The integrated stress response pathway controls cytokine production in tissue-resident memory CD4+ T cells

T cell memory is a hallmark of the adaptive immune system that results from the antigen-specific activation, expansion and maintenance of T cells. Antigen-experienced memory T (T_M) cells elicit a faster and more potent immune response upon re-exposure to the same antigen and provide superior protection against reinfection than naive T (T_N) cells¹. Based on their specific homing capacity and effector functions, circulating T_M cells can be divided into central memory T (T_CM) and effector memory (T_EM) cells¹. In contrast, tissue-resident memory T (T_RM) cells do not recirculate and exhibit long-term persistence at sites of previous infections^2,3,4,5,6. T_RM cells provide long-term immunity against local reinfections^2,3,4,5,6, but they also contribute to immune-mediated inflammatory diseases of the gut⁷, central nervous system⁸, skin^9,10 and kidney^11,12,13. The local production of cytokines by CD4⁺ T_RM cells orchestrates protective tissue immunity against pathogens, but can also promote hyperinflammation and tissue damage in immune-mediated diseases^{14,15,16,17,18,19}.

Under homeostatic conditions, T_M cells in nonlymphoid tissues express mRNA encoding effector molecules such as IFNγ and TNF for rapid cytokine production through translation of the already-transcribed mRNA^20,21,22,23. Yet, whether different T_M cell subsets differentially store cytokine mRNA remains incompletely explored. To address this issue, we combined single-cell RNA sequencing (scRNA-seq) and cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) to analyze CD3⁺CD4⁺ and CD3⁺CD8⁺ T cells from renal tissue and matched blood of eight healthy donors (cohort 1, median age 60.5 years (interquartile range (IQR) 55–72), 62.5% male)¹³. mRNA for type 1 (IFNG, TNF, LTA and CSF2) and type 3 (IL17A, IL17F and IL22) cytokines was detected in CD4⁺ and CD8⁺ T cells in the kidney, but was expressed at much lower levels in CD4⁺ and CD8⁺ T cells from the blood (Fig. 1a,b and Extended Data Fig. 1a,b). RNA fluorescence in situ hybridization (FISH) analysis detected CD3⁺ T cells expressing IL17A, IFNG and CSF2 mRNA in healthy kidney tissue (Fig. 1c). Expression of IL17A and IL17F mRNA, and to a lesser degree of CSF2 and TNF mRNA, was higher in CD4⁺ T cells compared to CD8⁺ T cells in the kidney (Fig. 1b).

a, Uniform Manifold Approximation and Projection (UMAP) plots showing cytokine scores based on the expression of IL2, IL17A, IL17F, IL22, IFNG, TNF, CSF2 and LTA in CD4⁺ T cells and CD8⁺ T cells sorted from the blood and healthy kidney tissue of eight healthy donors (cohort 1). b, Heatmap showing the expression of the cytokine score genes in CD4⁺ T cells and CD8⁺ T cells from the blood and healthy kidney of donors in cohort 1. c, Representative images of mRNA FISH for IL17A, IFNG and CSF2 combined with immunofluorescence staining for CD3 in healthy kidney tissue from human donors in cohort 1. Scale bars, 50 μm. The experiment was repeated three times. d, UMAP plot of CD4⁺ T cells isolated from human healthy kidney tissue (cohort 1), showing the distribution of S1PR1⁻CD44⁺CD69⁺ T_RM cells, CCR7⁻S1PR1⁺CD44⁺ T_EM cells, CCR7⁺S1PR1⁺CD44⁺ T_CM cells and CD45RA⁺ T_N cells. e, Heatmaps showing the protein expression of CD45RA, CD44, CD62L, CCR7 and CD69 (top) and the mRNA expression of CD44, KLF2, S1PR1, SELL and CCR7 (bottom) in CD4⁺ T_RM, T_EM, T_CM and T_N cells (as in d). f, UMAP plot and violin plot showing cytokine scores in CD4⁺ T_RM, T_EM, T_CM and T_N cells (as in d). g, Heatmap showing the expression of the cytokine score genes in CD4⁺ T_RM, T_EM, T_CM and T_N cells (as in d). h, UMAP plot showing CCL5 mRNA expression in CD4⁺ T_RM, T_EM, T_CM and T_N cells (as in d). i, Representative flow cytometry plot showing the gating strategy of CD4⁺CD69⁺ T_RM cells and CD4⁺CD69⁻ non-T_RM cells and RT–qPCR analysis showing the expression of IL17A, IFNG and CSF2 mRNA in CD69⁺CD4⁺ T_RM cells and CD69⁻CD4⁺ non-T_RM cells sorted from the human healthy kidneys (n = 5, cohort 2). j,k, UMAP plots showing CD4⁺ T cells (left) and violin plots showing cytokine scores in SELL⁻CCR7⁻KLF2⁻ T_RM cells, SELL⁻CCR7⁻KLF2⁺ T_EM cells and SELL⁺CCR7⁺S1PR1⁺ T_CM cells (middle and right) in colon²⁴ (n = 4) (j) and lung²⁵ (n = 10) (k) (left) from healthy human donors. l, Percentage of polysome (translated) mRNA out of cytoplasmic RNA based on RT–qPCR quantification of IL17A, IFNG, CSF2 and ATF4 transcripts in CD4⁺ T_RM cells isolated from human healthy kidneys and left unstimulated or stimulated with PMA + Iono for 1 h (n = 5). Data are mean + s.e.m. Statistical analysis was performed using a one-way analysis of variance (ANOVA) with Tukey’s multiple comparison test (a,f,j) or unpaired two-tailed t-test with Welch’s correction (i,l); *P < 0.05, **P < 0.01. NS, not significant.

Full size image

Unsupervised clustering of human kidney CD4⁺ T cells identified CD45RA⁺ T_N, CCR7⁺S1PR1⁺CD44⁺ T_CM, CCR7⁻S1PR1⁺CD44⁺ T_EM and S1PR1⁻CD44⁺CD69⁺ T_RM cells (Fig. 1d,e and Extended Data Fig. 1c). CD4⁺ T_RM cells were the predominant cellular source of proinflammatory cytokine mRNA (for example, IL17A, IFNG and CSF2) and chemokine mRNA (for example, CCL4, CCL5 and CCL20) (Fig. 1f–h and Extended Data Fig. 1c,d). Quantitative PCR with reverse transcription (RT–qPCR) analysis indicated that IL17A, IFNG and CSF2 mRNA expression was significantly higher in CD69⁺CD4⁺ T_RM cells compared to CD69⁻CD4⁺ non-T_RM cells isolated from the kidney of five healthy donors (cohort 2, median age 59 years (IQR 55–64), 60% male) (Fig. 1i). To explore whether CD4⁺ T_RM cells from other tissues expressed cytokine mRNA, we analyzed public scRNA-seq datasets for healthy human colon²⁴ and lung²⁵. CD4⁺SELL⁻CCR7⁻KLF2⁻ T_RM cells were detected in the healthy colon (Fig. 1j) and lung (Fig. 1k), and expressed higher levels of cytokine mRNA, such as IFNG, TNF and CSF2, compared to other T_M cell populations (Fig. 1j,k and Extended Data Fig. 2a–d). These data showed that CD4⁺ T_RM cells expressed proinflammatory cytokine mRNA in various human tissues under noninflammatory conditions.

Production of inflammatory cytokines might lead to inflammation and immunopathology. In the kidney tissue where CD4⁺ T_RM cells were isolated, we did not observe any signs of active inflammation, such as immune cell infiltration (Extended Data Fig. 3a). To investigate whether cytokine mRNA expression in CD4⁺ T_RM cells led to cytokine production, we isolated CD4⁺ T cells from the blood and kidney of five healthy donors (cohort 2). These cells were treated with brefeldin A, an inhibitor of vesicle transport between the endoplasmic reticulum and the Golgi apparatus, in the presence or absence of phorbol 12-myristate 13-acetate (PMA) + ionomycin (Iono), which mimic T cell receptor activation, for 6 h. Stimulated kidney CD4⁺ T cells, but not unstimulated ones, produced interleukin (IL)-17A, interferon (IFN)γ and granulocyte–macrophage colony-stimulating factor (GM-CSF) protein (Extended Data Fig. 3b). To investigate whether expression of proinflammatory cytokine mRNA in T cells gave rise to tissue responses under homeostatic conditions, we analyzed the European Renal cDNA Bank database²⁶, which encompasses kidney biopsies from healthy individuals (living donor kidney transplantation) and from patients with autoimmune kidney diseases, such as anti-neutrophil cytoplasmic antibody-associated glomerulonephritis (ANCA-GN), lupus nephritis and IgA nephropathy. We found high expression of genes induced by IL-17, IFNγ or GM-CSF signaling, such as CXCL1 and CXCL5 (IL-17-responsive genes), CXCL9 and CXCL10 (IFNγ-responsive genes) and CCR2 and CCL22 (GM-CSF-responsive genes) in the kidney tissue from autoimmune kidney disease patients^27,28,29 (Extended Data Fig. 3c,d). In contrast, expression of these genes was significantly lower in healthy kidney tissue (Extended Data Fig. 3c,d), suggesting that the kidney does not exhibit a tissue response to cytokines under homeostatic conditions.

To investigate the translation status of the cytokine mRNA in resting CD4⁺ T_RM cells, CD69⁺CD4⁺ T_RM cells were isolated from healthy human kidney tissue. These cells were left unstimulated or stimulated with PMA + Iono for 1 h and mixed with 2 million NIH/3T3 cells to achieve sufficient cell numbers to identify the border between the translated and untranslated mRNA fractions, referred to as the monosome peak. The cells were lysed and subjected to sucrose gradient centrifugation to separate the translated (polysome) and untranslated mRNA fractions³⁰ for RT–qPCR analysis. We found that only around 20% of IL17A, IFNG and CSF2 mRNA was detected in the translated mRNA fraction in unstimulated CD69⁺CD4⁺ T_RM cells, whereas approximately 80% was found in the translated mRNA fraction in CD69⁺CD4⁺ T_RM cells stimulated with PMA + Iono (Fig. 1l), indicating that proinflammatory cytokine mRNAs were not translated in CD69⁺CD4⁺ T_RM cells at steady state. ATF4 mRNA, which is translated under compromised cap-dependent translation³¹, was efficiently translated in unstimulated CD69⁺CD4⁺ T_RM cells (Fig. 1l), indicating that the translation of cytokine mRNAs was actively suppressed in CD4⁺ T_RM cells under homeostatic conditions.

Mice maintained under specific-pathogen-free conditions harbor few T_RM cells in nonbarrier organs such as the kidney. To perform an in-depth functional analysis of CD4⁺ T_RM cells, we induced kidney T_RM cells¹³, particularly IL-17A-producing T_RM (T_RM17) cells, by systemically infecting Il17a^CreR26^eYFP mice³², in which IL-17A-producing cells are labeled with yellow fluorescent protein (YFP), with intravenously injected Staphylococcus aureus. Starting at day 7 post-infection, we treated the mice with antibiotics for 2 weeks to clear the infection. Mice were analyzed 2 months after the end of the antibiotics treatment^13,33. We refer to these mice as memory-induced IL-17A reporter mice (hereafter IL-17A^MI mice). Immunofluorescence staining combined with mRNA FISH detected numerous CD3⁺ T cells in the kidney of IL-17A^MI mice, with some cells expressing Il17a, Ifng or Csf2 (Fig. 2a). scRNA-seq analysis on YFP⁺CD4⁺ T cells sorted from the kidneys of IL-17A^MI identified Sell⁺S1pr1⁺Klf2⁺ T_CM cell (hereafter CD4⁺ TCM17 cell), Sell^–S1pr1⁺Klf2⁺ T_EM cell (CD4⁺ T_EM17 cell) and Sell⁻S1pr1⁻Klf2⁻ T_RM cell (CD4⁺ T_RM17 cell) subsets (Fig. 2b and Extended Data Fig. 4a). CD4⁺ T_RM17 cells highly expressed mRNA for Il2, Il17a, Il17f, Il22, Ifng, Tnf, Csf2 and Lta, while CD4⁺ T_CM17 cells expressed only Csf2 and CD4⁺ T_EM17 cell expressed only Il22 and Lta (Fig. 2b,c and Extended Data Fig. 4b). RT–qPCR confirmed the higher expression of Il17a mRNA in kidney YFP⁺CD4⁺CD44⁺CD69⁺CD62L⁻ T_RM17 cells compared to YFP⁺CD4⁺CD44⁺CD69⁻CD62L⁻ T_EM17 cells, YFP⁺CD4⁺CD44⁺CD69⁻CD62L⁺ T_CM17 cells, YFP⁻CD4⁺CD44⁺CD69⁺CD62L⁻ T_RM cells and CD4⁺CD44⁻ T_N cells (Extended Data Fig. 4c). Gene Ontology (GO) pathway analysis indicated that pathways associated with cytoplasmic translation and ribosome biogenesis were significantly downregulated in CD4⁺ T_RM17 cells compared to CD4⁺ T_EM17 cells (Fig. 2d). In the scRNA-seq data of YFP⁺CD4⁺ T cells, we observed the coordinated downregulation of ribosomal protein genes (RPGs)³⁴, such as Rpsa, Rps3 and Rpl3 in CD4⁺ T_RM17 cells compared to CD4⁺ T_EM17 cells and CD4⁺ T_CM17 cells (Fig. 2e,f and Extended Data Fig. 4d), indicating the suppression of global mRNA translation in CD4⁺ T_RM17 cells. Moreover, the protein–protein interaction enrichment analysis³⁵ indicated suppression of the ribosomal scanning and start codon recognition processes, which are crucial steps in translation initiation, in CD4⁺ T_RM17 cells compared to CD4⁺ T_EM17 cells and CD4⁺ T_CM17 cells (Extended Data Fig. 4e). These findings indicated that CD4⁺ T_RM17 cells from the kidney of IL-17A^MI mice express cytokine mRNA while suppressing mRNA translation under homeostatic conditions.

a, Representative images of mRNA FISH of Il17a, Ifng and Csf2 mRNA combined with immunofluorescence CD3 staining in the kidney of IL-17A^MI mice. Scale bars, 50 μm. The experiment was repeated three times. b, UMAP plots showing the distribution of Sell⁻S1pr1⁻Klf2⁻ T_RM17 cells, Sell⁻S1pr1⁺Klf2⁺ T_EM17 cells and Sell⁺S1pr1⁺Klf2⁺ T_CM17 cells in YFP⁺CD4⁺ T cells isolated from the kidneys of IL-17A^MI mice and analyzed by scRNA-seq (left) and violin plot showing cytokine scores calculated based on the expression of Il2, Il17a, Il17f, Il22, Ifng, Tnf, Csf2 and Lta in the YFP⁺CD4⁺ T_RM17, T_EM17 and T_CM17 cell clusters (middle and right). c, Heatmap showing the expression of Il2, Il17a, Il17f, Il22, Ifng, Tnf, Csf2 and Lta in YFP⁺CD4⁺ T_RM17, T_EM17 and T_CM17 cell clusters as in b. d, GO analysis displaying pathways downregulated in YFP⁺CD4⁺ T_RM17 cells compared to YFP⁺CD4⁺ T_EM17 cells as in b. e, Violin plot showing the ribosomal protein gene score calculated based on the expression of all RPGs in YFP⁺CD4⁺ T_RM17, T_EM17 and T_CM17 cell clusters as in b. f, Volcano plot showing differential gene expression between YFP⁺CD4⁺ T_RM17 and YFP⁺CD4⁺ T_EM17 cells as in b. RPGs downregulated in T_RM17 cells are shown in orange. g, Time course of IL-17A production in YFP⁺CD4⁺CD44⁺CD69⁺ T_RM17 cells, YFP⁺CD4⁺CD44⁺CD69^–CD62L⁻ T_EM17 cells, YFP⁺CD4⁺CD44⁺CD62L⁺ T_CM17 cells and YFP⁻CD4⁺CD44⁻ T_N cells isolated from the kidneys of IL-17A^MI mice and stimulated with PMA + Iono for 30 min, 1 h and 3 h (n = 6). h, Representative flow cytometry plots showing IL-17A production in YFP⁺CD4⁺CD44⁺CD69⁺ T_RM17 cells and YFP⁺CD4⁺CD44⁺CD69⁻CD62L⁻ T_EM17 cells isolated from the kidneys of IL-17A^MI mice stimulated with PMA + Iono in the presence or absence of 20 μg ml⁻¹ actinomycin D for 4 h (n = 3). i, Representative flow cytometry plots showing IL-17A production in CD4⁺CD44⁺CD69⁺ T_RM cells and CD4⁺CD44⁺CD69^–CD62L⁻ T_EM cells isolated from the intestine of wild-type C57BL/6 mice stimulated with PMA + Iono in the presence or absence of 20 μg ml⁻¹ actinomycin D for 4 h (n = 4). Data are mean + s.e.m. Statistical analysis was performed using one-way ANOVA with Tukey’s multiple comparison test; **P < 0.01.

Full size image

Rapid production of cytokines by T_M cells is crucial for effective host defense. To investigate the kinetics of cytokine production, we stimulated YFP⁺CD4⁺CD44⁺CD69⁺ T_RM17 cells, YFP⁺CD4⁺CD44⁺CD69⁻CD62L⁻ T_EM17 cells, YFP⁺CD4⁺CD44⁺CD62L⁺ T_CM17 cells and YFP⁻CD4⁺CD44⁻ T_N cells from the kidneys of IL-17A^MI mice with PMA + Iono. CD4⁺ T_RM17 cells produced higher levels of IL-17A, IFNγ and GM-CSF at 30 min compared to CD4⁺ T_CM17 cells and CD4⁺ T_EM17 cells (Fig. 2g and Extended Data Fig. 4f). To investigate whether the rapid production of IL-17A in CD4⁺ T_RM17 cells was due to the translation of pre-transcribed mRNA, we stimulated CD4⁺ T_RM17 cells and CD4⁺ T_EM17 cells from the kidneys of IL-17A^MI mice with PMA + Iono in the presence or absence of actinomycin D (ActD), which blocks mRNA transcription. CD4⁺ T_RM17 cells and, to a lesser extent, CD4⁺ T_EM17 cells produced IL-17A upon stimulation with PMA + Iono in the absence of ActD (Fig. 2h). When stimulated with PMA + Iono + ActD, CD4⁺ T_RM17 cells (but not CD4⁺ T_EM17 cells) still produced IL-17A protein, although at lower levels than those stimulated with PMA + Iono alone (Fig. 2h). Similar results were observed in CD4⁺CD44⁺CD69⁺ T_RM cells and CD4⁺CD44⁺CD69⁻CD62L⁻ T_EM cells from the small intestine of healthy wild-type mice (Fig. 2i). These findings indicated that transcribed Il17a mRNA pools stored in CD4⁺ T_RM cells were sufficient for immediate cytokine protein production upon stimulation.

To understand the molecular pathways that regulated mRNA translation in human CD4⁺ T_RM cells, we analyzed the scRNA-seq dataset of kidney T cells (cohort 1) and published datasets of colon²⁴ and lung²⁵ T cells. GO term pathway analysis and ribosomal protein gene expression analysis revealed the downregulation of translation and ribosome biogenesis pathways in healthy human CD4⁺ T_RM cells isolated from the kidney (Fig. 3a,b and Extended Data Fig. 5a), colon and lung (Extended Data Fig. 5b–e) compared to other healthy human CD4⁺ T_M cell subsets. In contrast, Reactome pathway analysis found upregulation of pathways linked to cellular responses to stress in human kidney CD4⁺ T_RM cells (Fig. 3c). Cellular stress responses are associated with the integrated stress response (ISR) pathway, a biological process activated by various stimuli, such as hypoxia and unfolded protein response^31,36. The ISR phosphorylates eIF2α to decrease cap-dependent translation initiation and downregulate global mRNA translation³¹. Similarly, analysis of the scRNA-seq dataset in the IL-17A^MI mice also revealed the downregulation of the start codon recognition and translation initiation process in CD4⁺ T_RM17 cells compared to other CD4⁺ T_M17 cells (Extended Data Fig. 4e). Consistent with these findings, scRNA-seq analysis of human CD4⁺ T cells from the healthy kidney, colon and lung showed that human CD4⁺ T_RM cells expressed high levels of ISR-associated genes, such as PPP1R15A, ATF3, ATF4 and ATF5 (refs. ^{31,37,38,39,40,41}) (Fig. 3d and Extended Data Fig. 5f–i). eIF2α was phosphorylated in unstimulated human kidney CD4⁺CD69⁺ T_RM cells, but was dephosphorylated after stimulation of these cells with PMA + Iono for 3 h or CD3⁺ CD28 Ab for 2 days (Fig. 3f,g and Extended Data Fig. 5j,k). In a puromycin-uptake assay, unstimulated human kidney CD4⁺CD69⁺ T_RM cells exhibited lower levels of global protein synthesis compared to CD4⁺CD69⁺ T_RM cells stimulated with PMA + Iono for 2 h (Extended Data Fig. 5l).

a, GO analysis showing pathways downregulated (left) and upregulated (right) in CD4⁺SELL⁻S1PR1⁻CD69⁺ T_RM cells compared to CD4⁺CD69⁻ non-T_RM cells isolated from human healthy kidneys (scRNA-seq data of cohort 1 as in Fig. 1d). b, Violin plot showing the ribosomal protein gene score calculated based on the expression of all RPGs in CD4⁺ T_RM, T_EM, T_CM and T_N cell clusters (as in a). c, Reactome analysis displaying pathways upregulated (left) and downregulated (right) in CD4⁺SELL⁻S1PR1⁻CD69⁺ T_RM cells compared to CD4⁺CD69⁻ non-T_RM cells in human healthy kidneys (as in a). d, Heatmap showing the expression of the ISR-associated genes PPP1R15A, PPP1R15B, ATF3, ATF4, ATF5, ATF6, DDIT3, DNAJB1, DUSP1, FOS, JUN, FOSB and JUNB in CD4⁺ T_RM, T_EM, T_CM and T_N cell clusters (as in a). e, Violin plot showing the ISR-associated gene scores in CD4⁺ T_RM, T_EM, T_CM and T_N cell clusters (as in a). f, Representative immunocytochemistry images of CD4⁺CD69⁺ T_RM cells sorted from healthy human kidney and stained with antibodies specific for phosphorylated eIF2α (p-eIF2α) and 4,6-diamidino-2-phenylindole (DAPI) and quantification of p-eIF2α expression in CD4⁺CD69⁺ T_RM cells sorted from healthy human kidney and stimulated or not with PMA + Iono for 3 h (n = 3). Scale bars, 2 μm. g, Representative histogram showing p-eIF2α levels in CD4⁺CD69⁺ T_RM cells with or without CD3 + CD28 antibody stimulation for 2 days (representative of two experiments). Data are mean + s.e.m. Statistical analysis was performed using one-way ANOVA with Tukey’s multiple comparison test (b,e) or unpaired two-tailed t-test with Welch’s correction (f); *P < 0.05, **P < 0.01.

Full size image

Similar to human CD4⁺ T_RM cells, unstimulated kidney CD4⁺CD69⁺ T_RM cells from memory-induced wild-type C57BL/6 mice (hereafter WT^MI mice) showed high p-eIF2α levels during homeostasis, while following stimulation with PMA + Iono for 3 h or CD3 + CD28 antibody for 2 days, these cells exhibited eIF2α dephosphorylation (Fig. 4a,b and Extended Data Fig. 6a). In CD4⁺CD69⁺ T_RM cells stimulated with PMA + Iono, global protein synthesis significantly increased compared to unstimulated controls (Extended Data Fig. 6b). In time kinetic experiments using kidney CD4⁺CD69⁺ T_RM cells from WT^MI mice, eIF2α dephosphorylation was observed 15 min after PMA + Iono stimulation (Fig. 4c). Cytokine production was detected 30 min after stimulation (Fig. 4c). p-eIF2α levels were significantly lower in IL-17A- or IFNγ-producing CD4⁺CD69⁺ T_RM cells stimulated with PMA + Iono compared to unstimulated controls (Extended Data Fig. 6c). Additionally, mouse CD4⁺CD69⁺ T_RM cells had higher p-eIF2α levels than CD4⁺CD44⁺CD69⁻CD62L⁺ T_CM cells, CD4⁺CD44⁺CD69⁻CD62L⁻ T_EM cells and CD4⁺CD44⁻ T_N cells (Extended Data Fig. 6d). Following ISR activation, untranslated mRNAs are sequestered into stress granules, which are membrane-less cytoplasmic organelles³⁰. IL17a mRNA FISH analysis combined with immunocytochemistry for the stress granule marker Tia1 revealed increased colocalization of cytoplasmic Il17a mRNA and Tia1 in unstimulated YFP⁺CD4⁺CD69⁺ T_RM cells from IL-17A^MI mice compared to YFP⁺CD4⁺CD69⁺ T_RM cells stimulated with PMA + Iono for 3 h (Fig. 4d), indicating the release of Il17a mRNA from the stress granules in stimulated T_RM cells.

a, Representative immunocytochemistry of CD4⁺CD69⁺ T_RM cells isolated from the kidneys of C57BL/6 WT^MI mice and stained with antibodies specific for phosphorylated eIF2α (p-eIF2α) and DAPI, and quantification of p-eIF2α expression in CD4⁺CD69⁺ T_RM cells isolated from the kidneys of C57BL/6 WT^MI mice stimulated or not with PMA + Iono for 3 h (n = 4). Scale bars, 2 μm. b, Representative histogram and quantification of p-eIF2α in CD4⁺CD69⁺ T_RM cells isolated from the kidneys of WT^MI mice and stimulated or not with CD3 + CD28 antibody for 2 days (n = 4). c, Time course plot showing the frequency of IFNγ⁺ and IL-17A⁺ in CD4⁺CD69⁺ T_RM cells and p-eIF2α expression in CD4⁺CD69⁺ T_RM cells isolated from the kidneys of WT^MI mice and stimulated with PMA + Iono for 0, 15, 30 and 60 min (n = 4). d, Representative mRNA FISH for Il17a mRNA combined with immunocytochemistry Tia1 staining in YFP⁺CD4⁺CD69⁺ T_RM17 cells sorted from the kidneys of IL-17A^MI mice and stimulated or not with PMA + Iono for 3 h (left), and quantification of Il17a mRNA and Tia1 colocalization in YFP⁺CD4⁺CD69⁺ T_RM17 cells sorted from the kidneys of IL-17A^MI mice and stimulated or not with PMA + Iono for 3 h (right). Scale bars, 2 μm. Quantification of four experiments is shown. e, Expression of p-eIF2α and frequency of IL-17A⁺ cells in CD4⁺CD69⁺ T_RM cells isolated from the kidneys of WT^MI mice and stimulated with PMA + Iono in the presence of vehicle, 500 μM NaAsO₂, 10 μM Sal003, 20 μM Sephin1 or 20 μM Raphin1 for 3 h (n = 4). f, Frequency of IL-17A⁺ cells in kidney CD4⁺CD69⁺ T_RM cells (left), glomerular crescent score (middle) and urinary albumin to creatinine ratio (right) in Rag1 KO mice adoptively transferred with eIF2α-S51A⁺ and eIF2α-S51D⁺ T cells (n = 6). g, Representative histochemistry images of periodic acid-Schiff (PAS)-stained kidney sections from cGN^MI mice treated with vehicle (n = 8) or raphin1 (n = 10) twice a day through oral gavage starting at day 0 of cGN for 10 days until analysis. Arrows indicate glomerular crescents. Scale bars, 50 μm. h, Frequency of crescentic glomeruli and the urinary albumin to creatinine ratio in cGN^MI mice treated with vehicle (n = 8) or raphin1 (n = 10) as in g. Data are mean + s.e.m. (*P < 0.05, **P < 0.01). Statistical analysis was performed using one-way ANOVA with Tukey’s multiple comparison test (c,e) or unpaired two-tailed t-test with Welch’s correction (a,b,f,h); *P < 0.05, **P < 0.01. NS, not significant.

Full size image

To investigate the role of the ISR pathway in proinflammatory cytokine production in CD4⁺CD69⁺ T_RM cells from WT^MI mice, we used sodium arsenite, which activates the ISR pathway by oxidative cues³⁰. Sodium arsenite treatment significantly increased eIF2α phosphorylation (Fig. 4e) and suppressed the production of IL-17A, IFNγ and GM-CSF in CD4⁺CD69⁺ T_RM cells from WT^MI mice stimulated with PMA + Iono (Fig. 4f and Extended Data Fig. 6e,f). Sodium arsenite treatment did not affect cell viability or intracellular Ca²⁺ concentration increase upon T cell stimulation (Extended Data Fig. 6g,h), indicating no signs of toxicity or inhibition of activation. These observations indicated that the ISR controlled the translation of cytokine mRNA.

We next tested whether blocking eIF2α dephosphorylation suppresses cytokine mRNA translation in activated T_RM cells. eIF2α is dephosphorylated by the phosphatase complexes PPP1R15A-PP1 and PPP1R15B-PP1 (ref. ³¹). The pan-eIF2α phosphatase inhibitor Sal003, which blocks both PPP1R15A and PPP1R15B, significantly suppressed eIF2α dephosphorylation and the production of IL-17A, IFNγ and GM-CSF in CD4⁺CD69⁺ T_RM cells from WT^MI mice stimulated with PMA + Iono (Fig. 4e and Extended Data Fig. 6e, f). Sephin1, a selective inhibitor of PPP1R15A, did not suppress eIF2α dephosphorylation or the production of IL-17A, IFNγ and GM-CSF in CD4⁺CD69⁺ T_RM cells (Fig. 4e and Extended Data Fig. 6e,f). In contrast, raphin1, a selective inhibitor of PPP1R15B, significantly reduced eIF2α dephosphorylation and suppressed the production of those cytokines (Fig. 4e and Extended Data Fig. 6e,f), indicating eIF2α dephosphorylation was predominantly driven by PPP1R15B. None of these compounds suppressed intracellular Ca²⁺ concentration increase or shedding of surface CD62L upon T cell activation (Extended Data Fig. 6h,i), indicating that T cell activation was not affected. In addition, CD4⁺CD69⁺ T_RM cells from WT^MI mice treated with ISRIB, which binds to and activates eIF2B to facilitate mRNA translation during the ISR, showed a moderate increase in IL-17A and IFNγ production compared to untreated WT^MI CD4⁺CD69⁺ T_RM cells in the absence of TCR stimulation (Extended Data Fig. 6j). We also analyzed CD8⁺CD69⁺ T_RM cells isolated from the intestine of wild-type C57BL/6 mice. eIF2α was phosphorylated in unstimulated CD8⁺CD69⁺ T_RM cells, while after stimulation with PMA + Iono for 3 h, CD8⁺CD69⁺ T_RM cells dephosphorylated eIF2α and produced of IFNγ (Extended Data Fig. 6k). Sodium arsenite treatment increased p-eIF2α levels and inhibited IFNγ production in these cells (Extended Data Fig. 6k). These findings indicated that the ISR pathway might regulate cytokine mRNA translation in CD8⁺ T_RM cells.

To further test whether eIF2α phosphorylation was required for the suppression of cytokine production, we transduced CD4⁺CD69⁺ T_RM cells from WT^MI mice with a constitutively active (eIF2α-S51A) or a constitutively inactive (eIF2α-S51D, a phosphomimetic variant) form of eIF2α⁴². Production of IL-17A was significantly suppressed in eIF2α-S51D-transduced T cells compared to eIF2α-S51A-trandsuced CD4⁺CD69⁺ T_RM cells (Fig. 4f), indicating that eIF2α phosphorylation was sufficient to suppress IL-17A production in mouse CD4⁺CD69⁺ T_RM cells. Next, we transferred either eIF2α-S51D-transduced or eIF2α-S51A-transduced CD4⁺CD69⁺ T_RM cells into T cell-deficient Rag1 KO mice. One day after transfer, the mice were intraperitoneally injected with nephrotoxic serum to induce autoimmune glomerulonephritis (crescentic GN; cGN)¹³. At day 10 post-cGN induction, mice that received eIF2α-S51D-transduced CD4⁺ T cells showed a milder renal disease phenotype, as assessed by glomerular crescent scoring and albuminuria measurement, compared to mice that received eIF2α S51A-transduced CD4⁺ T cells (Fig. 4f). These findings indicated that targeting the ISR pathway in CD4⁺ T cells ameliorated tissue inflammation and cGN progression.

To further investigate the therapeutic effects of targeting eIF2α dephosphorylation in autoimmune disease, we induced cGN in WT^MI mice (hereafter cGN^MI mice) and treated them with raphin1 (ref. ⁴³) via twice-daily oral gavage. At day 10 of cGN, the levels of p-eIF2α in kidney CD4⁺CD69⁺ T_RM cells were higher in raphin1-treated compared to vehicle-treated cGN^MI mice (Extended Data Fig. 7a). Raphin1-treated cGN^MI mice exhibited reduced glomerular crescent formation and albuminuria, indicating improved kidney function compared to vehicle-treated cGN^MI mice (Fig. 4g,h). Raphin1 treatment did not improve kidney function in nephritic wild-type mice in which CD4⁺CD69⁺ T_RM cells were not induced by infection (Extended Data Fig. 7b), indicating that CD4⁺CD69⁺ T_RM cells were required for the therapeutic effect. These findings suggested that targeting the ISR–eIF2α pathway may represent a therapeutic strategy in immune-mediated diseases.

Finally, to explore the relevance of our findings in human autoimmune diseases, we performed scRNA-seq analysis on CD4⁺ T cells isolated from kidney biopsies of five patients with ANCA-GN (cohort 3, median age 67 years (IQR 61–72), 60% male). We compared these cells with CD4⁺ T cells from healthy donors (three kidney biopsies from cohort 1, which were processed using the same protocol as the samples from patients with ANCA-GN) (Fig. 5a). The expression levels of mRNA encoding type 1 (for example, IFNG and TNF) and type 3 (for example, IL17A and IL17F) cytokines in S1PR1⁻KLF2⁻CD69⁺ T_RM cells were similar between patients with ANCA-GN and healthy controls (Fig. 5b). Reactome pathway analysis revealed downregulation of the cellular response to stress pathway in kidney CD4⁺ T_RM cells from patients with ANCA-GN compared to kidney CD4⁺ T_RM cells from healthy controls (Fig. 5c). Moreover, kidney CD4⁺ T_RM cells from patients with ANCA-GN exhibited significantly lower expression of ISR-associated genes (for example, PPP1R15A, PPP1R15B, ATF3 and ATF4) compared to kidney CD4⁺ T_RM cells from healthy controls (Fig. 5d and Extended Data Fig. 8a), suggesting CD4⁺ T_RM cells in patients with ANCA-GN could have higher mRNA translation efficiency. Of note, cytokine-responsive genes, such as CXCL1, CXCL5, CXCL9 and CXCL10 (Extended Data Fig. 3c), were highly expressed in the kidney tissue from patients with ANCA-GN compared to healthy controls (Fig. 5e). These findings suggested that enhanced cytokine mRNA translation contributed to T cell-mediated tissue damage in glomerulonephritis. Similarly, scRNA-seq analysis of CD8⁺ T cells from these patients found a lower ISR score in kidney CD8⁺ T_RM cells from patients with ANCA-GN compared to kidney CD8⁺ T_RM cells from healthy controls (Extended Data Fig. 8b,c), indicating increased CD8⁺ T_RM cell activation in kidneys from patients with ANCA-GN. Additionally, we analyzed a public scRNA-seq dataset of colon CD4⁺ T cells from healthy donors and patients with Crohn’s disease, an inflammatory bowel disease in which type 1 (for example, TNF) and type 3 (for example, IL-17) cytokines play key roles⁴⁴. We identified two CD4⁺SELL⁻CCR7⁻S1PR1⁻KLF2⁻ T_RM cell clusters, a TNF^hi CD4⁺ T_RM cell subset and an IL17A^hi CD4⁺ T_RM cell subset (Fig. 5f). The expression of mRNA encoding inflammatory cytokines (IL2, IL17A, IL17F, IL22, IFNG, TNF, CSF2 and LTA) in CD4⁺ T_RM cells was comparable between healthy controls and patients with Crohn’s disease (Fig. 5g). Both TNF^hi CD4⁺ T_RM cells and IL17A^hi CD4⁺ T_RM cells from patients with Crohn’s disease exhibited downregulation of the cellular response to stress pathway and lower ISR scores compared to healthy controls (Fig. 5h,i and Extended Data Fig. 8d). In the colon tissue from patients with Crohn’s disease, various cytokine-responsive genes were upregulated compared to healthy controls (Fig. 5j), indicating active cytokine mRNA translation in inflamed colon tissue. These findings suggested the potential downregulation of the ISR pathway as a mechanism to enhance the production of effector molecule proteins, resulting in increased tissue inflammation.

a, UMAP plot showing the distribution of CD4⁺KLF2⁻CD69⁺ T_RM cells, CD4⁺SELL⁻ KLF2⁺CD69⁻ T_EM cells and CD4⁺SELL⁺KLF2⁺CD69⁻ T_CM/T_N cells isolated from the kidney of healthy controls (n = 3) and patients with ANCA-GN (n = 5) (left) and heatmap showing the expression of CD69, CCR7, CD62L and CD45RA protein and KLF2, S1PR1, CCR7, SELL, IFNG, TNF, IL17A, IL17F and CSF2 mRNA in the CD4⁺ T_RM, T_EM and T_CM/T_N cell clusters defined in the UMAP (right). b, Violin plot showing cytokine scores based on the expression of IL2, IL17A, IL17F, IL22, IFNG, TNF, CSF2 and LTA in kidney CD4⁺ T_RM cells from healthy controls and patients with ANCA-GN as in a. c, Reactome analysis showing downregulated pathways in kidney CD4⁺ T_RM cells from patients with ANCA-GN compared to kidney CD4⁺ T_RM cells from healthy controls. d, Violin plot showing ISR-associated gene scores based on the expression of PPP1R15A, PPP1R15B, ATF3, ATF4, ATF5, ATF6, DDIT3, DNAJB1, DUSP1, FOS, JUN, FOSB and JUNB in kidney CD4⁺ T_RM cells from patients with ANCA-GN and healthy controls. e, Violin plot showing cytokine-responsive gene scores based on transcriptome data of kidney biopsies from patients with ANCA-GN (n = 22) and healthy controls (n = 21)²⁶. f, UMAP plot showing the distribution of CD4⁺S1PR1⁻IL17A^hi T_RM cells, CD4⁺SELL⁻S1PR1⁻TNF^hi T_RM cells, CD4⁺SELL⁻S1PR1⁺ T_EM cells and CD4⁺SELL⁺S1PR1⁺ T_CM/T_N cells isolated from the colon of healthy controls (n = 2) and patients with Crohn’s disease (n = 2) (left) and heatmap showing the expression of KLF2, S1PR1, CCR7, SELL, RORC, IL17A, IL22 and TNF mRNA in CD4⁺ T_CM/T_N, T_EM, TNF^hi T_RM and IL17A^hi T_RM cells (right). g, Violin plot showing cytokine scores in colon CD4⁺TNF^hi and CD4⁺IL17A^hi T_RM cells from healthy controls and patients with Crohn’s disease as in f. h, Reactome analysis showing pathways downregulated in colon CD4⁺TNF^hi and CD4⁺IL17A^hi T_RM cells from patients with Crohn’s disease compared to colon CD4⁺TNF^hi and CD4⁺IL17A^hi T_RM cells from healthy controls as in f. i, Violin plots showing ISR-associated gene scores in colon CD4⁺TNF^hi and CD4⁺IL17A^hi T_RM cells from healthy controls and patients with Crohn’s disease as in f. j, Violin plot showing cytokine-responsive gene scores based on transcriptome data of colon biopsies from patients with Crohn’s disease (n = 92) and healthy controls (n = 55). Data are mean + s.e.m. Statistical analysis was performed using unpaired two-tailed t-test with Welch’s correction (d,e,i,j); **P < 0.01.

Full size image

In this study, we identified CD4⁺ T_RM cells as the primary T_M cell subset that stored untranslated cytokine mRNA. We showed that the ISR pathway and its mediator p-eIF2α facilitated the mRNA storage thereby suppressing its translation, and that the stored mRNA was sufficient for the rapid cytokine production upon activation. Different mechanisms of translation regulation have been reported. How the ISR pathway synergizes with other mechanisms, such as RNA-binding proteins^21,45, the mTOR pathway⁴⁶ or regulation of mRNA half-life^21,47, remains unclear. While enhancing ISR with chemical compounds or gene overexpression significantly reduced cytokine production in mouse kidney CD4⁺ T_RM cells, the marginal effects of ISRIB on cytokine production in these cells suggested that other pathways also regulate cytokine mRNA translation in CD4⁺ T_RM cells. Overall, our findings suggested that the ISR–eIF2α pathway regulates cytokine mRNA translation in CD4⁺ T_RM cells during homeostasis, and identified this pathway as a potential therapeutic target in T cell-mediated inflammatory diseases where ISR-mediated regulation is impaired.

Human studies

Human kidney tissue was obtained from n = 8 donors (included in the cohort 1) and n = 5 donors (included in the cohort 2) registered to the Hamburg GN Registry; n = 21 healthy donors, n = 22 patients with ANCA-GN, n = 32 patients with lupus nephritis and n = 27 patients with IgA nephropathy registered to the European Renal cDNA Bank²⁶; and from n = 5 patients (included in cohort 3) registered to the Clinical Research Unit 228 (CRU 228) ANCA-GN cohort. For the eight donors in cohort 1 and the five donors in cohort 2, matched blood samples from the same donors were also analyzed for this study. Kidney T cells isolated from the healthy part of kidneys removed as part of tumor nephrectomy were considered as T cells at steady state. The histological examination confirmed that the healthy part of the kidney did not exhibit any signs of tumor invasion or active immune cell invasion. The research studies were conducted with the approval of the Ethik-Kommission der Ärztekammer Hamburg (the local ethics committee of the Chamber of Physicians in Hamburg) and in compliance with the ethical principles outlined in the Declaration of Helsinki. All participating patients gave informed consent for the research.

Mouse experiments

Experiments with mice followed the national guidelines, and local ethics committees (Behörde für Justiz und Verbraucherschutz Hamburg) approved the research protocols. Mice were housed under specific-pathogen-free conditions to ensure the validity and reliability of the experiments. Age-matched C57BL/6 male mice (between 8 and 16 weeks of age) were used in all experiments to minimize the variability between subjects. Il17a^CreR26^eYFP male mice³² between 8 and 16 weeks of age were used to identify T_RM17 cells. For the induction of IL-17A^MI and WT^MI mice models, Il17a^CreR26^eYFP or wild-type C57BL/6 mice were intravenously injected with 1 × 10⁷ S. aureus (strain SH1000¹³) in 100 μl sterile PBS. For pathogen and inflammation clearance, ampicillin was given in drinking water for 2 weeks¹³ and mice were analyzed 2 months after the completion of the antibiotic treatment. Experimental crescentic glomerulonephritis was induced by intraperitoneal administration of sheep nephrotoxic serum targeting the glomerular basement membrane¹³. The cGN^MI mice model was induced by administration of the sheep nephrotoxic serum into WT^MI mice models. Raphin1 was reconstituted in distilled water with 0.5% methylcellulose (Sigma, M0512) and administered twice a day by oral gavage.

Histopathology, immunohistochemistry and immunofluorescence

To evaluate glomerular crescent formation, 30 glomeruli per mouse were assessed in a blinded manner in PAS-stained paraffin sections of kidneys¹³. For immunofluorescence staining, primary antibodies, including CD3 (A0452, Dako), Phospho-eIF2α (MA5-32021, Invitrogen), total eIF2α (D7D3, Cell Signaling) and Tia1 (ab140595, Abcam) were incubated in blocking buffer at 4 °C for 1 h. Following PBS washing, fluorochrome-labeled secondary antibodies were applied, and the staining was visualized using an LSM800 with Airyscan and the ZenBlue software (all Carl Zeiss).

To detect mRNA (FISH) in mice and human kidney sections, RNAscope Multiplex Fluorescent Assay (Advanced Cell Diagnostics) was employed. mRNA detection (FISH) in sorted T cells was carried out using the ViewRNA Cell Plus Assay kit (cat. no. 88-19000, Invitrogen). The slides were imaged using a Zeiss LSM800 confocal microscope and analyzed with ZEN software (Carl Zeiss). To evaluate Il17a mRNA in stress granules, the percentage of Il17a signals overlapping with Tia1-positive granules was calculated.

Isolation and flow cytometric analysis of human and murine leukocytes

Single-cell suspensions were obtained from kidney and blood samples to isolate and analyze human leukocytes. Kidney tissue was enzymatically digested with collagenase D at 0.4 mg ml⁻¹ (Roche) and DNase I (10 μg ml⁻¹, Sigma-Aldrich) in RPMI 1640 medium at 37 °C for 30 min, followed by dissociation with gentleMACS (Miltenyi Biotec). Blood samples were separated using Leucosep tubes (Greiner Bio-One). Samples were filtered through a 30-μm filter (Partec) before antibody staining and flow cytometry.

Cells from murine spleens were isolated by squashing the organ through a 70-µm cell strainer. Erythrocytes were lysed using a lysis buffer (155 mM NH₄Cl, 10 mM KHCO₃ and 10 µM EDTA, pH 7.2). To isolate renal lymphocytes from mice, kidneys were enzymatically digested with 400 µg ml⁻¹ collagenase D (Roche) and 10 U ml⁻¹ DNase I (Sigma-Aldrich) at 37 °C for 30 min. Subsequently, leukocytes were isolated by density gradient centrifugation using 37% Easycoll (Merck Millipore) and a filtration step using a 30-µm cell strainer (Partec). T cell isolation from the intestine is described previously. In brief, murine intestine was cut longitudinally after removing Peyer’s patches and adventitial fat. To collect intraepithelial lymphocytes, the intestine tissue was incubated in HBSS containing 1 mM dithiothreitol (DTT) followed by a dissociation step using 1 mM EDTA for 20 min at 37 °C. To collect lamina propria lymphocytes, the tissue was cut into small pieces and incubated for 45 min at 37 °C in HBSS supplied with 1 mg ml⁻¹ collagenase and 10 U ml⁻¹ DNase I. Leukocytes were further enriched by Percoll gradient. Intraepithelial cells and lamina propria lymphocytes were pooled for analysis.

Cells were surface stained with fluorochrome-conjugated antibodies (human, CD45 (HI30, BD Biosciences), CD3 (OKT3, eBioscience), CD4 (RPA-T4, BD Biosciences), CD8 (RPA-T8, BD Biosciences), CD69 (FN50, BD Biosciences), CD45RA (HI100, BD Biosciences), GM-CSF (VD2-21C11, BD Biosciences), IL-17A (BL168, BioLegend), IL-17F (SHLR17, BioLegend), IFNγ (4S.B3, eBioscience), TNF (MAB11, BD Biosciences), CCR7/CD197 (Go43H7, BD Biosciences); mouse, CD45 (30-F11, BD Biosciences), CD3 (145-2C11, BD Biosciences), CD4 (RM4-5, BD Biosciences), CD8 (53-6.7, BD Biosciences), GM-CSF (MP1-22E9, BD Biosciences), IL-17A (TC11-18H10, BD Biosciences), IL-17F (9D3.1C8, BioLegend), IL-22 (poly5164, BioLegend), IFNγ (XMG1.2, BD Biosciences), TNF (MP6-XT22, BD Biosciences)) and a fixable dead-cell stain (Molecular Probes) to exclude dead cells from the analysis. For intracellular staining, samples were processed using Cytofix/Cytoperm (BD Biosciences) according to the manufacturer’s instructions. All antibodies were diluted at a ratio of 1:100 to 1:200.

To assess the phospho-eIF2α and total eIF2α levels, isolated T cells were stimulated with CD3/28 antibodies for 2–3 days, and transferred to unstimulated conditions. For re-stimulation, T cells were further stimulated with PMA + Iono or CD3 + CD28 antibodies. Stimulated and unstimulated T cells underwent surface staining and were subsequently fixed and permeabilized using Foxp3/Transcription Factor Staining Buffer (eBioscience). The cells were then incubated with primary antibodies directed against phospho-eIF2α or total eIF2α (1:1,000 dilution) for 40 min. After two rounds of washing, the cells were exposed to a secondary antibody conjugated with AF647 against Rabbit IgG (1:1,000 dilution) for 40 min.

Ca²⁺ live-cell imaging in murine T cells

Freshly isolated primary murine CD4⁺ T cells were loaded with Fura2-AM (4 µM) and incubated for 40 min at 37 °C. Following the loading process, the cells were washed twice and resuspended in Ca²⁺ buffer comprising 140 mM sodium chloride (NaCl), 5 mM potassium chloride (KCl), 1 mM magnesium sulfate (MgSO₄), 1 mM calcium chloride (CaCl₂), 20 mM 4-(2-hydroxyethyl)-1 piperazine ethanesulfonic acid (HEPES), 1 mM sodium dihydrogen phosphate (NaH₂PO₄) and 5.5 mM glucose at pH 7.4. Imaging was conducted using a Leica IRBE microscope with a ×40 magnification and an exposure of 25 ms. The microscope was equipped with a Sutter DG-4 as a light source and an electron-multiplying charge-coupled device camera (EMCCD; 13, Hamamatsu). The images were acquired in 16-bit mode with an acquisition rate of one frame every 2 s, using Volocity software (v.6.6.2; PerkinElmer). The following filter set for Fura2 was used (excitation filters, HC 340/26, HC 387/11; beamsplitter, 400 DCLP; emission filter, 510/84, all in nanometers). At 5 min before acquisition, the T cells were incubated with either 10 μM Sal003 (Sigma-Aldrich, S4451), 20 μM Sephin1 (Sigma-Aldrich, SML1356), 20 μM Raphin1 (Sigma-Aldrich, SML2562), 500 μM NaAsO₂ (Sigma-Aldrich, S7400) or Ca²⁺ buffer and dimethylsulfoxide (DMSO) (as control). T cell stimulation was initiated through the addition of soluble anti-CD3 antibody (10 µg ml⁻¹) after 1 min. Subsequently, the background correction, splitting of fluorescence channels and selection of cells into regions of interest were performed with Fiji v.2. Furthermore, to obtain the Ca²⁺ concentration from ratio values, a calibration was calculated by measuring the maximal ratio value (Rmax) and the minimal ratio value (Rmin). The mean Ca²⁺ concentration over time as well as the mean basal (after 30 s), peak (at 260 s) and plateau (at 520 s) Ca²⁺ concentration were calculated using Prism10.

Flow cytometry and cell sorting

Samples were measured with LSR II or Symphony A3 (both BD Biosciences). Data analysis was performed using FlowJo software (TreeStar). FACS sorting was performed with an AriaFusion or AriaIIIu (BD Biosciences).

Quantitative RT–PCR

T cells were subjected to RNA extraction utilizing the RNeasy Micro kit (QIAGEN). RNA from the renal cortex was isolated with the NucleoSpin kit (Macharey-Nagel) in compliance with the manufacturer’s recommended protocol. Subsequently, the RNA underwent reverse transcription using the High-Capacity cDNA Reverse Transcription kit (Thermo Fisher), and the StepOnePlus Real-Time PCR system (Thermo Fisher) was utilized for measurement. The Taqman primers used for human IL17A, IFNG, CSF2, ZC3H12D, ATF4, 18S rRNA and murine Il17a were procured from Life Technologies.

In vitro stimulation of cells

Human and murine T cells were classified based on the surface markers and FACS sorted before stimulation. T_RM cells were characterized as CD45⁺CD4⁺CD44⁺CD69⁺CD62L⁻; T_EM cells as CD45⁺CD4⁺CD44⁺CD69⁻CD62L⁻; T_CM cells as CD45⁺CD4⁺CD44⁺CD69⁻CD62L⁺; and naive T cells as CD45⁺CD4⁺CD44⁻CD69⁻CD62L⁺. The cells were exposed to PMA (50 ng ml⁻¹, Sigma-Aldrich) and Iono (1 μM, Sigma-Aldrich) for T cell activation. To detect cytokine production, brefeldin A (10 μg ml⁻¹, Sigma-Aldrich) was added to inhibit cytokine secretion from cells. After 3–6 h, cytokine production was measured using intracellular staining and flow cytometry techniques. To inhibit de novo mRNA transcription, 20 µg ml⁻¹ actinomycin D (A1410, Sigma-Aldrich) was added to the medium 10 min before T cell stimulation. For ex vivo stimulation to measure eIF2α levels, 1–2 × 10⁶ T_RM cells were cultured in a volume of 300 μl IMDM containing 10% FCS, streptomycin and penicillin in a 96-well plate pre-coated with anti-CD3 antibody (2 μg ml⁻¹, BioLegend 100360), along with the addition of 5 μg ml⁻¹ anti-CD28 antibody (BioLegend 102122) for a duration of 2 days. To induce the ISR, T cells were incubated with 10 μM Sal003 (Sigma-Aldrich, S4451), 20 μM Sephin1 (Sigma-Aldrich, SML1356), 20 μM Raphin1 (Sigma-Aldrich, SML2562) or 500 μM NaAsO₂ (Sigma-Aldrich, S7400). For the experiment using ISRIB, ex vivo-cultured T_RM cells isolated from the kidney were treated with 500 nM ISRIB (Sigma, SML0843) overnight and analyzed for the production of cytokines.

Puromycin incorporation assay

T cells were cultured in the presence or absence of PMA (50 ng ml⁻¹, Sigma-Aldrich) and Iono (1 μM, Sigma-Aldrich) for 1 h. Puromycin (5 μg ml⁻¹, Sigma-Aldrich) was added in the last 10 min. Cells were washed and stained for surface markers and fixable dead-cell stain and then fixed and permeabilized using Foxp3/Transcription Factor Staining Buffer (eBioscience). Puromycin was detected using anti-puromycin-AF647 (MABE343 clone 12D10, Sigma-Aldrich).

Polysome profiling for mRNA translation efficiency analysis

Polysome profiling and mRNA extraction were conducted as described previously³⁰. Human T_RM cells sorted from healthy human kidney tissue were mixed with two million NIH/3T3 cells because the number of human T_RM cells was insufficient to show monosome peaks. We relied on monosome peaks from a murine cell line because monosome peaks were observed with the same kinetics in humans and mice. Cells were lysed in 250 μl cell lysis buffer (10 mM Tris-HCl (pH 7.4), 5 mM MgCl₂, 100 mM KCl, 1% Triton X-100, 2 mM DTT and 100 μg ml⁻¹ cycloheximide). Cells were shear-opened with a 26-gauge needle by passing the lysate eight times through it. A quantity of 300 μl of the cell lysate was loaded onto a 5-ml sucrose gradient (60% to 15% sucrose) dissolved in polysome buffer (50 mM HEPES-KOH, pH 7.4, 5 mM MgCl₂, 100 mM KCl, 2 mM cycloheximide and 2 mM DTT) and separated by ultracentrifugation at 148,900g (Ti55 rotor, Beckman) for 1.5 h at 4 °C. Polysome fractions and nonpolysome fractions were collected separately. RNA was extracted from each fraction by adding 0.1 volume of 10% SDS and one volume of acidic phenol–chloroform (5:1, pH 4.5), incubated at 65 °C for 5 min, and centrifuged at 21,000g for 5 min at 4 °C to separate different phases. Equal volumes of acid phenol–chloroform were added to the aqueous phase, separated by centrifugation and supplemented with an equal volume of chloroform:isoamyl alcohol (24:1). Upon separation, the aqueous phase was supplemented with 0.1 volume 3 M NaOAc (pH 5.5) and an equal volume of isopropanol, precipitated at −20 °C for 3 h. RNA was pelleted at 21,000g at 4 °C, and the dried pellets were resuspended in water. RNA was subjected to RT–qPCR analysis and the original mRNA amount in polysome and nonpolysome fractions was calculated by RT–qPCR using Taqman primers specific to human mRNA. mRNA translation efficiency was defined as the amount of mRNA in polysome fraction divided by the total mRNA amount of nonpolysome and polysome fractions.

Retrovirus transduction and cell transfer into Rag1 KO mice

MIGR1 plasmid⁴⁸ (produced by W. Pear, Addgene plasmid #27490) was used to overexpress eIF2α variants. Both eIF2α variants, S51A and S51D, were inserted into EcoRI-digested MIGR1 backbone. To produce retrovirus, HEK293T cells were seeded into a six-well plate and, the next day, transfected with MIGR1-eIF2α variant plasmid and pCL-Eco plasmid⁴⁹ (Produced by Inder Verma, Addgene plasmid #12371) using Lipofectamine 3000 (Invitrogen L3000001). Murine T cells were stimulated in a 24-well plate (2 × 10⁶ cells per well) for 2 days using anti-CD3 antibody (2 μg ml⁻¹, BioLegend, 100360) and anti-CD28 antibody (5 μg ml⁻¹, BioLegend, 102122). After the 2-day stimulation, the medium was replaced with virus-containing medium with supplementation of Polybrene. Cells were centrifuged at 1,000g, room temperature for 1 h. The next day, green fluorescent protein-positive T cells were sorted using FACS and transferred into Rag1 KO mice (10⁵ cells per mouse). Following T cell reconstitution, mice were challenged by the cGN model.

Single-cell RNA sequencing

To carry out scRNA-seq, single-cell suspensions were derived from human and mouse kidney samples. Cell hashing was implemented according to the manufacturer´s instructions (BioLegend). For CITE-seq, monoclonal antibodies with corresponding barcodes were applied to samples (BioLegend). FACS-sorted CD45⁺ cells or CD3⁺ T cells were subjected to droplet-based single-cell analysis and transcriptome library preparation using Chromium Single-Cell 5′ Reagent kits v2 according to the manufacturer’s protocols (10x Genomics). The generated scRNA-seq libraries were subjected to sequencing on a NovaSeq6000 system (100 cycles) (Illumina).

Alignment, quality control and pre-processing of scRNA-seq data

Quality control and scRNA-seq pre-processing were carried out as previously described¹³. In brief, the Cell Ranger software pipeline (v.5.0.1, 10x Genomics) was utilized to perform the demultiplexing of cellular barcodes and mapping of reads to the reference genome (refdata-cellranger-hg19-1.2.0 (human) or refdata-gex-mm10-2020-A (mouse)). Seurat (v.4.0.2) demultiplexing function HTODemux was used to demultiplex the hash-tag samples. We filtered out the cells with <500 genes, >6,000 genes or >5% mitochondrial genes. For CITE-seq, a pseudo-reference genome was built with cellranger mkref function in Cell Ranger. CITE-seq raw data were aligned to this pseudo-reference genome using Cell Ranger function cellranger count. The antibody-derived tags data were integrated to scRNA-seq data using Seurat. Further information is described previously¹³.

Dimensionality reduction, clustering, enrichment analysis and scores

The Seurat package (v.4.0.2) was used to conduct unsupervised clustering analysis on scRNA-seq data. In brief, gene counts for cells were normalized by library size and log-transformed. To reduce batch effects, we employed the integration method implemented in the latest Seurat v.4 (function FindIntegrationAnchors and IntegrateData, dims of 1:30). The integrated matrix was then scaled by the ScaleData function (default parameters). To reduce dimensionality, principal-component analysis was performed on the scaled data (function RunPCA, npcs of 30). Thirty principal components were determined using the ElbowPlot function to compute the KNN graph based on the Euclidean distance (function FindNeighbors), which then generated cell clusters using the function FindClusters. UMAP was used to visualize clustering results. The top differential expressed genes in each cluster were found using the FindAllMarkers function (min.pct of 0.1) and running Wilcoxon rank-sum tests. The differential expression between clusters or groups was calculated by the FindMarkers function (min.pct of 0.1), which also included Wilcoxon rank-sum tests. For enrichment analysis including GO, Reactome, Canonical pathway and protein–protein interaction analyses, metascape⁵⁰ was used. To calculate scores (RPG score and ISR score) in scRNA-seq data, Seurat function AddModuleScore was used. For both scores, all RPGs³⁴ and ISR-associated genes^{31,38,39,40,41} listed in heatmaps (Extended Data Fig. 5) were included.

Data analysis, statistics and reproducibility

Statistical analysis was performed using GraphPad Prism. The results are shown as mean ± s.e.m. when presented as a bar graph or as single data points with the mean in a scatter-plot. Differences between the two individual groups were compared using a two-tailed t-test. In the case of three or more groups, a one-way ANOVA with Tukey’s multiple comparisons test was used. The correlation coefficient r was calculated using a Pearson correlation, and the corresponding P value was based on a t-distribution test. All experiments were repeated at least three times, except for some experiments with human cells, which were repeated twice as stated in the figure legends. No statistical methods were used to predetermine sample sizes but our sample sizes are similar to those reported in previous publications¹³. Data distribution was assumed to be normal but this was not formally tested. Data collection was not performed but data analysis was performed blind to the conditions of the experiments. We did not exclude any data points. For bioinformatics analysis, RStudio and Jupyter Notebook were used to execute code.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.