elevated in CD patients compared to healthy controls (17.7% of all CD CD4⁺ cells vs. 11.1% of all control CD4⁺ cells, p<0.0001). Similarly, the relative number of Th17 (24.1% vs. 21.3%, p<0.0001) and Th1 cells (3.6 vs. 2.1%, p<0.0001) was also increased in cells isolated from diseased individuals. Furthermore, CD was associated with lower proportions of follicular T helper (21.8% vs. 27.7%, p<0.0001) and effector memory cells (4.3% vs. 8.9%, p<0.0001). Principal Component Analysis demonstrates that cellular phenotypes (symbol specific color) cluster well across the healthy control subjects (circles), and in several instances appear distinct from analogous cell phenotypes (symbol color) derived from Crohn’s patients (triangles, ie TH17 cells, Treg cells, and central memory cells) (Fig. 1D, Extended Data Fig. 2A). As a group, the central memory cell population clustered separately from other effector cell populations. To evaluate gene expression changes underlying these observations, we identified genes induced in CD compared to controls by cell type. As expected, inflammation-associated genes including cytokines and histocompatibility antigen-encoding factors were upregulated (Fig. 1E). In addition, several cell type-specific markers were induced in CD, such as FOXP3 in Tregs and SELL in memory cells. Together, these data represent the first in-depth transcriptional assessment of CD4⁺ cells driving chronic ileal inflammation.

Fig. 1: scRNA-seq reveals cellular dynamics in Crohn’s disease.

(A) Terminal ileal biopsies were extracted from age- and sex-matched CD patients (sequenced samples: female n=3, male n=2) and healthy individuals (sequenced samples: female n=3, male n=3). CD4⁺ cells were purified via flow cytometry and analyzed by scRNA-seq. (B) UMAPs displaying eight distinct cell populations identified in healthy and CD samples. Pie charts indicate the proportion of cells originating from healthy (light blue) and CD (rose) biopsies per cell type. (C) Total numbers per cell type within the healthy (11,520 cells) and CD (8,686 cells) populations and their relative abundance (%) per group (Fischer’s exact test *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001). (D) PCA plot displaying the distinct expression patterns in healthy (circle) and CD (triangle) cells. Unknown and CD4^low populations were not included in this analysis. (E) Top 10 upregulated (red) genes per cell type. Log₂FC >1.0, padj <0.05.

Identification of unique Treg subpopulations in CD

Given the increase in Treg numbers in CD patients and based on our recent findings on the BMI1-dependent modulation of Treg identity in IBD⁸, we analyzed cells expressing the Treg marker gene FOXP3 in more detail. We identified all FOXP3-expressing cells which, as expected, were primarily found in our previously defined Treg cluster (Fig. 2A). To confirm the identity of Tregs, their expression pattern was compared to Tregs in publicly available datasets, showing significant overlap (Extended Data Fig. 3). The re-clustering of FOXP3⁺ cells now demonstrated that cells isolated from healthy individuals were clearly distinct from those of CD patients (Fig. 2B). In agreement with our previous observation, within the entire Treg population more cells originated from CD patients (66.4%, 1,044 cells) than from controls (33.6%, 529 cells). Based on their gene expression profile, Tregs were clustered into five distinct subpopulations (Fig. 2C). While healthy Tregs were assigned primarily to cluster 1 (healthy: 351 cells (66.3% of all healthy cells), CD: 25 cells (2.4% of all CD cells)), cluster 2 was highly heterogeneous (healthy: 110 cells (20.8%), CD: 313 cells (30.0%)) and cluster 3 contained predominantly CD Tregs (healthy: 21 cells (4.0%), CD: 617 cells (59.1%)). In general, cluster 4 and 5 contained low cell numbers and while proportions in cluster 4 were similar (healthy: 46 cells (8.7 %), CD: 36 cells (3.4%)), cluster 5 almost exclusively consisted of CD Tregs (healthy: 1 cell (0.2 %), CD: 53 cells (5.1%)). While Treg clusters 1–4 were highly heterogeneous and populated by cells of both sexes, cluster 5 was dominated by one specific CD patient (Extended Data Fig. 2B,C). To identify unique transcriptomic features, the most differentially expressed genes were determined (Fig. 2D, E; Extended Data Fig. 4). The CD-associated cluster 3 displayed an induction of inflammation-associated genes (TNFRSF4, TNFRSF18, CTLA4, TYMP). Furthermore, several Heat Shock Protein (HSP)-encoding genes were found to be differentially expressed. Intriguingly, the inducible variant HSP90AA1 was upregulated in the healthy cluster 1 while the constitutive form HSP90AB1 was equally expressed in both conditions. To assess potential sex-associated differences, we analyzed cells isolated from female and male individuals separately and detected only minor differences (Extended Data Fig. 5). In summary, our analyses reveal the presence of distinct Treg subpopulations in CD patients bearing transcriptional markers of active inflammatory pathways.

Fig. 2: Identification of unique Treg subpopulations in Crohn’s Disease.

(A) All FOXP3-expressing cells (teal; 1,573 cells) were identified and (B) re-clustered. 33.6% (529 cells) of these cells were detected in the healthy and 66.4% (1,044 cells) in the CD population. (C) Based on their expression patterns, FOXP3-expressing cells were divided into five distinct subpopulations following UMAP calculation. (D) The 25 most differentially expressed genes were determined within the five Treg subpopulations. (E) UMAPs displaying the minimum (gray) and maximum (red) expression of these differentially expressed genes within all Tregs.

Pseudotemporal progression of healthy Tregs into distinct Treg subpopulations in CD

Conceptualizing that cells may undergo transcriptome-wide changes during disease development and progression, we investigated the relative gene expression changes of the Treg population along this presumptive biological process using pseudotime. This computational analysis tool infers the ordering of cells along a lineage progression derived from their expression profile related to time¹⁴. In agreement with the hypothesis that Tregs display transcriptomic changes during chronic intestinal inflammation, healthy Tregs in cluster 1 as well as the mixed population in cluster 2 were predicted to represent the origin of pseudotemporal development (Fig. 3A, B, Extended Data Fig. 6A, B, Extended Data Fig. 7). Cells were predicted to undergo gene expression changes represented by cells detected in cluster 2 (heterogeneous population) before transitioning to one of two fates: either cluster 4 (heterogeneous) or cluster 3 (predominantly CD Tregs). A small subpopulation of cells (cluster 5; CD) was found to diverge from cluster 3. The expression patterns of significantly regulated genes during pseudotemporal development revealed the induction of various cytokines, cytokine receptors and NF-κB members (Fig. 3C). Subsequently, the expression patterns of a subset of significantly differentially regulated genes were visualized revealing the transcriptomic differences in distinct subpopulations (Fig. 3D). These findings appear to be independent of the sex of healthy individuals and CD patients (Extended Data Fig. 8, 9).

Fig. 3: Developmental trajectories of healthy Tregs into distinct Treg subpopulations in CD.

(A) Based on gene expression patterns, the developmental dynamics of Tregs was evaluated using trajectory interference¹⁴, indicating early (purple) and late (yellow) fates. Initial state: red arrowhead. (B) Pseudotemporal development of Tregs considering five distinct subpopulations. (C) Heatmap depicting genes differentially expressed during pseudotemporal progression (q-val <0.1; blue=low expression, red=high expression). (D) A subset of highly regulated DEGs was depicted in independent graphs.

Tregs isolated from CD patients display a unique responsiveness to TNFα signaling

To identify signaling networks associated with the gene expression changes observed in disease-associated Tregs, we performed pathway enrichment analysis. Interestingly, this analysis suggested that gene expression profiles of Tregs isolated from CD patients display a high overlap with transcriptomic changes associated with TNFα signaling via Nuclear Factor kappa-lightchain-enhancer of activated B cells (NF-κB; Fig. 4A). This finding was further supported by CellChat analysis, a program in which the model output predicts cell-cell communication patterns based on the expression of receptors and ligands¹⁵. Tregs were the only cell population within our dataset predicted to be strong receivers of the TNFα signaling network (Fig. 4B–D). As a complementary strategy, we analyzed publicly available microarray profiling datasets in which human Tregs were treated with TNFα for 2 or 24 hours in vitro and compared to untreated control cells¹⁶ (GSE18893). Significant overlap was observed when comparing differentially expressed genes (DEGs) from our dataset to the microarray results (Fig. 4E). We then identified TReg-specific genes Induced in Crohn’s (TRICs) by overlapping (I) genes upregulated in Tregs compared to all other cell populations with (II) genes upregulated in CD Tregs compared to healthy Tregs (Fig. 4F). A list of ten genes was obtained: BATF, GADD45A, DUSP4, IL2RA, ARID5B, CTLA4, FOXP3, S100A4, TNFRSF18, TNFRSF4. To observe the expression changes of those genes in vitro, we isolated CD45RA⁺ naïve T cells from healthy donors and differentiated them into Tregs (Fig. 4G). Upon differentiation, Tregs were treated with TNFα and other combinatorial experimental cytokine conditions. Notably, the induction of the majority of TRICs was observed upon individual or combination treatments with TNFα, IL12 and IL21 (Fig. 4H). These findings validate responsiveness of genes differentially expressed in Tregs isolated from CD patients to TNFα and support the bioinformatic interpretation that Tregs are strong receivers of the TNFα signaling pathway in CD.

Fig. 4: Tregs isolated from CD patients display a unique responsiveness to TNFα signaling.

(A) Genes upregulated in Tregs isolated from CD patients were compared to the Human Molecular Signatures Database (MSigDB). This analysis uncovered highly significant enrichment of TNFα signaling via the NF-κB pathway (MSigDB_Hallmark_2020). (B) CellChat revealed that Tregs display a particularly high incoming rather than outgoing interaction strength for the TNF signaling pathway network. (C) This signaling towards Tregs is mainly mediated by Th17 (red), Th1 (blue) and effector memory cells (orange). (D) Role of the respective CD4⁺ cell types as senders, receivers, mediators, and influencers in TNFα signaling (color intensity corresponds to interaction strength; light orange: low involvement, dark brown: high involvement). (E) Genes upregulated in FOXP3⁺ T cells isolated from CD lesions were compared to genes induced in a publicly available microarray dataset of human Tregs treated with TNFα in vitro and compared to untreated control cells¹⁶ (GSE18893). A significant overlap was observed between both datasets (Fisher’s exact test, p=0.0095; Odds ratio: 3.65). (F) Ten Treg-specific genes induced in Crohn’s (TRICs) were defined by overlapping genes upregulated in Tregs compared to all other cell populations with genes upregulated in CD Tregs. (G) To model the induction of TRICs in vitro, we isolated CD45RA⁺ naïve T cells from healthy donors and differentiated them into FOXP3⁺ Tregs. These cells were treated with designated cytokines for 48 h. (H) In three independent experiments (each n=3), a substantial induction of TRICs was confirmed after TNFα treatment using qRT-PCR.

Pro-inflammatory FOXP3-positive T cells are highly HDAC-dependent and can be targeted via vorinostat treatment

After modeling the induction of TRIC expression patterns in vitro, we evaluated our scRNA-seq data to identify a therapeutic strategy to reduce the expression of this gene set. For this purpose, we utilized A Single-cell Guided pipeline to Aid Repurposing of Drugs (ASGARD), a tool identifying candidate drugs from a database containing 21,304 FDA-approved compounds¹⁷. Hereby, drugs were selected which were predicted to significantly reverse gene expression patterns from CD Tregs compared to control Tregs. This tool identified the histone deacetylase (HDAC) inhibitor (HDACi) vorinostat and the mammalian Target of Rapamycin (mTOR) inhibitor (mTORi) temsirolimus as putative therapeutics (Fig. 5A). In fact, vorinostat and temsirolimus were predicted to affect gene expression patterns of the TNFα and NF-κB signaling pathways, two networks upregulated in CD Tregs (Fig. 5B, Extended Data Fig. 10A). As proof of concept, we compared genes upregulated in Tregs isolated from CD patients to expression profiles from Tregs isolated from mice with a Treg-specific Hdac1 or Hdac2 deletion, respectively (GSE139480). Our rationale was that if our CD Treg DEG gene set was largely dependent upon HDAC activity, we should find enrichment within our gene set of Treg-specific, HDAC-dependent genes. Thus, we generated gene sets containing all genes which were downregulated in Tregs isolated from Foxp3-Cre, Hdac1^fl/fl or Hdac2^fl/fl mice and performed GSEA using our dataset. Indeed, HDAC-dependent genes were significantly enriched in CD Tregs (Fig. 5C, D), including TRICs such as Ctla4, Tnfrsf4 and S100a4. Based on the results of our in vitro cytokine screen, we focused our subsequent therapeutic approach on cytokines which induced a reliable upregulation of TRICs. Upon induction of inflammation-associated gene expression patterns via TNFα, IL-12 and IL-21, cells were treated with vorinostat, (Fig. 5E), temsirolimus (Extended Data Fig. 10B) or DMSO (negative control). As expected, TRIC levels were induced after cytokine treatment, especially after the addition of TNFα alone. Importantly, while temsirolimus treatment only affected a subset of TRICs, vorinostat drastically diminished gene expression. In summary, we were able to demonstrate that unique Treg clusters diverge from healthy Tregs during chronic ileal inflammation and that these proinflammatory gene pathways are highly HDAC-dependent and can be targeted by vorinostat treatment.

Fig. 5: CD Tregs can be targeted via vorinostat treatment.

(A) Using the ASGARD package, the most promising drugs to reverse expression patterns of CD Tregs were determined. (B) As predicted by ASGARD and displayed using Cytoscape, vorinostat treatment impacts multiple inflammation-associated proteins and pathways, including chemokine, NF-κB and TNFα signaling. (C) Publicly available gene expression profiles from Tregs isolated from mice with a Treg-specific Hdac1 or (D) Hdac2 deletion (GSE139480), were downloaded to perform GSEA using FOXP3⁺ cells from our own dataset. HDAC-dependent genes were significantly enriched in CD Tregs. (E) Human PBMCs were differentiated into Tregs in vitro and different cytokines were added for 24 h to induce the expression of TRICs (U: untreated; T: TNFα; 12,21: IL12 and IL21; T12,21: TNFα, IL12 and IL21). Afterwards, cells were treated with vorinostat for 24h. qRT-PCR analysis revealed that the upregulation of TRICs upon cytokine treatment was rescued by the addition of vorinostat. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. Mann-Whitney test.

The functional capacity of inflammatory FOXP3⁺ T cells is rescued by vorinostat treatment in vitro

As a functional readout to assess the suppressive activity of Tregs after exposure to TNFα and/or Vorinostat, an in vitro suppression assay was performed. Compared to untreated control cells, the ability of Tregs to inhibit the proliferation of CD4⁺CD25⁻⁻ T responder cells was reduced in TNFα-treated Tregs (54.4% vs 86.6%, padj=0.0004). Notably, additional treatment with vorinostat restored the suppressive capacity of TNFα-treated Tregs (95.4% vs 54.4%, padj<0.0001, Fig. 6A). Next, to determine their effect on intestinal epithelial cells, Tregs were co-cultured with human colon organoids in vitro. Upon maintenance for seven days, immunofluorescent staining for ZO-1 and Ki67 revealed that tight junctions and proliferative capacity were lost in organoids in the presence of TNFα-treated Tregs (Fig 6B, Extended Data Fig 11). Notably, the presence of those markers was preserved when these Tregs were additionally treated with vorinostat. Similarly, the gene expression patterns of MKI67, MLKL2, RIPK3 and LGR5 were rescued upon the addition of vorinostat (Fig. 6C). Finally, the supernatants of these co-cultures were analyzed for the presence of cytokines. In agreement with our previous findings, exposing FOXP3⁺ T cells to TNFα resulted in a broad deregulation while the additional treatment with vorinostat was able to normalize cytokine levels. Utilizing standard (in vitro suppression assay) and novel (organoid co-culture system) functional read out methodology, we have demonstrated for the first time a pathophysiologic role for TNFα exposed Tregs, putative responsible gene pathways, and promising therapeutic approaches.

Fig. 6: Vorinostat rescues the suppressive function of inflammatory FOXP3⁺ T cells in vitro.

(A) To perform in vitro suppression assays, CD25⁻⁻ cells were isolated from the same patients as CD4⁺ naïve T cells and frozen until CD4⁺ naïve T cells were successfully differentiated to Tregs and treated with TNFα and/or vorinostat. Tregs were washed and cultured in 4 different dilutions (1:1, 1:2, 1:4 and 1:8) with 5,000 CD25⁻ responder cells. Assessment of proliferation revealed that the suppressive capacity of TNFα-treated Tregs was reduced but could be rescued by HDACi in four independent experiments (each n=3, One-way ANOVA, multiple testing, Tukey, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001). (B) Human colon epithelial organoids were co-cultured with Tregs. Whole-mount organoids were subjected to immunofluorescent staining for Ki67 (red), ZO-1 (green) and DAPI (blue) in duplicates in two independent experiments. Scale bar: 10μm. (C) Gene expression of MKI67, TJP1, MLKL2, RIPK3 and LGR5 in co-cultured cells was evaluated using qRT-PCR. (D) Cytokine levels in supernatants of co-cultured cells was determined using a Human Cytokine 71-Plex Discovery Assay (Eve Technologies). *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. Mann-Whitney test.

Discussion

Despite the generation and optimization of various therapeutic regimens in IBD, more than half of patients do not achieve complete remission and therapeutic failure remains common¹⁸. While the etiology of IBD is still not completely understood, the pathogenic heterogeneity is expected to underlie this group of diseases^19,20. Analysis of state-of-the-art scRNA-seq data allows for the identification of cell populations and distinct subtypes contributing to this heterogeneity. In this study we detected a subpopulation of proinflammatory FOXP3⁺ T cells in CD patients.

In the past, IBD-related scRNA-seq studies focused on the broad characterization of transcriptome-wide changes underlying these diseases and the identification of cell populations driving the pathogenesis of IBD. Due to the unequal distribution of cell types within the analyzed tissues, the total numbers of distinct populations were insufficient to allow in-depth analyses. To overcome this issue and given the known relevance of CD4⁺ cells in intestinal inflammation, we specifically purified CD4⁺ cells in human biopsies. One limitation of our work with FACS-based purification of CD4⁺ cells is the possible capturing of transient CD4-expressing cells. In fact, when assigning cell populations in our dataset, we detected the presence of CD4⁺ cells (“unknown”) that displayed highly distinct expression patterns and did not cluster with other cell types. In addition, cells with low CD4 expression were detected. The identity of those cells could not be determined due to a lack of the expression of distinct marker genes. Similar to CD4, other genes that were used to classify cell populations in our study could have been transiently upregulated by other cell types. In fact, the selection of genes varies among studies and can significantly affect the results of scRNA-seq studies. Previous IBD-related scRNA-seq studies utilized normal adjacent tissue (NAT), uninvolved tissue in close proximity to the inflamed lesion, as controls. As outlined in other disease contexts, NAT frequently displays a unique intermediate state between healthy and diseased²¹. Therefore, in this study, ileal specimens from healthy individuals and CD patients were compared in order to understand cellular as well as transcriptional differences. Indeed, as revealed by PCA, the respective cell populations displayed only low variation when comparing gene expression signatures between healthy and CD samples. This finding clearly suggests that healthy individuals are suitable controls, and avoids the risk of utilizing an intermediate state between normal and inflamed ileal tissue. While surgical resection specimens can be obtained in IBD patients, only small biopsies are isolated from healthy individuals undergoing routine endoscopic examinations. Here, we aimed to avoid the induction of potential artifact caused by differences in sample isolation and handling in CD patients and controls. Therefore, we collected biopsies along the mucosa of the resection specimens from CD patients using biopsy forceps, aiming to minimize processing effects.

Our gene expression data suggested that CD FOXP3⁺ T cells diverge from normal Tregs and are strong receivers of TNFα signaling. Subsequently, we generated an in vitro model which displayed reduced suppressive capacity and showed a decrease in the tight junction marker ZO-1 in co-cultures with human colon organoids. Tight junction dysfunction was associated with the disruption of intestinal barrier integrity and believed to contribute to the pathogenesis of IBD^22,23. Indeed, ZO-1 has been demonstrated to be essential for mucosal repair in experimental models for IBD but appears to be dispensable for epithelial barrier function. Therefore, it is assumed that reduced ZO-1 levels result in impaired mucosal healing in inflamed intestinal areas²⁴. Here, we demonstrated that the HDAC inhibitor (HDACi), vorinostat, rescued the loss of ZO-1 expression in epithelial cells and the supressive dysfunction associated with inflammatory FOXP3⁺ T cells.

More than a decade ago, researchers observed a beneficial effect of HDAC inhibitors in experimental models of colitis²⁵. Vorinostat (SAHA), an hydroxamic acid, is an FDA-approved pan-HDACi licensed for the treatment of patients with refractory cutaneous T-cell lymphoma (CTCL). Vorinostat was recently described to attenuate dextran sodium sulfate-induced (DSS) colitis in mice by suppressing local secretion of pro-inflammatory cytokines and chemokines and by preventing the accumulation of inflammatory cells²⁶. While there are clinical trials evaluating the benefit of this inhibitor in CD patients (https://clinicaltrials.gov/ct2/show/NCT03167437), vorinostat has been associated with severe adverse effects in phase I and II studies. In fact, low tolerability was demonstrated in colorectal, breast and thyroid cancer patients, and in individuals with CTCL, diffuse large B-cell lymphoma or non-small-cell lung cancer²⁷. Therefore, to increase the clinical applicability of our finding, we aim to test the effect of more specific, and potentially more tolerable, HDACi in future studies. In fact, our results revealed a high overlap of genes upregulated in Tregs isolated from CD patients and mice with Treg-specific deletions of Hdac1 or Hdac2, respectively. While the potential of vorinostat to reverse expression patterns of pro-inflammatory Tregs was predicted using primary CD patient material, this in silico finding was verified using induced Treg cells isolated from PBMCs of healthy donors. The translation of this finding to primary intestinal Treg cells or natural thymic derived Treg cells is, as of yet, uncertain and requires further experimentation.

Together, we describe a novel, pro-inflammatory FOXP3⁺ T cell subpopulation in CD patients which is responsive to TNFα signaling. In addition, we generated an in vitro model which displayed reduced suppressive capacity and reduced the levels of the tight junction marker ZO-1 which was described to be critical for mucosal healing. Additional, and larger studies of intestinal FOXP3+ cells should allow future correlations with patient demographics (ie sex, age, etc) and clinical characteristics (ie medication). Based on primary patient material, we developed a pipeline to specifically target inflammatory FOXP3⁺ T cells in CD patients using the FDA-approved drug, vorinostat.

Extended Data

Extended Data Fig. 1: Distribution of marker genes and sample composition.

(A) Heatmap depicting the top 10 marker genes across all distinct eight clusters within human CD4⁺ cells. (B) Proportions of cell types per healthy individual or CD patient and (C) proportions of the respective samples contributing to the total number of each cell type.

Extended Data Fig. 2: Sample clustering and composition of Treg subtypes.

(A) PCA plot depicting the sample distribution of healthy controls (circle) and CD patients (triangle). (B) Tregs belonging to clusters 1–4 were represented by both sexes in all tissue samples. However, cells from cluster 5 were predominantly present in sample GEX_RST10175. (C) UMAP of FOXP3-positive cells indicating patient ID.

Extended Data Fig. 3: Comparison between cells classified as Tregs in our study and in a colorectal cancer study by Lee et al.

(A) DotPlot of all marker genes for each cluster defined by Lee et al. and our study (PMID: 32451460). (B) UMAPs indicating the successful harmonization of our data with the dataset generated by Lee and colleagues (C) The three key markers for Tregs (CD4, IL2RA, FOXP3) were plotted on the harmonized UMAP together and (D) separately. (E) The visualization of cells positive for all three Treg markers demonstrates the high overlap between Tregs in both studies.

Extended Data Fig. 4:

The top 10 up- and downregulated genes per Treg subpopulation.

Extended Data Fig. 5: Most differentially expressed genes among Treg subtypes separated by sex.

(A) UMAPs highlighting the most differentially expressed genes in cells isolated from females and the corresponding dotplot. UMAPs display the minimum (gray) and maximum (red) expression of these differentially expressed genes within all Tregs. (B) UMAPs and dot plot of those genes in cells isolated from males. Overall, we detected only marginal differences when comparing both sexes.

Extended Data Fig. 6: Pseudotemporal development of Tregs determined using monocle2.

(A) Using moncle2, four branch points and a bifurcating developmental trajectory were calculated, with (B) cluster 0 depicting the earliest (start) cluster.

Extended Data Fig. 7: Trajectory interference determined using RNA-velocity.

RNA-velocity estimation as basis of trajectory interference. While the trajectory in cluster 1 is heterogeneous, a continuous transition from cluster 2 to cluster 3 and 4 was visualized in an unbiased manner using scVelo and the algorithms published by La Manno et al. (PMID: 30089906).

Extended Data Fig. 8: Developmental trajectories of healthy Tregs into distinct Treg subpopulations in female CD patients.

(A) Trajectory interference indicating early (purple) and late (yellow) fates. (B) Pseudotemporal development of five distinct Treg subpopulations. (C) Heatmap depicting genes differentially expressed during pseudotemporal progression (q-val <0.1; blue: low expression, red: high expression). (D) The expression of highly regulated DEGs was depicted in independent graphs.

Extended Data Fig. 9: Developmental trajectories of healthy Tregs into distinct Treg subpopulations in male CD patients.

(A) Early (purple) and late (yellow) fates of Tregs isolated from male individuals. (B) Pseudotemporal development of five distinct Treg subpopulations. (C) DEGs displayed in a heatmap (q-val <0.1; blue: low expression, red: high expression) and in independent graphs.

Extended Data Fig. 10: Temsirolimus treatment only partially rescues the proinflammatory effect of TNFα-treated FOXP3⁺ cells in vitro.

(A) Multiple inflammation-associated factors and pathways were predicted to be affected by Temsirolimus treatment as suggested by ASGARD and displayed using Cytoscape. (B) Human PBMCs were differentiated into Tregs in vitro and cytokines (U: untreated; T: TNFα; 12,21: IL12 and IL21; T12,21: TNFα, IL12 and IL21) were added for 24 h to induce the expression of TRICs. Upon the treatment with Temsirolimus, qRT-PCR analysis revealed only a partial rescue. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. Mann-Whitney test.

Extended Data Fig 11: Co-culture of normal colon organoids with T-regs.

Human colon epithelial organoids were co-cultured alone (A), with Tregs (B), and with Tregs stimulated with TNFa (C). Whole-mount organoids were subjected to immunofluorescent staining for CD45 (red), ZO1 (green) and DAPI (blue).

Background and Context:

Crohn’s disease is a highly heterogenous disease; yet molecular insights at the level of the single cell may provide new insights into disease pathogenesis, behavior, and treatment.

New Findings:

Profiling of isolated CD4+ cells allows deep transcriptional characterization of unique cell types associated with Crohn’s disease, in particular a FOXP3+ lymphocyte with a pro-inflammatory TNFα responsive gene signature.

Limitations:

Modeling of presumed in vivo cellular function using ex vivo systems may not entirely recapitulate the physiologic state.

Clinical Research Relevance:

The important role of TNFα signaling in Crohn’s disease may be extended to T regulatory cell biology.

Basic Research Relevance:

The mechanisms of TNFα pathway impinging on T regulatory cell function are likely relevant in mucosal inflammatory disorders.