Identification of a unique subset of tissue-resident memory CD4⁺ T cells in Crohn’s disease

Significance

The impact of tissue-resident memory T cells (Trm) on the pathogenesis of inflammatory bowel disease remains controversial due to their heterogeneity. In particular, the pathological function of CD4⁺Trm, as opposed to their CD8⁺counterparts, is largely unknown. Here, using a comprehensive analytical approach, we found that CD103⁺CD4⁺Trm with an inflammatory phenotype were increased in the gut of Crohn’s disease (CD) patients but not in ulcerative colitis patients. Further profiling revealed that a subset of CD103⁺CD4⁺Trm, expressing CD161 and CCR5, were specific to CD patients and that this subset exerted cytotoxic activity. CD103⁺CD161⁺CCR5⁺CD4⁺Trm were predominant producers of proinflammatory cytokines, highlighting the importance of this specific subset in the pathogenesis of CD.

Abstract

T cells differentiate into highly diverse subsets and display plasticity depending on the environment. Although lymphocytes are key mediators of inflammation, functional specialization of T cells in inflammatory bowel disease (IBD) has not been effectively described. Here, we performed deep profiling of T cells in the intestinal mucosa of IBD and identified a CD4⁺ tissue-resident memory T cell (Trm) subset that is increased in Crohn’s disease (CD) showing unique inflammatory properties. Functionally and transcriptionally distinct CD4⁺ Trm subsets are observed in the inflamed gut mucosa, among which a CD-specific CD4⁺ Trm subset, expressing CD161 and CCR5 along with CD103, displays previously unrecognized pleiotropic signatures of innate and effector activities. These inflammatory features are further enhanced by their spatial proximity to gut epithelial cells. Furthermore, the CD-specific CD4⁺ Trm subset is the most predominant producer of type 1 inflammatory cytokines upon various stimulations among all CD4⁺ T cells, suggesting that the accumulation of this T cell subset is a pathological hallmark of CD. Our results provide comprehensive insights into the pathogenesis of IBD, paving the way for decoding of the molecular mechanisms underlying this disease.

Inflammatory bowel disease (IBD), which consists of Crohn’s disease (CD) and ulcerative colitis (UC), involves chronic inflammation of the digestive tract. Accumulating evidence indicates the heterogeneity of IBD, exhibiting different genetic and environmental backgrounds, which complicates our efforts to understand the disease. Genome-wide association studies (GWAS) have identified a substantial number of IBD risk loci encoding a diverse array of immune system components, many of which are shared across both subtypes of IBD and other immune-mediated diseases (1, 2). In this regard, major triggers by which patients are skewed toward either subtype of IBD lie beyond the genetic predisposition.

The gastrointestinal tract contains a vast number and a diverse array of immune cells, which together form the largest immune system in the body. Inflammatory pathways in the gut mucosa are finely controlled by the stringent mechanisms under homeostatic conditions. In the gut mucosa of IBD, excessively activated innate immune cells present antigens and secrete inflammatory cytokines, resulting in the aberrant activation of adaptive immune cells, which in turn leads to the further infiltration of inflammatory immune cells into inflamed sites and exacerbates tissue damage (3, 4). Regardless of the genetic and environmental variances of the patient cohort, dysregulated immune responses may finally converge on the common T cell–mediated inflammatory cascades, which are the direct drivers of IBD.

Despite a plethora of studies emphasizing the involvement of T cells in the pathogenesis of IBD (5, 6) and extensive efforts to develop T cell–directed strategies (7), no disease-specific T cell subset that plays an essential role in the progression of IBD has been identified. The classically proposed theory of T helper1 (Th1) and Th2 imbalance in IBD is based on the cytokine context. Given the lack of successful anticytokine therapies other than anti-Tumor necrosis factor (TNF) and anti-Interleukin (IL-12/23) antibodies (4, 8–10), detailed profiling of immune cells that goes beyond cytokines is required for future therapeutic approaches. T cells are highly plastic, which may arise from a flexible genetic program in response to local environmental cues or heterogeneity of the cells (11, 12). It is thus important to resolve how immune cell populations display pronounced diversity in the pathological settings of IBD and which of these cells contribute to pathogenesis. To address these issues, significant recent technological advances could be used to achieve breakthroughs and advance our understanding of complex biological systems. Recently, the single-cell atlas of immune cells in IBD was reported (13–15); however, to the best of our knowledge, no comprehensive study has identified CD4⁺ T cells differentially involved in CD and UC, and therefore, their biological properties have not been characterized.

In contrast to circulating T cells, which patrol the entire body via lymphatic and blood flow, tissue-resident T cells (Trm) remain within nonlymphoid barrier tissues and serve as the frontline of host defense, which allows for the efficient elimination of invasive pathogens (16). The contribution of Trm to the pathogenesis of IBD has been controversial, with conflicting results often being obtained in human and animal studies (17–21). This controversy is due not only to differences in patients’ backgrounds but also to heterogeneity of Trm. In particular, the functional properties of CD4⁺ Trm in pathological settings have remained largely unknown, as opposed to CD8⁺ Trm, which have been extensively studied. Furthermore, despite the emerging importance of Trm in mouse models, it remains unclear whether animal models recapitulate the human immune responses (22). There is thus a need to analyze the functional diversity of CD4⁺ Trm in humans and their impacts on the pathophysiology of IBD.

Here, we set out to identify T cell subsets and biological networks specifically wired in the gut in each subtype of IBD. By analyzing human gut samples, we identified the T cell subpopulations abundant in the gut in CD and UC that are differentially controlled by key transcriptional programs. Most importantly, we found that a pathogenic CD4⁺ Trm subset with diverse properties localized adjacent to the epithelia was specifically enriched in CD and orchestrated the local inflammatory response.

Results

CD and UC Are Characterized by Distinct T Cell Subsets.

To explore the specific molecular properties of each subtype of IBD, we applied mass cytometry (cytometry by time of flight: CyTOF) panels, which allow us to capture various types of T cell subsets (SI Appendix, Table S1). Peripheral blood mononuclear cells (PBMC) of healthy individuals and lamina propria mononuclear cells (LPMC) of the normal intestinal mucosa revealed contrasting features, highlighting the importance of profiling immune cells localized to the lesions (SI Appendix, fig. S1 A–C). We profiled the cryopreserved intestinal LPMC isolated from the inflamed mucosa of patients with CD, UC, and unaffected mucosal regions of colon cancer patients as a control (SI Appendix, Table S2). Patients were subdivided into two cohort groups, with validation of the data from the first small cohort (n = 8, each group) being achieved via the second large cohort (n = 18, each group). The first cohort was randomly divided into two batches and analyzed independently to minimize the batch effect. Two-dimensional visualization of the mass cytometry data using a clustering algorithm phenograph (23) revealed T cells with distinct expression patterns of cell surface markers (fig. 1 A and B and SI Appendix, fig. S2 A and B and Table S3). Hierarchical clustering of mean marker expression revealed 19 and 16 T cell subclusters in the first and second batches, respectively. CD3⁺ T cells of batch 1 comprised 12 CD4⁺ and 4 CD8⁺ T cell subclusters, with a few smaller unannotated clusters, whereas batch 2 comprised 11 CD4⁺ and 5 CD8⁺ T cell subclusters. Overall, CD and UC were clearly distinguished from each other by the marked changes in the specific clusters. CD patients were uniquely characterized by the massive expansion of CD4⁺ Trm (fig. 1 A and C and SI Appendix, fig. S2A), which were defined by the high expression of the resident marker CD103 and the low expression of the effector memory markers CCR7 and CD27 (fig. 1D and SI Appendix, fig. S2C). In general, CD69 is also used as a marker of tissue retention (22, 24). In the gut, however, CD69 is widely expressed throughout the T cells with no cluster specificity, and 96% of CD103-expressing cells express CD69 (SI Appendix, fig. S2 D and E). Thus, tissue residency in the gut was defined solely by the expression of CD103. Heat maps of the mean expression levels of T cell markers revealed that a particular fraction of CD4⁺ Trm abundant in the gut of CD exhibited high CD161 and CCR5 compared with other T cell fractions (fig. 1D and SI Appendix, fig. S2C). The proportion of CD45RA⁻ CCR7⁻ CD4⁺ effector memory T cells (Tem) was significantly higher in control than in IBD (fig. 1 A and C and SI Appendix, fig. S2A). In contrast, UC samples were significantly enriched in CD45RA⁺ CCR7⁺-naive CD4⁺ T cells and CXCR5⁺ CD4⁺ T follicular helper cells (Tfh) (fig. 1 A, C, and D and SI Appendix, fig. S2 A and C), which is consistent with a previous report showing that Tfh increased in the gut in patients with UC (25). CD25^high CD127⁻ regulatory T cells (Treg) were elevated in both subtypes of IBD (SI Appendix, fig. S2F). Meanwhile, the proportion of CD45RA⁻ CCR7⁺ CD4⁺ central memory T cells (Tcm) was comparable between the groups. In contrast to the marked changes in the CD4⁺ T cell composition, there was no pronounced change of CD8⁺ T cells in IBD gut: The proportions of naive CD8⁺ T cells, CD8⁺ Tem, and γδT cells among CD3⁺ T cells were comparable between the groups, while only a significant increase in CD8⁺ Trm in CD compared with that in UC, but not with that in control, was observed (SI Appendix, fig. S2F). This is partly in line with a previously reported single-cell atlas on UC gut that showed a decrease in CD8⁺ Trm in UC (13). In contrast, the expansion of CD8⁺ Trm in the UC gut has been shown (25). This difference may reflect the heterogeneity of CD8⁺ Trm detected in each study. Because the involvement of CD4⁺ Trm on CD pathogenesis has been little studied, we further characterized CD4⁺ Trm.

Fig. 1.

Proteomic profiles of intestinal lymphocytes in IBD revealed by mass cytometry. (A) Phenograph-based visualization on the tSNE plot of CD3⁺ T cells from frozen LPMC of the first batch of the first cohort of patients. Annotation of each subcluster on merged tSNE. CD4⁺ and CD8⁺ T cells are circled by red and blue dashed lines, respectively, and Trm by solid lines in their respective colors. Each sample (n = 4 per group) was concatenated together. (B) Heat map of mean expression values of T cell markers. (C) Cell type proportions of CD3⁺ T cells in the first cohort of patients (n = 8 per group, batches 1 and 2). *P < 0.05, **P < 0.01, ****P < 0.0001, mean ± SEM, one-way ANOVA (Tukey’s multiple comparison test). (D) Expression level of each cell surface marker associated with CD- and UC-predominant CD4⁺ T cell subsets was overlaid on the tSNE plot. Representative markers high in CD- and UC-predominant T cell subsets are shown in Upper and Lower rows, respectively. (E) Proportions of CD4⁺ Trm, CDpop, and the rest of CD4⁺ Trm (non-CDpop Trm) among CD4⁺ T cells in each sample. *P < 0.05, **P < 0.01, ***P < 0.001, n = 18 per group, one-way ANOVA (Tukey’s multiple comparison test). (F) UMAP of the second cohort of patients (n = 18 per group). tSNE overlay of selected markers is shown. (G) Proportion of IFN-γ⁺ and GZMA⁺ cells in CD4⁺ Trm and CDpop derived from CD patients after stimulation with PMA/ionomycin for 2 h. **P < 0.01, n = 6 per group, mean ± SEM, two-tailed paired Student’s t test. (H) Serum CRP levels of CDpop^high (n = 11) and CDpop^low (n = 15) patients before surgery. *P < 0.05, mean ± SEM, two-tailed paired Student’s t test.

Analysis of the larger second cohort revealed that CD4⁺ Trm were significantly increased in CD, confirming the first cohort data (fig. 1E). We next examined whether we could deduce the cellular markers that would enable us to enrich the potentially detrimental lymphoid fraction. The results of the mean value of each marker in cohort 1 (fig. 1B) were replicated by the expression marker analysis of cohort 2: CD4⁺ Trm showed higher expression of CCR5, CD161, CD127, CD69, and CCR6 accompanied by lower expression of CCR7 and CD27 when compared with the other CD4⁺ T cell subsets (SI Appendix, fig. S3A). Furthermore, CD4⁺ Trm–expressing CCR5 or CD161 produced significantly higher Interferon (IFN)-γ levels upon phorbol 12-myristate 13-acetate (PMA)/ionomycin stimulation than their counterparts that did not express these markers. Meanwhile, no differences were observed between CD127⁺ Trm or CCR6⁺ Trm and their respective negative counterparts (SI Appendix, fig. S3B). On the basis of these results, we gated the CD161⁺ CCR5⁺ CD4⁺ Trm, which we named “CDpop” (CD–enriched population) (fig. 1F). Indeed, CDpop, but not non-CDpop Trm (CD161⁻ or CCR5⁻ CD4⁺ Trm), was more abundant in CD than in UC and control (fig. 1E). IFN-γ and granzyme A (GZMA) secretion induced by PMA/ionomycin was higher in CDpop than in CD4⁺ Trm, confirming that CDpop is potentially a more activated cytotoxic subset among CD4⁺ Trm (fig. 1G). According to a previous report, the number of CD103-expressing T cells shows a gradual decline along the intestine (17). Because CD, but not UC and colon cancer, frequently affects the proximal colon rather than the distal colon, it was inevitable that our samples would be unevenly distributed along the entire colon between the groups. Nevertheless, the abundance of CDpop was significantly higher in CD than in UC and control, even when confined to the distal colon (SI Appendix, fig. S3C), ruling out the location having an effect on our results.

Substantial expansion of CDpop was found in a subset of CD patients (fig. 1E), and therefore, we stratified patients by the frequency of CDpop, with a cutoff value of 4% of CD4⁺ T cells, which is the overall mean in frozen samples, and investigated whether it correlated with their clinical history. The medical records of CD patients showed that serum CRP levels before surgery were significantly higher in the CDpop^high than in the CDpop^low group (fig. 1H). Further analysis using fresh colonic biopsy specimens from another cohort of CD patients revealed a positive correlation between %CDpop and International Organization for the Study of Inflammatory Bowel Diseases scores, a measure of disease activity in CD (26) (SI Appendix, fig. S3D). Altogether, these findings suggest that an increase in local CDpop is associated with more severe disease activity of CD.

Single-Cell Transcriptional Landscape of IBD Recapitulates Mass Cytometry Results.

We next performed single-cell targeted RNA-sequencing (scRNA-seq) analysis of CD3⁺ T cells separated by fluorescence-activated cell sorting (FACS) to identify the transcriptional landscape of IBD gut. Given the predominance of specific T cells with unique cell surface marker sets in each IBD subtype, we applied the BD AbSeq system, which enables the simultaneous detection of targeted protein and mRNA (SI Appendix, Table S4A). T cells freshly isolated from the inflamed colonic mucosa of IBD patients and unaffected normal regions of the gut mucosa from colon cancer patients were analyzed (SI Appendix, Table S2). A total of 85,524 cells (comprising 25,171 for controls, 27,779 for CD, and 32,574 for UC) from 12 patients (n = 4 per group) were encapsulated for cDNA synthesis with the BD Rhapsody system. Following batch effect correction of the data by the Scanorama algorithm, we projected cells in two dimensions using Uniform Manifold Approximation and Projection (UMAP) (fig. 2A). CD3⁺ lymphocytes were subdivided into 17 subclusters comprising nine CD4⁺ T cells, six CD8⁺ T cells, CD4/CD8 double-positive cells, and a mixed subcluster of CD4 and CD8 single-positive cells; each subcluster was assigned to the major cell types by its unique transcriptional signature (fig. 2A and SI Appendix, Figs. S4 and S5 and Table S5), and clusters were ranked in their order of abundance. First, transcriptional and translational levels of selected markers correlated well with each other, except for ITGAE encoding CD103 (SI Appendix, fig. S6 A and B). In contrast to the highly exclusive distribution of CD103, ITGAE was ubiquitously expressed across the subclusters, indicating the posttranscriptional regulation of the expression of this molecule. Tem, naive T cells, and Tcm were annotated based on the combination of expression levels of CD45RA and CCR7: CD45RA⁻ CCR7⁻ T cells as Tem, CD45RA⁺ CCR7⁺ as naive T cells, and CD45RA⁻ CCR7⁺ as Tcm. CD4⁺ Tem did not display discrete clusters of T helper subtypes, such as Th1 and Th2, but was rather defined by the continuous activation and effector status of each cluster, as previously described (27, 28). The largest T cell subpopulation was CD161^high CD4⁺ Tem (clusters 1, 5, and 6) (fig. 2A and SI Appendix, fig. S6A), which indicated its potential polarization toward Th17 lineages (29). Cluster 21 was enriched in cytotoxic genes such as GZMs and NKGs, implying the cytotoxic nature of this cluster, which was compatible with cytolytic CD4⁺ T cells (CD4⁺ CTL) (30) (SI Appendix, Figs. S4 and S5). This subset was not detected by the mass cytometry analysis, probably due to the lack of appropriate cell surface markers. Remarkably, the transcriptional landscape of CD, UC, and control displayed substantial differences between the groups, which strengthened the results of mass cytometry analysis (fig. 2 B and C). In line with the CyTOF results (fig. 1C and SI Appendix, fig. S2F), CD4⁺ Tem (clusters 1, 5, and 6) was predominant in control samples, while CD4⁺ Trm (cluster 12) was elevated in CD samples. Furthermore, Tfh (cluster 8) highly expressing master transcription factor transcripts BCL6 and TCF7 (31–33) together with CXCL13, encoding a ligand for CXCR5 (34), tended to increase in UC.

Fig. 2.

Single-cell transcriptional landscape of colonic lymphocytes in IBD. (A) UMAP plot of FACS-sorted CD3⁺ T cells in fresh colonic samples (n = 4 per group). Cluster number is in parentheses. Data from CD, UC, and controls were merged. (B) UMAP plot of FACS-sorted CD3⁺ T cells in each group. (C) Cell type proportions of each CD4⁺ T cell subset in CD3⁺ T cells. *P < 0.05, **P < 0.01, ****P < 0.0001, mean ± SEM, one-way ANOVA (Tukey’s multiple comparison test). (D) Volcano plot of genes differentially expressed between CD and UC CD4⁺ T cell subclusters. ADT: antibody-derived tag. (E) Gene expression plot of representative transcription factor transcripts that were significantly up-regulated in CD (PRDM1) or in UC (LEF1, TCF7, and BCL6) was overlaid on the UMAP plot. (F) MRtree that reveals the transcriptional distinctions and similarities between cell types. The pie charts on tree nodes represent the cell type composition in each cluster.

CD and UC Are Transcriptionally Wired by Programs Acting in Opposing Directions.

To understand the transcriptional divergence underpinning disease-specific adaptive immunity, we compared the transcriptional landscape of CD and UC CD4⁺ lymphocytes and found that several transcription factor transcripts were differentially expressed (fig. 2 D and E). For instance, PRDM1 encoding B Lymphocyte-Induced Maturation Protein (BLIMP-1), which is dominantly expressed in effector T cells (35) and regulates a transcriptional program of tissue residency (36), was up-regulated in CD compared with the level in UC. In contrast, BCL6 and TCF7, whose protein products are reciprocally antagonistic transcription factors of BLIMP-1 (31–33), were enriched in UC. LEF1, which like TCF7 was highly expressed in both naive T cells and Tfh, contributes to the differentiation of Tfh through the regulation of Bcl6 (32, 33).

Multiresolution Reconciled Tree (MRtree) constructs trees that effectively reflect the extent of transcriptional distinctions among cell groups and align with the levels of functional specialization (37). MRtree applied to scRNA-seq datasets revealed that the first split created two major groups: the right major branch and the left major branch (fig. 2F). Importantly, CD4⁺ Trm belonged to the right branch consisting primarily of the CD8⁺ subsets rather than the left branch consisting of the CD4⁺ subsets, indicating that the transcriptional signature of CD4⁺ Trm is more closely related to that of CD8⁺ T cells. Indeed, gene enrichment analyses of differentially expressed genes showed that the right major branch, including CD4⁺ Trm, was enriched in the terms “natural killer cell–mediated cytotoxicity” and “granzyme-mediated programmed cell death signaling pathway,” whereas genes related to the terms “regulation of alpha–beta T cell activation” and “Th17 cell differentiation” were enriched in the left major branch (SI Appendix, fig. S7 A and B).

Trajectory inference methods order cells based on similarity in their gene expression patterns, which allow us to infer cell differentiation, cell activation, and the phase of cell cycle transition (38). Cell lineage and pseudotime inference analysis using Slingshot (39) by Seurat divided CD4⁺ lymphocytes into four lineages (SI Appendix, fig. S8A). Lineage 1 was directed to CD4⁺ Trm, lineage 2 to CD4⁺ Tem, lineage 3 to Tfh, and lineage 4 to CD4⁺ CTL, among which CD, control, and UC deviated toward lineages 1, 2, and 3, respectively (SI Appendix, fig. S8B). As such, CD and UC are transcriptionally controlled by programs acting in conflicting directions, and thus, an imbalance between two mutually opposing differentiation programs might underpin the unique phenotype of each IBD subtype.

Tissue-Resident Memory T Cells Are Reprogrammed in CD Gut to Initiate Effector and Innate Functions.

We next attempted to define the transcriptional signature of a unique CD4⁺ Trm subset, CDpop. By plotting CD4⁺ CD103⁺ CD161⁺ CCR5⁺ CD27⁻ CCR7⁻ cells on the UMAP (fig. 3A), CDpop formed a well-defined cluster as it was almost exclusively mapped on the portion of the CD4⁺ Trm (cluster 12) (fig. 3B). Cells within the CDpop highly expressed GZMA and IFNG under unstimulated conditions (fig. 3C). Genes differentially expressed between CDpop and the rest of the CD4⁺ T cells revealed that transcripts of chemokines and IFNG were significantly up-regulated in CDpop, whereas genes related to a resting phenotype were up-regulated in the rest of the CD4⁺ T cells (fig. 3D). Pathway analysis of these up-regulated genes included “IL-12 pathway” and “inflammatory response” (SI Appendix, fig. S9). Together with the results of MRtree that revealed that the transcriptional signature of CD4⁺ Trm is closer to that of CD8⁺ T cells than to those of other CD4⁺ T cell subsets, these results indicate that CDpop is a colitogenic T cell subset that potentially secretes inflammatory cytokines and cytotoxic molecules in pathological settings. On the basis of these findings, we further subdivided CDpop to see if we could narrow it down into a more inflammatory CD-specific CD4⁺ Trm fraction. Indeed, reclustering of CDpop clearly demarcated CD-specific CD4⁺ Trm (CDtrm: cluster 2 in fig. 3E) and the rest of the CDpop (fig. 3 E and I). This newly emerged population in CD appeared and accumulated specifically in CD but not in control and UC (fig. 3 E and H). CDtrm displayed the higher expression of GZMA and IFNG than the rest of CDpop (fig. 3 F and G). Genes differentially expressed between CDtrm and the rest of CDpop revealed that IFNG and a substantial number of natural cytotoxicity-related genes were up-regulated in CDtrm (fig. 3J). Gene ontology and pathway analyses of the up-regulated genes revealed particular association with natural killer cell–mediated cytotoxicity and “IL-12 pathway” (fig. 3K). Thus, by combining scRNA-seq datasets with the selection of appropriate surface marker sets in mass cytometry, we identified a unique CD-specific CD4⁺ Trm with innate-like properties.

Fig. 3.

Transcriptional signatures of CD-predominant CD4⁺ Trm. (A) Protein expression of cell surface markers defining CDpop. (B) CDpop gated by the positive expression of CD4, CD103, CCR5, and CD161, and negative expression of CD27 and CCR7 was overlaid on cluster 12 of the UMAP plot. (C) Gene expression plot of secretory proteins overlaid on CD3⁺ T cells (Upper row) and cluster 12 (Lower row) of the UMAP. (D) Volcano plot of genes differentially expressed between CDpop and the rest of the CD4⁺ T cells. (E) CDpop was reclustered (upper row), and CDpop in each condition was overlaid on its UMAP (Lower row). (F) Gene expression plot of GZMA and IFNG overlaid on the UMAP. (G) Violin plot of GZMA and IFNG in each cluster. (H) Proportion of CDtrm in CD4⁺ T cells. (I) Diagram showing the relationships between CD4⁺ Trm, CDpop, and CDtrm. (J) Volcano plot of genes differentially expressed between CDtrm and the rest of CDpop. (K) Bar graph of the top 20 enriched terms across genes that were significantly up-regulated in CDtrm compared with the levels in the rest of CDpop.

CD-Predominant T Cell Subset Secretes Inflammatory Cytokines Independent of Antigen Recognition.

Transcriptional profiles revealed innate-like aspects of CDtrm, and therefore, we hypothesized that CDtrm is poised for full activation and thus quickly acquires colitogenic properties upon stimulation. We investigated whether CDtrm is responsive to cytokine stimuli independent of T cell receptor (TCR) ligation. CDpop, which contains CDtrm and can readily be isolated by cell surface marker sets, was used for the analysis. On the basis of the transcriptional profile of CDpop that partially displays Th1 signatures, such as high IFNG and IL-12 signaling pathways (fig. 3D and SI Appendix, fig. S9), we stimulated LPMC with IL-12 and IL-18, which resulted in the release of IFN-γ from CDpop (fig. 4A). Additionally, a high expression level of IL-7 receptor subunit in this cluster (SI Appendix, Figs. S4 and S6A) prompted us to stimulate cells with common gamma-chain (γc) family cytokines IL-7 and IL-15, which are necessary for intrinsic TCR tuning and sensitization (40, 41). Although homeostatic cytokines alone induced little cytokine secretion from CDpop, synergistic enhancement of IFN-γ secretion was observed by the exposure to γc cytokines in combination with IL-12 and IL-18 (fig. 4A). This result was replicated in an experiment using FACS-sorted CD3⁺ T cells to exclude the effect of other cells contained in LPMC (SI Appendix, fig. S10A), which confirmed the validity of the direct effect of exogenous cytokines on CDtrm.

Fig. 4.

Responsiveness of CD-predominant CD4⁺ Trm to cytokines. (A) Left: Proportion of IFN-γ⁺ cells in CDpop under cytokine stimulation in CD samples. *P < 0.05, n = 8 per group, mean ± SEM, two-way RM ANOVA (Tukey’s multiple comparison test). Right: Representative images of flow cytometry analysis. (B) Proportions of IFN-γ⁺ cells with or without stimulation in CDpop and CDpop with negative selection of MAIT (TCRVα7.2⁺) and γδT cells. *P < 0.05 and **P < 0.01, n = 5 per group, mean ± SEM, RM one-way ANOVA (Tukey’s multiple comparison test). (C) Left: Proportions of IFN-γ⁺ cells in CDpop and CD4⁺ T cells after cytokine stimulation. Right: The corresponding samples are connected with a line. *P < 0.05, n = 7 per group, mean ± SEM, two-tailed paired Student’s t test. (D) Left: Proportions of IFN-γ⁺ cells in CDpop and CD4⁺ T cells after TCR or cytokine cocktail. Right: The corresponding samples are connected with a line. *P < 0.05, n = 6 per group, mean ± SEM, two-way RM ANOVA (Sidak’s multiple comparison test). (E) Left: Cytokine/TCR ratio for IFN-γ secretion of each sample. Right: The corresponding samples are connected with a line. *P < 0.05, n = 6 per group, mean ± SEM, two-tailed paired Student’s t test. (F) Left: Proportions of TNF-γ⁺ cells in CDpop and CD4⁺ T cells after TCR or cytokine stimulation. Right: The corresponding samples are connected with a line. *P < 0.05, n = 6 per group, mean ± SEM, two-way RM ANOVA (Sidak’s multiple comparison test). (G) Phenograph-based visualization on the tSNE plot of frozen colonic CD4⁺ T cells from LPMC from four CD patients and two controls (n = 6 per condition and n = 12 merged) stimulated with or without cytokine cocktail. Heat map of mean expression values of T cell markers is shown on the right. (H) tSNE overlay of selected surface marker expression on merged plot. (I) tSNE overlay of cytokine markers in stimulated and unstimulated samples. (J) Proportion of each cytokine-positive cell in total CD4⁺ T cells, CD103⁻ counterpart of CDpop, and CDpop after stimulation by cytokine cocktail. ***P < 0.05, **P < 0.01, ***P < 0.001, n = 6 per group, mean ± SEM, two-way RM ANOVA (Sidak’s multiple comparison test).

Unconventional T cells, such as mucosa-associated invariant T cells (MAIT), NK-T cells, and γδT cells, contribute to both innate and adaptive arms of the immune system and thus secrete cytokines in a TCR–independent manner (42, 43). In general, γδT cells and MAIT are relatively abundant in the gut, whereas NK-T cells are rare among the human T cell population (44). Although most of the unconventional T cells are positive for CD8, a minor portion of unconventional T cells are CD4⁺ T cells. Indeed, 3.2% and 6.5% of LP CD4⁺ T cells were positive for TCR Vα7.2 and γδTCR, respectively, and a portion of CDpop accounting for 7.0% and 10.4% of the total overlapped with TCR Vα7.2 and γδTCR, respectively (SI Appendix, fig. S10B). To rule out the possible effects of unconventional T cells contained in CDpop, MAIT and γδT cells were negatively selected from CDpop, and the proportion of IFN-γ–producing cells was subsequently evaluated (fig. 4B). The rate of IFN-γ⁺ cells in CDpop increased after the exclusion of unconventional T cells, confirming that contaminated unconventional T cells were not the major source of IFN-γ secreted from CDpop. The proportion of IFN-γ–producing cells among CDpop was significantly higher than that of the entire CD4⁺ T cells (fig. 4C), emphasizing the cytokine-sensitive properties of this specific subset.

We next examined whether CDpop with negative selection of unconventional T cells was differentially responsive to TCR and cytokine stimulation. TCR stimulation by anti-CD3/CD28 antibodies resulted in the pan activation of CD4⁺ T cells, and TCR stimulation alone showed IFN-γ production in CDpop comparable to that in the entire CD4⁺ T cells (fig. 4D). Indeed, the cytokine/TCR response ratio for IFN-γ secretion in each sample was significantly higher in CDpop than in CD4⁺ T cells, reflecting the cytokine-sensitive nature of CDtrm (fig. 4E). Thus, the secretion of IFN-γ from CDpop was specifically and maximally enhanced by cytokine stimulation. Secretion of tumor necrosis factor-α (TNF-α, another Th1 cytokine) was significantly lower in CDpop than in total CD4⁺ T cells stimulated with TCR (fig. 4F). These findings indicate that CDtrm possesses innate cell-like properties in addition to Th1 properties.

CD-Predominant T Cell Subset Is the Main Source of TCR-Independent IFN-γ Production.

To gain broader insight into cytokine-responding T cells, we performed mass cytometry analysis in frozen LPMC stimulated by a cytokine cocktail. Seven probes against intracellular cytokines were added to the previous surface marker antibody panel (SI Appendix, Table S6). Samples were stimulated overnight with or without cytokines. Cell surface markers were used for the clustering process, and the cytokine plot was subsequently overlaid on the map. A phenograph revealed that a mixed population of stimulated and unstimulated CD4⁺ T cells was subdivided into 17 subclusters, some of which responded well to the cytokine cocktail, as shown by the upregulation of CD25 and CD28 accompanied by the downregulation of CD127 after stimulation (fig. 4G). As observed in the previous flow cytometry analysis (fig. 4A), cytokine cocktail stimuli drove CDpop, corresponding to a portion of cluster 5 (fig. 4 G and H), to secrete high levels of IFN-γ and IL-2 (fig. 4 I and J). A heat map of mean marker expression revealed that cluster 16 is CD161⁺ CD4⁺ Tem, which were distinguished from CDpop by the negative expression of CD103. This subset was predominant in controls in CyTOF analysis (fig. 1 A and B and SI Appendix, fig. S2 A and B) and secreted only a small amount of IFN-γ (fig. 4 I and J). In contrast, TNF-α, IL-17, and IL-4 production did not differ between the subsets. These findings were reproduced in the same experiments with fresh samples, ruling out the possibility that the freezing and thawing process itself stimulates cells to secrete inflammatory cytokines (SI Appendix, fig. S10C). Because the cytokine cocktail is more potent at stimulating type 1 cytokine secretion, we next examined the cytokine-producing potential of CDpop by applying stimulation with PMA/ionomycin to further examine potential cytokine production. Similar to the effects of the cytokine cocktail, PMA/ionomycin also induced significantly higher levels of IFN-γ and IL-2 secretion from CDpop compared with the levels in other CD4⁺ T cell fractions (SI Appendix, fig. S10D). Additionally, GZMA was significantly higher in CDpop than in other CD4⁺ fractions in both unstimulated and stimulated conditions, which is in line with the scRNA-seq results. IL-17 and TNF-α, which were not secreted in response to cytokine cocktail stimuli, were induced by PMA/ionomycin treatment and were significantly elevated in CDpop compared with the levels in other CD4⁺ fractions, implying the pleiotropic nature of CDpop with innate and Th1/Th17 signatures. Thus, in the cytokine milieu of IBD, CDpop may produce proinflammatory cytokines continuously even in the absence of specific antigen stimulation, which may in turn contribute to the persistence of chronic inflammation.

Restricted Clonality of CD-Predominant T Cells Is Observed in a Portion of CD Patients.

To elucidate the clonality of expanded CDpop, we next performed TCR-seq on FACS-sorted CDpop from 6 CDpop^high patients. The mean percentage of CDpop in CD4⁺ T cells of the subjects was 7.1 ± 1.8%, which was higher than the cutoff value of 4% of frozen CD4⁺ T cells. Overall, the clonal structure of CDpop from CD patients was highly diverse and polyclonal, and only a limited number of clones were shared between the patients (SI Appendix, fig. S11A and Table S7). The clonotypes in each patient used a broad spectrum of TCR Vα and Vβ genes (SI Appendix, fig. S11B). In patient 1, the specific clonotype 1 and the top 10 clonotypes accounted for 11% and 35% of the total, respectively. Meanwhile, even the highest ranked clones in patients 2 to 6 accounted for 0.9 to 3.4%. We finally assessed the antigen specificity of the top 10 CD4⁺ T cell clonotypes in each patient by referring to the McPAS-TCR database (45). Although none of the CDR3 sequences matched perfectly to known pathogen- or immune disease-associated sequences, the clonotype 1 sequence was analogous to that of previously reported IBD-associated clonotype (46), with two amino acid substitutions. As such, polyclonal proliferation underlies the expansion of CDpop in CD patients, but restricted clonality of CDpop was observed in a portion of patients.

Cytotoxic Effect of CD-Predominant T Cells on the Epithelial Layer Is Maximized by Their Localization Properties.

The spatial properties of cells can elicit cellular functions, and therefore, we next investigated the pathological relevance of CDpop localization in LP. Immunohistochemical evaluation of CD4 and CD103 double-positive cells including CDpop in LP revealed that they localized adjacent to the epithelia along with other CD103-expressing cells (fig. 5A). In accordance with this unique spatial distribution of CD4⁺ Trm in LP, we investigated whether CDpop can directly damage the gut epithelial layer upon stimulation. IFN-γ treatment alone injured the human gut spheroids in a dose-dependent manner (fig. 5B). Notably, coculture of human-derived spheroids with FACS-sorted CDpop gated by CD4⁺ CD103⁺ CD161⁺ CCR5⁺ CD27⁻ CCR7⁻, stimulated with a cytokine cocktail for 72 h, resulted in epithelial damage accompanied by the elevation of Lactate Dehydrogenase (LDH) levels in the culture supernatant that is comparable to the findings observed when IFN-γ was added to the spheroids (fig. 5 B–D). In contrast, coculture with CDpop without cytokine cocktail did not disrupt the spheroid structure or induce LDH release. Furthermore, IFN-γ neutralizing antibody canceled the effect of cytokine-stimulated CDpop on the gut spheroids (fig. 4 E and F). Thus, IFN-γ produced by CDpop exerts its colitogenic effects by directly damaging the adjacent epithelial barrier. Together with the previous findings that IL-12, IL-15, and IL-18 are highly expressed in the affected LP of IBD (47–49), these observations support the notion that CDpop can be maximally enhanced both by the unique cytokine milieu in IBD gut and by the localization site in vivo.

Fig. 5.

Cytotoxic effect of CD-predominant CD4⁺ Trm on epithelia. (A) Representative images of immunohistochemistry of LP from CD patients. (Scale bar, 20 μm.) Red, CD4; green, CD103; and blue, DAPI. White arrows show CD4⁺ CD103⁺ Trm in LP. (B) Percentage of LDH release in the culture supernatant. *P < 0.05, n = 4 per group, mean ± SEM. shown, two-way ANOVA (Sidak’s multiple comparison test). (C) Average spheroid damage scores of four independent experiments. *P < 0.05, n = 4 per group, mean ± SEM. shown, two-way ANOVA (Sidak’s multiple comparison test). (D) Representative images of spheroids cocultured with or without CDpop under the cytokine stimuli. (Scale bar, 200 μm.) (E) The effect of IFN-γ neutralizing antibody (nAb) on cytokine-stimulated CDpop evaluated by relative LDH release in the culture supernatant. *P < 0.05, n = 4 per group, mean ± SEM. shown, Dunn’s multiple comparison test. (F) Representative images of spheroids cocultured with or without CDpop and nAb. (Scale bar, 150 μm.)

Discussion

Using single-cell technologies on LPMC derived from human samples, we identified a CD4⁺ Trm subset that was specifically increased in CD. This CD-specific T cell subset, reprogrammed into effector and innate-like phenotypes, is capable of directly damaging adjacent epithelial cells, thereby exerting a pathogenic effect. Trm is characterized by the expression of C-type lectin-like molecules CD69 and CD103, which are the αE subunits of αEβ7 integrin involved in the adhesion of lymphocytes to epithelial cells. Although CD69⁺ T cells have been reported to exhibit tissue-localized properties, CD103⁺ T cells were specifically referred to as Trm in this study because the majority of lymphocytes express CD69 in the gut (24, 50). The present study showed that the expression of CD103 is mainly regulated at the posttranscriptional level. Therefore, deep profiling of Trm is possible only by combining scRNA-seq analysis with protein analysis, which allowed us to identify disease-specific CD4⁺ T cells. By taking full advantage of mass cytometry and scRNA-seq analysis, we successfully narrowed down CD4⁺ Trm to CDtrm, and this specific subset indeed expanded drastically in CD patients.

Whether Trm plays a central role in the pathology of IBD is controversial (17–19), probably due to the difference in methodologies used in each study, high variance of the patient cohort, and the heterogeneity of Trm at the molecular level. Furthermore, most of these previous studies focused on CD8⁺ Trm, and little is known about the pathological properties of CD4⁺ Trm in IBD, especially in humans. The lack of appropriate cell surface markers that define CDtrm with high purity led us to perform biological validation using CDpop. Future studies may establish a more sophisticated set of markers for CDtrm, which could allow us to extend our study to future therapeutic strategies. Dissecting the diversity of CD4⁺ Trm and targeting the inflammatory CD4⁺ Trm fraction will further enhance the effectiveness and minimize the side effects of therapeutic interventions targeting the β7 integrin for IBD, which is already in phase III clinical trials (51).

In mice, colitogenic effects of Trm are attributed to the transcriptional regulators of Trm development (18). In addition to the regulatory function on tissue homing, BLIMP-1 encoded by Prdm1 is responsible for the acquisition of innate-like activity by T helper cells in response to the γc cytokine IL-2 (52). Thus, the functional dichotomy of the key transcription factor may control tissue residency and innate-like activities in CDtrm. Although many studies have reported on animal models of enteritis, the findings cannot be applied directly to human pathology because the expression of molecules is not completely conserved across species, and there are no appropriate animal models that exactly reflect each subtype of IBD.

Another important aspect is that CDtrm is poised for rapid execution of effector functions by cytokine stimulation without TCR ligation. IL-12 and IL-18 synergistically induce IFN-γ production from T cells via the upregulation of IL18R and transcriptional enhancement of IFNG by transcription factors induced by both signaling pathways (53, 54). Intriguingly, we found that the synergistic effect of γc cytokines further reinforced the effector function of these cells. Indeed, increased levels of IL-12, IL-15, and IL-18 have been observed in the LP of IBD gut in previous studies (47–49). Given the proximity of Trm to epithelia, it is reasonable that CD4⁺ Trm responds to IL-15, which can be secreted by the epithelia in concert with other inflammatory cytokines abundant in the LP of CD patients, resulting in epithelial cytotoxicity.

GWAS revealed that CD- or IBD-associated loci contain genes that encode IFN-γ and its receptor (1), highlighting the importance of IFN-γ in disease pathogenesis. Indeed, IFN-γ has been shown to be increased in the LP of CD patients (55, 56) and is required for inducing inflammation in colitis models (57, 58). IFN-γ regulates intestinal epithelial homeostasis (59, 60) and drives IBD pathogenesis through vascular barrier disruption (61, 62). Despite this plethora of findings indicating that IFN-γ is profoundly involved in intestinal inflammation, its specific neutralizing antibody fontolizumab has had limited efficacy in clinical trials (10), suggesting that additional factors are required for disease development. Intriguingly, IFN-γ, in synergy with TNF-α, promotes the apoptosis of intestinal epithelia (59). This is noteworthy because a neutralizing antibody specific for TNF-α resulted in remarkable clinical improvement (63), in contrast to the poor clinical efficacy of many other anticytokine antibodies (4, 8, 9). Furthermore, Th17 cells, which have recently received attention as the core of IBD pathogenesis, have considerably high plasticity (64, 65). In this context, the increased rate of IFN-γ–producing Th17 cells within CDpop upon PMA/ionomycin stimulation and high expression of CD161 in CDpop indicate that IFN-γ produced by CDpop may be a marker of pathogenic Th17 cells. Taking these findings together, the properties of CDtrm, that is, IFN-γ secretion and spatial proximity to the gut epithelia, should be related to susceptibility to mucosal injury.

CD and UC display distinct patterns of immune networks, which are transcriptionally wired by the programs acting in opposing directions. In contrast to the accumulation of CD4⁺ Trm in CD, TCF7 ^high Tfh were increased in a proportion of UC patients. The TCF7 product TCF-1 induces GATA3 expression required for Th2 differentiation while blocking IFN-γ–producing Th1 cell fate (66); thus, differential TCF7 expression may reflect the classically proposed Th1–Th2 deviation as a mechanism of IBD. This finding points to the different inflammatory gut environments that tilt the balance between these T cell lineages.

In previous studies in which single-cell analysis of immune cells in IBD was conducted, Trm expressing high levels of Th1/Th17-related cytokines (14) and intraepithelial Th17 cells (67) were described. However, the functional role and pathological relevance of these cells have not yet been explored. Furthermore, most previous studies focused on one of the two subtypes of IBD, preventing a disease-specific program of T cell differentiation from being revealed. By the comprehensive analysis of human samples, we showed that distinct immune networks were rewired in each IBD subtype. In CD, gut-resident T cells were expanded. This included a subset of CD4⁺ Trm, which we identified to be the most pathogenic and poised to exert effector and cytotoxic responses upon inflammatory stimuli. In contrast, the shift toward the Tfh differentiation program in a proportion of UC patients is possibly associated with the increased pathological IgG⁺ plasma cells in the UC setting (25, 68) and basal plasmacytosis observed in UC histology (69). In summary, our findings shed light on the transcriptional and proteomic landscape of adaptive immunity in IBD, highlighting the potential roles of a specific T cell subset in the pathogenesis of this disease.

Materials and Methods

Detailed methods are available in SI Appendix, Methods.

LPMC Isolation.

The normal intestinal mucosa was obtained from macroscopically intact areas in patients with colorectal cancer. The inflamed intestinal mucosa was obtained from surgically resected specimens from patients with IBD. Briefly, intestinal epithelial cells were dissociated by shaking in 5 mM ethylenediaminetetraacetic acid (EDTA) in Hanks’ Balanced Salt Solution, followed by removal of the muscle layer. The mucosal layer was cut into pieces and digested with 1 mg/mL collagenase type 1 (Sigma Aldrich) mixed with 0.07 IU/mL DNase in RPMI 1640 for 30 min at 37 °C. Cells were centrifuged at 800 g, dispersed in EDTA solution, and washed in phosphate-buffered saline (PBS). LPMC were frozen in liquid nitrogen until analysis. The same procedure was followed for cell isolation from biopsy samples, except that removal of the muscle layer was not required.

Mass Cytometry.

Cryopreserved LPMC and PBMC were barcoded with a combination of anti-CD45 antibodies (89Y, 115In, and 175Lu) for 30 min at RT and washed with Maxpar Cell Staining Buffer (Fluidigm). After centrifugation, cells were stained with an antibody mix (SI Appendix, Tables S1 and S5) for 45 min at RT. After washes, cells were incubated with cisplatin (1:1,000, Cell-ID™ Cisplatin-¹⁹⁸Pt; Fluidigm) for 5 min at RT, quenched with Maxpar Cell Staining Buffer, and fixed with 2% PFA for 30 min at 4 °C. For intracellular staining, eBioScience Fix/Perm Buffer (Invitrogen) was added to cells and centrifuged, stained with markers dissolved in Fix/Perm Buffer for 45 min at 4 °C, and washed with Fix/Perm Buffer. After centrifugation, the pellet was incubated with iridium (1:1,000, Cell-ID™ Intracalator-Ir; Fluidigm) overnight at 4 °C, washed, and acquired on CyTOF Helios (Fluidigm). EQ beads were used for normalization.

Single-Cell RNA-Seq and BD AbSeq.

LPMC were freshly prepared from surgically resected samples and stained with FITC anti-human CD3 antibody (clone HIT3a; BioLegend) for 30 min at 4 °C, followed by dead cell staining, after which cells were sorted on a Becton, Dickinson and Company (BD) FACSAria. Sorted live CD3⁺ cells were incubated with BD AbSeq antibodies (BD Biosciences) conjugated with oligonucleotides (SI Appendix, Table S4) for 30 min at 4 °C and washed in PBS. Subsequently, a single-cell suspension was loaded onto BD Rhapsody (BD Biosciences), followed by the loading of a bead library onto the cartridge. The cells were lysed to hybridize mRNAs onto barcoded capture oligos on the beads. The beads were collected into a single tube for the subsequent steps of complementary DNA (cDNA) synthesis, exonuclease I digestion, and multiplex PCR-based library construction. The BD Rhapsody Immune Response Panel for human and a custom primer set (SI Appendix, Table S4) were used for library generation. Libraries were sequenced on the Illumina HiSeq 3000 platform (2 × 100 bp).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Single-cell RNA data are available on website with the accession number GSE218000: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE218000. All other data are included in the article and/or SI Appendix.