T Cell Plasticity in Systemic Lupus Erythematosus Revealed by Large‐Scale T Cell Receptor Repertoire and Transcriptome Studies

Abstract

Objective

We aimed to characterize CD4⁺ T cell plasticity in human systemic lupus erythematosus (SLE) by leveraging T cell receptor (TCR) repertoire features as markers of prior lineage states, integrating TCR and transcriptomic profiling to delineate plasticity patterns and evaluate their association with clinical disease activity.

Methods

We used TCR repertoire data as molecular signatures alongside a transcriptomic data set. Using a large‐scale ImmuNexUT database of patients with autoimmune disease, including 117 with SLE, we quantified T cell plasticity across 13 fine‐grained T cell types. We analyzed 6,392 samples in total. We defined “cell type” and “disease” signatures and evaluated plasticity by correlations between these signatures and by within‐donor TCR clonotype overlap. Replication was performed in independent bulk and single‐cell cohorts.

Results

We identified two orthogonal signatures of the repertoire and transcriptome, the cell type and disease signatures, allowing us to investigate CD4⁺ T cell plasticity comprehensively. Among all possible patterns, the strongest signal was observed between effector Treg cells (eTreg cells) and Th1 cells, and this was replicated in an independent cohort. SLE Th1 cells exhibited Treg‐like TCR features and transcriptomic profiles, and eTreg cells showed increased clonotype sharing with Th1 cells compared with healthy controls. The Th1 “Tregness” score positively correlated with SLE disease activity.

Conclusion

Our study identifies a Treg‐associated Th1 state in human SLE, consistent with Treg cell to Th1 cell plasticity.

INTRODUCTION

Systemic lupus erythematosus (SLE) is a chronic autoimmune disease marked by a myriad of clinical symptoms and autoantibodies. ¹ Many lines of evidence support the important role of CD4⁺ T cells in SLE pathogenesis. CD4⁺ T cells guide B cells to produce autoantibodies, ² , ³ infiltrate into multiple tissues, ³ , ⁴ , ⁵ show disrupted abundance in peripheral blood, ² , ³ , ⁶ , ⁷ , ⁸ and produce dysregulated inflammatory cytokines. ⁷ , ⁹ B cell–directed cellular therapies (chimeric antigen receptor T cell therapies and T cell engagers) achieve high response rates; however, relapses occur in a subset of patients. ¹⁰ , ¹¹ , ¹² Although these approaches effectively reset B cell repertoires, their impact on T cell receptor (TCR) repertoires appears limited. ¹³

Human CD4⁺ T cells include transcriptionally and functionally distinct cell types, such as Th1, Th2, Th17, follicular helper T (Tfh), and Treg cells. ¹⁴ , ¹⁵ In addition to CD45RA⁺CD25^mid resting Treg cells (Fr. I) and CD45RA⁻CD25^high activated Treg cells (Fr. II), a subset of CD45RA⁻CD25^mid activated Treg‐like cells with low FOXP3 expression levels (named Fr. III) has been described. ⁶ Fr. III T cells are not immunosuppressive and produce more interleukin‐17 than other CD4⁺ T cell types. ⁶ Intriguingly, the transcriptional state of each T cell is not fixed but rather plastic. For example, Treg cells transition into effector Th17 or Th1 cells in mice under experimental inflammatory conditions. ¹⁶ , ¹⁷ This shift involves the loss of the characteristic Treg cell FOXP3 expression and immunosuppressive function. In patients with untreated multiple sclerosis, an increase in Th1‐like, interferon‐γ–secreting FOXP3⁺ T cells with reduced in vitro suppressive activity has been reported. ¹⁸ Nonetheless, there are scarce data about Treg cell plasticity in patients with SLE.

The largest obstacle to investigating human T cell plasticity is the lack of a reporter system that can track previous cellular identities, such as the FOXP3‐GFP marker in mice. ¹⁶ , ¹⁷ However, when a T cell proliferates, its unique TCR nucleotide sequence is replicated in daughter cells. Thus, the TCR sequence, a product of stochastic gene recombination, is a unique molecular “fingerprint” that can track T cells and their progeny through clonal proliferation. We and others have observed that T cell types have characteristic TCR sequence patterns; for example, CDR3 amino acid hydrophobicity is a major distinguishing characteristic of Treg cells. ¹⁹ This TCR feature may contribute to Treg cell fate, given the stochastic nature of TCR gene recombination before Treg cell differentiation. Although T cell types are typically inferred through transcriptional data, the T cell transcriptome is not inherently constant due to T cell plasticity. Therefore, TCR and transcriptome data together provide complementary evidence of previous T cell states. In the therapeutic context, scalable assessment of T cell plasticity abnormalities is needed to refine next‐generation therapies.

In this study, we sought to obtain a global picture of the T cell plasticity in patients with SLE. To do so, we revisited the ImmuNexUT (Immune Cell Gene Expression Atlas from the University of Tokyo) database: a large‐scale bulk RNA sequencing (RNA‐seq) database of fine‐grained immune cell types. ²⁰ , ²¹ We identified TCR and transcriptome profiles for each T cell type, quantified TCR sequence overlap between T cell types, and examined how these results differed in patients with SLE compared to healthy controls (HCs) (Figure 1, Supplementary Table 1).

Figure 1.

Study overview. (A) We recruited 6,392 T cell RNA‐seq samples from 568 individuals, including 134 healthy controls (HCs) and 117 patients with systemic lupus erythematosus (SLE), and (B) performed T cell receptor (TCR) repertoire and transcriptome analyses. In the TCR analysis, we selected 159 TCR features encompassing both V and J gene usage and CDR3 region amino acid usage. (C) We characterized the cell type and disease signatures. Cell type signatures are specific to each T cell type and were calculated using only HC samples. Disease signatures are case‐control signatures of the patients with SLE in each T cell type. (D) We aimed to investigate the plasticity of CD4⁺ T cells through three approaches: (i) unsupervised canonical correlation analysis of TCR and transcriptome data sets, (ii) supervised correlation analysis of the two identified signatures, and (iii) TCR clonotype overlap analysis. (iv) We further evaluated the clinical relevance of the identified plasticity‐related state using the Th1 Tregness score. (E) The signatures and clinical association were externally validated using public data sets. T cell type abbreviations are summarized in Supplementary Table 3. eTreg, effector Treg cells; nTreg, naive Treg cells; scRNA‐seq, single‐cell RNA sequencing; SLEDAI‐2K, Systemic Lupus Erythematosus Disease Activity Index 2000; Tfh, follicular helper T cells.

MATERIALS AND METHODS

We analyzed bulk RNA‐seq data from the ImmuNexUT cohort, which included 117 patients with SLE, 317 patients with other immune‐mediated diseases (IMDs), and 134 HCs. Peripheral blood mononuclear cells were sorted into fine‐grained CD4⁺ and CD8⁺ T cell types according to the standardized ImmuNexUT framework, and RNA‐seq libraries were generated as previously described. ²¹ After quality control, 6,392 T cell samples from 568 donors were included. TCR α‐ and β‐chain sequences were extracted from RNA‐seq reads using MiXCR, and sample‐level TCR repertoire features were defined by V/J gene usage and CDR3 amino acid usage.

To characterize shared structure between TCR repertoire and transcriptome data, we performed regularized canonical correlation analysis (CCA). Cell type signatures were estimated in HCs using linear mixed models comparing each T cell type with the remaining T cell types, adjusting for age, sex, batch, and donor effects. Disease signatures were estimated by comparing each patient with IMD with HCs within each T cell type, adjusting for age, sex, and batch; HLA alleles associated with TCR features were additionally included as covariates in TCR disease signature analyses.

Plasticity‐related relationships were evaluated by correlations between cell type and disease signatures and by within‐donor TCR clonotype overlap between CD4⁺ T cell subsets. TCR clonotype overlap was modeled using generalized linear mixed models. A transcriptome‐based Th1 Tregness score was calculated from the Fr. II effector Treg (eTreg) cell type signature and tested for association with the Systemic Lupus Erythematosus Disease Activity Index 2000 (SLEDAI‐2K) score and clinical manifestations. Replication analyses were performed using independent bulk RNA‐seq and public single‐cell data sets. All statistical tests were corrected for multiple testing where appropriate. Additional details are provided in the Supplementary Methods.

The cell type and SLE disease signature data are provided as tables in the Supplementary Material. The data sets generated during this study are available at the National Bioscience Database Center with the study accession code hum0214 and the data set code JGAD000903. We used publicly available software for the analyses. Custom codes are available at https://github.com/YNagafuchi/ImmuNexUT_TCR.

RESULTS

Covariations between TCR and transcriptome data sets

ImmuNexUT has 117 patients with SLE, 317 patients with nine other IMDs, and 134 HCs (N = 568 in total; Supplementary Table 2). We sorted nine CD4⁺ T cell types and four CD8⁺ T cell types, targeting 5,000 cells for each sample, and conducted RNA‐seq (Figure 1A, Supplementary Table 3). In sorting the effector CD4⁺ T cell types, CD25⁺ Treg cells (Fr. I and Fr. II) and Treg‐like Fr. III cells were excluded, whereas the remaining memory and effector CD4⁺ T cells were designated as “Mem CD4.” This gating strategy excluded not only canonical CD25⁺ Treg cells but also the FOXP3^low CD45RA⁻ Fr. III population, which broadly corresponds to cells with intermediate rather than high CD25 expression. These cells were further subdivided into Th1, Th2, Th17, and Tfh cells based on canonical surface markers: CXCR3 for Th1 cells, CCR4 for Th2 cells, CCR6 for Th17 cells, and CXCR5 for Tfh cells. The analyzed CD4⁺ and CD8⁺ subsets were defined according to the standardized ImmuNexUT sorting framework described previously. ²¹ Following quality controls, 6,392 samples from 568 donors remained in total, and we prepared TCR and transcriptome data for the downstream analyses (see details in the Supplementary Note). We extracted RNA‐seq reads covering TCR messenger RNA and successfully identified TCR amino acid sequences from each sample. On average, we found 876 and 1,034 distinct functional clonotypes from alpha and beta chains, respectively. Considering that we sorted at most 5,000 cells ie, the theoretical maximum number of unique clonotypes assuming no clonal expansion), our pipeline has an efficient coverage of the TCR repertoire. From alpha and beta chains, we extracted 159 TCR repertoire sequence features: 33 CDR3 middle position amino acid frequencies and 69 V and 57 J gene usage frequencies ²² (Figure 1B, Materials and Methods).

We recently reported a substantial covariation between paired single‐cell TCR and transcriptome data sets, ²² and we first sought to replicate these findings in our bulk data sets. We conducted regularized CCA using a paired data set of TCR and transcriptome features (Figure 2A–D). CCA is a dimension‐reduction method that quantifies the covariation between two data sets and accordingly projects each data set into low‐dimensional space (ie, canonical variates [CVs]). CCA is clearly different from principal components analysis (PCA) (Supplementary Figure 1), another popular dimension‐reduction strategy, quantifying the variation of a single data set. We only included HC samples to avoid any influences from IMD disease status (1,366 samples from 134 donors, “inter–cell type” CCA). We conducted CCA in three conditions, including all T cells, CD4⁺ T cells, or CD8⁺ T cell samples. Strikingly, CCA found substantial covariations between TCR and transcriptome data sets (Figure 2E and F, Supplementary Figure 2A, Supplementary Methods). Compared with the principal component (PC) scores calculated by TCR or transcriptome data separately, the CVs offered clear advantages for separating T cell types. In TCR data analysis, CVs yielded much finer separation of cell types than PCs, suggesting that TCR data alone captures some cell type information but benefits from additional transcriptome data reflecting comprehensive cellular biologic processes (Figure 2B–D, Supplementary Figure 1). Similarly, transcriptome data also benefited from integrating TCR information. As a notable example, the separation of Th2 cells was much more distinct in CVs than in PCs, suggesting that TCR data augmented cell type signatures that are hard to purify using transcriptome data alone. To quantify the separation of the T cell types in the CCA space, we applied weighted variance partitioning analysis (Supplementary Methods). Cell type labels explained a substantial proportion of the overall variance in CCA: 67.5%, 28.9%, and 33.2% for all T cells, CD4⁺ T cell samples, and CD8⁺ T cell samples (Figure 2G). These results demonstrated that TCR and transcriptome data sets possess shared cell type signatures, mutually complementing each other. We provide further investigation on the biologic processes inferred from PCA and CCA in the Supplementary Note.

Figure 2.

Canonical correlation of the T cell receptor (TCR) and transcriptome data sets. (A) Schematic representation of canonical correlation analysis (CCA). CCA seeks to re‐express the TCR and transcriptome data sets as pairs of canonical variates (CV) that are highly correlated with each other. The resulting CV is composed of the weighted sum of the original variables by the loading vectors. (B–D) Inter–cell type CCA of TCR and transcriptome data sets was performed using all healthy control samples (B) and separately for CD4⁺ T cell (C) and CD8⁺ T cell (D) samples. The first two TCR CVs of the samples are depicted. (E) Canonical correlation of the first three TCR and transcriptome CVs. (F) Representative canonical correlation of TCR CV1 and transcriptome CV1. (G) Weighted variance partitioning analysis of the inter–cell type CCA. The explained variance under each of the three analysis conditions is depicted. Error bars indicate the 95% confidence interval of the jackknife resampling. Together, these analyses show that paired TCR and transcriptome data capture shared cell type structure and improve separation of T cell types compared with either data set alone. eTreg, effector Treg cells; Mem CD4, memory CD4 T cells; nTreg, naive Treg cells; Tfh, follicular helper T cells. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.70218/abstract.

Cell type signatures of TCR features and transcriptome separating CD4 ⁺ T cell type

We next took a supervised approach to infer cell type signatures of TCR features and transcriptome in CD4⁺ T cells. As in the CCA, we only included HC samples to avoid any influences from IMD disease status. In both TCR and transcriptome analyses, we applied a linear mixed model analysis of specific T cell types, such as Fr. II eTreg cells, compared to all other CD4⁺ T cell types while conditioning for the effects of age, sex, and donor batch effects. This model showed improved power compared with the conventional linear models, and the permutation test confirmed well‐calibrated P values (Supplementary Figure 3A–F).

In the TCR feature analysis, we found all TCR features were associated with at least one cell type: 81 features for each cell type on average and all 159 features for at least one cell type (n = 159 features × 13 cell types, 2,067 tests in total, false discovery rate (FDR) < 0.05; Figure 3A; Supplementary Table 4). The results were well replicated in an independent data set we reported recently ²³ (Supplementary Figure 4A–E). The Takeshima et al ²³ validation data set was generated as a separate study and was independent at the sample level, although limited patient‐level overlap cannot be excluded. In alignment with the inter–cell type CCA that captured the Treg signatures on the CV1 axis and the Th2 signatures on the CV2 axis, we observed many TCR features associated with Treg cells and Th2 cells. For example, CDR3 β chains of the Fr. II eTreg cells showed higher frequency of hydrophobic amino acids than other CD4⁺ T cell types, consistent with our previous report ¹⁹ (Figure 3B, E, and F). We further confirmed the increased hydrophobicity also in α chains. In Th2 analysis, basic amino acids were more preferentially used than other effector cell types, whereas vice versa for acidic amino acids (Figure 3C–E and G; Supplementary Figure 5). Interestingly, Th2 cells also showed the increased CDR3 hydrophobicity (Figure 3C, E, and F). These results suggest that the physicochemical amino acid features of TCR CDR3, shaped through random recombination during thymic development, likely affect a T cell's propensity for Treg and Th2 cell differentiation.

Figure 3.

Cell type signatures of CD4⁺ T cell types. (A) Number of significant cell type TCR signatures in CD4⁺ T cell types (FDR < 0.05). TCR features were classified by TCR α and β chains and by feature category, including V gene usage, J gene usage, and CDR3 amino acid usage. (B–D) TCR CDR3 amino acid usage signatures across T cell types. Amino acids were grouped by physicochemical properties, and signature betas were compared between TCR α and β chains using Pearson correlation coefficients. (E) Heatmap of TCR β‐chain CDR3 amino acid usage across CD4⁺ T cell types. Amino acid usage was normalized, and Z scores were hierarchically clustered by T cell type. (F and G) Representative boxplots of TCR β‐chain CDR3 amino acid usage. *FDR < 0.05 in the cell type signature analysis. (H) Heatmap of the strongest cell type signature genes across CD4⁺ T cell types. Expression of the top five genes was normalized and displayed as Z scores. Top genes for memory CD4 T cells (Mem CD4), a mixture of effector CD4⁺ T cell types, were excluded. (I and J) Representative boxplots of FOXP3 and NELL2 expression. P values were derived from the cell type signature analysis. CPM, counts per million; eTreg, effector Treg cells; FDR, false discovery rate; nTreg, naive Treg cells; TCR, T cell receptor; Tfh, follicular helper T cells. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.70218/abstract.

In the transcriptome analysis, we identified 9,753 transcriptome signature genes for each T cell type on average (n = 13,483 genes × 13 cell types, 175,279 tests in total, 72% of the tested genes at FDR < 0.05; Supplementary Figure 3G; Supplementary Table 5). Again, we confirmed an excellent replicability in an independent cohort (Supplementary Figure 4A–D). This analysis validated our sorting strategy: the expression of well‐established marker genes was mostly restricted to the corresponding cell types, for example, CCL5 and GZMK for Th1 cells; SEMA5A and PTGDR2 for Th2 cells; RORC and CCR6 for Th17 cells; CXCR5 and RASSF6 for Tfh cells; and FOXP3, TRIB1, and TNFRSF9 for Treg cell types, aligning with prior literature findings ²⁴ , ²⁵ (Figure 3H and I). This analysis also identified differentially expressed genes separating closely related cell types, for example, down‐regulated NELL2 differentiated Fr. II eTreg cells from the other two Treg or Treg‐like cell types (Figure 3J). To explore the dysregulated biologic processes in SLE T cells, we conducted gene set enrichment analysis (GSEA). This analysis showed that transcriptional pathways, such as oxidative phosphorylation and MTORC1 signaling, are up‐regulated in Fr. II eTreg cells, in alignment with literature findings ²⁶ , ²⁷ (Supplementary Figure 6A).

We finally sought to confirm that TCR and transcriptome cell type signatures reflected common inter–cell type similarities, as we demonstrated in CCA. For this aim, we assessed the inter–cell type correlation of both signatures (Supplementary Figure 6B). Globally, both signatures showed similar correlation patterns. For example, the signatures of Fr. III T cells showed a tight correlation with that of Fr. II eTreg cells (ρ = 0.83, P < 2.2 × 10⁻¹⁶ in the TCR analysis; ρ = 0.90, P < 2.2 × 10⁻¹⁶ in the transcriptome analysis). However, we also noted several distinct correlation patterns. A notable example was that Fr. I naive Treg (nTreg) cells, a subpopulation of Treg cells possessing a naive phenotype, shared transcriptome signatures with naive CD4⁺, more than other Treg or Treg‐like cell types. In contrast, Fr. I nTreg cells clustered with other Treg cell types in TCR features. Considering that the Treg cell is a distinct T cell type developed through a unique thymic selection process, TCR features may provide more reliable information regarding T cell identity.

To further validate our cell type TCR signatures, we independently estimated them using the public single‐cell RNA‐seq data set from the COMBAT consortium data. We detected distinct clusters of T cell types (Supplementary Figure 7A and B) and estimated cell type–specific TCR signatures using a pseudo‐bulk approach (Supplementary Methods). Although ImmuNexUT and COMBAT consortium data were generated using completely different experimental and analytical pipelines, both cell type TCR signatures were substantially correlated with each other (Supplementary Figure 7C). These findings further confirmed the robustness of our signatures.

Disease signatures separating patients with IMDs and HCs in CD4 ⁺ T cells

We continued a supervised approach to infer “disease signatures” based on transcriptome and TCR features (Figure 1C; Supplementary Methods). In this analysis, we applied a linear model to compare each patient with IMD with HCs within each T cell type, for example, we compared SLE Th1 cells with HC Th1 cells to obtain the SLE Th1 disease signature. We controlled the effects of age, sex, and study batch (Supplementary Methods). Importantly, HLA alleles can be a strong confounding factor in TCR analysis because they correlate with both IMDs’ susceptibility and TCR repertoire sequences ²⁸ , ²⁹ (Supplementary Note). Therefore, we additionally included HLA alleles as covariates.

In the transcriptome analysis for SLE, we detected an average of 7,022 genes with differential expression, constituting 52% of the tested genes at FDR < 0.05 (n = 13,483 genes × 13 cell types, 175,279 tests in total; Figure 4A, Supplementary Figure 8A, Supplementary Table 6). The transcriptome signatures of SLE were replicated very well in the independent replication data set (Supplementary Figure 4F and G). SLE Th1 cells showed the most pronounced dysregulation in the transcriptome (Figure 4A). Variance partitioning analysis confirmed that the disease condition had the strongest impact on Th1 cells (Supplementary Note). For example, the expression of interferon signaling genes, particularly IFI27 and RSAD2, exhibited the most substantial up‐regulation in SLE Th1 cells (Figure 4B). Additionally, Th1 cell data showed a noteworthy increase in FOXP3 expression within the top one percentile of the disease signatures (Figure 4B, Supplementary Table 6). In contrast, IL2RA expression did not show a corresponding increase, which is evidence against the possibility that this signal is explained simply by contamination with canonical CD25‐expressing Treg cell populations or the previously defined FOXP3^low/CD25^mid Fr. III population (Supplementary Table 6). To explore the dysregulated biologic processes in SLE T cells, we conducted GSEA. As expected, this analysis revealed up‐regulated interferon signaling pathways in many cell types (Figure 4C, Supplementary Figure 9). In addition, multiple other pathways showed dysregulation in SLE, for example, oxidative phosphorylation and MTORC1 signaling genes, which were pathways up‐regulated in Treg cells (Supplementary Figure 6A), exhibited the strongest enrichment in SLE Th1 cells.

Figure 4.

SLE disease signatures. (A) The number of significant transcriptomic SLE disease signatures in CD4⁺ T cell types (at FDR < 0.05). (B and D) The disease signature of SLE Th1 cells by transcriptome analysis (B) and TCR analysis (D). Genes with an FDR <0.05 are noted in red. (C) A heatmap showing the NES of selected pathways from the SLE disease signature gene set enrichment analysis. (E and F) Representative boxplots showing the TCR gene usage of TRBV18 and TRAV13‐2 in Th1 cells. *FDR < 0.05 in the SLE disease signature analysis. eTreg, effector Treg cells; FDR, false discovery rate; HC, healthy controls; Mem CD4, memory CD4 T cells; NES, normalized enrichment score; nTreg, naive Treg cells; SLE, systemic lupus erythematosus; TCR, T cell receptor; Tfh, follicular helper T cells. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.70218/abstract.

In the TCR analysis for SLE, we detected on average 10.5 TCR features with differential usage for SLE, representing 6.6% of the tested features at FDR < 0.05 (n = 159 features × 13 cell types, 2,067 tests in total; Figure 4D, Supplementary Figure 8B, Supplementary Table 7). Although this analysis had only moderate statistical power, the results were replicated well in an independent replication data set (Supplementary Figure 4F and H). The TCR repertoire of SLE Th1 cells demonstrated increased use of TRBV18 and decreased use of TRAV13‐2 (Figure 4D–F). For other IMDs, the number of differentially expressed TCR features was notably limited, and hence we restricted our analysis to SLE disease signatures for the downstream analyses (Supplementary Figure 8).

To assess the effect of sample size on the number of disease‐associated TCR features, we performed an additional sensitivity analysis in which the SLE, idiopathic inflammatory myopathy, and systemic sclerosis cohorts were each repeatedly resampled to 50 cases. We then recalculated the number of significant TCR disease signature features 1,000 times for each disease cohort. Even after sample size matching, SLE retained a significantly greater number of disease‐associated TCR features in naive and effector CD4⁺ T cell subsets, suggesting that the stronger disease‐specific TCR signal in SLE is not explained by power alone (Supplementary Figure 10).

Correlation analysis between cell type and disease signatures

To investigate T cell plasticity in SLE, we tested correlations between CD4⁺ T cells’ cell type and disease signatures. The rationale behind our strategy is that when T cell plasticity is characterized by a transition from cell type A to cell type B, the cell type B disease signature should resemble the cell type A signature. We considered both TCR and transcriptome signatures for this analysis. We found significant positive correlations between cell type signatures of Treg and Treg‐like cell types (Fr. II eTreg cells and Fr. III T cells) and SLE disease signatures of effector CD4⁺ T cells (Th1, Mem CD4, and Tfh cells) in both the TCR and transcriptome spaces (Supplementary Figure 11, Supplementary Table 8). Among all significant correlations of the Fr. II eTreg cell type signature, the SLE Th1 disease signature exhibited the largest correlation in both the TCR and transcriptome analyses (Figure 1D, Figure 5A–C).

Figure 5.

Treg‐associated Th1 state in SLE. (A) Comparison of Spearman's correlation coefficients between transcriptome and TCR analyses in the ImmuNexUT data set. The x‐axis displays Spearman's correlation coefficients with the 99% confidence interval, focusing on the correlation between the transcriptome Fr. II eTreg cell type signature and the SLE disease signature in CD4⁺ T cell types. The y‐axis represents the TCR counterparts. (B and C) The correlation between the Fr. II eTreg cell type signature and the SLE disease signature using transcriptome (B) and TCR data sets (C) from ImmuNexUT. The gray bar in each dot indicates ±2 SE ranges. (D) Comparison of Spearman's correlation coefficients between transcriptome and TCR analyses in the Takeshima et al data set. ²³ (E) The odds ratio of TCR clonotype overlap with Fr. II eTreg cells in patients with SLE in comparison to HCs in CD4⁺ T cell types based on logistic mixed effect analysis. Error bars indicated 95% confidence intervals. Red color indicates P values <0.05. *P < 0.05; ***P < 0.001; and ****P < 0.0001. Together, these analyses provide complementary transcriptomic, TCR, and clonotype‐based support for a Treg‐associated Th1 state in SLE. eTreg, effector Treg; HC, healthy controls; Mem CD4, memory CD4 T cells; nTreg, naive Treg cells; SLE, systemic lupus erythematosus; TCR, T cell receptor; Tfh, follicular helper T cells. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.70218/abstract.

To test the replicability of our findings, we repeated the identical analysis in the replication cohort and confirmed the strongest correlation between Fr. II eTreg cell type signature and the SLE Th1 disease signature in both the transcriptome and TCR analyses (Figure 5D, Supplementary Figure 4I–K). Importantly, we observed a much weaker correlation in the opposite direction, specifically between the Th1 cell type signature and the SLE Fr. II eTreg disease signature, in both the TCR and transcriptome analyses (Supplementary Table 8). Together, these results mean that effector CD4⁺ T cells in SLE possess Treg‐like TCR and transcriptome features and are consistent with, but do not directly demonstrate, Treg cell to effector CD4⁺ T cell lineage conversion in SLE.

Shared TCR clonotypes between SLE cell types

To obtain additional evidence relevant to plasticity between CD4⁺ T cell types in SLE, we quantified the TCR clonotype overlap between CD4⁺ T cell types in each donor (Supplementary Methods). We first restricted our analysis to HC samples and asked whether closely related T cell types exhibited more TCR clonotype overlap than unrelated cell types. As expected, we observed higher TCR clonotype overlap between Fr. II Treg cells and Fr. III T cells and between effector CD4⁺ T cell types (Supplementary Figure 12A). This finding was consistent with our previous comparison between cell type signatures (Supplementary Figure 6B). We next performed the same analysis in SLE samples and found a similar overall pattern of TCR clonotype overlap across CD4⁺ T cell types (Supplementary Figure 12B).

To assess the significance and magnitude of TCR clonotype overlap in SLE, we applied a generalized linear mixed model approach in which we treated each TCR clonotype as an independent observation and modeled the clonotype overlap as a binary event (Supplementary Methods). We first focused on Fr. II eTreg cells highlighted in the signature correlation analysis. In patients with SLE, Fr. II eTreg cells showed significantly more frequent overlap with effector CD4⁺ T cell types compared with HCs (Mem CD4, Th1, Th2, and Tfh cells; Figure 5E). Among the effector cell types, Th1 and Th2 cells showed the most frequent overlap (odds ratio [OR] 3.1, P = 9.1 × 10⁻⁶ in Th1 cells; OR 3.1, P = 2.3 × 10⁻¹⁵ in Th2 cells; Figure 5E). Enhanced TCR overlaps between Fr. II eTreg, Th1, and Th2 cells were bidirectional, showing increased overlap both from Fr. II eTreg cells to Th1 cells and from Th1 cells to Fr. II eTreg cells (Supplementary Figure 12C). Hence, the analysis of TCR clonotype overlap also provided additional support for a close relationship between Fr. II eTreg cells and effector CD4⁺ T cells, including Th1 cells.

Clinical relevance of Th1 Tregness in SLE

The aforementioned analyses identified the Tregness of SLE Th1 cells, possibly reflecting Fr. II eTreg cell to Th1 cell plasticity in SLE. We next sought to understand the clinical relevance of this Treg‐associated Th1 state. For this purpose, we first developed a Treg score based on the transcriptome Fr. II eTreg cell type signature calculated by the ImmuNexUT HC samples (Supplementary Methods). We applied the Tregness scoring system to the validation cohort and confirmed its robustness (Figure 6A and B). We then evaluated the SLE‐associated Tregness of each cell type by comparing the score of patients with SLE and HCs. Consistent with the results in Figure 6, Tregness was prominent in Th1 cells and other effector CD4⁺ T cell types in both the ImmuNexUT and the replication data sets.

Figure 6.

Treg score and SLE disease activity. (A and B) Treg score in CD4⁺ T cell types from HCs in the ImmuNexUT discovery cohort (A) and the Takeshima et al validation cohort ²³ (B). The difference in Treg scores between HCs and patients with SLE was tested with Student's t‐tests. *P < 0.05; ****P < 0.0001. On the right, the delta difference in the mean Treg score between patients with SLE and HCs is shown. (C) Spearman's correlation coefficients with the 99% confidence intervals between the Treg score in each CD4⁺ T cell type and the SLEDAI‐2K score in the ImmuNexUT cohort. (D and E) The correlation between the Th1 Treg score and the SLEDAI‐2K score was evaluated with Spearman's correlation coefficients in the ImmuNexUT cohort (D) and the Takeshima et al cohort ²³ (E). (F) Boxplots comparing the Th1 Treg scores by the positivity of the SLEDAI‐2K components in the ImmuNexUT data set. Student's t‐test was performed in the comparison of negative and positive groups. Of the 24 SLEDAI‐2K components, only 5 components were significantly different with P values <0.05. These analyses indicate that increased Th1 Tregness is associated with higher SLE disease activity and clinically relevant manifestations. eTreg, effector Treg cells; HC, healthy control; Mem CD4, memory CD4 T cells; ns, not significant; nTreg, naive Treg cells; SLE, systemic lupus erythematosus; SLEDAI‐2K, Systemic Lupus Erythematosus Disease Activity Index 2000; Tfh, follicular helper T cells. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.70218/abstract.

We previously defined a B cell receptor (BCR) repertoire naiveness score in SLE. ³⁰ In the present cohort, the Th1 interferon‐α signature, the Th1 Treg score, and the BCR naiveness score were positively correlated, suggesting a link between interferon signaling, T cell plasticity, and B cell abnormalities in SLE (Supplementary Figure 13A–E). Additionally, the interferon signature levels in other IMDs were associated with the correlation between the Fr. II eTreg cell type signature and the Th1 disease signature of these IMDs (Supplementary Figure 13F). This suggests that the interferon signature may play a role in Treg cell to Th1 cell plasticity in other IMDs as well.

Finally, we evaluated the association between Th1 Tregness and SLE clinical phenotypes. Among all CD4⁺ T cell types, the Th1 Treg score exhibited the strongest association with the SLEDAI‐2K score (Figure 6C). In both cohorts, the Th1 Treg score showed a positive association with the SLEDAI‐2K score, a composite score based on multiple clinical phenotypes, indicating a link to global disease activity (Figure 6D and E). When we investigated each component of clinical phenotypes, the Th1 Treg score was significantly associated with active symptoms such as fever, nephritis, and rash (Figure 6F). Mediation analysis revealed that Th1 Tregness mediated the effects of the interferon signature on the SLEDAI‐2K score (Supplementary Figure 13G and H). These findings highlight a link between T cell plasticity and clinical heterogeneity in SLE and suggest the utility of the Tregness score of Th1 cells as an SLE biomarker.

DISCUSSION

In this study, we thoroughly investigated the cell type and disease signatures of purified T cell types in SLE and other IMDs by combining TCR repertoire and transcriptome analyses. Comprehensive correlation analysis between cell type and disease signatures provided evidence that Th1 cells in SLE possess Treg‐like features, consistent with Treg cell to Th1 cell plasticity in SLE. The TCR clonotype overlap analysis further supports a close relationship between these cell types. We further conducted intensive replication analyses using single‐cell and bulk paired TCR and transcriptome data. We finally demonstrated the substantial correlation of Th1 Tregness with clinically relevant SLE phenotypes and suggested the clinical utility of our finding as a biomarker of SLE in the clinical setting.

To place these findings in the broader context of our study, we next considered how the TCR sequence features associated with specific T cell types may help explain the differentiation‐related structure observed across T cell states. This study extends our previous work reporting TCR cell type signature of Treg cells. ¹⁹ A notable example is the unique Th2 TCR features characterized by hydrophobic and basic CDR3 amino acid compositions, implying a distinct differentiation machinery of Th2 cells compared to other effector CD4⁺ T cell types. Semi‐unsupervised analysis through CCA also supported the existence of unique Treg and Th2 cell type signatures. The TCR repertoire is formed through the random rearrangement in the thymus preceding the encounters with cognate antigens. Therefore, the TCR sequence features we identify may be associated with the probability of thymic or peripheral T cell differentiation, although the precise mechanism remains to be determined. These associations may also reflect antigen‐driven selection, clonal expansion, or survival advantages rather than lineage commitment alone. One possible interpretation is that intrinsic TCR sequence features may influence the probability of Th2 cell differentiation by shaping the quality or intensity of TCR signaling. Alternatively, such features may favor antigen‐recognition contexts associated with Th2‐polarizing priming, although this remains speculative. A similar interpretation may also apply to the separation of CD8 memory subsets, as recent work has shown that intrinsic TCR sequence features can influence the transition from naive to memory T cell states, suggesting that TCR properties may bias the probability of adopting particular CD8 memory or effector‐memory phenotypes. ²²

Transcriptome disease signatures, or differentially expressed genes, of SLE have been extensively researched, revealing overall up‐regulation of interferon signatures and various cell type–specific signatures of the disease. ²⁰ , ³¹ , ³² , ³³ Our study uniquely adds cell type–specific “TCR” disease signatures of SLE, which can serve as a tracking signature for T cell plasticity. In our analyses, we carefully controlled for the effects of the HLA alleles, which can be a major confounder associated with both IMD susceptibility and TCR repertoire patterns. ²⁸ , ²⁹ Even using our stringent approach, we observed a substantial degree of TCR abnormality in SLE. Previous SLE TCR repertoire analyses reported different V and J gene usage in unfractionated whole blood samples. ³⁴ , ³⁵ We successfully extended the investigation of SLE TCR abnormality to fine‐grained T cell types, which shed light on T cell plasticity in SLE.

To date, there has been scarce evidence showing T cell plasticity in human autoimmunity. Our results suggest that the strongest plasticity‐related signal in SLE is observed between Treg‐associated and Th1‐like states. Previous mouse studies have revealed the plasticity of Treg cells to Teff cells using genetically engineered tracking systems of FOXP3 protein expression. ¹⁶ , ¹⁷ These studies showed that ex‐FOXP3 Teff cells were pathogenic in autoimmune disease models. However, other studies using inducible Foxp3 fate‐mapping in mice indicated that the Treg cell lineage is largely stable, ³⁶ , ³⁷ and the extent to which Treg cell plasticity contributes to autoimmune pathogenesis remains debated. In our study, Treg cell type TCR signatures, which are closely linked to Treg cell fate, served as a tracking signature of Treg cells in SLE. We demonstrated the association between Treg‐like Th1 cells, BCR repertoire abnormalities, and disease activity in SLE. Our study provides evidence for a Treg‐associated Th1 state in human SLE that may be pathogenic and is compatible with prior reports of Treg cell plasticity in mice. At the same time, the observed overlap may partly reflect alternative, non–mutually exclusive explanations, including transcriptional convergence under inflammatory conditions, shared antigen‐driven clonal expansion, the relative proximity of activated/eTreg states to Th1‐like effector CD4⁺ T cell subsets, or activation state rather than lineage identity alone. However, because the observed relationship was also supported by TCR repertoire features and clonotype overlap, a purely activation‐based explanation is unlikely to fully account for the observed pattern. Recently, cytotoxic CD4⁺ T cells were proposed to be the human counterpart of ex‐FOXP3 Treg cells. ³⁸ Considering the similarity between Th1 and cytotoxic CD4⁺ T cells, ³⁹ , ⁴⁰ such as high expression of T‐bet and TCR clonotype overlap, the Treg cell to Th1 cell plasticity could be a universal phenomenon in the human immune system.

Our analytical approach to TCR data is unique in that we go beyond the molecular barcode approach to investigate individual components of the TCR sequence, which can be represented with continuous physicochemical values. ⁴¹ This approach is clearly different from a popular approach that treats TCR sequence information as a cellular barcode, focuses only on the small fraction of TCR clonotypes shared between the cell types, and does not analyze the majority of the TCR repertoire. ³⁸ , ⁴² , ⁴³ Our approach provided insight into the asymmetric relationship suggestive of T cell plasticity in SLE. In contrast, conventional approaches relying on clonotype overlap do not provide comparable information on this asymmetry (Supplementary Figure 12C).

Translationally, these data nominate the Th1 Tregness score as a scalable biomarker of T cell plasticity abnormality in SLE. By correlating with interferon‐α signatures and SLEDAI‐2K scores and replicating across cohorts, this metric may support patient stratification and response monitoring, providing a practical basis for refining therapeutic strategies.

Although we discovered multiple novel aspects of SLE T cell biology, we need to acknowledge some critical limitations in this study. First, we could not determine whether the T cell plasticity suggested by our data is causal in SLE pathogenesis or a result of the disease. To address this point, we need future longitudinal studies following TCR, transcriptome, and clinical path. Second, we did not conduct functional investigations to delve into the mechanisms underlying cell type signatures and the proposed plasticity. Third, further exploration is needed in other IMDs to determine whether the plasticity‐related findings observed here are specific to SLE or more prevalent across other IMDs. Fourth, our findings are limited to peripheral blood; tissue‐resident T cells in lymphoid organs or affected tissues may show different characteristics, requiring future integrated studies. Fifth, our gating strategy defined canonical Treg cells by CD25 positivity and excluded CD25⁺ cells from the effector CD4⁺ gates, including the CD45RA⁻CD25^mid Fr. III population. Moreover, increased FOXP3 expression without a corresponding increase in IL2RA expression in SLE Th1 cells is evidence against simple contamination by canonical CD25‐expressing Treg cells. Nevertheless, we cannot fully exclude other CD25^low or CD25⁻ transitional Treg‐related cells, which would still be compatible with Treg cell plasticity accompanied by CD25 down‐regulation. Sixth, although our sorting strategy makes major contamination by innate‐like T cell populations unlikely, we cannot completely exclude a minor contribution from rare CD4⁺ innate‐like populations, such as iNKT cells, particularly in the interpretation of Th2‐associated TCR features. Seventh, the Th1, Th2, Th17, and Tfh cell populations were defined using canonical surface markers as an enrichment strategy, not as strictly lineage‐committed populations. Accordingly, overlap between marker‐defined subsets, particularly in inflammatory conditions such as SLE, may have contributed in part to the observed inter‐subset similarity. Eighth, the current Tregness score is transcriptome‐based and therefore does not fully disentangle lineage‐related identity from activation‐associated transcriptional programs, particularly in SLE, in which strong interferon‐driven activation may influence the observed signatures.

Our study provides a comprehensive picture of plasticity‐related relationships among T cell types from the combined viewpoint of TCR and transcriptome data at an unprecedented scale and sheds new light on their role in the pathogenesis of SLE.

AUTHOR CONTRIBUTIONS

All authors contributed to at least one of the following manuscript preparation roles: conceptualization AND/OR methodology, software, investigation, formal analysis, data curation, visualization, and validation AND drafting or reviewing/editing the final draft. As corresponding authors, Drs Fujio and Ishigaki confirm that all authors have provided the final approval of the version to be published and take responsibility for the affirmations regarding article submission (eg, not under consideration by another journal), the integrity of the data presented, and the statements regarding compliance with institutional review board/Declaration of Helsinki requirements.

ROLE OF THE STUDY SPONSOR

Chugai Pharmaceutical Co, Ltd had no role in the study design or in the collection, analysis, or interpretation of the data, the writing of the manuscript, or the decision to submit the manuscript for publication. Publication of this article was not contingent upon approval by Chugai Pharmaceutical Co, Ltd.

Supporting information

Disclosure Form:

Data S1. Supporting Information.

Data S2. Supplementary Tables.