The ubiquitin ligase KLHL6 drives resistance to CD8+ T cell dysfunction

Abstract

The multifaceted dysfunction of tumour-infiltrating T cells, including exhaustion and mitochondrial dysfunction, remains a major obstacle in cancer immunotherapy^1-6. Transcriptomic and epigenomic regulation of T cell dysfunction have been extensively studied^7-9, but the role of proteostasis in regulating these obstacles remains less defined. Here we combined computational analyses of atlases of T cell exhaustion and mitochondrial fitness with performed targeted in vivo CRISPR screens, which identified the E3 ubiquitin ligase KLHL6 as a dual-negative regulator of both T cell exhaustion and mitochondrial dysfunction. Mechanistically, KLHL6 expression promoted TOX poly-ubiquitination and subsequent proteasomal degradation, thereby attenuating the transition of progenitor exhausted T cells towards terminal exhaustion. Simultaneously, KLHL6 maintained mitochondrial fitness by constraining the excessive mitochondrial fission that occurs during chronic T cell receptor stimulation by means of post-translational regulation of the PGAM5-Drp1 axis. However, KLHL6 is naturally downregulated by T cell receptor ligation, mitigating its potentially beneficial ubiquitin ligase activities during exposure to chronic stimulation. Enforcing KLHL6 expression in T cells markedly improved efficacy and long-term persistence against tumours and during viral infections in vivo. These findings uncover KLHL6 as a multifunctional, clinically actionable target for cancer immunotherapy, and highlight the potential of modulating proteostasis and ubiquitin modification to improve immunotherapy.

Fig. 1. An integrative computational analysis-guided CRISPR screen identifies post-translational regulators of both exhaustion and mitochondrial fitness in T cells.

a, Schematic illustration of workflow for computational analysis of RNA-seq atlases of T cell exhaustion. b,c, Samples from two published datasets (GSE89307, GSE86881) were projected onto a two-dimensional map defined by computationally derived gene modules. b, Antigen-specific TCR-transgenic T cells collected across acute Listeria infection and tumour progression at matched time points following adoptive transfer. c, Antigen-specific naive and chronically exhausted T cells isolated during late-stage LCMV infection. d, GSEA enrichment of proteostasis-associated pathways in modules 1 and 2. e, Experimental schematic of CRISPR screen for E3 ligases that regulate T cell exhaustion and mitochondrial function. f,g, Rank plots (left) of gene-level enrichment scores in exhausted versus non-exhausted T cells (f) and in T cell populations with dysfunctional versus functional mitochondria (g). The CRISPR enrichment scores (log₂ fold change of PD-1⁺TIM-3⁺ versus PD-1⁻TIM-3⁻ or (MTDR/MTG)^lo versus (MTDR/MTG)^hi) were determined by comparing the indicated subsets for target genes from Cas9⁺sgRNA (mCherry)⁺ cells isolated from tumours on day 7 after ACT. The x axis shows targeted genes; the y axis shows the CRISPR enrichment score of each targeted gene; the dot colour represents false discovery rate (FDR). Distribution of several top-hit sgRNAs (right). Axis represents log₂ fold change (FC). The histogram shows distribution of all sgRNAs. Red bars represent targeted sgRNAs, grey bars represent all other sgRNAs. h,i, Representative plots (left) and quantification (right) of the proportions of PD-1⁺TIM-3⁺ (h, n = 7 mice) and (MTDR/MTG)^lo (i, n = 6 mice) TILs in sgCtrl-transduced or sgKlhl6-transduced Cas9⁺ OT-I T cells from B16-OVA tumours obtained on day 7 after ACT. Diagram in a created in Biorender. Li, G. (2025) https://BioRender.com/d5c767f. Diagram in e created in BioRender. Li, G. (2025) https://BioRender.com/dw0yfsp. Data are presented as mean ± s.e.m. Statistical analyses were determined by two-way ANOVA with Tukey’s multiple-comparisons test (h,i). *P < 0.05, ***P < 0.001 and ****P < 0.0001. Source data

Fig. 2. KLHL6 deficiency promotes T cell exhaustion and impairs mitochondrial function.

a,b, CD45.2⁺ C57BL/6N mice were subcutaneously injected with B16-OVA melanoma cells and, 9 days later, treated by ACT with 3 × 10⁶ CD45.1/2⁺ Klhl6^+/+ (WT) or Klhl6^−/− (KO) OT-I T cells (n = 14 mice). Tumour weights (a) and the numbers of transferred OT-I T cells (b) were assessed at day 14 post-ACT. c, Percentages of PD-1⁻TIM-3⁻, PD-1⁺TIM-3⁻ and PD-1⁺TIM-3⁺ subsets among CD45.1/2⁺ TILs (n = 14 mice). d, Cytokine production (TNF and IFNγ) in transferred CD45.1/2⁺ TILs was determined (n = 14 mice). e–i, For cotransfer experiments, CD45.1/2⁺ WT and CD45.1⁺ KO OT-I T cells were mixed at a 1:1 ratio and then adoptively transferred into tumour-bearing CD45.2⁺ mice 9 days after tumour inoculation, and the mice were euthanized for analysis at day 14 after ACT (n = 7 mice). Experimental design (e), the proportions of WT and KO OT-I cells in tumour, dLN and spleen (f), the percentages of PD-1⁺TIM-3⁺ population (g) and the expression levels of PD-1 and TIM-3 (h) and TOX (i) in cotransferred WT and KO TILs. j, Heat map of exhaustion-associated genes in WT and KO TILs at day 14 post-ACT (n = 3 independent samples). k, GSEA plots of signatures of mitochondrial OXPHOS and mitochondrial respiratory chain complex assembly in WT versus KO TILs (n = 3 independent samples). GSEA was performed using a one-sided, permutation-based modified K–S test with adjustments for multiple comparisons. l–n, Cotransferred WT and KO TILs were sorted for Seahorse assays on day 14 post-ACT. OCR (l), SRC (m) and mitochondrial ATP production (n) were measured (n = 8 independent tests; 20 mice). o,p, The frequencies of (MTDR/MTG)^lo subsets in transferred WT and KO OT-I TILs (o) and their distribution among exhausted (PD-1⁺TIM-3⁻ and PD-1⁺TIM-3⁺) and non-exhausted (PD-1⁻TIM-3⁻) populations (p) at day 14 after ACT (n = 8 mice). Diagram in e created in BioRender. Li, G. (2025) https://BioRender.com/5se3g09. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (a,b,f–i,m–o) or two-way ANOVA with Sidak’s multiple-comparisons test (c,d,l,p). *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001; NS, not significant. FCCP, carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone; MFI, mean fluorescence intensity; NES, normalized enrichment score; R&A, rotenone and antimycin A. Source data

Fig. 3. KLHL6 restrains the transition of Tpex cells to terminal differentiation and enhances anti-tumour immunity.

a–e, In total 3 × 10⁶ Control or KLHL6-OE OT-I T cells were transferred into tumour-bearing CD45.2⁺ mice 9 days after tumour inoculation, and the mice were euthanized for analysis at day 14 after ACT (n = 7 mice). Percentages (a) and numbers (b) of transferred CD45.1⁺ OT-I TILs, cytokine production (TNF and IFNγ) (c), and frequencies (d) and numbers (e) of Tpex (Ly108⁺TIM-3⁻) and Tex^term (Ly108⁻TIM-3⁺) subsets in Control and KLHL6-OE OT-I TILs were determined. f, Survival curves of tumour-bearing mice receiving Control or KLHL6-OE CD45.1⁺ OT-I T cells (n = 10 mice). Mice with tumour volumes greater than 1,500 mm³ were euthanized and this was defined as death. g, Transferred Control and KLHL6-OE OT-I TILs at day 14 post-ACT were subjected to scRNA-seq analysis. UMAP embedding showing clusters of all transferred CD8⁺ OT-I TILs (left) and their relative proportions (right); cluster annotations: 0, proliferation; 1, proinflammation; 2, exhaustion; 3, progenitor and 4, early activation. h, UMAP density plots comparing Control and KLHL6-OE TILs. i, Violin plots of Tpex and Tex^term gene signatures; the boxplot spans from the first to the third quartile of the distribution, with the median positioned in the centre and whiskers representing the minimum and maximum values, excluding outliers. Values plotted represent cells from a single replicate. j, Dot plots showing expression of signature genes, with both colour and size indicating effect size. k–m, CD45.1⁺ Control or KLHL6-OE OT-I Tpex cells were transferred into CD45.2⁺ congenic mice that had been subcutaneously implanted with B16-OVA cells 2 days earlier, and the mice were euthanized for analysis on days 8 and 16 post-ACT. k, Experimental design. l,m, Representative plots (l) and percentages (m) of Ly108⁺TIM-3⁻, Ly108⁺TIM-3⁺ and Ly108⁻TIM-3⁺ subsets among transferred cells before and after transfer (n = 7 mice). Diagram in k created in BioRender. Li, G. (2025) https://BioRender.com/rxkc0bf. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (a,b), two-way ANOVA with Sidak’s multiple-comparisons test (c–e), two-way ANOVA with Tukey’s multiple-comparisons test (m) and log-rank (Mantel–Cox) test (f). *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Source data

Fig. 4. TOX acts as a downstream target of KLHL6 and promotes terminal differentiation of Tex cells.

a, E-STUB assay and label-free mass spectrometry were used to identify KLHL6-proximal ubiquitylated substrates. Fold changes in protein abundance between KLHL6-BirA and Empty-BirA groups were calculated by two-sided moderated t-test (limma). b, Interaction between endogenous KLHL6 and TOX in human T cells. c, Cycloheximide (CHX) chase assay showing TOX degradation in Jurkat cells transduced with empty or KLHL6-Myc vectors (n = 3 independent samples). d, TOX levels in WT or Klhl6^−/− OT-I T cells with or without anti-CD3 restimulation (n = 3 independent samples). e, Quantification of endogenous TOX protein in Jurkat cells transduced with an empty vector or KLHL6-Myc, with or without MG132 treatment (10 μM, 6 h; n = 3 independent samples). f,g, Ubiquitination of endogenous TOX in KLHL6-overexpressing human T cells (f) and in mouse Klhl6^−/− CD8⁺ T cells (g). h, Violin plots showing the TOX and TCF-1 signatures in transferred Control and KLHL6-OE TILs from scRNA-seq data in Fig. 3g. The boxplot spans from the first to the third quartile of the distribution, with the median positioned in the centre. Whiskers represent the minimum and maximum values, excluding outliers. Values plotted represent cells from a single replicate. i–k, OT-I CD8⁺ T cells from CD45.1⁺ WT or KO donor mice were transduced with shCtrl or shTox retrovirus and adoptively transferred (4 × 10⁶) into B16-OVA tumour-bearing mice. The mice were euthanized at day 14 for analysis after adoptive transfer (n = 8 mice). Percentages (i) and absolute numbers (j) of Tex^term (Ly108⁻TIM-3⁺) and Tpex (Ly108⁺TIM-3⁻) subsets, and tumour weights (k) were assessed. l, Correlation of KLHL6 expression with Tpex, Tex^term and TOX signatures in human CD8⁺ TILs from human pan-cancer scRNA-seq data. For immunoblot source data, see Supplementary Fig. 1. Illustration in l created in BioRender. Li, G. (2025) https://BioRender.com/p3754eu. Data in b,f,g are representative of three independent experiments. Data are presented as mean ± s.e.m. Statistical analyses were determined by two-way ANOVA with Tukey’s multiple-comparisons test (c–e,i–k). *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001; NS, not significant. IP, immunoprecipitation. Source data

Fig. 5. PGAM5 serves as another downstream target of KLHL6 governing T cell mitochondrial fitness to modulate the anti-tumour responses.

a, Representative mitochondrial morphology, individual mitochondrial area (n = 218) and total crista length per mitochondrion (n = 60 cells) in WT and KO OT-I T cells. b–f, CD8⁺ WT and KLHL6 KO OT-I T cells transduced with shPgam5 (shP5) or shCtrl retrovirus were cultured for 6 days in vitro. b, Immunoblots of indicated proteins. c–f, OCR (c,d), SRC (e) and mitochondrial ATP production (f) were assessed (n = 9 tests). g–i, CD45.1/2⁺ WT and CD45.1⁺ KO OT-I T cells transduced with either GFP-shCtrl or Thy1.1-shPgam5 (shP5) were mixed equally (1:1:1:1) and cotransferred into CD45.2⁺ mice bearing B16-OVA tumours. Mice were analysed on day 14 post-ACT (n = 7 mice). Percentages of TNF⁺IFNγ⁺CD8⁺ TILs (g), frequencies (h) and cell numbers (i) of Tpex (Ly108⁺TIM-3⁻) and Tex^term (Ly108⁻TIM-3⁺) subsets were assessed. j,k, Tumour volume (j) and survival curves (k) of tumour-bearing mice following separate transfer of 5 × 10⁶ indicated cells were recorded (n = 10 mice). Mice with tumour volumes greater than 1,500 mm³ were euthanized and this was defined as death. l,m, 4 × 10⁶ WT + Empty, WT + PGC1α, KO + Empty or KO + PGC1α OT-I T cells were adoptively transferred into B16-OVA tumour-bearing mice. Mice were analysed at day 14 post-ACT for the frequencies and numbers of Tpex and Tex^term cells (l), and cytokine production (m) (n = 6 mice). n,o, CD45.1⁺ WT and KO OT-I T cells were transduced with either shCtrl, shPgam5 (shP5), shTox or a combination of shPgam5 and shTox (shP5 + shTox) retrovirus. A total of 4 × 10⁶ cells from each group were transferred into CD45.2⁺ B16-OVA tumour-bearing mice. Tpex and Tex^term subsets (n) and cytokine production (o) in CD8⁺ TILs were assessed on day 14 post-ACT (n = 6 mice). For immunoblot source data, see Supplementary Fig. 1. Experiments in a,b were repeated three times. Data in n,o are representative of two independent experiments. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (a), two-way ANOVA with Tukey’s multiple-comparisons test (d–j,l–o), and log-rank (Mantel–Cox) test (k). *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001; NS, not significant. Scale bar, 1 μm. Source data

Extended Data Fig. 1. Visualization of transcriptome deconvolution into 2 gene modules.

a, Application of surprisal analysis to the RNA-seq data atlases related to T cell exhaustion. Samples from two representative T cell exhaustion studies (GSE89307 and GSE86881) were shown antigen-specific CD8⁺ T cells from acute Listeria infection and tumour progression (right), and from chronic LCMV infection (left). In the acute/tumour model, TCR-transgenic naïve T cells were transferred prior to pathogen or tumour induction, with effector/memory and early/late exhausted states sampled across defined time points. In the chronic model, naïve and late exhausted T cells were isolated from spleens following persistent LCMV infection. The combination of the gene expression baseline and the first two gene modules accurately recapitulated the experimentally measured transcriptome profiles. Heatmaps display genes positively or negatively associated with Module 1 or Module 2 (red = high, blue = low). Sample annotations follow standard naming conventions. b,c, GSEA enrichment of pathways based on genes positively or negatively associated in Module 1 (M1, b) and Module 2 (M2, c). GSEA uses a one-sided, permutation-based modified K–S test with adjustments for multiple comparisons. d,e, Module scores for M1 and M2 were calculated using a time-series T cell RNA-seq dataset of acute and chronic LCMV infection (GSE41867) (d, n = 4 independent samples), and projected onto a two-dimensional map with the x- and y-axes representing the respective module scores (e). f, Heat map showing expression of E3-related genes in adoptively transferred T cells from spleen, and in double negative (DN, PD-1⁻TIM-3⁻), single positive (SP, PD-1⁺TIM-3⁻), and double positive (DP, PD-1⁺TIM-3⁺) cells from B16-OVA tumours at day 14 after ACT (n = 3 independent samples). g, Percentages for E3 ligase-related genes predicted to be correlated (orange) or anti-correlated (blue) with exhaustion are shown. h, Volcano plot showing differential expression of E3-related genes between PD-1⁻TIM-3⁻ and PD-1⁺TIM-3⁺ TILs. The x-axis represents the log₂transformed fold change (FC) values, and the y-axis represents the negative log₁₀ of adjusted P values (n = 3 independent samples). i, GSEA of ubiquitin-associated pathways in 400,000 human TILs with functional versus dysfunctional mitochondria (defined by the activity of mitochondrial complex I) across 21 cancer types. j, GSEA of ubiquitin-associated pathways in tumour-specific TILs with (MTDR/MTG)^hi versus (MTDR/MTG)^lo mitochondria from a mouse ACT model. k,l, Venn diagrams showing 78 E3 ligase genes negatively linked to T cell exhaustion, identified by overlapping differentially expressed genes from two gene modules (M1 and M2) with a curated E3 ligase list (k), and 133 E3 ligase genes positively associated with mitochondrial function, shared between mouse and human T cells (l).

Extended Data Fig. 2. KLHL6 deletion drives T cell exhaustion, but does not impact thymocyte development or peripheral T cell homeostasis.

a, Sanger sequencing of the Klhl6 locus in Cas9⁺sgRNA⁺ OT-I T cells transduced with sgCtrl, sgKlhl6-1, or sgKlhl6-2 in vitro. Primer sequences are listed in Supplementary Table 6. b,c, Mice were subcutaneously injected with 2×10⁵ B16-OVA melanoma cells, followed by transfer of 3×10⁶ Cas9⁺CD8⁺ OT-I T cells expressing control sgRNA or sgKlhl6 (two targets) on day 9. Tumours were collected and analyzed 7 days later. Experimental design (b). The number of transferred Cas9⁺ OT-I T cells in the tumour (c, n = 7 mice). d,e, Representative plots (left) and proportions (right) of LAG-3⁺TIM-3⁺ TILs (d), and cell numbers of indicated subsets (e) among sgCtrl- or sgKlhl6-transduced Cas9⁺CD8⁺ OT-I TILs (n = 7 mice). f,g, Proportions of PD-1⁺TIM-3⁺ (f) and (MTDR/MTG)^lo (g) TILs among sgCtrl- or the indicated sgRNA-transduced Cas9⁺CD8⁺ OT-I T cells from B16-OVA tumours (n = 6 mice). h, Schematic of CRISPR/Cas9-mediated generation of Klhl6^−/− mice. Guide RNA sequences are provided in Supplementary Table 6. i, Genotyping results for Klhl6^+/+ (WT) or Klhl6^−/− (KO) alleles. j,k, Percentages of thymocyte subsets: CD4⁻CD8⁻ double-negative (DN), CD4⁺CD8⁺ double-positive (DP), CD4⁺ single-positive (CD4SP) and CD8⁺ single-positive (CD8SP) (j), and CD44⁺ single-positive (DN1), CD44⁺CD25⁺ double-positive (DN2), CD25⁺ single-positive (DN3) and CD44⁻CD25⁻ double-negative (DN4) subpopulations (k) in DN cells of WT and KO mice (n = 6 mice). l, Numbers of CD19⁺, CD4⁺, and CD8⁺ cells in spleens of WT and KO mice (n = 6 mice). m, Percentages of naïve (CD44^loCD62L^hi), central memory (CD44^hiCD62L^hi; CM) and effector/effector memory (CD44^hiCD62L^lo, EFF/EM) CD8⁺ and CD4⁺ T cells in spleens of WT and KO mice (n = 6 mice). n,o, Proliferation of naïve WT and KO CD8⁺ T cells following anti-CD3/CD28 activation measured by CFSE dilution (n), and expression of activation markers (CD44, CD69 and CD25) at days 1-2 in vitro (o). p, Experimental design related to Fig. 2a–d. q-s, Percentages of transferred CD45.1/2⁺ OT-I T cells in dLN and spleen (q), and quantification of PD-1, TIM-3, LAG-3, and TOX expression OT-I TILs (r,s) at day 14 post-ACT (n = 14 mice). t, Survival curves of tumour-bearing mice after transfer of 5×10⁶ CD45.1/2⁺ WT and KO OT-I T cells (n = 13 mice). Mice with tumour volumes >1500 mm³ were euthanized and defined as death. Diagram in b created in BioRender. Li, G. (2025) https://BioRender.com/aicmpdn. Diagram in p created in BioRender. Li, G. (2025) https://BioRender.com/0feebot. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (q-s), two-way ANOVA with Tukey’s multiple-comparisons test (c-g), two-way ANOVA with Sidak’s multiple-comparisons test (j-m), and Log-rank (Mantel-Cox) test (t). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data

Extended Data Fig. 3. KLHL6 deficiency impairs mitochondrial fitness.

a,b, Heat maps of differentially expressed genes (a) and stemness-associated genes (b) between transferred WT and KO CD8⁺ T cells isolated from B16-OVA tumour-bearing mice at day 14 after ACT (n = 3 independent samples). c, Gene ontology analysis of differentially expressed genes in WT versus KO CD8⁺ TILs. Red and blue indicate molecular functions enriched in WT and KO cells, respectively. d,e, GSEA plots of memory and effector signatures (d) or cell cycle arrest and apoptosis signatures (e) for WT versus KO TILs. NES, normalized enrichment score. GSEA uses a one-sided, permutation-based modified K–S test with adjustments for multiple comparisons. f, Distinct mitochondrial biological processes in WT versus KO TILs at day 14 after ACT. g, WT and KO OT-I T cells were activated with anti-CD3/CD28 for 6 days. Mitochondrial mass and potential were measured using MTG (250 nM) and TMRE (50 nM), respectively. Relative fold changes of mitochondrial mass, membrane potential and TMRE/MTG ratio were calculated (n = 3 independent samples). h-k, Seahorse assays of OCR (h,i), SRC and mitochondrial ATP production (j), and glycoPER (k) in WT and KO OT-I T cells at day 6 after activation in vitro (n = 10 tests). l, Experimental design. Equal numbers (2×10⁶ each) of CD45.1/2⁺ WT and CD45.1⁺ KO OT-I T cells were cotransferred into CD45.2⁺ mice bearing B16-OVA tumour (day 9 post-inoculation) and analyzed at day 14 after ACT. m, Analysis of glycoPER in WT and KO OT-I TILs from B16-OVA tumours at day 14 after ACT (n = 8 independent tests; 20 mice). n, Gating strategy for identifying (MTDR/MTG)^hi and (MTDR/MTG)^lo populations among cotransferred WT and KO OT-I TILs. Diagram in l created in BioRender. Li, G. (2025) https://BioRender.com/5se3g09. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (g,j) and two-way ANOVA with Sidak’s multiple-comparisons test (i,m). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Source data

Extended Data Fig. 4. TCR stimulation downregulates KLHL6 level.

a, Relative Klhl6 mRNA levels in adoptively transferred OT-I T cells from spleens and PD-1⁻TIM-3⁻, PD-1⁺TIM-3⁻ or PD-1⁺TIM-3⁺ CD8⁺ TILs from B16-OVA tumours at day 14 after ACT (n = 3 independent samples). b,c, Schematic of OT-I T cell restimulation in vitro. OT-I T cells were isolated from spleens, activated with anti-CD3/CD28, and cultured in mIL-2-containing media for 4 days. Cells were restimulated on day 4 with anti-CD3 (2 μg/mL, 24 h), OVA (1 μg/mL, 24 h), or PMA (50 ng/mL, 12 h) and compared to the un-restimulated Controls (b). Immunoblot and quantification of KLHL6 protein level in treated cells. Actin was used as a loading control (c, n = 3 independent samples). d,e, OT-I T cells were activated and expanded in the media containing mIL-2 for 4 days, then subjected to either repeated anti-CD3 restimulation every 2 days until day 9 (three restim) or a single restimulation on day 4 followed by resting in mIL-2 media until day 9 (one restim) (d). Western blot and quantification of KLHL6 expression were performed (e, n = 3 independent samples). f, Transcriptional expression of Tcf7, Pdcd1, and Klhl6 in P14 T cells during acute and chronic LCMV infection (GSE41867) (n = 4 independent samples). The boxplot spans from the first to the third quartile of the distribution, with the median positioned in the center. Whiskers represent the minimum and maximum values, excluding outliers. g, Klhl6 mRNA expression in antigen-specific T cells over time during acute infection and tumour progression (GSE89307) (n = 2-3 independent samples). h-j, Naïve CD45.1⁺P14⁺CD8⁺ T cells were adoptively transferred into CD45.2⁺ recipients (5×10³ cells/mouse for LCMV-Clone 13 or 5×10⁴ cells/mouse for LCMV-Armstrong). One day later, mice were infected with either virus. CD45.1⁺P14⁺ T cells were sorted from the spleen at the indicated time points post-infection for analysis. Experimental design (h). mRNA (i, n = 3 independent samples) and protein (j) levels of KLHL6 in CD45.1⁺P14⁺ T cells were analyzed at indicated time points. Diagram in b created in BioRender. Li, G. (2025) https://BioRender.com/9zoxyuh. Diagram in d created in BioRender. Li, G. (2025) https://BioRender.com/5se3g09. Diagram in h created in BioRender. Li, G. (2025) https://BioRender.com/l0bgjp0. Data are presented as mean ± s.e.m. Statistical analyses were determined by two-way ANOVA with Tukey’s multiple-comparisons test (a,c,e). **P < 0.01, ***P < 0.001, and ****P < 0.0001. Source data

Extended Data Fig. 5. Enforced expression of KLHL6 restricts T cell exhaustion and promotes mitochondrial function.

a, Experimental design, related to Fig. 3a–e. b, Percentages of transferred Control and KLHL6-OE CD45.1⁺ OT-I T cells detected in the spleen and dLN on day 14 post-transfer (n = 7 mice). c, CellTrace Violet (CTV)-labeled CD45.1/2⁺ Control and CD45.1⁺ KLHL6-OE OT-I T cells were cotransferred into B16-OVA tumour-bearing recipient mice. TILs were analyzed on day 4 post-ACT (n = 5 mice). d-f, Proliferation (Ki-67^hi) (d), apoptosis (Annexin V⁺) (e), and expression of apoptotic molecules (Bcl-2, Bcl-XL, and Bim) (f) in transferred Control and KLHL6-OE TILs at day 14 post-ACT (n = 6 mice). g,h, Percentages (g) and cell numbers (h) of PD-1⁻TIM-3⁻, PD-1⁺TIM-3⁻ and PD-1⁺TIM-3⁺ populations in transferred CD45.1⁺CD8⁺ TILs (n = 7 mice). i,j, TOX (i) and TCF-1 (j) expression in CD8⁺PD-1⁺ TILs (n = 7 mice). k, Percentages of CD44⁺CD62L⁺ cells in spleen and dLN (n = 7 mice). l,m, Frequencies (l) and numbers (m) of Tpex (Ly108⁺TIM-3⁻) and Tex^term (Ly108⁻TIM-3⁺) subsets in WT and KO TILs (n = 14 mice), related to Extended Data Fig. 2p. n,o, TOX (n) and PD-1 (o) expression in the Tpex and Tex^term subsets at day 14 post-ACT (n = 7 mice). p-t, CD45.1/2⁺ KLHL6-OE and CD45.1⁺ Control OT-I T cells (2 × 10⁶ each) were cotransferred into CD45.2⁺ mice bearing B16-OVA tumour. At day 14 post-ACT, TILs were isolated for metabolic analysis, including OCR (p), SRC (q), mitochondrial ATP production (r), glycoPER (s), and the percentage of (MTDR/MTG)^lo cells (t) (Seahorse: n = 6 independent tests, 20 mice; (MTDR/MTG)^lo subset: n = 8 mice). u-y, CD8⁺ OT-I T cells transduced with KLHL6-OE or Control retrovirus were analyzed on day 6 for MTG and TMRE staining (u, n = 3 independent samples), as well as OCR (v,w), SRC (x), and mitochondrial ATP production (y) (n = 10 tests). Diagram in a created in BioRender. Li, G. (2025) https://BioRender.com/aicmpdn. Data are presented as mean ± s.e.m. Statistical analyses were performed by unpaired two-tailed Student’s t-test (b-f,i-k,q,r,t,u,x,y) or two-way ANOVA with Sidak’s multiple-comparisons (g,h,l-p,s,w). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data

Extended Data Fig. 6. Enforced KLHL6 expression enhances mouse and human T cell anti-tumour immunity.

a-d, Violin (a), dot (b), and UMAP (c,d) plots showing effector, exhaustion, and stem-like gene signatures, together with representative gene expression patterns and Tex^term (upper) and Tpex (lower) profiles from scRNA-seq in Fig. 3g. The boxplot spans from the first to the third quartile of the distribution, with the median positioned in the center. Whiskers represent the minimum and maximum values, excluding outliers. Values plotted represent cells from a single replicate. e, Gene ontology analysis of differentially expressed genes between Control and KLHL6-OE TILs from scRNA-seq data. f,g, Experimental design as in Fig. 3k. Numbers of transferred Control and KLHL6-OE OT-I T cells in B16-OVA tumours on days 8 and 16 post-ACT (f) and tumour growth curves in B16-OVA-bearing mice receiving PBS, Control, or KLHL6-OE Tpex cells (g) (n = 7 mice). h-n, CD45.2⁺ mice were subcutaneously injected with B16-OVA melanoma cells. After 9 days, CD45.1⁺ Control and CD45.1/2⁺ KLHL6-OE OT-I T cells were mixed at a 1:1 ratio up to 5 million total cells and adoptively transferred into the tumour-bearing mice. Mice were sacrificed at days 21/28 post-transfer (n = 7 mice). Experimental design (h). Percentages and numbers of transferred T cells in tumour, dLN, spleen, and blood were measured (i-k). MFI of PD-1, TIM-3, and Ki-67 (l), cytokine production (m), and percentages of T_CM (CD44⁺CD62L⁺) and T_SCM (CD44⁻CD62L⁺CD95⁺) subsets (n) were determined at day 28. Polyfunctional subset indicates simultaneous expression of IL-2, IFNγ, and TNFα. o-t, NCG mice were subcutaneously implanted with HepG2-ESO cells. After 12 days, 6×10⁶ activated 1G4 TCR-T cells transduced with KLHL6-OE or Control retrovirus were adoptively transferred. Mice were analyzed at day 16 post-transfer (n = 8 mice). Experimental design (o). Tumour weights (p), numbers of TCR-T cells in tumour and blood (q), MFI of LAG-3, TIM-3 and TOX (r) and cytokine production (TNFα and IFNγ) (s) in TCR-TILs, and frequencies of T_CM (CD62L⁺CD45RA⁻) and Tem (CD62L⁻CD45RA⁻) subsets from spleens (t) were evaluated. Diagram in h created in BioRender. Li, G. (2025) https://BioRender.com/5se3g09. Diagram in o created in BioRender. Li, G. (2025) https://BioRender.com/5se3g09. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (l-n,p-r), two-way ANOVA with Sidak’s multiple-comparisons test (f,i-k,s,t), or two-way ANOVA with Tukey’s multiple-comparisons test (g). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data

Extended Data Fig. 7. KLHL6 loss promotes Tex^term cell differentiation during chronic infection.

a, Naïve Klhl6^+/+ (WT) and Klhl6^−/− (KO) CD8⁺ P14 T cells were mixed at a 1:1 ratio (5×10³ for LCMV-CL13 or 5×10⁴ for LCMV-Arm) and transferred into CD45.2⁺ recipients, followed by intravenous LCMV infection one day later. Activation markers in transferred WT and KO P14 T cells from spleen were assessed at days 4 and 6 post-infection (p.i.). b,c, Naïve CD45.1/2⁺ WT and CD45.1⁺ KO P14 cells were cotransferred (2.5×10³ each) into CD45.2⁺ mice, followed by LCMV-CL13 infection one day later. Mice were sacrificed on days 8 and 28 p.i. (n = 6 mice). Experimental design (b), and percentages (left) and absolute numbers (right) of WT and KO P14 cells in spleen (c). d,e, Frequencies and numbers of “effector-like” (TCF-1⁻GzmB⁺ or Ly108⁻TIM-3⁺) and Tex^prec (TCF-1⁺GzmB⁻ or Ly108⁺TIM-3⁻) subsets in transferred WT and KO P14 T cells from spleens at day 8 p.i. (n = 6 mice). f-h, Percentages of TNFα⁺IFNγ⁺ cells (f), (MTDR/MTG)^lo subsets (g), and apoptotic cells (Annexin V⁺PI⁺; Annexin V⁺PI⁻) (h) in transferred WT and KO P14 cells from spleens at day 8 p.i. of CL13-infected mice (n = 6 mice). i, Quantification of TIM-3, PD-1, TOX and GzmB expression and the percentage of TCF-1⁺ cells at day 28 p.i. (n = 6 mice). j,k, Frequencies and numbers of Tpex (Ly108⁺CX3CR1⁻), Tex^int (Ly108⁻CX3CR1⁺) and Tex^term (Ly108⁻CX3CR1⁻) subsets (j), and Bcl-2/Bim expression in these subsets (k) at day 28 p.i. (n = 6 mice). l-n, Frequencies and numbers of Tpex1 (Ly108⁺CD69⁺), Tpex2 (Ly108⁺CD69⁻), Tex^int (Ly108⁻CD69⁻), and Tex^term (Ly108⁻CD69⁺) subsets (l), TOX expression in these subsets (m), and cytokine production (TNFα and IFNγ) after PMA/BFA stimulation (n) in transferred WT and KO P14 cells at day 28 p.i. (n = 6 mice). o, Mice receiving PBS, WT P14, or KO P14 cells were infected with LCMV-CL13 one day after transfer, and LCMV viral loads in liver and lungs were measured on day 15 p.i., normalized to the PBS group. LCMV titers were assessed by qPCR relative to HPRT (PBS, n = 6 mice; WT and KO, n = 8 mice). Diagram in b created in BioRender, Li. G (2025) https://BioRender.com/l0bgjp0. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (c-n), or two-way ANOVA with Tukey’s multiple-comparisons test (o). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data

Extended Data Fig. 8. Enforced KLHL6 expression restrains exhaustion and boosts anti-viral T cell responses.

a, Experimental design. Activated CD45.1⁺ P14 T cells transduced with KLHL6-OE or Control retrovirus were separately transferred into CD45.2⁺ recipients (5×10³ cells/recipient), followed by LCMV-CL13 infection one day later. Mice were sacrificed at days 8 and 21 p.i. b,c, Representative plots and frequencies of “effector-like” (TCF-1⁻GzmB⁺ or Ly108⁻TIM-3⁺) and Tex^prec (TCF-1⁺GzmB⁻ or Ly108⁺TIM-3⁻) subsets in transferred Control and KLHL6-OE P14 T cells from spleens at day 8 p.i. (n = 6 mice). d, Numbers of TCF-1⁺GzmB⁻, TCF-1⁻GzmB⁺ and total P14 T cells from spleen at day 8 p.i. (n = 6 mice). e, Quantification of Annexin V⁺ P14 cells at day 8 p.i. (n = 6 mice). f,g, Cytokine production (TNFα and IFNγ) (f) and relative MFI of TIM-3, PD-1 and TCF-1 (g) in KLHL6-OE versus Control P14 T cells from spleens at day 21 p.i. (n = 6 mice). h-j, Frequencies (h) and absolute numbers (i) of Tpex1, Tpex2, Tex^int and Tex^term subsets, and TOX expression in these subsets (j) (n = 6 mice). k,l, Frequencies (k) and numbers (l) of Tpex (Ly108⁺CX3CR1⁻), Tex^int (Ly108⁻CX3CR1⁺), and Tex^term (Ly108⁻CX3CR1⁻) subsets in spleens at day 21 p.i. (n = 6 mice). m, Mice receiving PBS, Control, and KLHL6-OE P14 T cells were infected with LCMV-CL13 one day after adoptive transfer, and LCMV viral loads in liver and lungs were measured on day 15 p.i., normalized to the PBS group. LCMV titers were assessed by qPCR relative to HPRT (PBS, n = 5 mice; Control and KLHL6-OE, n = 8 mice). Diagram in a created in BioRender. Li, G. (2025) https://BioRender.com/l0bgjp0. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (b-l), or two-way ANOVA with Tukey’s multiple-comparisons test (m). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data

Extended Data Fig. 9. KLHL6 mediates Lys48-linked poly-ubiquitination and degradation of TOX.

a, Schematic of the E-STUB system. KLHL6 was fused to the biotin ligase BirA and co-expressed with BAP-tagged ubiquitin in cells, enabling biotinylation of substrates proximal to BirA-KLHL6 in a ubiquitin-specific manner. b, Rank-ordered plot of MS results showing streptavidin-enriched proteins from an E-STUB assay in Jurkat cells expressing KLHL6-BirA versus those expressing Empty-BirA. c, Jurkat cells co-expressing BAP-tagged ubiquitin with either KLHL6-BirA or Empty-BirA were treated with or without biotin (50 μM) for 15 min. Biotinylated proteins were then enriched using streptavidin beads and detected by immunoblotting. Braces indicate bands related to the target proteins. d, Interaction between KLHL6-Myc and endogenous TOX in Jurkat and EL4 cells. e,f, Schematic of full-length TOX protein (WT) and truncated mutants (ΔC1, ΔC2, and ΔN) (e), and requirement of the C-terminal domain for KLHL6 interaction (f). g, Western blot of Flag-TOX degradation in HEK293T cells transfected with empty vector or different dosages (0.5, 1, and 1.5 μg) of KLHL6-Myc. h, Endogenous TOX levels in human T cells transduced with KLHL6-Myc and cultured with or without anti-CD3 stimulation (n = 3 independent samples). i, Ubiquitination of exogenous TOX in HEK293T cells cotransfected with HA-Ub and KLHL6-Myc. j, Ubiquitination of endogenous TOX in Jurkat cells with KLHL6 overexpression. k, Ubiquitination analysis of endogenous TOX in activated mouse and human T cells upon anti-CD3 restimulation. l, Measurement of TOX ubiquitination with mutant Ub (K48R, K63R) in HEK293T cells. m, Identification of the key TOX sites responsible for KLHL6-mediated degradation. n-p, Ubiquitination (n), interaction (o) and degradation (p, n = 3 independent samples) of a quadruple mutation (4KR) of TOX in HEK293T cells cotransfected with KLHL6-Myc. A triple mutant (K245, K246, and K248) was defined as ‘3KR’, and the quadruple mutant (K245, K246, K248, and K323) defined as ‘4KR’. q, Evaluation of degradation of the TOX quadruple mutant (4KR) in HEK293T cells treated with CHX (50 µg/mL) for the indicated times (n = 3 independent samples). Diagram in a created in BioRender. Li, G. (2025) https://BioRender.com/va0cyy2. Data in (c,d,f,g,i-o) are representative of two independent experiments. Data are presented as mean ± s.e.m. Statistical analyses were determined by unpaired two-tailed Student’s t-test (p,q), or two-way ANOVA with Tukey’s multiple-comparisons test (h). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data

Extended Data Fig. 10. A TOX mutant resistant to KLHL6-mediated degradation reinforces the exhaustion phenotype in T cells.

a, Gating strategy, histogram plots, and quantification of TIM-3 and TCF-1 levels in TOX^loPD-1^lo, TOX^loPD-1^hi, and TOX^hiPD-1^hi OT-I TIL populations from B16-OVA tumours at day 14 post-transfer (n = 7 mice). b, Percentages of Tpex (Ly108⁺TIM-3⁻) and Tex^term (Ly108⁻TIM-3⁺) populations in TOX^lo and TOX^hi TILs at day 14 (n = 7 mice). c,d, Numbers of CD45.1⁺ OT-I TILs (c) and TOX expression levels (d) in WT+shCtrl, KO+shCtrl, and KO+shTox groups at day 14 after ACT (n = 8 mice). e, Correlation between KLHL6 expression and the indicated genes in human TILs from melanoma, renal cancer, breast cancer, and nasopharyngeal cancer, using human pan-cancer scRNA-seq data. CD8⁺ TILs were stratified into KLHL6^hi and KLHL6^lo groups. Selected genes were quantified and visualized with red indicating highly expressed and blue indicating lower expressed. f-k, Naïve CD8⁺ OT-I T cells were activated with anti-CD3/CD28 for 24 h, and transduced with TOX(WT) (TOX^WT-OE), TOX(4KR) (TOX^4KR-OE) or Empty vector (Control). Cells were cultured for 5 days in vitro before subsequent analyses. Experimental design (f), percentages of PD-1⁺LAG-3⁺ populations (g), TOX, PD-1, and LAG-3 levels (h), cytokine production (i), heatmap of differentially expressed genes (j), and GSEA for exhaustion and effector signatures (k) (n = 3 independent samples). GSEA uses a one-sided, permutation-based modified K–S test with adjustments for multiple comparisons. l-o, 3×10⁶ Control, TOX(WT), or TOX(4KR) OT-I T cells were adoptively transferred into B16-OVA tumour-bearing mice. Mice were sacrificed on day 14 after ACT. TOX expression (l), percentages of PD-1⁻TIM-3⁻, PD-1⁺TIM-3⁻ and PD-1⁺TIM-3⁺ populations (m), MFI of LAG-3, TIM-3, and PD-1 (n), and cytokine production (o) in CD45.1⁺CD8⁺ TILs were assessed (n = 6 mice). Diagram in e created in BioRender. Li, G. (2025) https://BioRender.com/p3754eu. Diagram in f created in BioRender. Li, G. (2025) https://BioRender.com/ahcb92h. Data are presented as mean ± s.e.m. Statistical analyses were performed by two-way ANOVA with Sidak’s multiple-comparisons test (b) or with Tukey’s multiple-comparisons test (a,c,d,g-i,l-o). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data

Extended Data Fig. 11. KLHL6 modulates mitochondrial fitness and anti-tumour T cell responses via ubiquitinating PGAM5.

a, Immunoblot of mitochondrial fusion/fission proteins in WT or KLHL6 KO OT-I T cells activated by anti-CD3/CD28 for indicated times (n = 3 independent samples). b, Immunoblot of p62 and LC3 (upper band: LC3-I; lower band: LC3-II) in WT and KO OT-I T cells at day 5 post-activation. c, CD45.2⁺ B16-OVA-bearing mice received 4×10⁶ WT or KO CD45.1⁺ OT-I cells pretreated with DMSO or M1 (20 μM) + Mdivi-1 (10 μM) (MM) for 3 days. Survival of mice was monitored (n = 8 mice). d, Interaction between introduced KLHL6-Myc and endogenous PGAM5 in Jurkat cells. e, Ubiquitination of endogenous PGAM5 in Jurkat cells transduced with KLHL6-Myc. f,g, Immunoblot analysis of endogenous PGAM5 in Jurkat cells expressing an empty vector or KLHL6-Myc with or without MG132 treatment (f) and in WT and KO T cells collected at different time points after activation (g). h,i, WT and KLHL6 KO OT-I T cells transduced with shPgam5 (shP5) or shCtrl retrovirus were cultured in vitro for 5 days. TMRE/MTG ratio (h, n = 5 independent samples) and mitochondrial morphology and area (i; scale bar, 1 μm; n = 60 cells) were analyzed. j-m, CD45.1/2⁺ WT and CD45.1⁺ KLHL6 KO OT-I T cells transduced with GFP-shCtrl or Thy1.1-shPgam5 (shP5) were mixed at a 1:1:1:1 ratio and cotransferred into CD45.2⁺ B16-OVA tumour-bearing mice. Mice were sacrificed for analysis on day 14 after ACT. Experimental design (j), CD8⁺ TIL numbers (k), PD-1 and TIM-3 expression in TILs (l), and numbers of CD44⁺CD62L⁺ cells in dLN and spleen from the four groups (m) (n = 7 mice). n-r, WT and KO OT-I T cells were activated and cultured for 3 days in vitro, then treated with DMSO or the PGAM5 inhibitor LFHP-1c (P5i, 2 μM) for another 3 days before analysis. Immunoblotting of Mfn2, Opa1, Drp1, p-Drp1^S637, PGAM5, and Actin (n). OCR, SRC, mitochondrial ATP production (o-q, n = 10 tests), and TMRE/MTG ratio (r, n = 3 independent samples) were assessed. s-y, CD45.2⁺ mice bearing B16-OVA tumours received 4×10⁶ activated WT or KO CD45.1⁺ OT-I cells pretreated with DMSO or P5i. The mice were sacrificed for analysis at day 14 after ACT. Schematic of the experiment (s), tumour weights (t), total numbers (u), cytokine production (v), PD-1 and TIM-3 levels (w), and cell numbers of indicated subsets (x) of OT-I TILs were evaluated (n = 6 mice); percentages of T_CM populations in dLN and spleen (y, n = 6 mice). Diagram in j created in BioRender. Li, G. (2025) https://BioRender.com/md3c1bz. Diagram in s created in BioRender. Li, G. (2025) https://BioRender.com/ap5vq6h. Data in (b,d-g,n) are representative of three independent experiments. Data are presented as mean ± s.e.m. Statistical analyses were determined by two-way ANOVA with Tukey’s multiple-comparisons test (a,h,i,k-m,o-r,t-y) and Log-rank (Mantel-Cox) test (c). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data

Extended Data Fig. 12. TOX and mitochondrial fitness are key mediators of KLHL6-driven anti-tumour immune responses.

a,b, CD8⁺ WT and KLHL6 KO T cells transduced with either Empty vector (Empty) or PGC1α-overexpression (PGC1α) plasmids were cultured for 6 days. OCR (a, n = 10 tests) and SRC and mitochondrial ATP production (b, n = 10 tests) were measured. c-e, Tumour weights (c), absolute numbers of transferred CD8⁺ TILs (d), and proportion of damaged mitochondria in transferred CD8⁺ TILs (e) were assessed in the indicated groups (n = 6 mice), related to Fig. 5l,m. f-i, Experimental design related to Fig. 5n,o (f), tumour weights (g), absolute numbers of total (h) and Tpex (Ly108⁺TIM-3⁻) and Tex^term (Ly108⁻TIM-3⁺) subsets (i) of transferred CD8⁺ TILs among the five groups at day 14 after ACT (n = 6 mice). j-n, CD8⁺ OT-I T cells transduced with either TOX-overexpressing (TOX-OE) or Empty vector (Control) retrovirus were cultured for 5 days in vitro, and then restimulated with or without CD3 antibody for 24 h before analysis. Representative plots and quantification of mitochondrial depolarization (TMRE/MTG)^lo (j) and MitoSOX level (k) were assessed (n = 3 independent samples). OCR (l), SRC (left) and mitochondrial ATP production (right) (m), and glycoPER (n) were measured (n = 10 tests). o,p, Activated WT and KO CD8⁺ OT-I T cells were transduced with either shCtrl or shTox retrovirus at 24 h post-activation and cultured for an additional 5 days prior to analysis. OCR (o), SRC (left) and mitochondrial ATP production (right) (p) were measured (n = 8 tests). Diagram in f created in BioRender. Li, G. (2025) https://BioRender.com/md3c1bz. Data in (f-i) are representative of two independent experiments. Data are presented as mean ± s.e.m. Statistical analyses were determined by two-way ANOVA with Tukey’s multiple-comparisons test (a-e,g-p). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant. Source data