Single-cell transcriptomic signatures of T cells associated with rituximab responsiveness in idiopathic nephrotic syndrome

Summary

Rituximab (RTX) is increasingly used in steroid-dependent minimal change disease, a leading cause of idiopathic nephrotic syndrome (INS), yet predictors of response remain unclear. Although RTX primarily targets B cells, T cells are also implicated in INS pathogenesis. We investigated RTX-induced T cell dynamics by single-cell RNA sequencing/T cell receptor profiling of peripheral T cells from three responders and three non-responders before and after RTX treatment. Responders exhibited RTX-driven broad transcriptomic remodeling, accompanied by reduced exhausted CD8⁺ T cells and clonal expansion of CD4⁺ cytotoxic T cells with enhanced oxidative phosphorylation (OXPHOS) signatures, whereas non-responders showed marginal changes. In a separate cohort, reactive oxygen species (ROS) levels tended to be higher in responders, and declined post-RTX, collectively associating enhanced mitochondrial activity with favorable response. Reanalysis of childhood INS data supported altered B–T cell crosstalk and T cell metabolism. Together, T cells may influence RTX responsiveness, warranting further biomarker studies.

Graphical abstract

Highlights

• •T cell signatures differentiate RTX responders from non-responders in MCD (INS)
• •Responders show RTX-driven broad transcriptomic changes and distinct ROS regulation
• •CD4⁺ cytotoxic T cells undergo clonal expansion with enhanced OXPHOS in responders
• •Transcriptomic changes by RTX are modest in non-responders of steroid-dependent MCD

Immunology; Systems biology; Experimental models in systems biology

Introduction

Minimal change disease (MCD) is a common cause of idiopathic nephrotic syndrome (INS) in both children and adults. Although MCD responds well to oral corticosteroids,^1,2 approximately 80–90% of children with steroid-sensitive nephrotic syndrome (SS-NS) experience relapses, and approximately 40% progress to frequently relapsing and/or steroid-dependent nephrotic syndrome (FR/SD-NS).³ In adults, the relapse rate is 30–75%, with 10–30% developing FR/SD-NS.⁴ Since the discovery of the efficacy of rituximab (RTX) on INS,⁵ RTX has been established as an effective treatment of FR/SD-NS in both children^6,7,8,9 and adults.^10,11

Immune-mediated mechanisms involving T and B cells are implicated in the pathogenesis of MCD. Earlier studies have highlighted T cell dysfunction and alterations in humoral factors, which could cause increased protein permeability of the glomerular tufts.¹² In addition, SS-NS and SD-NS show distinct T cell profiles.¹³ On the other hand, discovery of circulating autoantibodies against podocyte proteins, particularly anti-nephrin antibodies, has provided evidence for an autoreactive B cell component in INS pathogenesis.¹⁴ Recent identification of an extrafollicular B cell response, including expansion of atypical B cells (atBCs), marginal zone-like B cells, and antibody-secreting cells (ASCs) in steroid-free childhood INS,¹⁵ and differences in B cell subsets between SS-NS and SD-NS,¹⁶ further supported the B cell involvement in INS.

RTX is thought to induce remission of MCD through depletion of these pathogenic B cell populations. As a typical clinical response, single-dose RTX induces full depletion of B cells (<1% of lymphocytes) that persists for 4–6 months,^17,18 and after that, approximately 10–40% of patients with MCD¹⁹ eventually relapse. As relapse reportedly occurs shortly after B cell recovery,^17,20 several studies have suggested that post-RTX relapses of INS are associated with a resurgence of class-switched memory B cells²¹ or nascent re-engagement of memory B cells in the extrafollicular pathway.¹⁵ However, the therapeutic mechanisms may extend beyond simple B cell depletion. Notably, 25–63% of patients maintain remission during a one-year observation period,^17,18 even after B cell recovery. Additionally, relapses can even occur during B cell depletion in some patients.^18,22 Therefore, it is largely unclear which patient characteristics predict RTX responsiveness, limiting our ability to identify patients most likely to benefit from treatment.^23,24,25 Although several studies have proposed the proteinuria selectivity index,²⁶ memory B cells, NK cells, and serum IgG levels²⁷ as potential biomarkers for RTX responsiveness, the supporting evidence remains limited. The inability to distinguish potential responders from non-responders creates clinical challenges: prolonged corticosteroid monotherapy in steroid-dependent patients leads to unnecessary exposure to steroid-related toxicity, while empirical RTX use in non-responders carries risks of infection and increases healthcare burden without therapeutic benefit.

To fully understand the impact of B cell depletion, it is also necessary to focus on non-B cell lineages, particularly T cells. Indeed, RTX not only depletes B cells but also modulates T cell homeostasis in NS,²⁸ as well as other kidney diseases^29,30 and autoimmune diseases.^31,32,33 Additionally, several features of T cell profiling have been associated with RTX responsiveness in NS²⁸ and type 1 diabetes.³⁴ In several mouse models, B cell depletion reportedly affects both CD4⁺ and CD8⁺ T cells.^35,36 However, systematic T cell profiling has not been performed in patients with SD-NS treated with RTX. Single-cell profiling enables resolution of T cell states and clonal dynamics that are not captured by bulk assays. In this study, we investigated whether RTX-induced modulatory effects on T cells are linked to RTX responsiveness. To identify transcriptomic signatures associated with RTX responsiveness, we conducted a comprehensive analysis of peripheral blood T cells using single-cell RNA sequencing (scRNA-seq) and T cell receptor (TCR) profiling in three patients who achieved long-term remission following RTX administration (RTX responders) and three patients who relapsed within 6 months of any RTX dose or remained steroid-dependent (non-responders). We implemented an integrative computational framework by combining denoised metabolic state inference, TCR clonality analysis, gene regulatory network analysis, and B cell–T cell communication analysis in public single-cell datasets of pediatric INS. Although it is difficult to collect such rare cases—patients with long-standing, often decade-long, steroid-dependent disease who recently received RTX treatment—and further analyses in larger cohorts would be necessary, our exploratory analysis suggested that RTX responsiveness is associated with differential responses of T cell transcriptome.

Results

Single-cell transcriptome and TCR profiling of T cells in RTX responders and non-responders

We performed scRNA-seq and single-cell TCR profiling of peripheral blood T cells from six patients with SD-MCD (Figures 1A–1C, S1 and S2; Tables 1, S1 and S2). These patients had been FR/SD for approximately a decade and received maintenance RTX therapy at Nagoya University (Figure 1A). Most patients were in clinical remission at the time of sampling and had a peripheral B cell proportion of ≥10% among lymphocytes before RTX treatment.

Figure 1.

Study design and T cell transcriptomic profiling in patients with steroid-dependent MCD treated with RTX

(A) Schematic of study design. Patients achieve steroid-free sustained remission after RTX administration were defined as responders (ptG1–3); others as non-responders (ptP1–3).

(B) Clinical courses of individual patients.

(C) Workflow from blood collection to scRNA-seq of CD3⁺ T cells.

(D) UMAP embedding of scRNA-seq data, showing 17 T cell subsets (MC00–MC16) across all samples.

(E and F) Differentiation trajectories of CD4⁺ (E) and CD8⁺ (F) T cells.

(G and H) Boxplots compare cell proportions before and after RTX in responders and non-responders: (G) CD4⁺ T cells, (H) CD8⁺ T cells. Unpaired t test was used for statistical analysis.

(I) Five subclusters of the MC01 subset (memory CD4⁺ T cells).

(J) Boxplots compare cell proportions of MC01 subsets before and after RTX in responders and non-responders. Images created in BioRender were used in (A) and (B). ∗p < 0.05 by Wilcoxon signed-rank test (G, H). Each dot represents the percentage of the subset for each patient (G, H, and J). Abbreviations: TEMRA, terminally differentiated effector memory cells; CM, central memory; EM, effector memory; MAIT, mucosal-associated invariant T cell; Treg, regulatory T cell; CTL, cytotoxic T lymphocyte.

Table 1.

Patient characteristics for scRNA-seq

	RTX responders (n = 3)	RTX non-responders (n = 3)	p
Age (y)	23.0 [19.5–31.5]	48.0 [38.0–49.5]	0.2
BMI (kg/m2)	22.4 [21.7–22.9]	24.7 [23.8–27.0]	0.2
Male (%)	100	33.3	0.4
prescribed PSL (mg/day)	25 [15–42.5]	15 [8.5–32.5]	0.7
prescribed CyA (%)	33.3	33.3	1
prescribed Tac (%)	0.0	0.0	1
prescribed MZR (%)	33.3	33.3	1
Period from the previous relapse to the rituximab administration (days)	30 [25–192]	72 [67–418]	0.4
Period from the first onset to the rituximab administration (days)	7426 [3966–8779]	4232 [2705–7208]	1
Total counts of rituximab administration during the follow-up period(times)	4.0 [3.0–4.0]	4.0 [3.5–5.5]	0.643
Rituximab dosing interval (days)	183 [169–191]	182 [182–215]	1

Median [IQR], Wilcoxon signed-rank test.

Abbreviations: BMI, body mass index; PSL, prednisolone; CyA, cyclosporine A; Tac, tacrolimus; MZR, mizoribine.

The clinical characteristics of the patients are summarized in Table 1. Details of individual patient characteristics and changes in laboratory parameters before and after RTX administration are included in Tables S1 and S2. Despite a median disease duration of 10 years with multiple prior treatments, the two groups remained relatively well balanced for analysis. The interval from the previous relapse and concomitant immunosuppressant use were comparable.

Among the six patients, three patients (ptG1–3) that achieved steroid discontinuation and sustained remission were defined as RTX responders, whereas the others (ptP1–3) that could not discontinue steroids during the observation period (ptP1) or relapsed within six months of any RTX treatment (ptP2–3) were defined as RTX non-responders (Figures 1B and S1; Tables 1, S1 and S2). CD3⁺ T cells were isolated from peripheral blood mononuclear cells (PBMCs) obtained before and one month after the first RTX administration (Figures 1C and S2, see STAR Methods). We focused on T cells in this study, since B cell viability was modestly affected by standard cryopreservation methods in our pilot study (data not shown). As B cells are typically depleted within one month after RTX administration, the present analysis was conducted at one month after RTX treatment. We confirmed that RTX depletes B cells (0%) at one month after the first RTX administration in all patients (Tables S1 and S2).

RTX tends to reduce exhausted CD8⁺ T cells in RTX responders

51,919 high-quality cells were obtained from six patients. Integrated clustering analysis revealed 17 distinct subsets (MC00–MC16) (Figure 1D), all of which were uniformly represented in RTX responders and non-responders, with or without RTX treatment (Figure S3). These 17 cell subsets were annotated based on the expression of canonical marker genes (Figure S4). The annotations were validated through trajectory analysis using SlingShot, showing a standard transition from naive to mature cells within both CD4⁺ and CD8⁺ T cell compartments (Figures 1E and 1F).

Compositional changes in the CD4⁺ and CD8⁺ T cells revealed that exhausted CD8⁺ T cells (MC07) in RTX responders tended to be more abundant at baseline than in RTX non-responders and significantly decreased following RTX treatment (Figures 1G and 1H). Although not statistically significant, baseline trends of reduced naive CD4⁺ T cells (MC00), reduced memory CD4⁺ T cells (MC01), increased CD4⁺ cytotoxic T lymphocytes (CTLs) (MC05), and decreased KIR⁺ CD8⁺ T cells (MC08) were observed in RTX responders (Figures 1G and 1H). Reanalysis of CD4⁺ T helper cells (MC01) showed a non-significant trend toward Th1- and Th17-skewed polarization over Th2 phenotypes in RTX responders (Figures 1I, 1J and S5). These findings suggest that RTX may alleviate exhaustion-prone environments in RTX responders.

RTX elicits widespread transcriptomic changes in RTX responders

We next performed differential gene expression analysis between pre- and post-treatment with RTX for each T cell subset in RTX responders and non-responders. By using pseudobulk methods,³⁷ we identified 1,415 differentially expressed genes (DEGs) (|log₂(fold change)| > 0.6, FDR q < 0.1) in at least one subset in RTX responders (Figure 2A, left). This change was substantially broader than that observed in non-responders, with only 77 DEGs. The transcriptional differences were relatively uniformly broad in major T cell subsets (MC00–MC10, Figure 2A, right), except for a few subsets with small cell numbers (MC11–14), which showed a small number of DEGs. Additional single-cell data-based differential gene expression analysis using the model-based analysis of single-cell transcriptomics (MAST)³⁸ framework reproduced similar results (Figures S6A and S6B).

Figure 2.

Widespread transcriptomic alterations in T cells following RTX treatment in responders

(A) Volcano plots of DEGs before and after treatment across all T cell subsets (MC00–MC14) in responders and non-responders. Bar graphs to the right indicate the number of upregulated (red) and downregulated (blue) genes per subset. Statistical threshold for DEG selection: FDR q < 0.1 and |log₂(fold change)| > 0.6.

(B) Heatmap of eight DEG groups (Up_C1–C4, Down_C1–C4) identified by k-means clustering of aggregated gene sets, showing z-scaled log₂(CPM+1) expression values. Samples are grouped by response status (left: responders; right: non-responders), T cell subset, and treatment condition. The z-scores are taken over samples.

(C) Heatmap of representative genes from the clusters shown in (B), all of which are included in the Hallmark gene sets presented in (E). The z-scores are taken over samples.

(D) Gene Ontology Biological Process enrichment of DEG groups, visualized with Cytoscape. Each node represents a GO term, with edges connecting semantically related terms. Node color indicates regulation direction (red: upregulated; blue: downregulated), and each node is divided into quadrants representing groups C1 to C4 in clockwise order. FDR q < 0.1.

(E) Hallmark pathway enrichment of DEG groups. Abbreviations: DEG, differentially expressed gene; CPM, counts per million; FDR, false discovery rate.

Transcriptional signatures of RTX response in T cells

Given that the upregulated and downregulated DEGs showed minimal overlap and were substantially shared in multiple T cell subsets (Figure S6C), we further performed cluster analysis of these upregulated and downregulated DEGs and identified eight DEG groups: four upregulated (Up_C1–C4) and four downregulated (Down_C1–C4) following RTX treatment (Figures 2B and 2C). These DEG groups exhibited distinct expression dynamics along T cell differentiation, with their expression enriched in one of the three major functional states of T cells: naive (MC00–02), activated (MC03–05 and MC08–09), and exhausted or terminally differentiated (MC06–07) (Figure 2B), as supported by independent signature score analysis (Figures S7 and S8).

Two upregulated DEG groups, Up_C1 and Up_C2, were upregulated in the activated T cell subsets (MC03–05 and MC08–09), and their expression levels were further enhanced by RTX in RTX responders (Figures 2B, S7 and S8). The Up_C1 group was associated with oxidative phosphorylation (OXPHOS), mitochondrial translation, and oxidant detoxification (Figures 2D, 2E and S9). Up groups were also associated with response and signaling of several cytokines, including IFN-γ, IFN-α, and IL-2. In contrast, the downregulated DEG groups (Down_C1–Down_C4) were linked to inflammation and cellular stress, including hypoxia-related process (Down_C1 and C2), TNF-α signaling via the NF-κB (Down_C1, C2, and C4), and the p53 pathway (Down_C1). These findings suggest that RTX dynamically changes T cell status in responders by enhancing mitochondrial respiration—specifically OXPHOS—and cytokine signaling and suppressing the hypoxic response pathway.

RTX upregulates OXPHOS genes in mature T cells in RTX responders

T cells undergo distinct metabolic reprogramming depending on their differentiation state; naive T cells rely on OXPHOS, whereas effector T cells use glycolysis to support activation.^39,40 While chronically stimulated T cells and exhausted T cells exhibit similar glycolytic phenotypes, chronic T cell stimulation induces inhibition of mitochondrial OXPHOS capacity and increased generation of reactive oxygen species (ROS) in these cells, leading to suppressed proliferation and self-renewal.⁴¹ To explore potential metabolic shifts associated with RTX responsiveness, we visualized the joint distribution of OXPHOS and glycolysis status by performing gene set variation analysis (GSVA) with custom gene sets for OXPHOS and glycolysis from a previous study⁴² (Figures 3A, 3B, S10 and S11). The Markov Affinity-based graph imputation of cells (MAGIC) pipeline was applied to reduce technical noise and enhance gene-gene correlations prior to GSVA. Based on the pathway activation score (PAS) for each cell, the cells were classified into four metabolic subgroups: subgroup A, low OXPHOS; subgroup B, moderate OXPHOS; subgroup C, high OXPHOS and low glycolysis; and subgroup D, high OXPHOS and high glycolysis. This classification revealed different trends between the three responders and three non-responders. RTX responders exhibited a marked shift; while pre-treatment cells were primarily in subgroup A, post-treatment cells were redistributed toward subgroups C and D (Figure 3A), indicating enhanced OXPHOS. In contrast, in RTX non-responders, pre-treatment cells were predominantly enriched in subgroup D, showing the trends of treatment-associated redistribution toward OXPHOS-low subgroup B (Figure 3A). Enhanced OXPHOS trends in responders and differences between responders and non-responders were supported by patient-level visualization despite of inter-patient variability (Figure S10).

Figure 3.

Enhanced oxidative phosphorylation in T cells of RTX responders

(A) Contour plots show joint distributions of PAS for OXPHOS and glycolysis in T cells before and after RTX treatment, based on merged data from three samples normalized to 10,000 cells each. Cells were classified into four metabolic subgroups based on PAS: subgroup A (low OXPHOS), subgroup B (moderate OXPHOS), subgroup C (high OXPHOS and low glycolysis), and subgroup D (high OXPHOS and high glycolysis).

(B) Sankey diagram illustrates the redistribution of metabolic subgroups (A–D) across T cell subsets (MC00–14) before and after RTX treatment in responders and non-responders.

(C) Bar plots show subset-wise proportions of cells in each metabolic subgroup before and after RTX treatment, separately for responders and non-responders. Bars represent the mean proportions across three patients in each group, with error bars indicating standard deviation. Proportions of metabolic subgroups A and B are displayed in the upward direction, while proportions of metabolic subgroups C and D are displayed in the downward direction. Abbreviations: PAS, pathway activation score; OXPHOS, oxidative phosphorylation.

Three major functional states of T cells, namely, naive, activated, and exhausted (terminally differentiated), were associated with distinct metabolic subgroups (Figure 3C). The majority of naive cells (MC00–02) were in subgroup C or D and showed high OXPHOS. The activated cells corresponded to various metabolic subgroups, whereas the majority of exhausted cells (MC06-07) belonged to subgroup A. In RTX responders, TEMRA CD8⁺ T cells (MC03), central memory (CM) CD8⁺ T cells (MC04), and CD4⁺ CTLs (MC05) tended to display post-treatment increases in subgroup C, suggesting enhanced OXPHOS (Figure 3C). This trend was particularly pronounced in CD4⁺ CTLs (MC05) in RTX responders (Figure S11). Collectively, the favorable RTX response appears to be characterized by increased OXPHOS in mature CD8⁺ T cells and CD4⁺ CTLs.

In our analysis, we failed to detect high glycolytic signals in activated T cells or the typical OXPHOS-glycolysis switch along T cell maturation. This may be partly explained by previous findings showing that glucocorticoids preferentially inhibit glycolysis in T cells.^43,44 Thus, under long-term steroid treatment, T cells, especially activated T cells, may primarily co-opt the OXPHOS pathway to satisfy metabolic requirements and modulate OXPHOS in response to RTX treatment.

Baseline transcriptional differences between RTX responders and non-responders

The analysis of baseline differences between responders and non-responders using the MAST approach revealed 4,925 DEGs (Figure 4A), which were further classified into four groups showing high expression (“Up”) and four groups showing low expression (“Down”) in RTX non-responders (Figures 4B, S12, S13, and S14). We adopted a naming convention for DEG groups such that those exhibiting similar expression patterns in Figures 2B and 4B were assigned the same names. An overlap analysis of the DEG groups defined in Figures 2B and 4B revealed substantial gene sharing (Figure S15). This suggests that many of the DEGs that differed between responders and non-responders at baseline were shared with those that changed after RTX treatment in responders.

Figure 4.

Baseline transcriptional profiles associated with RTX responsiveness and ROS status

(A) Volcano plot shows DEGs in T cells at baseline between RTX responders and non-responders. “Up” indicates higher expression in non-responders; “Down” indicates higher expression in responders. Right image shows the number of Up- and Down-regulated DEGs for each T cell subset (MC00–MC14). DEGs were identified using a threshold of Benjamini-Hochberg adjusted p < 0.05 and |log2(fold change)| > 0.6.

(B) Heatmap of z-scaled expression for DEGs grouped into eight groups (Up_C1–C4 and Down_C1–C4) by k-means clustering (k = 4). Note that the DEG groups in this analysis are independently derived and not identical to those in Figure 2B. The z-scores are taken over samples.

(C) Gene Ontology Biological Process enrichment of DEG groups, visualized with Cytoscape, as in Figure 2D. FDR q < 0.05. Note that DEGs included several major histocompatibility complex (MHC) class II genes, including HLA-DMA, DRA, DPB1, DRB5, DRB1, DQA1, DQA2, and DQB1.

(D) Intracellular ROS levels in peripheral T cells, shown as boxplots of MFI in healthy control (n = 1), RTX responders (n = 3), and RTX non-responders (n = 5), before and after RTX treatment. ∗p < 0.05 by unpaired t test was used for two-group comparisons, or one-way ANOVA was used for comparisons across four groups. Abbreviations: ROS, reactive oxygen species; MFI, mean fluorescence intensity; HC, healthy control.

Among the “Up” groups, the Up_C1 group was associated with aerobic respiration and OXPHOS, suggesting that baseline OXPHOS and respiratory status in T cells are compromised in RTX responders and/or more competent in RTX non-responders (Figures 4C and S15). In addition, several “Up” clusters (Up_C2, C3, and C4) were associated with cytokine signaling and immune responses. The “Down” groups were also associated with immune response-related terms, such as TNF-α-NF-κB signaling (Figures 4C and S15).

RTX responsiveness and ROS status

To investigate whether differences in metabolic genes affect ROS regulation, we directly measured intracellular ROS levels in an independent patient cohort (Figures 4D and S17; Table S3, and see STAR Methods). This analysis showed that, prior to RTX administration, responders tended to exhibit higher ROS levels than non-responders, in total, CD4⁺, and CD8⁺ T cells. Following RTX treatment, ROS levels tended to decrease in responders, whereas they increased in non-responders. These trends were confirmed in various subsets of CD4⁺ and CD8⁺ T cells (Figures S17B and S17C). These findings suggest that elevated ROS levels at baseline, followed by a decrease after treatment, may be associated with favorable RTX responsiveness. Together with scRNA-seq results, RTX may ameliorate mitochondrial dysfunction along with upregulation of OXPHOS genes and downregulation of ROS accumulation in responders.

Clonal expansion of CD4⁺ CTL following RTX treatment in RTX responders

To delineate the relationship between T cell clonality and RTX response, we performed scTCR analysis and extracted cells with both TCR α-chain and β-chain. Expanded clonotypes were primarily localized within CD8⁺ effector T cells and CD4⁺ CTLs (Figure 5A). Mature T cell subsets, including TEMRA CD8⁺ T cells (MC03), CM CD8⁺ T cells (MC04 and MC09), and CD4⁺ CTLs (MC05), harbored markedly larger clones (Figures 5B and S18). MC04 and MC05—encompassing central memory-like CD4⁺ T cells and CD4⁺ CTLs, respectively—displayed a significant post-treatment increase in the proportion of highly expanded clonotypes (clone size ≥20) in RTX responders (Figure 5C).

Figure 5.

Clonal expansion of CD4⁺ cytotoxic T cells in RTX responders

(A) UMAP of T cells (left) and UMAP overlaid with TCR clonotype size, visualized as red circles scaled by log₂(clone size) (right).

(B) Stacked bar plots show the distribution of clonotype size categories across transcriptionally defined subsets (MC00–MC09) in responders (left) and non-responders (right). For each subset, bars extending upward from the horizontal axis represent pre-treatment values and bars extending downward represent post-treatment values. Bars represent the mean proportions across three patients in each group, with error bars indicating standard deviation. For each sample, the total number of T cells in MC00–MC09 subsets was set to 1. Colors indicate clonotype size categories: NA (white), = 1 (light green), 2–19 (green), and ≥20 (orange). Individual patient-level data are shown in Figure S18.

(C) Stacked bar plots show the clonal ratio—the proportion of cells within each clone size category—for selected subsets (MC03, MC04, MC05, and MC09) extracted from (B). The ratios were normalized to 100% per 3 samples. Colors represent clone size categories: NA (white), = 1 (light green), 2–19 (green), and ≥20 (orange). The proportion of highly expanded clonotypes (clone size ≥20) for each patient is displayed as individual dots, stratified by treatment status (pre/post) and response (RTX responders vs. non-responders). Statistical comparisons between pre- and post-treatment groups were performed using the chi-squared test (∗p < 0.05).

(D) Scatterplots display OXPHOS PAS (x axis) and glycolysis PAS (y axis), with each dot representing a cell in MC05. The dot size and red color intensity scale with log₂(clone size). Plots are separated by treatment (pre/post-RTX) and response group (responders/non-responders).

(E) Boxplots show the distribution of OXPHOS PAS in MC05 cells stratified by clone size (=1, 2–19, 20–99, ≥100). ∗∗p < 0.001 and ∗p < 0.05 by Wilcoxon signed-rank test. In boxplots, the central line indicates the median, and the whiskers represent the interquartile range (IQR).

(F) Network plots of TCR repertoire similarity based on CDR3 amino acid sequences of both TCRα and TCRβ chains. Clonotype clusters with clonal size ≥20 are shown for each sample, stratified by prognosis, treatment phase, and individual patient.

We further assessed the relationship between clonal size and metabolic state. In CD4⁺ CTLs (MC05) from responders, larger clones tended to show enhanced OXPHOS following RTX treatment (Figure 5D), while inter-patient variability was observed in patient-level visualization (Figure S19). Further, stratification by clone size revealed that the increase in OXPHOS pathway activity following RTX treatment in responders tended to be greater in clonotypes with larger clone sizes (Figure 5E): the hyperexpanded clonotypes (clone size ≥100) preferentially adopted an OXPHOS-high phenotype in response to RTX treatment in responders. This finding was consistent with a previous report showing that prevention of mitochondrial oxidative stress during chronic T cell stimulation allowed sustained T cell proliferation.⁴¹ The association between clonal expansion and OXPHOS enhancement was not evident in the other mature T cell subsets (MC03, 04, and 09) (Figure S20). In RTX non-responders, changes in clonal expansion and metabolic signatures were not observed (Figures 5C–5E, S19 and S20). Similarity network analysis of TCR sequences showed an oligoclonal expansion of T cells induced by RTX in responders, whereas the majority of these expanded clonotypes were patient-specific (Figure 5F).

Transcription factors and upstream regulators in RTX response

We next explored the dynamics of gene-regulatory networks and transcription factor (TF) activities in RTX responses using SCENIC analysis. An overview of the normalized regulon activity scores (regulon specificity scores; RSS) for all 136 detected TFs reveals differentiation- and RTX responsiveness-associated changes (Figure 6A). The selected 71 TFs, which exhibited changes in activity (Δz-scaled RSS ≥0.2 between post- and pre-treatment) in at least two of the predefined ten subsets (MC00, 02, 01, 03, 04, 05, 08, 09, 06, and 07) when analyzed in responders, were further classified into four clusters (A–D) (Figures 6B and S21). Cluster A, comprising MAFF, MAF, and TFEB, showed a consistent decrease in activity across most early-to-mature T cell subsets (MC00–MC09), specifically in RTX responders (Figure 6B). Albeit modest, Cluster B showed similar trends. In contrast, cluster C, including many cytokine signaling-related TFs (STATs and IRFs), showed a broad increase in activity in RTX responders (Figure 6B).

Figure 6.

Transcriptional regulators and upstream modulators of RTX response in responders

(A) Heatmap of z-scaled RSS for 136 TFs across all T cells, stratified by cluster identity (MC00–MC14), RTX treatment status (pre/post), and clinical response (RTX responders/non-responders).

(B) Matrix plot summarizes treatment-induced changes in TF activity (Δz-scaled RSS, post minus pre) in RTX responders and non-responders. A total of 71 TFs shows Δz-scaled RSS (post minus pre) ≥ 0.2 in at least two of the predefined ten subsets (MC00, 02, 01, 03, 04, 05, 08, 09, 06, and 07) in responders were selected and grouped into four clusters (A–D) according to their activity patterns.

(C) Proportion of predicted target genes for MAFF, MAF, and TFEB (Cluster A TFs) across the eight DEG groups (Up_C1–C4, Down_C1–C4) defined in Figure 2B. Bar plots show the relative contribution of each TF’s target genes to each DEG Group. Notably, target genes are observed only in the Down_C1–C4 DEG groups.

(D) Table summarizes the overlap of predicted target genes of Cluster A TFs (MAFF, MAF, and TFEB) with those of other TFs, based on 22 genes identified as the overlap between the 50 predicted target genes of MAF, MAFF, and TFEB from SCENIC regulon analysis (Figure S22B) and the DEGs from Figure 2B. Of these 22 genes, all were co-targeted by at least one additional TF, indicating potential coregulatory interactions.

(E) Top predicted upstream regulators inferred by Ingenuity Pathway Analysis (IPA) using DEGs from pre-vs. post-treatment responder T cells.

(F) Schematic model illustrates key changes induced by RTX in responders. Images created in BioRender were used in (F). Abbreviations: TF, transcription factor; RSS, regulon specificity scores.

We examined the overlap between the predicted target genes of the TFs and the eight DEG groups defined in Figure 2B (Up_C1–C4 and Down_C1–C4) (Figure S22A). As expected, the predicted targets of TFs in clusters A and B were enriched in downregulated DEGs, while the predicted targets of TFs in cluster C were enriched in upregulated DEGs (Figure S22A). As MAF reportedly drives T cell exhaustion,⁴⁵ we further focused on cluster A TFs. Among MAFF, MAF, and TFEB, MAFF had the largest number of targets within Down_C2 DEG group associated with TNF-α-NF-κB signaling and hypoxia-related pathways (Figure 6C). Furthermore, cluster A TFs form a broad co-regulatory network with other TFs, sharing many MAFF targets with FOSL2 (Figures 6D and S22B). These co-targets included ZFP36 and TIPARP, which play important roles in T cell homeostasis⁴⁶ and mitochondrial dysfunction,⁴⁷ respectively (Figure 6D).

Ingenuity pathway analysis (IPA) identified IL-2 as a candidate activated upstream regulator across a broad range of T cell subsets in RTX responders (Figure 6E). This was consistent with enrichment of IL-2–STAT5 signaling in the Up_C4 DEG group (Figure 2E). The predicted inhibited upstream regulators included IL-4, IL-5, and EGF, as well as TFs such as CEBPB, which were also highlighted in the co-regulation analysis (Figure 6D). Inhibition of dexamethasone was also predicted, potentially reflecting corticosteroid tapering or enhanced T cell activation.

Taken together, our exploratory study suggests that RTX treatment may induce widespread changes in TF activities involved in T cell activation and exhaustion in responders, ameliorate mitochondrial dysfunction by upregulating OXPHOS capacity, and reduce ROS levels (Figure 6F). These findings should be investigated by further functional and mechanistic investigation and validation in larger studies.

Widespread B-T cell crosstalk in childhood INS

While the present study is constrained by the limited sample size and patient heterogeneity, our results suggest that B cell depletion causes widespread transcriptional changes in T cells in RTX responders. This raises the possibility that B-T cell communication is involved in INS pathogenesis. To test this possibility, we re-analyzed a recent scRNA-seq dataset of PBMCs comprising T and B cells from four pediatric patients with INS and four healthy controls (HC).¹⁵ Because T cells were not investigated in detail in this study, T cells were re-clustered and aligned to the T cell subsets (MC#) identified in our dataset (Figures S23A–S23C). Although the proportion of exhausted CD8⁺ T cells (MC07) did not differ significantly between INS and HC (p = 0.29) due to a small sample size, three of four INS patients showed values exceeding the HC median (Figure S23D).

We next evaluated intercellular communication using CellChat analysis. We observed two intriguing findings: first, B cells showed increased signals to T cells in INS, and second, MC05 (CD4⁺ CTLs) acted as a major signal receiver from various cell types (Figure S24). Several pathways, including MHC-II, LCK, Galectin, CD23, ICAM, and SELPLG, were suggested to be more active in the INS (Figure S25). INS-specific communication revealed (1) increased MHC-II and IL-16 signaling from B cells to naive and memory CD4⁺ T cells, (2) enhanced CD23, ICAM, and BAG signaling from B cells to CD4⁺ CTLs and CD8⁺ T cells, and (3) elevated TGF-β signaling from B cells to exhausted CD8⁺ T cells (Figure S26). B–T cell TNF signaling was also prominent. Additionally, CD4⁺ CTLs appeared to send reverse signals to B cells via CADM and LAIR1 (Figure S26A).

Finally, we inferred the metabolic state of T cells (Figure S27). HC-derived T cells are dominated by high OXPHOS and low glycolysis status and display a standard metabolic shift from OXPHOS to glycolysis along T cell differentiation,³⁹ emphasizing the unique metabolic trajectories in our datasets under long-term steroid treatment (Figure 3). In contrast, INS-derived T cells exhibited moderate OXPHOS and elevated glycolysis. These analyses collectively suggest that pathogenic B cells intimately interact with T cells and cause their metabolic reprogramming, possibly toward exhaustion.

Discussion

In this study, we implemented an integrative computational framework combining DEG module discovery with SCENIC-based regulatory network inference, MAGIC-enhanced metabolic state classification, CellChat-based B–T communication analysis in pediatric INS datasets, and integration of TCR clonality with metabolic programs to link RTX responsiveness to transcriptional, metabolic, and clonal T cell states. While the sample size is limited, our analysis showed distinct transcriptomic responses of T cells to RTX treatment in RTX responders and non-responders, providing findings that warrant further investigation. In RTX responders, widespread transcriptomic remodeling after RTX treatment was observed and accompanied by elevated OXPHOS in activated T cell subsets, especially in CD4⁺ CTLs (MC05), reduced exhausted CD8⁺ T cells, and clonal expansion of CD4⁺ CTLs (Figure 6F). In addition, responders showed elevated ROS levels at baseline, followed by a decrease after treatment (Figure 6F). The reanalysis of childhood INS datasets also suggested abnormal B–T cell crosstalk and altered T cell metabolism.

The present study raises the possibility that RTX may exert ripple effects on T cells in responders; however, how pathogenic B cells differentially communicate with T cells in responders versus non-responders remains an open question. Addressing this issue will require more comprehensive analyses that include B cells. In the present study, such analyses were not feasible for two main reasons. First, reflecting the retrospective nature of this study, only cryopreserved samples were available rather than fresh specimens. Second, in our pilot study, B cell viability was modestly affected by standard cryopreservation methods, resulting in slightly less consistent detection by flow cytometry compared with T cells. Furthermore, although the present study analyzed samples obtained only one month after RTX treatment, longitudinal analyses will also be necessary to capture the dynamic temporal changes in B and T cells following RTX administration. Despite these limitations, by focusing on T cells, our findings suggest the potential importance of T cell-specific alterations associated with RTX responsiveness. To our knowledge, this is the first study to analyze immune cells at single-cell resolution in adult SD-MCD, which has historically been less characterized than pediatric INS due to its rarity. To more comprehensively analyze B–T cell crosstalk, we are currently conducting integrated B–T cell transcriptional profiling in an expanded adult MCD cohort using optimized cryopreservation protocols that preserve viability of B cells (data not shown). Such analyses will also be important for determining whether the pathogenic B cells in our adult cases correspond to the extrafollicular B cells recently identified in pediatric INS,¹⁵ although differences in patient characteristics between children and adults should be carefully considered. Furthermore, given that RTX fully depletes B cells over 4–6 months,^17,20 focusing on T cells as secondary effectors of RTX may be useful for developing surrogate markers of RTX responsiveness, especially during B cell depletion periods. Currently, the contribution of T cells in RTX non-responders remains largely unclear. Pathogenic mechanisms in RTX non-responders may be different and potentially overlap with those in focal segmental glomerulosclerosis (FSGS), which is refractory to RTX. In FSGS, RTX resistance has been attributed to circulating permeability factors⁴⁸ and persistent T cell activation that are not fully corrected by CD20⁺ B-cell depletion.⁴⁹ Likewise, the lack of transcriptomic reprogramming in T cells observed in our non-responders may reflect a shift toward a B-cell-independent, FSGS-like pathogenic process.

RTX is primarily understood to exert immunomodulatory effects through B cell depletion. Additionally, RTX reportedly alters T cell homeostasis in INS,⁵⁰ other autoimmune diseases,⁵¹ and autoimmune mouse models.^35,52 While RTX reportedly modulates regulatory T cells (Tregs) in INS or MCD,²⁸ we did not observe consistent changes in Treg abundance following treatment, possibly because of small sample sizes and low Treg abundance. In the present study, we observed that RTX induces upregulation of OXPHOS genes in mature T cells and decreases ROS accumulation in responders. These findings are consistent with a report showing that chronic antigenic stimulation inhibits coupled OXPHOS and drives ROS-generating mitochondrial dysfunction, which can be reversed by antioxidant treatment.⁴¹ Therefore, in the context of MCD, RTX appears to ameliorate mitochondrial dysfunction and oxidative stress under the exhaustion-prone conditions, which are potentially mediated by chronic interactions with pathogenic B cells and chronic antigen stimulation. Consistent with this interpretation, reanalysis of pediatric INS datasets demonstrated elevated MHC-II and IL-16 signaling between pathogenic B cells and CD4⁺ T cells (Figure S26A), which may drive persistent antigen stimulation and exhaustion-associated mitochondrial dysfunction in INS. This RTX-induced response is similar to the effects of blunting B-T cell crosstalk in chronic lymphocytic leukemia (CLL), in which depletion of CLL cells reverses oxidative stress and impaired glycolysis in CD4⁺ T cells.⁵³

The therapeutic significance of RTX-mediated ripple effects on T cells remains unclear. In other words, it is unclear whether T cell modulation is beneficial or a secondary effect. While further investigation is necessary, possible therapeutic scenarios may include (1) normalization of inflammatory conditions mediated by T cells with mitochondrial malfunction and/or exhausted T cells and (2) normalization of B-T crosstalk. In the latter scenario, clonal expansion of CD4⁺ CTLs may be important for the normalization of B-T crosstalk. CD4⁺ CTLs, which lack CD28 and express cytolytic molecules such as perforin and granzymes, are implicated in aging,⁵⁴ autoimmune diseases,^55,56,57 viral infections,⁵⁸ and cancer,⁵⁹ as well as the related subsets such as T peripheral helper (Tph) cells and age-associated helper T (ThA) cells. While these cells appear to have tissue-damaging phenotypes in several autoimmune diseases,^55,60 several reports have suggested disease-protective roles in cancer⁵⁹ and aging,^61,62 indicating a high context dependency. Given that CD4⁺ CTLs appeared to send reverse signals to ASCs, one of the pathogenic B cells, via CADM and LAIR1 in pediatric INS (Figure S26A), CD4⁺ CTLs may contribute to the maintenance of RTX response. While typical T-B cell interactions, especially those involving follicular helper T (Tfh) cells, are strongly associated with B cell follicles (“follicular pathway”),⁶³ recent studies have suggested an alternative mode of T-B cell interactions, which involves extrafollicular age-associated B cells and ThA cells and contributes to autoimmune pathogenesis (“extrafollicular pathway”).⁵⁵ Strong interactions between B cells and CD4⁺ CTLs (Figure S24A) may present similar extrafollicular crosstalk in the INS.

In summary, our scRNA-seq study explored transcriptome changes in T cells associated with RTX responsiveness in SD-MCD, which should be validated in larger studies. The potential features associated with favorable RTX response include enhancement of OXPHOS transcriptional signatures in activated T cell subsets, reduction in ROS levels, decrease in exhausted CD8⁺ T cells, and clonal expansion of CD4⁺ CTLs, whereas non-responders exhibited minimal such changes. While the causality and clinical utility of these features require validation in larger, independent cohorts, these features identified here may provide a clue for future investigation of T cell-based stratification strategies in steroid-dependent MCD.

Limitations of the study

This study has several limitations. Since steroid-dependent MCD with a long-term clinical course in adults is relatively rare, the limited sample size and heterogeneous patient characteristics restricted our ability to perform functional and mechanistic investigation and validation in large cohorts. Differences in age and gender between responders and non-responders also restricted our ability to rigorously assess interaction effects between these covariates and RTX responsiveness, particularly those related to pediatric-versus adult-onset disease. Therefore, analyses in larger cohorts will be required in future studies.

As described above, how pathogenic B cells differentially communicate with T cells in responders versus non-responders remains an open question. Addressing this issue will require more comprehensive analyses that include B cells and longitudinal analyses, although such analyses were not feasible in the present study. Although baseline differences in ROS levels between responders and non-responders were suggestive of predictive value, they did not reach statistical significance in our cohort and thus remain hypothesis-generating observations requiring validation in larger studies. In addition, more detailed investigations of immune cell function, including mitochondrial activity and redox regulation, will be necessary to further elucidate the underlying mechanisms.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Hiroshi I. Suzuki (hisuzuki@med.nagoya-u.ac.jp).

Materials availability

No new, unique reagents were generated in this study.

Data and code availability

• Data: The raw sequence data reported in this paper have been deposited in Gene Expression Omnibus: GSE305637.

• Code: This paper does not report original code.

• Additional information: Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We are grateful to Suzuki and Maruyama laboratory members for discussion on the project. We thank the Division for Medical Research Engineering, Nagoya University Graduate School of Medicine, for the technical support of the usage of LSRFortessa X-20 and FACSAria™ Fusion flow cytometer. We appreciate the support of the 2022 Young Researcher’s Award of the Japanese Society of Women Nephrologist (JSWN). This work was supported by grants from Japan Agency for Medical Research and Development (AMED) (JP24ek0109753 to EKI, AT, KF, and SM, JP22ama221111, JP23kk0305026, JP23tk0124003, JP24ck0106875, JP25nk0101727, JP25ck0106019, and JP25ak0101291 to HIS, JP23ck0106791 to HIS and KO, and JP25kk0305028 to KO); YOKOYAMA Foundation for Clinical Pharmacology (YRY-2412 to EKI); Aichi Kidney Foundation Grant (to EKI); JSPS Grant-in-Aid for Research Activity Start-up (JP24K23431 to EKI); JSPS Grant-in-Aid for Early-Career Scientists (JP25K19485 to YW and JP25K19486 to SK); JSPS Grant-in-Aid for Scientific Research (A) (JP24H00614 to HIS); JSPS Home-Returning Researcher Development Research (JP19K24694 to HIS); the Takeda Science Foundation (to HIS); Toray Science Foundation (22-6304 to HIS); Inamori Research Institute for Science (InaRIS) (to HIS); and JSPS fellowship for Young Scientists (to YS).

Author contributions

E.K.I., Y.W., C.O., K.F., and H.I.S. carried out study design; E.K.I., Y.W., C.O., A.H., K.F., and H.I.S. performed experiments and data analysis; A.H., K.O., S.Y., Y.S., S.K., and A.T. interpreted the data; E.K.I., Y.W., K.F., and H.I.S. wrote the manuscript; K.F., S.M., and H.I.S. supervised the project. All authors had final approval of the submitted versions.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used ChatGPT and Perplexity to assist with code and grammatical text edits. The authors reviewed and edited the content as needed and take full responsibility for the publication’s content.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anit-human-CD3-BV786 (Clone: SK7)	BD	Cat# 563800; RRID: AB_2744384
Anit-human-CD235-APC-Cy7 (Clone: HI264)	BioLegend	Cat# 349116; RRID: AB_2650978
Fixable Viability Stain 780	BD	Cat# 565388; RRID: AB_2869673
Anti-human-CD45- PerCP-CyTM5.5 (Clone: HI30)	BD	Cat# 564106; RRID: AB_2744405
Anti-human-CD8- APC-R700 (Clone: SK1)	BD	Cat# 565192; RRID: AB_2739104
Anti-human-CD45RA-BV510 (Clone: HI100)	BD	Cat# 566114; RRID: AB_2739516
Anti-human-CD197-PE-CF594 (Clone: 150503)	BD	Cat# 562381; RRID: AB_11153301
Anti-human-CD4-BV650 (Clone: SK3)	BD	Cat# 563875; RRID: AB_2744425
Chemicals, peptides, and recombinant proteins
10x Genomics Chromium Next GEM Single Cell 5′ Reagent Kits v2 (Dual Index)	10x Genomics	Cat# PN-1000263
Critical commercial assays
CellROX™ Green Reagent, for oxidative stress detection	Thermo Fisher Scientific	Cat# C10444
Ficoll Paque PLUS™	Cytiva	Cat#17144003
SepMate-50™	STEMCELL Technologies	Cat#ST-86450
Phosphate-buffered saline (PBS)	Cell Science & Technology Institute Inc.	Cat#1102P0
BAMBANKER®	LTC	Cat#CS-02-001
DNase I stock solution	QIAGEN	Cat#79254
BD Horizon™ Brilliant Stain Buffer	BD Biosciences	Cat#563794; RRID: AB_2869750
Deposited data
single-cell RNA sequencing data from samples of patients	This study	Gene Expression Omnibus: GSE305637
single-cell RNA sequencing data from four pediatric patients with idiopathic nephrotic syndrome and four healthy controls	Al-Aubodah et al.15	Gene Expression Omnibus: GSE233277
Software and algorithms
FlowJo™ software v.10.8.1	BD Biosciences	https://flowjo.com/previous-versions-flowjo
Cell Ranger 8.0.0	10x Genomics	https://www.10xgenomics.com/support/software/cell-ranger/latest
MAST	McDavid et al.38	https://bioconductor.org/packages/MAST/
Seurat (v5.1.0)	Hao et al.64	https://satijalab.org/seurat/
scDblFinder (v1.20.0)	Germain et al.65	https://bioconductor.org/packages/scDblFinder/
SlingShot (v2.14.0)	Street et al.66	https://bioconductor.org/packages/slingshot/
edgeR (v4.8.2)	Chen et al.67	https://bioconductor.org/packages/edgeR/
ComplexHeatmap (v2.21.1)	Gu et al.68	https://bioconductor.org/packages/ComplexHeatmap/
clusterProfiler (v4.14.0)	Wu et al.69	https://bioconductor.org/packages/clusterProfiler/
msigdbr (v7.5.1)	Dolgalev et al.70	https://CRAN.R-project.org/package=msigdbr
Cytoscape	Shannon et al.71	https://www.cytoscape.org/
MAGIC (v3.0.0)	van Dijk et al.72	https://github.com/KrishnaswamyLab/MAGIC
GSVA (v2.0.5)	Hanzelmann et al.73	https://bioconductor.org/packages/GSVA/
ggplot2	Wickham et al.74	https://CRAN.R-project.org/package=ggplot2
biomaRt	Durinck et al.75	https://bioconductor.org/packages/biomaRt/
Scirpy (v0.20.0)	Sturm et al.76	https://pypi.org/project/scirpy/
SCENIC/pySCENIC (v0.12.1)	Van de Sande et al.77	https://pyscenic.readthedocs.io/
Ingenuity Pathway Analysis (IPA)	QIAGEN	https://digitalinsights.qiagen.com/products-overview/discovery-insights-portfolio/analysis-and-visualization/qiagen-ipa/

Experimental model and study participant details

This study investigated the T cell transcriptomic responses to RTX therapy in six adult patients with SD-MCD. Responders maintained steroid-free remission for >6 months post-treatment, whereas non-responders relapsed within 6 months or could not discontinue steroids throughout the observation period. PBMCs were collected before and one month after RTX administration. scRNA-seq and TCR profiling were performed using 10x Genomics platform. Quality control excluded low-quality cells. The main endpoints were the changes in gene expression and TF activity after RTX treatment. All clinical and experimental procedures were approved by the Institutional Review Board of the Nagoya University Graduate School of Medicine (Nagoya Kidney Disease Registry; 2010-1135), and informed consent was obtained from all participants. All patients included in the scRNA-seq analysis were Asian adults, comprising four males and two females, with a median age of 35.5 years (IQR 23–49). The clinical courses and laboratory data of individual patients are summarized in Figures 1B and S1, Tables 1, S1 and S2. We also measured intracellular reactive oxygen species (ROS) levels in an independent patient cohort, including three responders and five non-responders. All patients in this cohort were Asian adults, comprising four males and four females, with a median age of 33 years (IQR 28–48). The characteristics of these patients are summarized in Table S3.

Method details

PBMC isolation and storage

Peripheral blood collected in heparinized tubes was centrifuged at 1500 rpm for 10 min at room temperature, and the PBMC fraction was obtained using density gradient separation with Ficoll Paque PLUS^TM (#17144003, Cytiva) by centrifugation at 1200 × g for 10 min at room temperature in SepMate-50^TM (#ST-86450, STEMCELL Technologies). Isolated PBMCs were washed twice with phosphate-buffered saline (PBS) (#1102P0, Cell Science & Technology Institute Inc.) at 1500 rpm for 10 min at room temperature. The PBMCs were dissolved in BAMBANKER® (#CS-02-001, LTC) and kept at -80°C for 1 day and then stored at -150°C. The experimental procedure is illustrated in Figure 1C.

Sorting, single-cell isolation, and single-cell RNA-seq library preparation

Frozen PBMC samples were thawed rapidly in trypsin-stopping solution (D-MEM Low Glucose, 10% fetal bovine serum (FBS), and 1% penicillin/streptomycin) at 37°C, centrifuged at 400 ×g for 10 min at 4°C and the supernatant was discarded. To release cell aggregation, 25 μL of DNase Booster Buffer and 10 μL of DNase I stock solution (#79254, QIAGEN) were added to the pellet and incubated at room temperature for 15 min. The samples were washed with a trypsin-stopping solution and diluted in PBS.

To isolate CD3⁺ T cells, PBMCs collected at both pre- and post-treatment with RTX were stained and sorted at 4°C using pooled monoclonal antibodies listed on Key resources table and sorted using FACSAria^TM Fusion flow cytometer (BD Biosciences). Non-specific binding was blocked with Fc-gamma receptor antibodies, and 20% BD Horizon™ Brilliant Stain Buffer (#563794, BD Biosciences) was added to prevent fluorescent dye interactions that may cause staining artifacts. Stained PBMCs were gated on side scatter (SSC)^hi/forward scatter (FSC)^hi, CD235^lo, FVS^lo, and CD3⁺. The gating strategy is shown in Figure S2. Single-cell 5′ and TCR libraries were prepared following the protocol provided by the 10x Genomics Chromium Next GEM Single Cell 5′ Reagent Kits v2 (Dual Index). Each single-cell 5′ and TCR libraries were sequenced using the Illumina Novaseq 6000 with 150 paired-end reads.

Flow cytometry for ROS measurement

PBMCs obtained immediately before the third administration of RTX and one month after treatment were available for eight patients. These patients were classified as RTX responders (n = 3) and RTX non-responders (n = 5) according to the criteria described above (Table S3). Changes in the mean fluorescence intensity (MFI) of ROS were compared between these two groups. After thawing the frozen PBMCs with the same procedure of FACS, staining for ROS was performed using CellROX™ Green Reagent for oxidative stress detection (Thermo Fisher Scientific, Cat# C10444), according to the manufacturer’s instructions. Multicolor flow cytometry on PBMCs was similarly performed using LSRFortessa™ X-20 (BD Biosciences) with pooled monoclonal antibodies (Key resources table) to define the appropriate lymphocyte subpopulations (Figure S17). The number and mean fluorescence index of each subpopulation were determined by gating using FlowJo™ software v.10.8.1 (BD Biosciences).

Quantification and statistical analysis

Single-cell RNA-seq data processing

scRNA-seq computational pipelines and analysis

Droplet-based sequencing data were aligned and quantified using Cell Ranger 8.0.0 (10x Genomics) against the GRCh38 human reference genome. The R package Seurat (v5.1.0)⁶⁴ was used for data scaling, clustering, dimensionality reduction, and visualization. Quality control was performed to remove low-quality cells, retaining those with between 200 and 3000 detected genes, mitochondrial gene content below 15%, and singlet detected using scDblFinder (v1.20.0).⁶⁵

Data integration was performed using IntegrateLayers() function with the reciprocal principal component analysis (RPCA) integration method. The data were normalized and scaled using Seurat function with default parameters. The UMAP dimension reduction was generated based on the first 20 RPCA-adjusted principal components. A nearest-neighbor graph using the first 20 RPCA-adjusted principal components was constructed using FindNeighbors() function, followed by clustering using FindClusters() function (resolution = 1.0).

Trajectory analysis

Single-cell trajectory analysis was performed using SlingShot (v2.14.0)⁶⁶ package. From the clustered dataset, CD4⁺ and CD8⁺ T cell subsets were extracted separately. For each subset, trajectories were inferred starting from naïve T cells (MC00 for naïve CD4⁺ and MC02 for naïve CD8⁺ cells). The results were visualized on a UMAP embedding.

Differential expression analysis

Differential expression analysis for pre- and post-RTX treatment comparisons was performed using edgeR (v4.8.2)⁶⁷ package. We made pseudobulk expression matrix using AggregateExpression() function provided by Seurat. For pre- versus post-RTX treatment comparisons, paired differential expression analysis was performed by including patient identity in the design matrix, followed by testing using glmLRT() function provided by edgeR. DEGs between pre- and post-RTX treatment within each subset were identified using a threshold of FDR < 0.1 and |log₂(fold change)| > 0.6. For comparisons between responders and non-responders at pre-RTX treatment, differential expression analysis was performed using the MAST approach³⁸ for the scRNA-seq datasets. DEGs were identified using a threshold of Benjamini-Hochberg adjusted p < 0.05 and |log2(fold change)| > 0.6.

DEGs from all subsets were aggregated and separated into upregulated and downregulated groups, and each group was subjected to k-means clustering (k = 4). Genes were grouped based on their expression patterns, and the resulting subsets were visualized using heatmaps. Clustering and heatmap generation were performed using ComplexHeatmap (v2.21.1)⁶⁸ package. Enrichment analysis was performed using clusterProfiler (v4.14.0)⁶⁹ package. Expression changes in the DEG groups upon RTX treatment were predominant in RTX responders but not in non-responders (Figures 2B, S7, and S8). Figure 2C shows the representative genes in the DEG groups identified in Figure 2B.

We downloaded ontology data (Hallmark, Gene Ontology Biological Process (GOBP) using msigdbr (v7.5.1)⁷⁰ package, and used enrichr() function based on the previously identified DEGs. DEGs and gene set analysis were visualized using Cytoscape.⁷¹

Overlap analysis between treatment-associated and baseline DEG groups

Gene overlap between DEG clusters from Figures 2B and 4B was assessed by pairwise hypergeometric testing. Resulting p-values were corrected for multiple comparisons using the Benjamini-Hochberg procedure, and overlaps with q < 0.01 were considered significant.

Pathway activity score (PAS) analysis

MAGIC software (v3.0.0)⁷² was applied to the scRNA-seq dataset. Based on the imputed expression matrix, gene set variation analysis (GSVA) scores were calculated using the GSVA package (v2.0.5)⁷³ in R. Pathway activity scores (PAS) for each cell were then derived from the GSVA scores according to the following formula: where max and min represent the maximum and minimum GSVA scores, respectively, across all the cells in the dataset, respectively. Contour plots of the PAS distributions were generated using the geom_density_2d() function from the ggplot2 package,⁷⁴ based on the PAS values of individual cells.

$$\text{PAS}=\frac{\text{GSVA}\text{score}-\mathit{min}\left(\text{GSVA}\text{score}\right)}{\mathit{max}\left(\text{GSVA}\text{score}\right)-\mathit{min}\left(\text{GSVA}\text{score}\right)}$$

Gene sets for OXPHOS and glycolysis were obtained from a previously published study,⁷⁸ with gene symbols converted from mouse to human using biomaRt package.⁷⁵ ROS gene sets were derived from the Hallmark Reactive Oxygen Species pathway in the MSigDB.⁷⁰ These methods were used to generate Figures 3, 5D, 5E, S10, S11, S19, and S20. Based on the PAS values, the cells were classified into four metabolic subgroups: subgroup A, low OXPHOS; subgroup B, moderate OXPHOS; subgroup C, high OXPHOS and low glycolysis; and subgroup D, high OXPHOS and high glycolysis (Figures 3A and S10). Additionally, the redistribution of metabolic status across T cell subsets is summarized in a Sankey diagram (Figure 3B), mapping each metabolic subgroup (A–D) to T cell subsets (MC00–MC14). In our analysis, in naïve cells (MC00 and 02), glycolysis PAS was higher in RTX non-responders than in responders (Figures 3A and S10). This baseline change in glycolysis PAS in naïve cells between RTX responders and non-responders may be associated with differences in the relative use of OXPHOS and glycolysis under potentially distinct pathological conditions.

TCR analysis

Droplet-based sequencing data for the TCR sequences were aligned and quantified using Cell Ranger 8.0.0 (10x Genomics) against the GRCh38 human VDJ reference genome. Filtered annotated contigs for TCR sequences were analyzed using Scirpy (v0.20.0).⁷⁶ For TCR analysis, we selected T cells that were included in the scRNA-seq data. TCR clone sizes were projected onto the UMAP embedding (Figure 5A). Clone size was defined as the number of cells sharing an identical TCR clonotype. The clonal ratio represents the proportion of cells belonging to each clone size category. We evaluated the association between TCR clone size and metabolic status by correlating OXPHOS and glycolysis PAS with clone size and comparing OXPHOS PAS distributions across clone size categories in CD4⁺ CTLs (Figures 5D, 5E and S19) and in the other T cell subsets (TEMRA CD8⁺ T cells and CM CD8⁺ T cells; Figure S20). The TCR repertoire similarity was quantified using ir_dist() function. We constructed similarity networks based on CDR3 amino acid sequences of both TCR α-chains and β-chains (Figure 5F). Clonotype assignment was performed using define_clonotype_clusters() function, and clonotype networks were generated using clonotype_network() function with parameters: min_cells = 10, sequence = “aa,” metric = “tcrdist.”

Transcription factor (TF)-gene regulatory network analysis

Gene-regulatory networks and regulon activities were inferred with single-cell transcription factor regulatory network inference and clustering (SCENIC) (pySCENIC v0.12.1),⁷⁷ implemented via Docker, using the default parameters. The target genes for each TF were generated using the pySCENIC ctx command. Regulon specificity scores (RSS) were calculated using regulon_specificity_scores() function based on the loom file output from the pyscenic aucell command. The scores were computed for the groups stratified according to prognosis, subset identity, and treatment status (pre- or post- RTX). Figure 6A shows a heatmap of the normalized RSS values for all 136 detected TFs, providing an overview of TF activity across the differentiation trajectory and treatment groups.

To identify TFs whose activities were significantly altered by RTX treatment, we selected those TFs that exhibited substantial shifts in RSS, especially among responders. Specifically, we focused on 71 TFs that showed a change in z-scaled RSS (post minus pre ≥0.2) in at least two of the predefined ten subsets (MC00, 02, 01, 03, 04, 05, 08, 09, 06, and 07) among responders, indicating that their activities were notably modulated following RTX treatment (Figures 6B and S21). For cluster A TFs (MAFF, MAF, and TFEB), we counted the number of their predicted target DEGs from each DEG group in Figure 2B (Figure 6C). Co-regulatory interactions between cluster A TFs and other TFs were investigated by analyzing the overlap between their predicted targets (Figures 6D and S22B). TF activity heatmaps were generated for all detected TFs, as well as for the group of TFs selected based on the above criteria using z-scaled RSS with ComplexHeatmap (v2.21.1).⁶⁸ Upstream regulator analysis was performed using Ingenuity Pathway Analysis (QIAGEN) based on DEGs between pre- and post-treatment T cells in RTX responders to explore the upstream modulators of these transcriptional changes.

Cell-cell interaction analysis

For the analysis of cell-cell communication, we utilized the pediatric INS scRNA-seq dataset of PBMCs (Gene Expression Omnibus: GSE233277) comprising T and B cells from four pediatric patients with INS and four healthy controls (HC).¹⁵ Among the former (patients with INS), one patient was at initial onset, one patient was immediately after induction of remission with steroids, and two patients had samples collected at the time of relapse during observation without treatment after achieving remission with steroids. Clustering was re-performed using Seurat package. Because T cells were not investigated in detail in this study, T cells were re-clustered and aligned to the T cell subsets (MC#) identified in our dataset using UMAP embedding and canonical marker gene expression (Figures S23A–S23C). Cell-cell communication analysis was conducted using CellChat (v2.1.2)⁴² package. Communication probabilities were computed using the computeCommunProb() function with the parameters: type = “truncatedMean”, trim = 0.1.

To compare the interaction profiles between the disease and control states, we calculated the differences in the interaction counts for each subset between the INS and HC groups. Figures S24B and S24C display the raw differential numbers of interactions and differential interaction strengths between the INS and HC samples in Figure S24A, respectively, before sorting. To identify the signaling pathways that mediated differential interactions between the INS and HC in Figure S24A, we calculated the pathway-specific communication strength using CellChat’s information flow metric (Figure S25A). A comprehensive overview of the incoming and outgoing signaling strengths across all pathways is provided in Figures S25B–S25E. In addition, we quantified the differences in importance scores (as defined by CellChat) for each subset across signaling pathways. Figure S26B shows the per-pathway signaling differences (INS minus HC) from which the nine pathways shown in Figure S26A were selected. To focus on INS-specific communication, we selected nine signaling pathways that demonstrated marked differences in information flow and visualized the changes in network centrality scores (Δimportance) between INS and HC in Figure S26A.

Statistical analysis

For comparisons of study participant parameters, the Wilcoxon signed-rank test was used for continuous variables, and the Mann–Whitney U test was applied for categorical variables. Differences in cell counts (Figures 1G, 1H, 1J, 4D and S17) were assessed using paired t-tests for before-and-after comparisons or unpaired t-tests otherwise. For differences in the MFI shown in Figure 4D, one-way ANOVA was also used to compare the four groups. Chi-square tests were used to analyze differences in TCR clonal ratios (Figure 5C). Wilcoxon signed-rank tests were used to analyze differences in PAS (Figures 5E and S20). The significance threshold for these tests was set at p < 0.05, and statistically significant comparisons are indicated by asterisks in the corresponding panels (Figures 1H, 4D, 5C, 5E and S20). Details of statistical tests applied in the single-cell analysis are described in the “single-cell RNA-seq data processing” section.