Single-cell RNA profiling of blood CD4⁺ T cells identifies distinct helper and dysfunctional regulatory clusters in children with SLE

Abstract

To characterize the complexity of the CD4⁺ T cell compartment in patients with systemic lupus erythematosus (SLE), we performed single-cell RNA sequencing of sorted blood CD4⁺ T cells from pediatric patients and healthy donors. We identified naive, memory, regulatory T (T_reg) cell, proliferative and interferon-stimulated gene-high (ISG-high) clusters. Within the memory compartment, both follicular and peripheral helper cells were expanded in patients with lupus nephritis and/or high disease activity. Cytotoxic signatures were enriched in effector memory T cells re-expressing CD45RA (TEMRA), as well as in two memory subclusters, one of which overlapped with T helper 10-like cells (T_H10). Notably, we observed an expansion of dysfunctional T_reg cells in patients with lupus nephritis, along with upregulation of TLR5 and FCRL3 in SLE-naive T_reg cells, suggesting a potential link with mucosal microbial dysbiosis. These findings highlight distinct CD4⁺ T cell subsets that may contribute to aberrant antibody responses and impaired immune regulation in SLE.

SLE is a systemic autoimmune disease characterized by clinical and molecular heterogeneity¹. Pathogenic hallmarks include nucleic-acid-specific autoantibodies and increased interferon (IFN) activity². Childhood-onset SLE accounts for up to 20% of SLE cases, and a higher number of pediatric patients (up to 85%) develop lupus nephritis (LN) compared to adults³.

Follicular helper T (T_FH) cells contribute to autoimmunity in both mice and humans⁴, but extrafollicular helper T cells also drive the expansion of autoreactive B cells⁵. Thus, CXCR5⁻PD-1^hi peripheral helper T (T_PH) cells were identified in both the synovium of patients with rheumatoid arthritis⁵ and blood of adults with SLE⁶. As T_FH cells, these cells recruit and activate B cells through CXCL13 and IL-21 production. In addition, CXCR5⁻CXCR3⁺PD-1^hi cells expressing IFNγ and IL-10 (T_H10) are expanded in the blood of children with SLE and the kidney of patients with LN⁷. Importantly, unlike T regulatory 1 (TR-1) cells⁸, T_H10 cells lack regulatory properties but support differentiation of naive and memory B cells into plasmablasts in vitro⁷.

Dysfunctional effector CD4⁺ T cells could therefore represent a therapeutic opportunity in SLE. To dissect their heterogeneity, we performed single-cell RNA sequencing (scRNA-seq) of sorted blood CD4⁺ T cells from patients with SLE and healthy donors (HD). These analyses mapped T_H10 and T_PH cells within distinct subclusters (SCs), defined unique CD4⁺ cytotoxic programs, and uncovered phenotypic and functional alterations in SLE T_reg cells, especially those from patients with LN.

Results

scRNA-seq of blood CD4⁺ T cells from HD and patients with SLE

CD4⁺ T cells were sorted from peripheral blood mononuclear cells (PBMCs) of 16 children (age 9–17 years, mean = 14 ± 2.5) with SLE, 11 of whom had biopsy-proven LN, along with 10 matched HD (Fig. 1a). Disease activity (DA) was measured using the SLE Disease Activity Index (SLEDAI), and patients were categorized into moderate/high (SLEDAI > 4; n = 6) and low (SLEDAI < = 4; n = 10) groups (Supplementary Table 1). Sorted CD4⁺ T cells were processed using the 10x Genomics Chromium Single Cell 3’ library, followed by whole-transcriptome sequencing (Supplementary Table 2 and Extended Data Fig. 1a–c). After data preprocessing, including cell quality filtering⁹ and batch integration¹⁰, 239,473 CD4⁺ T cells were used for further clustering and phenotyping.

Fig. 1 |. scRNA-seq reveals altered cellular composition of CD4⁺ T cells in SLE.

a, Workflow of scRNA-seq experiment starting from sorted CD4⁺ T cells. Guided clustering analysis was performed on total CD4⁺ T cells, followed by subclustering analysis. b, UMAP of total CD4⁺ T cells clustered into five major populations: naive and memory T cells, T_reg cells, proliferating cells and ISG-high cells. c, Violin plot of expression of selected markers for the five major CD4⁺ T cell clusters. d, Box plots of cell proportions for the five major clusters within total CD4⁺ T cells by disease group (top), LN (middle panel) and DA (bottom panel). The boundaries and center line of the box plot represent upper and lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5× interquartile range (IQR) from the upper and lower quartiles. All data points are overlaid on the box plot. Two-sided t-tests were performed to test differences in cell proportions between groups. e, Heatmap showing the average expression of ISGs and IFN-related genes across the five clusters of CD4⁺ T cells in HD and patients with SLE. The colors of the heatmap indicate mean gene expression. Colors in the top row represent modular annotation of ISGs and IFN-related genes. f, Scatter plot showing correlations between per-sample average expression of ISGs and IFN-related genes in total CD4⁺ T cells and SLEDAI. Pearson correlation coefficient (r) and its corresponding P value were calculated. The error band represents the 95% confidence interval for the fitted regression line. Pearson’s correlation (0.648) and its corresponding two-sided P value (0.00034) were calculated. Single-cell analysis included SLE LN (n = 11), SLE NoLN (without nephritis; n = 5) and HD (n = 10) samples from the sorted CD4⁺ scRNA-seq dataset. *P < 0.05, **P < 0.01, ***P < 0.001.

A multistep subclustering framework was used to perform an initial clustering of total CD4⁺ T cells, followed by subclustering of each main cellular compartment (Fig. 1a)¹¹. The initial total CD4⁺ T cell analysis yielded 15 clusters (Extended Data Fig. 1d), which were merged according to known CD4⁺ T cell markers into five main clusters (Fig. 1b,c and Extended Data Fig. 1e). Their cell identity assignment was independent of 10x batches, disease groups or samples (Extended Data Fig. 1f–h). High CCR7, LEF1 and TCF7 and low S100A4 expression characterized the naive cluster, whereas the opposite pattern identified the memory cluster^12,13 (Fig. 1c). FOXP3, IKZF2 and IL2RA (CD25) expression defined the T_reg cell cluster¹⁴. An additional cluster was defined based on the highest levels of IFN-stimulated gene (ISG) expression (ISG-high)¹⁵. Finally, a small cluster expressing cell cycle genes, such as MKI67 and PCNA, was annotated as ‘proliferating’.

Analysis of cell proportions revealed differences between patients with SLE and HD across the five major clusters, including a reduction in naive cell clusters and an expansion of ISG-high and T_reg cell clusters in SLE. Naive and T_reg cell changes distinguished LN from both HD and non-LN patients (Fig. 1d). Memory and ISG-high clusters separated LN from HD. The frequency of proliferating CD4⁺ T cells was similar in patients with SLE and HD.

We assessed ISG expression using modules of coexpressed transcripts (M1.2, M3.4, M5.12)¹ and IFN-related non-IFN-induced transcripts¹⁵. All SLE versus HD clusters showed increased ISG expression, ranging from mild (naive) to high (ISG-high, proliferating) expression levels (Fig. 1e and Extended Data Fig. 2a,b). The proliferating cluster uniquely showed upregulation of type II IFN genes (for example, genes encoding GBPs and IFNG) and downregulation of SOCS1 (Fig. 1e)¹⁶. ISG overexpression fluctuated across individual SLE samples (Extended Data Fig. 2a), and participant-specific ISG activity was correlated across the five main CD4⁺ clusters (Extended Data Fig. 2c) and with SLEDAI (Pearson’s correlation coefficient r = 0.648, P < 0.001; Fig. 1f and Extended Data Fig. 2d). A trend towards an increase in ISG expression was seen in LN, but this did not reach statistical significance (Extended Data Fig. 2e).

Thus, patients with SLE showed substantial shifts in major blood CD4⁺ T cell population frequencies according to DA and/or LN status. In addition, they showed overexpression of ISGs in correlation with DA rather than with LN.

Subclustering analysis of naive CD4⁺ T cells highlights five distinct cell states

Naive CD4⁺ T cells are more diverse and plastic than previously thought¹⁷. We uncovered six naive (CCR7^hiS100A4^low) SCs (SC0–SC5) (Extended Data Fig. 3a). SC0 to SC4 expressed bona fide naive CD4⁺ T cell transcripts, whereas cells within SC5 expressed higher levels of S100A4, KLRB1, ITGB1 and ANXA1 transcripts, characteristic of central memory T (T_CM) cells¹⁸ (Fig. 2a,b).

Fig. 2 |. Subclustering analysis of naive CD4⁺ T cells reveals five distinct transcriptional states.

a, Matrix plot showing top markers for each SC in naive CD4⁺ T cells. b, Heatmap of selected markers across the six naive and T_CM SCs. c, Heatmap of the naive markers in b, plotted for naive clusters from PBMC validation dataset 1. The colors of the heatmap indicate the average expression of the markers. The row colors give the cluster annotation. NAI, naive. d, Box plots of relative cell proportions within total naive CD4⁺ T cells from the sorted dataset for each SC by disease group (top panel), LN (middle panel) and DA (bottom panel). The boundaries and center line of the box plot represent the upper and lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5× IQR from the upper and lower quartiles. All data points are overlaid on the box plot. Two-sided t-tests were performed to test differences in cell proportions between groups. Single-cell analysis included SLE LN (n = 11), SLE NoLN (n = 5) and HD (n = 10) samples from the sorted CD4⁺ scRNA-seq dataset and additional SLE (n = 10) samples from PBMC validation dataset 1. *P < 0.05, **P < 0.01, ***P < 0.001.

SC0 expressed cytoskeletal and motility genes (ACTG1, ACTB, CORO1A)¹⁹ and CD7, linked to adhesion²⁰ (Fig. 2a,b). SC1 showed upregulation of transcriptional and cytokine signaling genes (MALAT1, ATM, IL6ST). SC2 was enriched for AP-1 complex transcripts, which are key to chromatin opening during T cell activation²¹. SC3 showed upregulation of cell cycle, apoptosis and JAK–STAT transcripts (BCL2, CDK6, PIM1, CISH, SOCS2). SC4 uniquely expressed SOX4, an early-life naive CD4⁺ T cell transcription factor²² also linked to CXCL13 production during inflammation²³.

Subcellular analysis showed that naive CD4⁺ T cell markers were mainly in the nucleoplasm and linked to RNA splicing and transcription (Extended Data Fig. 3b). Naive SCs lacked CD transcripts, except for CD7 and CD69, which was upregulated in SC0 and SC2 (Extended Data Fig. 3c), respectively. Naive marker fold changes (FCs) were modest, but their variability was not due to noise or randomness (P < 2.2 × 10⁻¹⁶), as confirmed by in silico simulations (Extended Data Fig. 3d).

The naive T cell SC classification was confirmed in an independent PBMC dataset (validation dataset 1; Supplementary Table 3). Within this dataset, five naive states expressed the markers described above (Fig. 2c and Extended Data Fig. 3e). T_CM CD4⁺ T cell markers did not define a unique SC but were highly enriched in the SC expressing the highest levels of FOS and JUN (Fig. 2c), which was mapped adjacent to the T_CM SC in the original sorted CD4⁺ dataset (Extended Data Fig. 3a).

Within total CD4⁺ T cells, the frequencies of naive cells and of each SC were decreased in patients with SLE compared to HD. Within only naive CD4⁺ T cells, however, SC5 and SC3 were relatively expanded in SLE (Fig. 2d). SC3 was expanded in patients with LN and high DA compared to HD, whereas SC5 frequency classified patients with SLE with and without LN (Fig. 2d). These data confirm the presence of unique naive CD4⁺ T cell transcriptional states and their frequency changes in SLE.

The CD4⁺ memory compartment includes ten transcriptional distinct subsets

Phenotypical differences in homing and effector functions, antigen dependence, and capacity to proliferate or self-renew are characteristics of memory T cells. Our main cluster analysis yielded 90,178 memory CD4⁺ T cells based on high expression of S100A4 and low (although variable) expression of CCR7, which were classified into ten distinct SCs (Fig. 3a,b).

Fig. 3 |. Subclustering analysis reveals phenotypically diverse memory CD4⁺ T cells.

a, UMAP plot representing the ten SCs of memory CD4⁺ T cells. b, Matrix plot showing the top five markers for each SC in the memory CD4⁺ T cells. c, Heatmap representing the selective CD4⁺ T helper cell markers across all memory SCs. d, Expression density plots of selected T helper cell markers across memory SCs. e, Frequency of PD-1⁺CD38⁺ cells within circulating CD4⁺CXCR5⁻CD45RA⁻ memory T cells from HD (n = 6 independent donors) and patients with SLE (n = 20 independent donors). Data are presented as the mean ± s.d. (P = 0.0009; two-sided Student’s t-test). f, Cytokine expression profile as assessed by quantitative PCR in CD4⁺CXCR5⁻CD45RA⁻ memory T cells sorted into PD-1⁻CD38⁺, PD-1⁺CD38⁺ and PD-1⁺CD38⁻ subsets following CD3/CD28 stimulation (n = 4 independent donors). Data are presented as the mean ± s.d. (IL10 P = 0.0006; IL21 P = 0.0123; CXCL13 P = 0.0067; two-sided Student’s t-tests). Single-cell analysis included SLE (n = 16) and HD (n = 10) samples from the sorted CD4⁺ scRNA-seq dataset. *P < 0.05, **P < 0.01, ***P < 0.001.

Blood CXCR5⁺CD4⁺ T cells are counterparts of lymphoid tissue T_FH cells²⁴. In our dataset, SC0, SC1 and SC5 expressed the highest levels of CXCR5, supporting a T_FH-like phenotype (Fig. 3c,d). CCR7 and PD-1 (PDCD1) transcripts segregated CXCR5⁺ cells into a CCR7^lowPD-1^hi T_FH SC (SC5) and two CCR7^hiPD-1^low SCs (SC0 and SC1) (Fig. 3c,d). Cells from SC0 transcribed higher levels of TCF7 and lower levels of S100A4. They also expressed SESN3, a marker of memory and effector CD4⁺ T cells found in rheumatoid arthritis synovium²⁵. Consistent with CCR7^lowPD-1^hi T_FH cells exhibiting an ‘effector’ phenotype²⁶, SC5 cells transcribed IL21 and CXCL13 (Fig. 3c,d). Of the CXCR5⁺ SCs, only SC5 was significantly expanded in both patients with SLE (Extended Data Fig. 4a,b) and those with LN compared to HD (Extended Data Fig. 4c). Further in silico expression-based gating analysis confirmed the expansion of CXCR5⁺PD-1⁺CCR7⁻ cells according to both DA and LN (Extended Data Fig. 4d). In summary, we identified three distinct SCs of blood CXCR5⁺ T_FH-like cells, including a CCR7^lowPD-1^hi subset that transcribed IL21 and CXCL13 in the steady state and was expanded in patients with SLE with high DA and LN.

Within blood CXCR5⁻ memory T cells, SC6 expressed the highest levels of key T helper 2 (T_H2) cell transcripts, including transcription factor GATA3 and its antisense, long noncoding RNA GATA3-AS1, as well as PTGDR2 (CRTH2), CCR3, IL17RB, and transcripts encoding effector cytokines IL-4, IL-5 and IL-13 (Extended Data Fig. 4e). SC2 expressed GATA3 but lacked GATA3-AS1 and T_H2 cytokine transcripts. Expression of CCR4, CCR6, CCR10 and IL22 and lack of IL17A and IFNG pointed to a T helper 22 (T_H22) cell phenotype²⁷ (Extended Data Fig. 4f). SC3 fitted the description of T helper 17 (T_H17) cells (RORC, CTSH, IL17A and IL22)²⁸ (Extended Data Fig. 4g). SC2 and SC3 were expanded in patients with SLE with high DA and those with LN compared to HD and non-LN patients (Extended Data Fig. 4a–c).

MHC class II-restricted CD4⁺ T cells with cytotoxic potential, referred to as CD4-CTLs, have been widely reported²⁹. In our dataset, SC4, SC7 and SC9 expressed cytotoxic genes and TFs (TBX21, EOMES and RUNX3) that promote cytotoxic programs³⁰ (Fig. 3c,d). SC4 and SC7 expressed IFNG-AS1, GZMK and GZMA but had low levels or absence of GZMB and PRF1. SC4 uniquely transcribed XCL1³¹. SC9 matched a TEMRA profile (high CX3CR1, GZMA and PRF1) and exclusively expressed GZMB, GZMH, GNLY and ZNF683 (Hobit)²⁹. Only SC7 was enriched in patients with high DA or LN (Extended Data Fig. 4a–c).

Importantly, cells in SC7 expressed IL10 and PDCD1 (Fig. 3d). They also upregulated chemokine receptor (CXCR3, CCR2 and CCR5) and TF (EOMES, RUNX3 and SLAMF7) transcripts, all of which are hallmarks of T_H10 cells⁷ (Fig. 3c,d). In line with a T_H10 proliferative nature and nonanergic state, SC7 cells showed upregulation of MKI67 and genes encoding HLA class II molecules (Fig. 3c,d). Moreover, they lacked CXCL13 expression^5,6,32 (Fig. 3c,d) and thus fully matched the T_H10 phenotype⁷.

CXCR5⁻PD-1⁺ T_PH cells facilitate B cell recruitment and activation through CXCL13 and IL-21⁵. In our dataset, SC8 included cells that expressed PDCD1 and lacked CXCR5, similar to SC7 (Fig. 3c,d). Notably, they expressed IL21 and the highest levels of CXCL13 in the absence of cytotoxic transcripts (Fig. 3c,d). Akin to T_PH⁵, they upregulated activation markers, including ICOS and HLA class II transcripts (Fig. 3c,d). Similar to T_H10 cells, SC8 cells were significantly expanded in SLE and in patients with high DA and LN compared to HD (Extended Data Fig. 4a–c). Further participant-level analysis showed that both T_H10 and T_PH cells coexisted in patients with SLE, especially those with high DA (Extended Data Fig. 4h).

As coexpression of PDCD1 and CXCR3 encompassed both SC7 (T_H10) and SC8 (T_PH) (Fig. 3c,d), we sought additional markers to enrich SC7 cells and identified CD38 as a candidate (Fig. 3d). PBMC flow cytometry analysis confirmed expression of CD38 in a fraction of memory CXCR5⁻PD-1⁺CD4⁺ T cells (Extended Data Fig. 5a) and significant expansion of these cells in patients with SLE (Fig. 3e). Consistently, IL10 was highly transcribed, and IL-10 protein was secreted by sorted CD45RA⁻CXCR5⁻PD-1⁺CD38⁺CD4⁺ T cells following short-term activation. By contrast, IL-21 and CXCL13 were predominantly found at the transcript and protein levels in activated T_PH-like CD45RA⁻CXCR5⁻PD-1⁺CD38⁻CD4⁺ T cells (Fig. 3f and Extended Data Fig. 5b). Of note, a fraction of SC7 cells transcribed IL21 in their resting state (Fig. 3d and Extended Data Fig. 5c), but T cell receptor (TCR) stimulation of CD38⁺ cells barely induced IL21 transcription and IL-21 protein secretion (Fig. 3f and Extended Data Fig. 5b). Furthermore, CD38⁺CD4⁺ memory T cells expressed higher levels of SC7-associated protein markers, including CCR2, Ki-67, Tbet and PRF1 (Extended Data Fig. 5d). Across the entire CD4⁺ T cell pool, CD38 surface protein expression was lower in naive T cells and highest in T_H10 cells, in which it followed a bimodal distribution (Extended Data Fig. 5e).

CD96^hiCD4⁺ memory T cells expressing a T_H22-related signature were recently found to be decreased in adult patients with SLE³³. In our study, the average CD96 expression across memory CD4⁺ T cells was comparable in HD and pediatric patients with SLE (Extended Data Fig. 5f). Furthermore, a signature of CD96^hi cells³³ was enriched across four clusters, including SC2 (T_H22), SC3 (T_H17), SC6 (T_H2) and SC8 (T_PH) (Extended Data Fig. 5g,h). Cells expressing the highest (top 1%) levels of this signature were mapped to SC2 and SC3 and were significantly expanded in our pediatric SLE cohort. By contrast, the frequency of CD96^hi cells within the SC8 (T_PH-like) cluster was similar between HD and patients with SLE (Extended Data Fig. 5i). We therefore mapped all known subsets of CD4⁺ memory T cells, including three subsets expressing cytotoxic programs. We also identified two distinct populations of peripheral helper cells that were expanded in the blood of patients with high DA and LN.

CD4⁺ memory T cells are defined by canonical chemokine receptors. Accordingly, we performed expression-based in silico gating analysis for T helper 1 (T_H1) (CCR6⁻CXCR3⁺CCR4⁻), T_H2 (CCR6⁻CXCR3⁻CCR4⁺), T_H17 (CCR6⁺CXCR3⁻), T_H22 (CXCR3⁻CCR10⁺) and T_FH (CXCR5⁺) cells³⁴. As described above, CXCR5 was restricted to SC0, SC1 and SC5 (Extended Data Fig. 6a,b), whereas CXCR3⁺ cells were enriched in SC4 and SC7–SC9 (Extended Data Fig. 6b). CCR6⁺CXCR3⁻ cells were mapped to SC3 and exclusively expressed RORC and IL17A (Extended Data Fig. 4g). The distribution of CCR6⁻CXCR3⁻CCR4⁺ cells spanned SC2 and SC6, but only SC6 expressed the T_H2 marker PTGDR2 (ref. 35) (Extended Data Fig. 4e), along with T_H2-specific cytokine transcripts. SC2 expressed the bulk of the T_H22-associated CCR4, CCR6 and CCR10, with a smaller number of cells expressing the genes encoding these receptors in SC8 (Extended Data Fig. 4f). Finally, the gene encoding gut-homing chemokine receptor CCR9 was enriched in SLE cells from SC5 and SC8 (Extended Data Fig. 6c). Overall, canonical chemokine receptor expression largely matched the memory SC identities described above, while highlighting CXCR5⁻CXCR3⁺ cells as the most heterogeneous among CD4⁺ memory T cells.

Validation of CD4⁺ memory T cell SCs using PBMC scRNA-seq

To validate the CD4⁺ T cell memory SC classification, we used an additional dataset (validation dataset 2) of PBMCs from nine patients with SLE and five HD (Supplementary Table 4). After subclustering, 12,296 CD4⁺ memory T cells were mapped to eight SCs, including two CXCR5⁺ T_FH-like SCs (Extended Data Fig. 7a,b). Consistent with the sorted CD4⁺ dataset, two CXCR5⁺ SCs differed in PDCD1 expression. Thus, the CXCR5⁺PD-1^hi SC expressed IL21 and relatively lower levels of CCR7, matching SC5 from the sorted dataset. Conversely, PD-1^lo cells expressed higher levels of CCR7 in the absence of IL21, resembling SC0 and SC1 in the sorted dataset. Within the CXCR5⁻ compartment, the T_H17 (RORC, CTSH), T_H2 (GATA3-AS1, PTGDR2), TEMRA (CX3CR1, NKG7, GNLY, GZMs), XCL1⁺ T_H1-like, and CCR10⁺ T_H22 SCs robustly reproduced our original SCs (Extended Data Fig. 7a,b). In the PBMC data, however, we identified one SC of CXCR5⁻CXCR3⁺PD-1^hi memory cells that included both T_PH (CXCL13, IL21) and T_H10 (GZMK, CD38, IFNG-AS1, MKI67, CCR5) cells, likely owing to the lower resolution and/or cell frequency within PBMCs (Extended Data Fig. 7a,b). In line with our sorted CD4⁺ T cell data, expression of CD96 was not decreased in SLE (Extended Data Fig. 7c), and the CD96^hi signature³³ was detected across the T_H22, T_H17 and T_PH/T_H10 SCs (Extended Data Fig. 7d). In addition, T_H22 and T_H17 cells expressing the top 1% of the CD96^hi signature were increased in patients with SLE compared to HD (Extended Data Fig. 7e). In conclusion, the characterization of the CD4⁺ memory T cell compartment in SLE and HD blood was validated using an independent, nonsorted PBMC dataset containing about 1/7 as many cells as the original sorted dataset.

The ISG-high cluster encompasses nT_reg, mT_reg and T_reg cells

Blood CD4⁺ T cells included cells overexpressing ISGs that were expanded in patients with SLE. In fact, their frequency correlated with the average ISG expression across participants (Fig. 4a). We identified four distinct SCs (Fig. 4b), of which SC0 and SC2 matched naive CD4⁺ T cells (Fig. 4c). SC0 and SC2 transcriptional programs overlapped with those of SOX4 and IL6ST naive SCs, respectively (Fig. 4c). SC1 exhibited a memory phenotype (S100A4 and KLRB1) with various SC-defining markers, including T_FH, T_H17, T_H2 and cytotoxic markers (Fig. 4c,d). Last, SC3 expressed FOXP3 and IKZF2 along with S100A4, pointing to memory T_reg (mT_reg) cells (Fig. 4c,d).

Fig. 4 |. The ISG-high cluster includes naive, memory and T_reg cells.

a, Scatter plot showing the correlation between average ISG and IFN-related gene expression and frequency of ISG-high cells in total CD4⁺ T cells at the participant level. The error band represents the 95% confidence interval for the fitted regression line. b, UMAP of the four ISG-high SCs. c, Matrix plot showing the top 10 markers for each SC in ISG-high CD4⁺ T cells. d, Heatmap showing expression of selective CD4⁺ T helper cell markers across the ISG-high SCs. e, Box plots of cell proportions within total CD4⁺ T cells for each ISG-high SC by disease group (top), LN (middle) and DA (bottom). The boundaries and center line of the box plot represent the upper and lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5× IQR from the upper and lower quartiles. All data points are overlaid on the box plot. Two-sided t-tests were performed to test differences in cell proportions between groups. f, Dot plots reporting top differentially expressed genes between patients with SLE and HD. The size of the dot indicates the percentage expression of each gene in each cluster. Genes with significant P values (P < 0.05) and |FC | > 4 are highlighted by bolded squares. Single-cell analysis included SLE LN (n = 11), SLE NoLN (n = 5) and HD (n = 10) samples from the sorted CD4⁺ scRNA-seq dataset. *P < 0.05, **P < 0.01, ***P < 0.001.

Although present in HD, all ISG-high SCs were expanded in patients with SLE, especially those with LN (Fig. 4e). Pseudobulk analysis identified differentially expressed transcripts in each of the four SCs compared to HD (Fig. 4f). In addition to ISGs, transcripts associated with T_reg differentiation, such as PTGER2 and RARA³⁶, or with T_reg function, including FAS, TNFRSF18 (GITR) and TNFRSF4 (OX40)³⁷, were enriched in SLE SC3 (Fig. 4f). FCRL3, encoding a receptor for secretory IgA³⁸, as well as transcripts associated with Toll-like receptors (TLRs) such as TLR5 and LY96, or their downstream signaling (RIPK1), were also overexpressed in SLE SC3 (Fig. 4f).

These findings highlight the heterogeneous composition of the blood ISG-high CD4⁺ T cell compartment in the steady state and its expansion in SLE.

T_reg cells are expanded in SLE and LN and exhibit decreased suppression

The frequency and role of T_reg cells in SLE remain controversial. Our analyses confirmed increased T_reg frequency in patients with SLE³⁹, especially those with LN. This was further supported by flow cytometry according to either CD25^hiCD127^low or CD25^hiFoxP3^hi protein expression within CD4⁺ T cells (Extended Data Fig. 8a).

To better elucidate this compartment, we reclustered it into two SCs. T_reg SC0 transcribed CCR7 and lower levels of S100A4 (Fig. 5a), whereas T_reg SC1 transcribed higher levels of S100A4, along with genes encoding activation markers such as HLA class II (Fig. 5a). Both SCs were expanded in patients with SLE, and in those with LN (Extended Data Fig. 8b). These SCs matched two functionally distinct human blood T_reg subpopulations: a CD25^lowCD45RA⁺ fraction (fraction I) consisting of ‘naive’ T_reg (nT_reg) cells; and a CD25^hiCD45RA⁻ fraction (fraction II) including ‘memory’ T_reg (mT_reg) cells³⁹ (Extended Data Fig. 8c). Using flow cytometry, we observed expansion of nT_reg cells but not of mT_reg cells in patients with SLE compared to HD (Extended Data Fig. 8c). This discrepancy with the scRNA-seq data (Extended Data Fig. 8b) may have arisen from the limited set of surface markers used in flow cytometry compared to the comprehensive scRNA-seq profiling. In addition, whereas nTreg cells could be reliably identified by flow cytometry using CD45RA and CD25³⁹, the gating of mT_reg cells (fraction II) and the adjacent non-T_reg cells (fraction III) was less well defined (Extended Data Fig. 8c). Analysis of the transcriptomes of sorted fraction I and fraction II populations using published data⁴⁰ confirmed the identities of T_reg SC0 and SC1 (Extended Data Fig. 8d).

Fig. 5 |. TLR5⁺ nT_reg cells are expanded in SLE, and TLR5 ligation increases nT_reg dysfunction.

a, UMAP of T_reg SCs (left) and expression density plots of T_reg-related markers (right). b, Heatmap showing expression of the four T_reg-specific differentially expressed genes across CD4⁺ T cell clusters separated by disease group. c, TLR5 expression density plot in total CD4⁺ T cells. d, Box plots of relative expression of TLR5 across all five major CD4⁺ T cell groups between patients with SLE and HD. The boundaries and center line of the box plot represent the upper and lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5× IQR from the upper and lower quartiles. All data points are overlaid on the box plot. Differential gene expression analysis was performed using edgeR, and adjusted P values were calculated. The asterisk denotes an adjusted P value less than 0.05. Exact adjusted P values for all cell groups were as follows: 0.178 (ISG-high), 0.283 (memory), 0.205 (naive), 0.869 (proliferating), 0.024 (T_{reg_}SC0) and 0.637 (T_{reg_}SC1). e, Percentages of TLR5⁺ cells within naive T (CD4⁺CD25⁻CD127⁺CD45 RA⁺), memory T_eff (CD4⁺CD25⁻ CD127⁺CD45RA⁻), nT_reg (CD4⁺CD25^lowCD127⁻CD 45RA⁺) and mT_reg (CD4⁺CD25⁺⁺CD127⁻CD45RA⁺) cells in HD (n = 9 independent donors) and patients with SLE (n = 19 independent donors). Data are presented as the mean ± s.d. (naive T P = 0.0803 (nonsignificant (NS)); naive T_eff P = 0.0635 (NS); nT_reg P = 0.017; mT_reg P = 0.0967 (NS); two-sided Student’s t-test). f, Percentage of suppression of responder cell proliferation by nT_reg or mT_reg cells, with or without flagellin stimulation, in samples from patients with SLE (n = 11 independent donors) and HD (n = 6 independent donors). Data are presented as the mean ± s.d. (HD fraction I (Fr. I) P = 0.8613 (NS); HD fraction II (Fr. II) P = 0.2993 (NS); SLE fraction I P = 0.0001; SLE fraction II P = 0.9339 (NS); two-sided paired Student’s t-test). Single-cell analysis included SLE (n = 16) and HD (n = 10) samples from the sorted CD4⁺ scRNA-seq dataset. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

To explore transcriptional differences in SLE and HD T_reg cells, we performed pseudobulk analysis and identified 289 and 46 differentially expressed genes (P < 0.05 and FC > 1.5) for the naive (SC0) and memory (SC1) T_reg SCs, respectively. Selecting transcripts with log₂FC ≥ 0.25, and a five-fold increased expression in T_reg cells compared to the remaining CD4⁺ T cells yielded four ‘T_reg-specific differentially expressed genes’ for SC0, including TIGIT, HDAC9, FCRL3 and TLR5 (Fig. 5b). Notably, TLR5, whose activation has been reported to increase total T_reg suppression⁴¹, was predominantly upregulated in SLE nT_reg cells (Fig. 5c,d and Extended Data Fig. 8e). In our validation PBMC dataset 2 comprising 2,375 T_reg cells, TLR5 and FCRL3 were also enriched in nT_reg cells, and both nT_reg and mT_reg cells were expanded in SLE (Extended Data Fig. 9a,b). Assessing TLR5 protein expression across CD4⁺ T cell subsets revealed increased TLR5 in all SLE populations, with nT_reg cells displaying the highest levels and the most significant difference compared to HD (Fig. 5e and Extended Data Fig. 9c). These findings identify TLR5 as the main TLR in human T_reg cells and emphasize its upregulation in SLE.

We next tested the baseline T_reg suppression function and the effect of flagellin, the main ligand for TLR5⁴², on nT_reg and mT_reg cells from HD and patients with SLE. At baseline, nT_reg and mT_reg suppression was compromised in ~40% of patients with SLE³⁹ (Extended Data Fig. 9d,e). Addition of flagellin did not have an effect on HD or SLE mT_reg cells. It did, however, impair the suppressive function of SLE nT_reg cells and, in some cases, even promoted effector T (T_eff) cell proliferation (Fig. 5f and Extended Data Fig. 9f). These results point to a potential microbial trigger of nT_reg dysfunction in SLE.

An integrated blood CD4⁺ T cell landscape in SLE

Using a recently described integration algorithm from the Ragas package¹¹, we mapped 23 transcriptionally distinct CD4⁺ T cell SCs and reintegrated them to generate a high-resolution view of the CD4⁺ T cell compartment (Extended Data Fig. 10). We classified naive, memory, regulatory, proliferating and ISG-high T cells and correlated their frequencies with both SLE DA and LN. Distinct naive CD4⁺ T cell transcriptional states, as well as a heterogeneous ISG-high compartment, were identified in the steady state of HD and patients with SLE. Importantly, we mapped T_FH, T_PH and T_H10 cells to adjacent CD4⁺ memory SCs and identified three SCs expressing cytotoxic programs. Finally, we uncovered SLE-specific T_reg cell transcriptional and functional alterations predominantly affecting patients with LN.

Discussion

CD4⁺ T cells contribute to the pathogenesis of SLE by helping autoreactive B cells, secreting proinflammatory cytokines, differentiating into cytotoxic effectors and maintaining pathogenic memory responses Here, we profiled the transcriptome of CD4⁺ T cells sorted from the blood of pediatric patients with SLE and HD at the single-cell level, mapped them to 23 transcriptionally distinct CD4⁺ T cell SCs and studied their association with SLE DA and LN.

Recent studies have highlighted the contribution of extrafollicular reactions to autoimmune diseases and the expansion of blood CD4⁺CXCR5⁻PD-1^hi memory T cells, including T_H10 cells and T_PH cells, in SLE^5–7. Our analyses mapped these two effector helper populations to neighboring SCs and detected additional surface markers, including CD38, to enable their identification. Notably, combining CD38 with additional T_H10 markers permitted us to identify cells that transcribed either IL10 or IL21 in the steady state but primarily secreted IL-10 upon TCR stimulation, a finding that warrants further study.

CD4-CTLs²⁹ mapped in our study to three different memory T_H1-like SCs, including TEMRA (SC9), T_H10 (SC7), and T_H1-like XCL1⁺ (SC4) cells. All three SCs expressed GZMA, but GZMK was restricted to SC7 and SC4. Within the CD8⁺ T cell compartment, GZMK mediates complement activation and contributes to tissue inflammation⁴³, but its role in CD4⁺ T cells remains to be addressed. GZMK also marks a population of clonally expanded cytotoxic CD4⁺ T cells found in inflamed tissues of patients with IgG4-related disease⁴⁴, where they associate with activated (including extrafollicular) B cells. In these patients, GZMK⁺ cytotoxic T cells express amphiregulin and TGFβ, which contribute to tissue fibrosis⁴⁴, as well as SLAMF7, a marker of cytotoxic cells⁴⁵. In our dataset, SLAMF7 was expressed in SC9 and (to a lesser extent) in SC7 but was absent from SC4.

T_H1-related transcripts, such as CXCR3 and IFNG, were detected in all three cytotoxic SCs, as well as in T_PH-like SC8. The cytokine and cytotoxic transcriptional profiles of cells in SC7 overlaps with both T_H10 and TR-1 cells^7,8. Functionally, however, we reported that SLE blood T_H10 cells are not anergic but proliferate upon activation and provide B cell help in an IL-10-dependent manner⁷. We now show that, in addition to CXCR3 and PD-1, cells within SC7 express activation (HLA-DR, CD38), proliferation (MKI67) and costimulation (ICOS) markers in the steady state, and we confirm the expression of unique chemokine receptors (CCR2 and CCR5)⁷. Importantly, we show that PD-1 and CD38 coexpression identifies blood CD4⁺ memory cells enriched in IL-10-secreting but not IL-21-secreting precursors, fitting the description of T_H10 cells⁷. CD38 expression has been linked to CD4⁺ and CD8⁺ recent thymic emigrant T cells that express SOX4 and decline with age⁴⁶. A recent study also reported CD38 as a marker of IL-2 immunotherapy-induced proliferative HLA-DR⁺CD38⁺CXCR3⁺ T_reg cells that could be recapitulated in vitro through TCR plus IL-2 signals and might be primed for migration to inflammatory sites in SLE⁴⁷. Finally, an induced T follicular regulatory subset was also recently identified in human tonsils as a distinct CXCR5⁺CD38⁺, T_FH-descended subset that gains suppressive function while retaining the capacity to help B cells. Notably, these cells express IL-21 and IL-10 together with cytotoxic markers such as KLRB1⁴⁸). The potential relationship between these cells and cells within blood SC7 deserves further study.

Key findings of our study are the expansion of T_reg cells in LN, the overexpression of TLR5 in SLE nT_reg cells and the potentiation of their dysfunction by flagellin. T_reg cells express a variety of TLRs⁴⁹, and TLR1/2 ligation promotes nT_reg and mT_reg differentiation into pathogenic T_H17-like cells in multiple sclerosis⁵⁰. TLR5 expression in HD T_reg cells was previously reported to enhance suppression upon flagellin binding in vitro. However, CD127^low T_eff cells were not excluded from the T_reg pool in that study⁴¹. In addition to TLR5, we found that SLE nT_reg cells overexpressed FCRL3, which has been shown to inhibit T_reg function and to promote the skewing of these cells toward a proinflammatory phenotype³⁸. Microbial orthologs of SLE autoantigens have recently been described to trigger autoimmunity in mice⁵¹, and antibodies against a restricted pool of microbiome strain-specific cell wall lipoglycans correlated with DA, particularly in patients with active LN⁵². Our studies further support the need to elucidate the complex interactions between T_reg cells and the microbiome, as well as their connection with specific organ involvement in patients with SLE.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41590-025-02297-2.

Methods

Human samples

Children and adolescents with SLE were enrolled through the pediatric rheumatology clinics at Texas Scottish Rite Hospital for Children and Children’s Medical Center in Dallas, Texas. All study procedures were conducted in accordance with protocols approved by the Institutional Review Boards (IRBs) at Weill Cornell Medicine (IRB no. 22–03024648, 17–11018757 and 0604008488–07), the University of Texas Southwestern Medical Center (IRB no. STU 092010–167) and the Hospital for Special Surgery (IRB no. 2019–2338). Consent or assent was obtained from patients between 0 and 25 years of age. All enrollees consented to have coded information from their samples shared with external entities via publications and/or other data sharing mechanisms. Blood was collected in ACD tubes (BD Biosciences), and laboratory measurements were recorded. Clinical DA was assessed using the SLEDAI 2000.

Flow cytometry and cell sorting

Cryopreserved PBMCs were thawed at 37 °C, washed once with phosphate-buffered saline (PBS), and resuspended in Flow buffer (PBS containing 2% fetal bovine serum (FBS) and 0.5 mM EDTA). Cells were treated with Fc receptor block (BD Pharmingen) for 5 min at 37 °C, followed by surface staining for 30 min on ice. For CD4 T cell staining, cells were incubated with anti-CD3 BUV737 (UCHT1; BD Pharmingen) and anti-CD4 BV510 (OKT4; BioLegend), washed and stained with 7-AAD (BioLegend) to exclude dead cells. For isolation of T_reg and T_eff cells, PBMCs were stained with the following antibody panel: anti-CD3 BUV737 (UCHT1; BD Pharmingen), anti-CD4 BV510 (OKT4; BioLegend), anti-CD25 allophycocyanin (BC96; BioLegend), anti-CD127 BV421 (A01905; BioLegend) and anti-CD45RA BV785 (H100; BioLegend). For identification and sorting of CD3⁺CD4⁺CXCR5⁻CD45RA⁻PD-1⁺CD38⁺ cells, the following antibodies were used: anti-CD3 BUV737 (UCHT1; BD Pharmingen), anti-CD4 BV510 (OKT4; BioLegend), anti-CD45RA BV785 (H100; BioLegend), anti-CD38 APC-Fire 810 (HIT2; BioLegend), anti-CXCR5 AF647 (RF8B2; BD Pharmingen) and anti-PD-1 BV421 (EH12.2H7; BioLegend). Additional surface markers included anti-CCR2 FITC (K036C2; BioLegend). Samples were acquired on an Aurora flow cytometer (Cytek) or sorted using Melody cell sorter (BD Pharmingen; 100-μm nozzle) or a Cytek Aurora Cell Sorter (70-μm nozzle). Data were analyzed using FlowJo v.10.8.1 (BD Biosciences). For intracellular staining, surface-stained cells were labeled with a fixable viability dye (for example, 7-AAD or Fixable Viability Dye eFluor 780; eBioscience), followed by fixation and permeabilization using a FoxP3/Transcription Factor Staining Buffer Set (Thermo Fisher). Cells were then stained with antibodies against intracellular markers, including anti-Ki-67 PE (Ki-67; BioLegend), anti-PRF1 PE (B-D48; BioLegend), anti-Tbet PE (eBio4B10; eBioscience) and anti-FoxP3 PE (259D; BioLegend). For quantification of nT_reg and mT_reg cells and TLR5 expression, PBMCs were stained with anti-CD3 BUV737, anti-CD4 BV510, anti-CD45RA BV785, anti-CD127 BV421, anti-CD25 allophycocyanin and anti-TLR5 Alexa Fluor 488 (S16021; BioLegend). Viability was assessed using 7-AAD.

Quantitative real-time PCR and cytokine assay

Sorted T cells (1 × 10⁵ cells) were stimulated overnight with Immuno-Cult (StemCell) following the manufacturer’s protocol. Supernatants were collected, and cytokines were measured using a CBA Flex Set (BD; for IL-10 and IL-21) or Human BLC/CXCL13 ELISA Kit (Thermo Fisher; for CXCL13). For quantitative PCR, a total of 1 × 10⁵ cells were lysed in 50 μl Cell-to-Ct lysis buffer (Thermo Fisher). Complementary DNA (cDNA) was synthesized directly from cell lysates using a Cell-to-Ct Kit (Thermo Fisher). Quantitative real-time PCR was performed with TaqMan Gene Expression Assays (Applied Biosystems) according to the manufacturer’s instructions.

Suppression assay

nT_reg cells (CD4⁺CD25^hiCD127^lowCD45RA⁺), mT_reg cells (CD4⁺CD25^{h i}CD127^lowCD45RA⁻) and naive T cells (CD4⁺CD25⁻CD127⁺CD45RA⁺) were sorted from PBMCs of pediatric patients with SLE or HD by fluorescence-assisted cell sorting. Antigen-presenting cells (APCs) from an allogeneic HD were isolated by depleting CD3⁺ cells using CD3 microbeads (Miltenyi) and then treated with mitomycin C (50 μg ml⁻¹). One million APCs were incubated with mitomycin C at 37 °C for 15 min and washed three times. Suppression assays were performed by coculturing nT_reg cells (20,000 cells) or mT_reg cells (20,000 cells) with CFSE-labeled naive T cells (50,000 cells per well in 200 μl). Naive T cells were stimulated with soluble anti-CD3 monoclonal antibody (OKT3, 1 μg ml⁻¹) in the presence of mitomycin-treated APCs (50,000 cells). Control wells contained naive T cells with anti-CD3 and APCs but without T_reg cells. Assays were conducted with or without flagellin (Salmonella typhimurium, 100 ng ml⁻¹; Invivogen). After 5 days, cells were harvested, stained with 7-AAD and anti-human CD4 antibody to exclude dead cells and APCs, and analyzed by flow cytometry. Proliferation was assessed by CFSE dilution. Suppression was calculated as follows: percentage of suppression = difference in the percentage of proliferation in the presence and absence of T_reg cells/percentage of proliferation of naive T cells alone × 100.

scRNA-seq data processing and analysis

Sample preparation.

For the sorted CD4⁺ T cell dataset, sorted live CD4⁺ T cells were resuspended in PBS supplemented with 0.2% nonacetylated bovine serum albumin (BSA; Miltenyi), and viability was determined using trypan blue staining and measured on a Countess FLII (Thermo Fisher). Then, 50,000 cells were loaded for capture on a 10x Chromium system using a v.3 Single Cell 3′ Reagent Kit for RNA sequencing (10x Genomics). After capture and lysis, cDNAs from the cells were amplified as per the recommendations for the 10x Chromium. The amplified cDNAs were loaded on a single lane of NovaSeq 6000 for a depth of 50,000 reads per cell.

For SLE PBMC validation dataset 1, PBMCs in 10% DMSO + 90% FBS were thawed quickly at 37 °C and placed in DMEM supplemented with 10% FBS. Cells exhibiting a viability rate less than 70% were excluded. Cells were quickly spun down at 400g for 10 min, washed once with 1× PBS supplemented with 0.04% BSA and finally resuspended in 1× PBS with 0.04% BSA. Viability was determined using trypan blue staining and measured on a Countess FLII. Then, 12,000 single cells were loaded into one lane of a 10x Genomics Chromium [Controller, X], targeting 6,000 barcoded single cells. Single-cell capture, barcoding and library preparation were performed using the 10× 3′ Chromium platform (v.3.1) chemistry. cDNA and libraries were checked for quality on an Agilent 4200 TapeStation, quantified by KAPA quantitative PCR, and sequenced on an Illumina NovaSeq 6000 targeting 100,000 raw read pairs per cell.

For the 5′ PBMC validation dataset 2, cryopreserved PBMCs were thawed and resuspended in RPMI containing 10% FBS for cell number and viability assessment. Live PBMCs were enriched using a Dead Cell Removal Kit (Miltenyi) and resuspended in PBS with 1% BSA for scRNA-seq processing using Chromium Single Cell 5’ V2 gene expression (10x Genomics). The samples were divided into four batches and loaded into different lanes of a 10x chip, aiming for a cell recovery of 10,000 per sample, following the manufacturer’s instructions. Paired-end sequencing was performed on an Illumina NovaSeq 6000.

Cell quality filtering, normalization and main cluster analysis for the sorted CD4⁺ T cell dataset.

Count data of 16 patients with SLE and 10 healthy controls processed by CellRanger⁵³ were merged using Seurat¹⁸ from R. Single-cell sequencing generated libraries of 9,398 ± 2,143 and 8911 ± 2,561 cells (mean ± s.d.) per sample from patients with SLE and HD, respectively. On average, >1,600 genes and/or features per cell were detected for both groups. Following the Seurat guided clustering workflow (https://satijalab.org/seurat/articles/pbmc3k_tutorial.html), cell quality filtering was performed to remove cells with abnormal number of features (<500 or >4,000) and/or high mitochondrial gene expression (>20%); this yielded 257,229 filtered cells. Subsequently, we performed normalization, variable feature selection, scaling (regressing out percentage mitochondrial expression and total RNA count per cell) and dimension reduction by principal component analysis (PCA). To remove batch effects from single-cell data, Harmony¹⁰ was applied to the principal components using sample as the grouping variable. The Harmony-corrected cell embeddings were further used for UMAP⁵⁴ and clustering analysis (resolution = 0.5), which generated an initial version of 16 clusters. Further investigation using immune cell markers in combination with automated single-cell annotation by Azimuth¹⁸ revealed several additional doublet clusters, including monocyte doublets and plasmacytoid dendritic cell doublets, and an extra cluster with abnormal mitochondrial activities; these were removed from the data. We also removed a small T cell cluster with high CD79A expression but low CD79B expression. Finally, we reperformed UMAP and clustering analysis on the remaining 239,473 cells and obtained 15 clusters for the total CD4 data.

Single-cell subclustering analysis for the sorted CD4⁺ T cell dataset.

Single-cell subclustering analysis was performed to improve the identification of subpopulations of CD4⁺ T cells. The 15 initial clusters from total CD4⁺ T cells were further grouped into naive T cells (clusters 0, 1, 5, 11, 12 and 13), memory T cells (clusters 2, 4, 6, 7 and 9), T_reg cells (clusters 3 and 10) and ISG-high CD4⁺ T cells (cluster 8), and subclustering analysis was performed for each subset separately. The subclustering analysis also followed the Seurat guided clustering workflow with some modifications for the T_reg cells and ISG-high T cells. Briefly, we reperformed variable feature selection, scaling, dimension reduction, Harmony batch integration, UMAP and clustering and removed any additional low-quality clusters with higher mitochondrial expression and low RNA content. For the ISG-high cell subset, to prevent formation of additional high-ISG SCs, the following steps were performed: (1) we calculated the average ISG expression using seven representative ISGs (ISG15, ISG20, IFI6, IFI44L, IFITM3, LY6E, MT2A), which were regressed out during the scaling step; and (2) we removed all ISGs and IFN-related genes¹⁵ from the variable feature list when rerunning PCA. For T_reg cells, we directly performed reclustering and UMAP analysis on the original Harmony embeddings from the total CD4 analysis without rerunning variable feature selection, PCA or batch integration to avoid overfitting to noise in the data. Seurat and Ragas¹¹ were used for downstream analysis, including marker analysis, differential state analysis and cell proportion analysis. The Nebulosa package⁵⁵ was used to construct expression density plots.

Analysis of naive CD4⁺ T cells from PBMC validation dataset 1.

Analysis of scRNA-seq data from SLE PBMCs was performed following a similar guided clustering procedure to that previously described. First, cells with nFeature_RNA smaller than 500 or greater than 5,000 or with a percentage mitochondrial expression greater than 15% were filtered. In addition, we applied Scrublet⁵⁶ to systematically identify doublets from the PBMC data, and cells with a Scrublet score >0.28 were deemed to be technical doublets and removed from further analysis. Normalization, variable feature selection, scaling by regressing out ‘percent.mt’ and ‘nCount_RNA’, PCA, Harmony batch integration by sample, UMAP and clustering were subsequently performed, yielding an initial version of 18 clusters. After removal of clusters with low nCount_RNA and/or high mitochondrial gene expression and a small low-diversity cluster with cells predominantly from a single patient with SLE, we obtained the final PBMC data with 39,133 cells that formed 15 clusters. To obtain accurate cell identities for naive CD4⁺ T cells, we performed a series of subclustering analyses starting from the total T cells. We extracted CD3⁺ T cells from PBMC clusters 0, 3, 4, 5 and 9 and performed subclustering analysis. After removing an additional doublet cluster and a mito-high/RNA-low cluster, we acquired nine final clusters for 18,940 total T cells. Among them, clusters 0, 3, 4, 7, and 8 were positive for CD4 expression and were extracted for a second round of subclustering analysis. Initial clustering of the CD4⁺ T cells yielded ten clusters, including one small cluster with residual CD8A and/or CD8B expression, which was removed from downstream analysis. Among the final nine clusters from 11,024 CD4⁺ T cells, clusters 0, 1, and 3 were naive CD4⁺ T cells (high CCR7 and low S100A4 expression). A third round of subclustering analysis was then performed to delineate the heterogeneity within the naive CD4⁺ T cell compartment, which yielded five final clusters.

Analysis of memory and T_reg SCs from PBMC validation dataset 2.

As validation dataset 1 only contained SLE samples, to validate memory and T_reg SCs and their cell proportion changes between SLE and HD from the sorted CD4⁺ dataset, we generated a second validation dataset using 10× 5′ chemistry on PBMCs from nine additional patients with SLE and five HD. Cell quality filtering (nFeature_RNA ≤500 or ≥3000, percentage of mitochondrial gene expression ≥10%) and doublet removal by Scrublet were first performed; this generated 139,482 total PBMCs, which were analyzed using the aforementioned Seurat guided clustering workflow. After cleaning up additional doublet clusters and clusters with high mitochondrial expression, we grouped and annotated the remaining clusters. From the total PBMCs, we further subsetted T cells, proliferating T cells and NK cells (CD56^bright and CD56^dim) and performed reclustering analysis. This led to 11 refined T/NK clusters spanning CD4⁺ T, CD8⁺ T, mucosal-associated invariant T/γδT and NK cell types. A total CD4⁺ T cell object with 32,850 cells containing naive, memory, regulatory and ISG-high subsets was constructed, replicating the main cell populations from the sorted CD4⁺ dataset. Of note, this total CD4⁺ T cell object did not contain a proliferating cluster, because proliferative CD4⁺ T cells formed a separate ‘proliferating’ cluster of its own with other proliferating NK and CD8⁺ T cells. To validate the CD4⁺ memory SCs, we combined 12,138 CD4⁺ memory T cells with 385 Azimuth-predicted CTL/T_CM/TEM cells from the ‘CD8_TEM/CD4_TEMRA’ cluster. After reclustering the 12,523 combined CD4⁺ memory and cytotoxic T cells, we further identified and removed a smaller cluster enriched for innated lymphoid cell markers (SOX4 and TRDC) and obtained 12,296 final CD4⁺ memory T cells that were grouped into eight subsets. Moreover, we also subsetted and reclustered 2,375 T_reg cells from the total CD4⁺ T cell object and divided them into nT_reg and mT_reg cells based on CCR7, S100A4, CD74 and HLA-DR expression.

Miscellaneous bioinformatics analysis.

To evaluate whether the naive SCs identified here represented real transcriptional diversity in the naive CD4⁺ T cell compartment or were simply driven by noise, we downsampled 50,000 cells from the total naive CD4⁺ T cells and permuted their raw expression counts to build ‘null’ data of single-cell expression. After data normalization, variable feature selection, PCA and clustering, we obtained five final SCs from the null data. Seurat function FindAllMarkers was used to identify markers for the null data and calculate their FCs. Subcellular location information was downloaded using R package HPAanalyze⁵⁷. We used the newly developed Ragas package to analyze scRNA-seq data at both main and SC levels. To identify SC-specific markers, the RunFindAllMarkers function was used, and the top markers were visualized with the RunMatrixPlot function. To identify differentially expressed genes between disease groups, differential state analysis⁵⁸ was performed using the RunPseudobulkAnalysis function with the edgeR method⁵⁹, and we corrected for batch effects between subsets of data. Differential analysis of cell frequencies was performed using the RunProportionPlot function, and the parent object was assigned based on the chosen cell compartment of interest. R package ggplot2⁶⁰ was used to produce basic R plots for scRNA-seq data analysis. To further characterize T_reg SC0 and SC1, we reanalyzed the bulk RNA-seq data from ref. 40, which profiled the expression of conventional CD4⁺ T cells and regulatory T cells, including CD25^lowCD45RA⁺ nT_reg cells and CD25^hiCD45RA⁻ mT_reg cells (also referred to as eT_reg cells by Cuadrado et al.). We performed hierarchical clustering followed by filtering of non-T_reg markers and identified 193 and 161 genes as markers for nT_reg and mT_reg cells, respectively. After calculating their expression FCs between SC0 and SC1 T_reg cells in our CD4⁺ T cell dataset and removing genes with a log₂FC smaller than 0.1, we obtained a final set of 39 T_reg marker genes. Pearson’s correlation was calculated between the log₂FC of the 39 genes for T_reg SC0 versus SC1 from our CD4⁺ scRNA-seq data and for nT_reg versus eT_reg from the bulk RNA-seq data.

Statistical analysis

All statistical analyses were performed based on the distribution of data and study design. All results from spectral flow cytometry data and cell-based assays are presented as mean ± s.e.m. Comparisons between two groups were performed using two-tailed Welch’s t-test. For cell-based suppression assays, paired t-tests were used to assess differences between control and treatment groups. Statistical analyses for flow cytometry data and cell-based assays were conducted using GraphPad Prism (v.9; GraphPad Software). Differential cell proportion analysis for scRNA-seq data was performed using a two-sample t-test implemented in the Ragas package¹¹, and two-sided raw P values were reported. For pseudobulk-based differential state analysis, two-sided adjusted P values were calculated for each gene.

Extended Data

Extended Data Fig. 1 |. Quality summary for 10X sequencing, clustering and batch integration.

(a-b) Bar plot and violin plot showing the distribution of number of cells per individual sample in HD (n = 10, in coral) and SLE (n = 16, in blue). (c) Histogram showing the distributions of number of detected genes per sample, grouped by disease. (d) UMAP of total CD4⁺ T cells grouped into 15 initial clusters. (e) Dot plot illustrating further re-grouping of the 15 initial clusters into five major CD4⁺ T cells clusters based on the selected markers. (f-h) Bar plots showing the distributions of cells in the five major CD4⁺ T cell populations after harmony batch correction based on (f) 10X batch, (g) disease group and (h) individual sample.

Extended Data Fig. 2 |. ISG expression across CD4⁺ T cell clusters and its association with LN and DA.

(a) Heatmap showing the average expression of ISGs and IFN-related genes from Nehar-Belaid et al., 2020 across the five major CD4⁺ T cell populations at the sample level. Row colors indicate modular annotation of ISGs and IFN-related genes. (b) Dot plot reporting the logFC and expression frequency from differential gene expression analysis between SLE patients and HD for the five major CD4⁺ T cell clusters across all ISGs and IFN-related genes. The size of the dot indicates percentage expression for each gene in each cluster. Genes with expression frequency > 0.1 are highlighted by bolded square. (c) Pairwise scatter plots and correlations comparing per sample average expression of ISGs and IFN-related genes among the five major CD4⁺ T cell clusters. (d-e) Box plots showing average gene expression of ISGs and IFN-related genes in total CD4⁺ T cells grouped by DA or LN status. The boundaries and center line of the box plot represent upper/lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5 IQR from the upper/lower quartiles. All data points were overlaid on the box plot. Two-sided t-tests were performed to test differences in cell proportions between groups. Exact p-values for all pairwise comparisons are: 0.00077 (SLEDAI high vs. none), 0.008 (SLEDAI low vs. none), 0.0267 (SLEDAI high vs. low), 0.00026 (LN vs. HD), 0.08865 (NoLN vs. HD), 0.245 (LN vs. NoLN). Single-cell analysis included SLE LN (n = 11), SLE NoLN (n = 5) and HD (n = 10) samples from the sorted CD4⁺ scRNA-Seq dataset. LN/NoLN: w/wo nephritis; High: SLEDAI > 4; Low: SLEDAI < = 4. Significance levels: * (p < 0.05), ** (p < 0.01), *** (p < 0.001).

Extended Data Fig. 3 |. Analysis of naïve CD4⁺ T cell SCs from the sorted CD4 dataset and PBMC validation dataset 1.

(a) UMAP showing six naïve/CM SCs obtained by subclustering the naïve cluster from the sorted CD4⁺ T cell dataset. (b) Bar plot showing the subcellular locations of the naïve/CM SC markers with FC > 1.5 from the sorted CD4⁺ T cell dataset. (c) Heatmap showing the average expression of CD69 across all naïve/CM SCs from the sorted CD4⁺ T cell dataset. (d) Bar plot showing the distribution of log₂FC of subcluster markers from the sorted CD4⁺ T cell dataset (pink) and the permuted dataset (cyan). (e) Subcluster analysis of SLE PBMCs from the validation dataset 1. PBMC panel: 15 clusters were identified from total PBMCs, including two B cell clusters (1, 7), CD14⁺ (2) and CD16⁺ monocytes (8), five T cell clusters (0, 3, 4, 5, 9), one NK cluster (6), one DC (13) and one pDC (14) cluster, one plasma cell cluster (10), platelets (12), and one cluster of proliferating cells (11); CD3⁺ T cell panel: the 1st round of subcluster analysis was performed on T cells from PBMC clusters 0, 3, 4, 5 and 9, which yielded nine T cell SCs, among them the CD4⁺ T cells (0, 3, 4, 7, 8); CD4⁺ T cell panel: CD4⁺ T cells were further analyzed and re-clustered, which led to nine final SCs. SCs with high CCR7 and low S100A4 (0, 1, 3) were annotated as naïve CD4⁺ T cells. SC7 was also naïve-like, but was high for ISG expression; Finally, five SCs were identified from the naïve CD4⁺ T cells. Single-cell analysis included SLE (n = 16) and HD (n = 10) samples from the sorted CD4⁺ scRNA-Seq dataset, and additional SLE (n = 10) samples from the PBMC validation dataset 1.

Extended Data Fig. 4 |. Cell proportion analysis for memory CD4⁺ T cells from the sorted dataset.

(a-c) Box plots of relative cell proportions within total CD4⁺ T cells for each memory SC by disease group (a), severity (b) and LN (c). The boundaries and center line of the box plot represent upper/lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5 IQR from the upper/lower quartiles. All data points were overlaid on the box plot. Two-sided t-tests were performed to test differences in cell proportions between groups. (d) Expression-based in silico gating analysis for T_FH subsets. Two CXCR5⁺ sub-populations, PD-1⁺ CCR7⁻ (yellow) and PD-1⁻ CCR7⁺ (blue), were defined based on single cell gene expression as shown in the UMAP (upper left panel). Box plots (upper right and lower panel) show the relative cell proportions of the two CXCR5⁺ subsets within total CD4⁺ T cells by disease, LN, or DA. Box plots represent the same summary statistics as previously described. Two-sided t-tests were performed to test differences in cell proportions between groups. (e-g) Expression density plots of selective markers specific for T_H2 (e), T_H22 (f), and T_H17 (g). (h) Proportion of T_H10 (SC7) and T_PH (SC8) cells in total CD4⁺ T cells per subject. Single-cell analysis included SLE LN (n = 11), SLE NoLN (n = 5) and HD (n = 10) samples from the sorted CD4⁺ scRNA-Seq dataset. LN/NoLN: w/wo nephritis; High: SLEDAI > 4; Low: SLEDAI < = 4. Significance levels: * (p < 0.05), ** (p < 0.01), *** (p < 0.001).

Extended Data Fig. 5 |. Flow cytometry characterization of the PD-1⁺ CD38⁺ memory subset.

(a) Gating strategy and representative flow cytometry dot plots showing the percentages of PD-1⁺ CD38⁺ in SLE patients. (b) Levels of IL10, IL21 and CXCL13 in the supernatants of PD-1⁺ CD38⁺ and PD-1⁺ CD38⁻ cells upon activation (n = 3 independent donors). Data are represented as the means ± s.d. (IL-10 *p = 0.0235; IL-21 **p = 0.0089; CXCL13 **p = 0.004. Two-sided Student’s t-tests). (c) Bar plot showing number of cells within SC7 that express either IL10, IL21 or both. (d) Representative flow cytometry means fluorescence index of T_H10 specific markers on CD4⁺ CXCR5⁻ CD45RA⁻ PD-1⁻ CD38⁺, CD4⁺ CXCR5⁻ CD45RA⁻ PD-1⁺ CD38⁺ and CD4⁺ CXCR5⁻ CD45RA⁻ PD-1⁺ CD38⁻ cells. (e) Bar graphs showing relative MFI (median fluorescence index) of CD38 expression on naïve CD4 (CD4⁺ CD25⁻ CD127⁺ CD45RA⁺ CCR7⁺ cells, T_H10 (CD4⁺ CD25⁻ CD127⁺ CD45RA⁻ CXCR5⁻ CCR6⁻ CXCR3⁺ PD1⁺) cells, T_FH (CD4⁺ CD25⁻ CD127⁺ CD45RA⁻ CXCR5⁺), nTreg cells (CD4⁺ CD25⁺ CD127⁻ CD45RA⁺) and mTreg cells (CD4⁺ CD25⁺ CD127⁻ CD45RA⁺) (n = 39 independent donors). Data are represented as the means ± s.d. (T_H10-Naïve CD4 **p = 0.0024; T_H10-T_FH **p = 0.0039; T_H10-nTreg **p = 0.0024; T_H10-mTreg **p = 0.0026. Two-sided Student’s t-tests). (f) Average expression of CD96 in memory CD4⁺ T cells. The boundaries and center line of the box plot represent upper/lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5 IQR from the upper/lower quartiles. All data points were overlaid on the box plot. (g-h) Module score of CD96^hi signatures in memory CD4⁺ T cells and across SCs. (i) Cell proportion changes for the CD96 signature^hi cells between SLE and HD. Box plots represent the same summary statistics as previously described. Two-sided t-tests were performed to test differences in cell proportions between groups. Single-cell analysis included SLE (n = 16) and HD (n = 10) samples from the sorted CD4⁺ scRNA-Seq dataset. Significance levels: * (p < 0.05), ** (p < 0.01), *** (p < 0.001).

Extended Data Fig. 6 |. In silico gating analysis of CD4⁺ T helper subsets based on chemokine receptor expression.

(a) UMAPs showing the presence (yes) or absence (no) of different CD4⁺ Th subsets based on gating of chemokine receptor gene expression from single cell data. (b) Bar plot showing the proportion of cells of CD4⁺ Th subsets inferred by expression-based gating across the ten memory CD4⁺ T cell SCs. (c) Expression density plot of CCR9 on memory SCs (left) and heatmap showing the average expression of CCR9 across memory SCs and disease groups (right). Single-cell analysis included SLE (n = 16) and HD (n = 10) samples from the sorted CD4⁺ scRNA-Seq dataset.

Extended Data Fig. 7 |. Analysis of the PBMC validation dataset 2.

(a) Workflow of subclustering analysis of the PBMC validation dataset 2. After T cells and NK cells were subsetted and reclustered, CD4⁺ memory T cells and Treg cells were further reclustered, yielding eight CD4⁺ memory T cell clusters and two Treg clusters. (b) Summarized heatmap showing expression of selected T helper cell markers across memory subsets. (c) Average expression of CD96 in CD4⁺ memory T cells. The boundaries and center line of the box plot represent upper/lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5 IQR from the upper/lower quartiles. All data points were overlaid on the box plot. (d) Module score of CD96^hi signatures across CD4⁺ memory T cell subsets. (e) Cell proportion changes for the CD96 signature^hi cells between SLE and HD (* indicates p < 0.05). Box plots represent the same summary statistics as previously described. Two-sided t-tests were performed to test differences in cell proportions between groups. Single-cell analysis included SLE (n = 9) and HD (n = 5) samples from the PBMC validation dataset 2.

Extended Data Fig. 8 |. Additional data for Treg subsets definition, validation, and their associated cell frequency changes.

(a) Representative flow cytometry gating of Treg cells (left), with corresponding percentages in the blood of SLE patients and HD (middle). Data are represented as the means ± s.d. (For Foxp3⁺ SLE n = 19; HD n = 10 ****p = 0.0001; for CD127⁻ SLE n = 16; HD n = 7 **p = 0.0029. Two-sided Student’s t-tests). Right: Frequencies of Foxp3⁺ cells in HD (n = 10), SLE patients without LN (n = 7), and those with LN (n = 12). Data are represented as the means ± s.d. (HD vs SLE NoLN ****p = 0.0001; HD vs SLE LN ***p = 0.0003. Ordinary one-way ANOVA). (b) Box plots showing relative proportions of each Treg SC within total CD4⁺ T cells by disease group (top), LN (middle) and DA (bottom). The boundaries and center line of the box plot represent upper/lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5 IQR from the upper/lower quartiles. All data points were overlaid on the box plot. Two-sided t-tests were performed to test differences in cell proportions between groups. (c) Gating strategy for nTreg and mTreg cells (left) and their percentages in total CD4⁺ T cells in the blood of HD (n = 10 independent donors) and SLE (n = 21 independent donors) (right). Data are represented as the means ± s.d. (Fr.I ***p = 0.0004; Fr.II n.s. p = 0.2044. Two-sided Student’s t-tests). (d) Scatter plot of log2FC of 39 nTreg/mTreg markers between Treg SCs (SC0 vs. SC1) and sorted nTreg and mTreg fractions. Pearson’s correlation (0.67) and its corresponding two-sided p-value was calculated. (e) Box plot showing the mean gene expression of all TLRs across the Treg SCs. Single-cell analysis in (b), (d) and (e) included SLE LN (n = 11), SLE NoLN (n = 5) and HD (n = 10) samples from the sorted CD4⁺. LN/NoLN: w/wo nephritis; High: SLEDAI > 4; Low: SLEDAI < = 4. Significance levels: * (p < 0.05), ** (p < 0.01), *** (p < 0.001), **** (p < 0.0001), n.s. (not significant).

Extended Data Fig. 9 |. Additional data for subcluster analysis of Treg cells from the PBMC validation dataset 2 and their frequency changes due to the addition of flagellin.

(a) UMAP showing total Treg cells are clustered into two subsets with the expression density plots of Treg-related genes. (b) Box plots showing relative proportions of Treg subsets within total CD4⁺ T cells by disease group. The boundaries and center line of the box plot represent upper/lower quartiles and the median of the data, respectively. The whiskers extend to the maximum and minimum data values within the 1.5 IQR from the upper/lower quartiles. All data points were overlaid on the box plot. Two-sided t-tests were performed to test differences in cell proportions between groups. (c) Representative flow cytometry density plots showing the expression of TLR5 on nTreg and mTreg cells gated on CD4⁺ CD25⁺ CD127^low cells from the blood of HD and SLE patients. (d) Percentage suppression of responder cells by nTreg or mTreg cells in the absence of flagellin in SLE patients (n = 10 independent donors) and HD (n = 10 independent donors). Data are represented as the means ± s.d. (Fr.I **p = 0.0027; Fr.II *p = 0.0199. Two-sided Student’s t-tests). (e) Percentage suppression of responder cells by nTreg or mTreg cells at varying responder cell:Treg ratios. (f) Representative histogram plots showing CellTrace dilution of responder cells cultured with HD or SLE nTreg cells, with or without flagellin. Single-cell analysis included SLE (n = 9) and HD (n = 5) samples from the PBMC validation dataset 2. Significance levels: * (p < 0.05), ** (p < 0.01), *** (p < 0.001).

Extended Data Fig. 10 |. Integrated CD4+ peripheral T cell landscape in SLE.

Key conclusions from the study are highlighted and illustrated. All subclusters from the naïve/CM (NCM), memory (MEM), regulatory (TREG) and ISG-high (ISGH) CD4⁺ T cell groups were integrated and re-projected to a new UMAP using the R package Ragas (sorted CD4⁺ T cell dataset, n = 26). Each subcluster was numerically labeled based on their original subcluster identity before integration. Graphics created using BioRender.com.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41590-025-02297-2.

Data availability

The sorted CD4⁺ T cell dataset (GSE285773) and two PBMC validation cohorts (GSE285774 and GSE298578) are available via GEO. The processed Seurat objects are available via Zenodo at https://doi.org/10.5281/zenodo.16385794 (ref. 61).

Code availability

The code used to generate all scRNA-seq analysis results presented in Figs. 1–5 is available via GitHub at https://github.com/jig4003/CD4-SLE.