Dysregulated IL-2 signaling and T-cell subset shifts are associated with eosinophilic esophagitis activity

Abstract

Eosinophilic esophagitis (EoE) is a chronic immune-mediated disorder of the esophagus. However, the T-cell transcriptional programs distinguishing active disease from remission remain incompletely understood. In this study, we applied single-cell RNA sequencing to esophageal biopsies obtained from normal subjects and patients in remission or with active EoE, to characterize T-cell heterogeneity, state-dependent transcriptional changes, and intercellular communication networks. After stringent quality control, no substantial batch effects were observed. We identified five major T-cell subsets, with natural killer T (NKT) cells representing the most abundant population. Notably, naïve T cells and T helper 2 (Th2) cells were detected exclusively in active EoE samples, whereas regulatory T (Treg) and Th17 cells were present in all groups. The abundance of Tregs surpassed that of Th17 cells in active EoE but was similar to Th17 levels in normal and remission tissues. Functional enrichment analysis revealed a preferential association of NKT cells with receptor-binding functions. Th2 and naïve T cells shared signatures related to ribosome biology; naïve T cells additionally exhibited 5′-UTR binding and translation-regulator activity, aligning with their maturation potential. Consistent with the non-malignant nature of EoE, copy-number variation signals were minimal. Cell-type-specific differential expression analysis uncovered activated immune programs in NKT cells from active EoE, enriched for cytokine activity, along with epigenetic and transcriptional alterations in Tregs, including demethylase-linked functions, and protein-folding-related changes in Th17 cells. Cell-cell communication inference indicated a substantial rewiring of interaction networks in active EoE, suggesting activation of IL-16, IL-10, TRAIL, and PECAM1 pathways and heightened outgoing signaling from Th2 cells. Across all conditions, NKT cells served as dominant signaling senders and receivers. Collectively, these results delineate the T-cell composition and signaling circuits specific to active EoE, offering mechanistic insights into the maintenance of disease activity.

Highlights

• •Single-cell RNA-seq reveals five T-cell subsets in eosinophilic esophagitis.
• •Naïve T and Th2 cells are uniquely enriched in active EoE.
• •Active EoE shows rewired IL-2-centered cell-cell communication networks.
• •NKT cells are dominant signaling senders and receivers across disease states.
• •Treg epigenetic programs and Th17 stress signatures differ by disease activity.

1. Introduction

Eosinophilic esophagitis (EoE) is a chronic, immune-mediated inflammatory disorder of the esophagus, characterized by eosinophil-predominant infiltration and clinically manifested by recurrent symptoms such as dysphagia and food impaction [1,2]. While eosinophilic esophagitis (EoE) is frequently framed within the context of allergic inflammation, its pathogenesis extends beyond eosinophilic activity alone. Rather, the disease arises from coordinated and dynamic crosstalk among multiple immune cell compartments within the esophageal mucosa [3]. Clinically, patients often cycle between phases of active inflammation and remission [4]. However, the precise cellular programs that differentiate these states, as well as the immune circuits that sustain or resolve mucosal inflammation, are still not fully elucidated. A more detailed dissection of the immune landscape across disease phases is therefore critical. Such analysis can clarify underlying mechanisms and identify biomarkers or therapeutic targets associated with disease activity.

T cells are pivotal orchestrators of mucosal immunity and have long been implicated in EoE, primarily via type 2 immune responses [5]. Yet the broad category T cells comprises functionally diverse subsets, including cytotoxic, regulatory, and helper lineages, each possessing distinct effector programs that may differ substantially between active inflammation and remission [6]. IL-2/CD25 signaling differentially tunes T-cell subsets, enabling tailored immunotherapy for autoimmunity and cancer [7]. IL-2 differentially modulates Treg and CD8⁺ T cells, including reversal of exhaustion, constituting a targetable axis for autoimmune and cancer immunotherapy, yet its opposing effects necessitate precise modulation [8]. IL-2-based strategies that selectively expand or potentiate regulatory T cells are emerging as clinically validated, targeted immunomodulatory approaches for restoring immune homeostasis in autoimmune disorders, transplant rejection, and eosinophilic esophagitis, offering a potential alternative to conventional broad-spectrum immunosuppressants [9]. Previous transcriptomic analyses have cataloged multiple T-cell subsets in esophageal biopsies [10]; however, subtype assignments often depend on the analytical framework used, and bulk-tissue profiling obscures subset-specific contributions. Moreover, shifts in T-cell composition, such as the emergence or loss of particular subsets, can profoundly remodel the inflammatory microenvironment even when global transcriptional signatures appear stable [11]. Thus, resolving T-cell heterogeneity within EoE tissues is a necessary step toward linking immune architecture to functional outputs and clinical disease states.

Single-cell RNA sequencing (scRNA-seq) offers a powerful, high-resolution approach to deconvolute immune heterogeneity in complex tissues and to uncover cell-type-specific transcriptional programs that are otherwise obscured in bulk analyses [12]. A key strength of scRNA-seq lies in its ability to: (i) perform unbiased identification and classification of T-cell subsets within tissue microenvironments [13]; (ii) enable quantitative comparison of subset frequencies across different clinical or experimental conditions [14]; and (iii) detect differential gene expression patterns within defined cell types that may drive disease activity or contribute to remission [15]. Beyond cataloging intrinsic transcriptional states, scRNA-seq data can also be leveraged to infer intercellular communication networks, which play a critical role in shaping inflammatory responses [16]. In such networks, individual cell types can dynamically alternate between sender and receiver roles depending on contextual signals from the microenvironment. Consequently, an integrative analytical framework, combining T-cell subset annotation, functional pathway enrichment, cell-type-specific differential expression, and cell–cell communication inference, provides a comprehensive strategy to bridge cellular architecture with immune function in EoE and to identify mechanistic drivers of disease phenotypes [17].

In this study, we employed a scRNA-seq–based analytical framework to profile T-cell heterogeneity in esophageal biopsies from individuals with normal histology, EoE in remission, and active EoE [18]. Following stringent quality control and confirmation that batch correction was unnecessary, we identified five major T-cell subtypes. These annotations were cross-referenced with previously reported T-cell clusters to ensure consistency and interpretability across datasets [19]. We subsequently quantified shifts in subset abundance across disease states, revealing condition-associated changes in the composition of regulatory and effector T-cell populations, as well as the presence of subsets uniquely enriched in active EoE. To uncover state-dependent immune programs, we performed cell-type-specific differential expression analysis within shared T-cell subsets, supported by functional enrichment and copy-number variation assessment to distinguish inflammatory transcriptional remodeling from potential tumor-like genomic alterations [20]. Finally, we reconstructed intercellular communication networks across conditions, which demonstrated extensive rewiring of signaling pathways, including IL2-mediated interactions. This systems-level analysis provides a comprehensive view of how coordinated changes in T-cell composition and communication dynamics may collectively drive EoE disease activity and resolution.

2. Materials and Methods

2.1. Data collection

Single-cell RNA-sequencing (scRNA-seq) data of esophageal T cells from patients with eosinophilic esophagitis (EoE) were obtained from the Gene Expression Omnibus (GEO) under accession GSE126250 [21]. Raw sequencing reads were downloaded from the Sequence Read Archive (SRA) under accession SRP184517.

2.2. scRNA-seq data preprocessing and quality control

Raw sequencing reads were quality-filtered and adapter-trimmed using fastp (v0.20.0) [22] with default parameters. Filtered reads were aligned to the human reference genome (Ensembl release 100) using HISAT2 (v2.2.1) [23]. Alignment files were processed using SAMtools (v1.14) [24]. Gene-level quantification was performed using featureCounts (v2.0.1) [25] with gene annotations from Ensembl release 100. Specifically, the 20% mitochondrial transcript threshold was selected based on established practices in scRNA-seq immune profiling, where mitochondrial content >20% typically indicates cellular stress, membrane permeabilization, or apoptotic cells that could introduce technical artifacts. This threshold is widely adopted in T-cell-focused scRNA-seq studies [[26], [27], [28]] and has been empirically validated to retain viable lymphocyte populations while excluding compromised cells. The 200-4000 detected gene range follows standard Seurat recommendations for immune cells: the lower bound (200 genes) excludes empty droplets and dead cells with minimal transcriptional content, while the upper bound (4000 genes) excludes potential doublets or multiplets, which typically exhibit abnormally high gene counts. For T lymphocytes specifically, this range captures the expected transcriptional complexity of resting and activated states without introducing doublet contamination. We also note that these cutoffs were visually validated against the distribution of nFeature_RNA and mitochondrial percentage plots (Fig. S1A–B). Downstream analyses were conducted in Seurat (v4.0.6) [29]. Cells were retained if they met the following quality-control criteria: (i) mitochondrial transcript proportion < 20%, and (ii) number of detected features (genes) between 200 and 4000. Cells passing QC were normalized and scaled using Seurat's standard workflow. Dimensionality reduction and visualization were performed using the Uniform Manifold Approximation and Projection (UMAP) algorithm. Clustering was carried out using FindClusters with a resolution of 0.5. Marker genes for each cluster were identified using FindAllMarkers with parameters min.pct = 0.75 and logfc.threshold = 0.25. Cell-type annotation was performed manually based on canonical marker genes (see Materials and Methods for details). Batch correction was not applied in this analysis, as quantitative assessments confirmed negligible batch effects. UMAP projection showed substantial intermingling of cells from different sample origins without discernible batch-related clustering (Fig. S1C); principal component analysis attributed <5% of total variance in the first 20 principal components to sample-level differences; the local neighborhood mixing score (scCustomizeCalculateMixingMetric) exceeded 0.85, indicating near-perfect distribution across UMAP neighborhoods; and canonical T-cell markers exhibited consistent expression patterns across samples. Given these findings and the absence of significant batch-associated separation, we performed integrated analysis without batch correction to avoid potential overcorrection that might remove genuine biological variation, following established recommendations for scRNA-seq datasets with minimal batch effects [30].

2.3. Copy number variation inference

To assess large-scale copy number variation (CNV) signals, we applied inferCNV (v1.8.1) [31]. Annotated normal cells were used as the reference population, and cells from the Active EoE and Remission groups were treated as test populations. Key parameters were set following the package recommendations: cutoff = 1, denoise = TRUE, and HMM = TRUE.

2.4. Functional enrichment analysis

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed using clusterProfiler (v4.3.1) [32]. For each cell type, marker genes identified by FindAllMarkers (Section 1.2) were analyzed separately, and genes with P < 0.05 were included. For cell-type-specific differentially expressed genes (DEGs), genes with adjusted P < 0.1 were included for enrichment analysis. The enrichGO and enrichKEGG functions were used with default settings unless otherwise specified. For marker gene identification (FindAllMarkers), genes with P < 0.05 were included for enrichment analysis. This relatively stringent threshold was applied to ensure that cluster-defining markers represented highly confident cell-type-specific signatures, given that marker genes are used for downstream annotation and cell-type identification, where specificity is paramount. For cell-type-specific differentially expressed genes (DEGs) comparing disease conditions, genes with adjusted P < 0.1 were included. The more lenient threshold for DEGs reflects the expected smaller effect sizes and reduced statistical power when comparing biological conditions within a single cell type, as opposed to comparing distinct cell types. This threshold (adjusted P < 0.1) is widely adopted in scRNA-seq differential expression studies for condition-based comparisons and appropriately balances discovery of biologically relevant transcriptional changes against false discovery control in the context of limited per-cell-type sample sizes.

2.5. Identification of cell-type-specific differentially expressed genes

Cell-type-specific DEGs were identified using Seurat's FindMarkers by comparing conditions (Active EoE vs Normal) and (Remission vs Normal) within each annotated cell type. Only cell types containing >30 cells in the Normal group were included in DEG testing. This threshold was selected to ensure sufficient statistical power for differential expression analysis using Seurat's Wilcoxon rank-sum test, as smaller cell populations (<30 cells per group) yield unstable variance estimates and increased false discovery rates in single-cell differential testing. The 30-cell cutoff represents a pragmatic balance between statistical robustness and biological inclusivity, consistent with recommendations for scRNA-seq DEG analysis in small subpopulations [33,34] and with previously published EoE single-cell studies.

2.6. Cell-cell ligand–receptor interaction analysis

Intercellular communication networks were inferred using CellChat (v1.1.3) [35]. Analyses were performed according to the official CellChat workflow (tutorial: https://github.com/sqjin/CellChat). We assessed and compared cell–cell communication patterns among T-cell subsets across Normal, Remission, and Active EoE conditions. CellChat analysis was performed following the official workflow (v1.1.3). Key parameter settings were as follows: Interaction probability threshold: ligand-receptor pairs with pval < 0.05 (derived from permutation testing, n = 100 permutations) were considered statistically significant and retained for downstream analysis. Minimum cell count per group: communication inference was performed separately for each condition (Normal, Remission, Active EoE) using all cells within each group, with no additional minimum cell count filtering applied to individual cell types to preserve rare subset representation. Communication strength cutoff: min.cells = 10 (default) was used to require at least 10 cells expressing a given ligand/receptor pair for interaction calculation, reducing spurious predictions from low-abundance transcripts. Net analysis: Pathway-level communication strength (information flow) was calculated using netAnalysis_computeCentrality with default parameters, and relative information flow was normalized to the maximum value across conditions for comparative visualization. All other parameters were kept at CellChat defaults as documented in the package tutorial (https://github.com/sqjin/CellChat).

2.7. Visualization and software

All preprocessing and downstream analyses were performed using R (v4.1.1) and Python (v3.8.8). In addition to figures generated by Seurat, inferCNV, clusterProfiler, and CellChat, custom visualizations were produced using ggplot2 (v3.3.5) in R and seaborn (v0.11.2) in Python.

3. Results

3.1. T cells in esophageal biopsies comprise distinct subtypes

After preprocessing the scRNA-seq dataset, low-quality cells and lowly expressed genes were removed (Fig. S1A and B). Dimensionality reduction indicated that batch correction was not required (Fig. S1C). We identified five major T-cell subtypes in esophageal biopsies (Fig. 1A). Among these, natural killer T (NKT)-like cells constituted the largest fraction. To facilitate comparison with prior annotations, we remapped the previously reported T1–T8 subsets to our identified cell types (Fig. S2). Notably, naïve T cells (corresponding to the previous T8 subset) and T helper 2 (Th2) cells (previous T7 subset) were detected exclusively in Active EoE samples (Fig. 1B; Fig. S2), whereas the overall “cell status” metrics did not show marked differences across subtypes (Fig. S3). Each subtype was defined by canonical marker genes (Materials and Methods) and displayed distinct expression patterns (Fig. 1C–E; Fig. S4). For example, GZMB, CCL5, CCL4, CCL4L2, and ITGA1 were selectively and highly expressed in NKT-like cells, supporting their distinct immune functional identity.

Fig. 1.

scRNA-seq analysis identifies major valve cell types. (A-B) Cell annotation by different information: (A) Specific cell types were defined; (B) Samples clinical information annotated. (C) Violine plot of expression of marker gene of different cell types. (D) Marker gene expression in the UMAP plot. (E) Heatmap of marker genes according to different cell types.

3.2. T-cell composition differs across disease conditions

We next compared T-cell subtype composition across Normal, Remission, and Active EoE conditions. Consistent with the subtype mapping, naïve T cells and Th2 cells were observed only in Active EoE and were absent from Normal and Remission samples (Fig. 2A). The proportions of Treg and Th17 cells were broadly comparable between Normal and Remission samples. In contrast, Treg cells outnumbered Th17 cells in Active EoE. Gene Ontology (GO) enrichment analysis revealed that NKT-like cells were preferentially enriched for receptor-binding-related functions (Fig. 2B and C). Th2 and naïve T cells shared enrichment in ribosomal structural components and rRNA binding, whereas naïve T cells additionally showed enrichment in mRNA 5′ UTR binding and translation regulator activity, consistent with a potential maturation-related transcriptional program (Fig. 2D and E). In line with the non-malignant nature of this inflammatory disease, copy number variation (CNV) signals were minimal (Fig. 2F).

Fig. 2.

Analysis of various cell types in scRNA-seq data. (A) The barplot of the proportion of various types of cells in different conditions. (B-C) GO function enrichment results of marker genes of different cell types. (B) Dotplot of significant enriched GO functions; (C) Network between different enriched GO functions in different cell types. (D-E) KEGG pathway enrichment results of marker genes of different cell types. (D) Dotplot of significantly enriched KEGG pathways; (E) Network between different enriched KEGG pathways in different cell types. (F) The copy number variation results inferred by InferCNV.

3.3. Cell-type–specific differential expression reveals condition-dependent programs

To characterize disease-associated molecular changes at single-cell resolution, we performed cell-type-specific differential expression analyses. Since naïve T and Th2 cells were restricted to Active EoE, differential expression comparisons against Normal were not possible for these two subsets (Fig. 3A). In contrast, robust condition-associated transcriptional differences were identified for NKT-like cells, Treg cells, and Th17 cells.

Fig. 3.

Cell type specifically expressed gene analysis. (A) UMAP plot of different cell proportions in different conditions. (B) Heatmap of differential expressed gene in NK T cells. Genes were clustered by whether genes were specific differential expressed in ActiveEOE or Remission group. (C-D) GO function enrichment results of differential expressed genes in Treg cell. (C) Dotplot of significant enriched GO functions; (D) Network between different enriched GO functions in different gene groups. (E-F) KEGG pathway enrichment results of differential expressed gene in NK T cells. (E) Dotplot of significantly enriched KEGG pathways; (F) Network between different enriched KEGG pathways in different gene groups.

For NKT-like cells (Fig. 3B), differential expression patterns could be classified into Active EoE–specific, Remission-specific, and shared changes relative to Normal. Among the shared group, USP9Y was upregulated compared with Normal. This gene has been reported as a biomarker in Sjögren's syndrome based on immune infiltration analyses [36]. In our dataset, USP9Y upregulation was largely confined to NKT-like cells, suggesting a potentially cell-type-restricted immune role in the esophageal mucosa. In Remission samples, CCL5 was selectively downregulated, consistent with prior evidence linking reduced CCL5 to a less activated NK/NKT-like state [37]. Active EoE–associated NKT-like DEGs were enriched for cytokine activity and cytokine receptor binding, indicating altered immune effector programs during active inflammation (Fig. S7).

In Treg cells, DEGs across conditions were consistently enriched for histone/protein demethylase activity, pointing to epigenetic and transcriptional regulation as a major axis distinguishing Treg states (Fig. S5). This pattern aligns with the known role of Tregs as key mediators of immune tolerance and their proposed application in immunosuppressive cellular therapy for autoimmunity and chronic inflammation (doi: 10.3389/fimmu.2020.609891). Indeed, many of the observed differences in our analysis were linked to transcription-related processes (Fig. 3C and D). By contrast, Th17 cells in Active EoE exhibited condition-associated changes more closely related to protein folding and protein binding, suggesting a distinct stress/processing signature in this subset (Figs. S6 and S8).

3.4. Intercellular communication shows extensive rewiring in Active EoE

Cell-cell communication is a key mechanism for maintaining tissue homeostasis within the esophageal mucosal microenvironment, and perturbations in signaling networks can contribute to inflammation [38]. We therefore inferred intercellular interactions across conditions using both bulk T-cell and subtype-level analyses (Fig. 4A and B). Several interaction programs, including MHC-related, IL2, and CCL signaling, were detected across Normal, Remission, and Active EoE groups. In contrast, additional pathways, such as IL16, IL10, TRAIL, and PECAM1, were preferentially activated in Active EoE (Fig. 4A). Subtype-level analysis further revealed that even interactions between the same pair of cell types differed substantially across conditions (Fig. 4B). For instance, the Th17 compartment acted predominantly as a receiver in Normal samples but shifted toward a sender role in Remission, whereas Th17 cells showed minimal inferred interactions in Active EoE.

Fig. 4.

Cell-to-cell intercellular interactions in different conditions. (A) Heatmap of activated intercellular interactions in different conditions cells. (B) Inferred interactions when considered all the cells (top left), normal cells (top right), remission cells (bottom left) and activeEOE cells (bottom right). (C) The outgoing signaling patterns in ActiveEOE cells. (D) Inferred incoming interaction strength and outgoing interaction strength when considered all the cells (top left), normal cells (top right), remission cells (bottom left) and activeEOE cells (bottom right). (E) The relative information flows comparing activeEOE cells and normal cells. (F) The expression of genes involved IL2 ligand-receptor interactions in different cell types and conditions.

Among outgoing signals, Th2 cells exhibited the strongest signaling output in Active EoE, with pathways dominated by COLLAGEN and IL2 signaling (Fig. 4C). Across conditions, NKT-like cells consistently displayed high communication activity as both senders and receivers (Fig. 4D; Fig. S9). While some pathways showed similar information flow between Normal and Active EoE (e.g., CD99, CCL, ADGRE5; Fig. 4E), many others were markedly rewired (e.g., JAM, IL2, CLEC). Notably, although IL2 ligand/receptor expression appeared broadly comparable in NKT-like, Treg, and Th17 cells, IL2 signaling differed across conditions. This difference was largely attributable to the presence of Active EoE–specific naïve T and Th2 populations (Fig. 4F). Together, these findings highlight that cellular composition and subtype-specific signaling jointly shape immune network behavior in EoE.

4. Discussion

Our single-cell transcriptomic profiling identified five distinct T-cell subtypes within esophageal biopsies. Among these, Naive T cells and Th2 cells were specifically enriched in Active EoE, reinforcing the established paradigm of eosinophilic esophagitis as a chronic, immune-mediated disorder driven predominantly by type 2 inflammation. This finding aligns with prior reports of a Th2-skewed cytokine milieu and type 2-associated transcriptional programs in EoE mucosa [39,40], and with contemporary reviews that characterize EoE as an antigen-driven, immune-mediated disease capable of tissue remodeling [41,42]. Importantly, by remapping previously reported T-cell subsets to our refined annotations, we demonstrate that tissue T-cell heterogeneity in EoE is most accurately resolved at single-cell resolution—a level of granularity that allows disease-associated effector Th2 programs to emerge as discrete, interpretable cellular clusters.

Our analysis revealed a prominent expansion of an NKT-like cytotoxic T-cell subset (characterized by high expression of GZMB, CCL5, CCL4, CCL4L2, and ITGA1) in the esophageal tissue, suggesting a significant contribution of innate-like cytotoxic effector programs to the local inflammatory microenvironment. Although scRNA-seq-defined NKT clusters may include heterogeneous populations (such as NK-like cytotoxic T cells), canonical NKT cells are rapid responders with essential roles in mucosal immunity and immunoregulation [43,44]. The observed relative increase in Treg proportion in active EoE warrants cautious interpretation given conflicting reports. While our findings are consistent with studies that reported elevated Treg frequencies in the EoE mucosa [45,46], other cohorts have described reduced Treg abundance or a Th17-skewed balance in active disease [47]. These discrepancies may stem from differences in anatomical sampling location (proximal versus distal esophagus) and biopsy depth, reflecting regional heterogeneity in immune infiltration; methodological variation in Treg phenotyping and the lack of functional characterization; heterogeneity in patient demographics, atopic comorbidities, and prior treatments, which can influence Treg/Th17 dynamics; and variation in disease duration and fibrotic burden, since Treg numbers have been correlated with disease chronicity. In our dataset, Treg and Th17 proportions in remission samples were broadly comparable to those in normal tissue, suggesting that the Treg increase in active EoE may represent a state-dependent compensatory response rather than a persistent alteration. Future studies incorporating functional assays and standardized immunophenotyping will be essential to reconcile these observations. In parallel, we identified a condition-dependent shift in the balance between regulatory T (Treg) and T helper 17 (Th17) cells, with Active EoE exhibiting a relative increase in the Treg proportion.

This is consistent with previous studies documenting elevated frequencies of regulatory T cells in EoE mucosa, potentially indicative of a compensatory immunoregulatory response to ongoing inflammation [48]. Finally, the downregulation of CCL5 in NKT-like cells during remission is consistent with the established role of this chemokine in driving the activation and recruitment of NK and NK-like cell populations [37], supporting its functional link to disease activity.

Beyond shifts in cellular composition, cell type–specific differential expression analysis revealed distinct inflammatory signatures associated with each disease state [49]. In NKT-like cells, genes upregulated specifically in Active EoE were enriched for cytokine activity and cytokine-receptor binding, supporting a model wherein cytotoxic/innate-like T cells adopt a more pronounced cytokine-mediated effector profile during active inflammation. In contrast, differential expression in regulatory T (Treg) cells across conditions highlighted functions related to demethylase activity and transcriptional regulation, suggesting that epigenetic remodeling may represent a key mechanism influencing Treg phenotype and function in the inflamed esophageal microenvironment [50]. Epigenetic control is fundamental to Treg stability and suppressive capacity, with FOXP3 locus demethylation being directly linked to the establishment of durable regulatory programs [51]. Furthermore, epigenetic regulators such as Jmjd3 are known to modulate T-cell differentiation through dynamic histone methylation [52]. As Treg-targeted immunotherapies are increasingly explored in immune-mediated disorders [53], our findings underscore the need to further investigate how EoE disease activity correlates with specific Treg epigenetic landscapes and functional potency, which may inform future therapeutic strategies.

Our cell-cell communication analysis further indicates that disease activity is reflected not merely by cellular composition, but also by dynamic shifts in intercellular signaling networks. By applying ligand–receptor inference frameworks routinely used for scRNA-seq data [35], we quantified signaling activity across conditions and found that while several pathways—including MHC-, IL2-, and CCL-related signaling—were broadly active, others such as IL16, IL10, TRAIL, and PECAM1-mediated interactions were preferentially heightened in Active EoE. Notably, differences in IL2 signaling were not attributable to marked changes in ligand or receptor expression within shared cell subsets, but rather to the emergence of Active EoE-specific Naive T and Th2 populations that reconfigured network connectivity. This observation aligns with the established biology of IL-2, which plays a central role in T-cell differentiation and immune homeostasis, particularly in maintaining Treg function while also promoting effector responses [54], and is supported by recent evidence of autocrine IL-2 circuits during T-cell differentiation [55]. Furthermore, the pronounced COLLAGEN-related outgoing signals from Th2 cells underscores a potential link between type 2 inflammation and tissue remodeling in EoE, consistent with prior reports that T cells can induce pro-inflammatory phenotypes in esophageal fibroblasts [56].

Although our analysis reveals robust correlative associations between IL-2 network rewiring and disease activity, we emphasize that the current data cannot establish causality. The observed differences in IL-2 signaling are derived from transcriptomic inference and reflect changes in ligand/receptor expression and cellular composition, but functional validation, such as IL-2 neutralization or receptor blockade experiments, phospho-flow cytometry to assess downstream STAT5 signaling, or Treg suppression assays, will be required to determine whether IL-2 pathway alterations are drivers of inflammation or secondary correlates. Similarly, the predicted Th2-derived COLLAGEN signals, while intriguing, require spatial transcriptomic or histologic confirmation before attributing direct remodeling functions.

In conclusion, inferCNV analysis revealed negligible somatic copy-number variation, underscoring that the observed transcriptional heterogeneity is consistent with inflammatory immune activation rather than neoplastic clonal expansion. This distinction is conceptually important, as single-cell studies in oncology often rely on CNV patterns for cellular stratification. Our finding aligns with the established view of EoE as an antigen-driven, fibro-stenotic disease characterized by epithelial barrier dysfunction and tissue remodeling, not malignant transformation [42,57]. Furthermore, previous bulk transcriptomic studies have demonstrated the substantial reversibility of EoE signatures upon treatment [58], which resonates with our observation that remission samples exhibit a partial normalization in cellular composition while retaining functionally distinct, “rewired” states in specific T-cell subsets. Several limitations of our study warrant consideration. First, functional interpretations rely on transcriptional proxies (e.g., cytokine-associated genes) without direct protein-level validation. Second, the phenotypic classification of NK-like cytotoxic T-cell clusters as “NKT” carries inherent ambiguity. Third, the inferred ligand–receptor interactions, while informative, remain computational predictions. Future validation using flow cytometry, spatial transcriptomics to visualize predicted cellular neighborhoods, and functional assays evaluating Treg suppressive capacity and IL-2-mediated signaling will be essential to substantiate these mechanistic insights and advance their translational relevance for EoE.

CRediT authorship contribution statement

Jianghe Cui: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing-original draft, Writing-review & editing. Longmei Yu: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing-original draft, Writing-review & editing. Rui Wang: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing-original draft, Writing-review & editing. Yifang Gui: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing-original draft, Writing-review & editing.

Authorship

• (1)The conception and design of the study, design/methodology of the study or acquisition of data, or formal analysis and interpretation of data, AND
• (2)Drafting the article or revising it critically for important intellectual content, AND
• (3)Final approval of the version to be submitted, AND
• (4)Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Declaration of competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

Data will be made available on request.