Targeting ZMYND8 unleashes IL-2 signalling to override T cell exhaustion

Persistent antigen exposure in chronic infection and cancer induces an exhausted program orchestrated by the transcription factor TOX in CD8⁺ T cells^4,5,6,7,8, which is a barrier for immunotherapies. Such hypofunctional programming is associated with progressively decreased expression of the high-affinity IL-2 receptor subunit IL-2Rα (also known as CD25), leading to intermittent IL-2 engagement and suboptimal STAT5 activation that promote exhaustion². The mechanisms underlying this impaired IL-2Rα expression are largely unknown, so it remains unclear how to engage high-affinity IL-2R signalling and reinvigorate exhausted T cell (T_ex) effector function for antiviral and antitumour immunity.

Epigenetic remodelling contributes to T_ex heterogeneity, with T_ex subpopulations exhibiting distinct epigenetic states and differential responses to immune checkpoint blockade (ICB)^1,3. Progenitor exhausted T cells (Tex^prog cells) respond to ICB by transiently developing into effector-like cells with increased cytotoxic function and later convert to terminally exhausted T cells (Tex^term cells)^9,10,11. Effector-like cells comprise intermediate exhausted T cell (Tex^int cell) and killer cell lectin-like receptor (KLR)-expressing T cell (Tex^KLR cell) subpopulations, with Tex^KLR cells displaying the strongest cytolytic activity^12,13. Uncovering mechanisms that promote effector-like cell accumulation or redirect differentiation into Tex^KLR over Tex^term states may unleash protective immunity against chronic infection and tumours.

To identify epigenetic regulators underlying T_ex heterogeneity, we applied complementary in vivo single-cell CRISPR (scCRISPR) and bulk CRISPR screens in Cas9-expressing P14 cells, which are antigen-specific CD8⁺ T cells that respond to chronic lymphocytic choriomeningitis virus (LCMV) clone 13 (Cl13) infection. Considering cell coverage limitations¹⁴, we designed a single guide RNA (sgRNA) library based on analysis of public datasets^11,15,16 (Extended Data Fig. 1a), resulting in a total of 91 epigenetic factors. Nine transcription factors known to regulate T_ex differentiation were also included in the library (Supplementary Table 1), which contained 3 sgRNAs per gene and 30 non-targeting control sgRNAs (sgNTCs) (Supplementary Table 2). Library-transduced P14 cells (Ametrine⁺) were transferred into Cas9⁺ mice, followed by LCMV Cl13 infection and analysis at 28 days post-infection (dpi) (Fig. 1a). Of the 330 total sgRNAs in the library, 326 were detected in the scCRISPR screen (Extended Data Fig. 1b). Known positive (for example, Batf^17,18,19) and negative (for example, Carm1²⁰) regulators of T cell accumulation were captured (Extended Data Fig. 1c).

We visualized single-cell transcriptome data using uniform manifold approximation and projection (UMAP) plots. Four clusters were identified and annotated as Tex^prog, Tex^int, Tex^KLR and Tex^term clusters based on their signatures (Extended Data Fig. 1d–f). Flow cytometry analysis revealed that Tex^KLR cells exhibited the highest expression of IFNγ (but not TNF), GZMA, GZMB and perforin (Extended Data Fig. 1g). Additionally, pseudotime analysis revealed a differentiation trajectory originating from Tex^prog cells, which progressed through the Tex^int state before diverging towards either the Tex^KLR or Tex^term state (Fig. 1b).

We next explored how epigenetic factors regulate gene programs in T_ex cells (Methods). This analysis identified nine co-functional modules (M1–M9; Fig. 1c and Supplementary Table 3) exhibiting differential regulatory effects, along with four co-regulated gene programs (G1–G4; Fig. 1c and Supplementary Table 4) that displayed unique molecular signatures (Extended Data Fig. 1h) and distributions on UMAP (Fig. 1d). We visualized the regulatory strength of each co-functional module on the four gene programs (Extended Data Fig. 1i) and highlighted the top five co-functional modules based on their regulatory effects (Fig. 1e). Two co-functional modules, M3 and M8, showed the strongest perturbation effects on these gene programs (Extended Data Fig. 1i). M3 promoted stemness-associated program G1 and exhaustion-associated program G4, but suppressed proliferation-associated program G2 and effector function-associated program G3 (Fig. 1e). By contrast, M8 suppressed G1 and G4 but promoted G2 and G3 (Fig. 1e), suggesting that M3 and M8 may antagonize each other in shaping downstream gene programs. Additionally, Zmynd8 and Hdac4 from M3 and Ep300 (encoding p300) from M8 represented putative positive and negative regulators of Tox expression, respectively (Extended Data Fig. 1j). Together, these results reveal effects of distinct co-functional modules on orchestrating gene programs and Tox expression in T_ex cells.

During chronic infection, P14 cells displayed low Il2ra expression, especially compared to Il2rb (Extended Data Fig. 1e). Furthermore, Il2ra expression and CD25⁺ cells, together with CD122 surface levels, were progressively decreased (Extended Data Fig. 1k,l), suggesting that T_ex cells receive suboptimal IL-2 signals at later stages of chronic infection. To test the contribution of IL-2 signalling, we performed another scCRISPR screen targeting regulators of IL-2 signalling in chronic infection. Loss of positive regulators of IL-2 signalling (JAK3 or STAT5A and STAT5B) reduced numbers of effector-like and total P14 cells (Extended Data Fig. 1m,n and Supplementary Table 5). By contrast, targeting negative regulators of IL-2 signalling (SOCS1 and SOCS3) increased accumulation of effector-like and total P14 cells (Extended Data Fig. 1m,n). Therefore, negative regulators of IL-2 signalling represent putative targets for enhancing the formation of effector-like states.

We next focused on putative epigenetic factors that suppress effector-like states in scCRISPR screening (Fig. 1f,g and Extended Data Fig. 1o–q). Notably, Zmynd8 was among the top-ranked negative regulators of effector-like cells (Fig. 1f) and positive regulators of Tex^term cells (Extended Data Fig. 1p and Supplementary Tables 6 and 7), with enrichment of the Tex^int or Tex^KLR versus Tex^term state observed in cells expressing sgRNAs targeting Zmynd8 (sgZmynd8) (Fig. 1g and Extended Data Fig. 1q). Similarly, in bulk CRISPR screening, Zmynd8 was the top-ranked negative regulator of effector-like relative to Tex^term cells, whereas Zmynd8 was the top-ranked positive regulator of Tex^term cells (Extended Data Fig. 1r,s and Supplementary Table 6), suggesting that ZMYND8 suppresses effector-like but promotes Tex^term cell accumulation during chronic infection.

ZMYND8 was also the top-ranked negative regulator of STAT5 activity (Fig. 1h). Gene set enrichment analysis (GSEA) revealed that a STAT5 constitutively active (STAT5CA) Up signature²¹ was increased in ZMYND8-deficient cells, whereas a STAT5CA Down signature²¹ was decreased (Fig. 1i). Moreover, sgZmynd8-expressing cells were enriched in Tex^int and Tex^KLR states, with sgZmynd8 and STAT5CA Up signature activity overlapping in the Tex^int and Tex^KLR clusters (Extended Data Fig. 1t–v). Similarly, individual sgRNAs targeting Zmynd8 were enriched in the Tex^KLR versus Tex^term state (Extended Data Fig. 1w,x and Supplementary Table 7). Collectively, ZMYND8 deficiency increases effector-like cell while dampening Tex^term formation, and is associated with enhanced STAT5 activity.

To establish the role of ZMYND8 in T_ex heterogeneity, we performed single-cell RNA-sequencing (scRNA-seq) analysis of P14 cells from a dual-colour transfer system^19,22 (Extended Data Fig. 2a). ZMYND8-deficient P14 cells showed a selective accumulation of Tex^int and Tex^KLR states (Fig. 2a,b and Extended Data Fig. 2b), corresponding to enrichment of an effector-like gene signature¹⁰ (Extended Data Fig. 2c). Furthermore, effector-associated genes (Cx3cr1, Gzmb and Zeb2) and natural killer (NK) cell-associated receptor genes (Klrd1, Klre1, Klrg1 and Klrk1) were increased in ZMYND8-deficient P14 cells, whereas exhaustion-related genes (Tox, Tox2 and Cxcr6) were reduced (Extended Data Fig. 2d), suggesting that ZMYND8 deficiency boosts effector and reduces exhaustion-associated features.

Flow cytometry analysis validated that ZMYND8 deficiency increased numbers of effector-like cells and reduced Tex^term cells at multiple time points after infection (Fig. 2c and Extended Data Fig. 2e–i). Tex^prog numbers were also modestly reduced at 21 and 28 dpi (but not at 90 dpi) (Extended Data Fig. 2j–l). Although ZMYND8 deficiency increased Ki-67 expression and reduced the frequency of active caspase-3⁺ cells among total P14 cells (corresponding to their increased accumulation), Ki-67 expression was largely unaltered in ZMYND8-deficient T_ex subpopulations (Extended Data Fig. 2m–o). By contrast, the frequency of active caspase-3⁺ cells among ZMYND8-deficient effector-like cells was modestly reduced (Extended Data Fig. 2n), indicating that increased cell survival is likely to contribute to accumulation of effector-like cells upon ZMYND8 deletion.

We compared the cellular effects of the three independent sgRNAs targeting Zmynd8 and found that P14 cells expressing sgZmynd8.g1 and sgZmynd8.g2 showed marked increases in numbers of effector-like cells and reductions in Tex^prog and Tex^term cells, with sgZmynd8.g3 exerting similar, albeit less pronounced, effects (Extended Data Fig. 2p,q), in line with its less potent deletion efficiency (Extended Data Fig. 1x and Supplementary Table 7). Also, we targeted Zmynd8 in naive P14 cells by delivering sgRNA with Cas9 ribonucleoproteins (RNPs) (Cas9-RNP) via electroporation¹⁴ (Extended Data Fig. 2r). ZMYND8 deficiency in naive P14 cells also increased effector-like cells and reduced Tex^term and Tex^prog cells (Extended Data Fig. 2s–u), indicating that ZMYND8 shapes T_ex heterogeneity in both naive and preactivated CD8⁺ T cells.

Next, we assessed the effects of ZMYND8 deficiency on CD8⁺ T cells in multiple lymphoid and non-lymphoid tissues during chronic infection. Upon ZMYND8 deficiency, effector-like cells were increased in all tissues examined, accompanied by reduced Tex^term cell accumulation (Extended Data Fig. 3a–c). Thus, ZMYND8 deficiency affects T_ex heterogeneity across multiple tissues. Nonetheless, ZMYND8 deficiency showed limited effect on blood accessibility¹¹ of CD8⁺ T cells (Extended Data Fig. 3d,e), suggesting that ZMYND8 deletion-induced effects on the T_ex states are largely independent of alterations in lymphoid tissue residency.

Pseudotime analysis revealed that sgZmynd8-expressing cells predominantly accumulated in the Tex^KLR clusters (Extended Data Fig. 3f). ZMYND8 deficiency resulted in a substantial increase in cells co-expressing the Tex^KLR markers CD94 and NKG2D among total P14 cells or T_ex subpopulations (Fig. 2d and Extended Data Fig. 3g,h). Similarly, ZMYND8 deficiency increased Tex^KLR cells at 90 dpi and in multiple lymphoid and non-lymphoid tissues (Extended Data Fig. 3i,j). We had similar observations using Cas9-RNP editing of naive P14 cells (Extended Data Fig. 3k,l). Furthermore, ZMYND8-deficient P14 cells showed an increase in the proportion of cells expressing NK1.1, or those co-expressing NKG2A and CD94 or CX3CR1 and KLRG1 (Extended Data Fig. 3m–o), all of which mark Tex^KLR cells^12,13. Such accumulation effects were observed mainly in effector-like cells (Extended Data Fig. 3p,q). Therefore, ZMYND8 opposes Tex^KLR cell accumulation.

We next tested whether ZMYND8 directly regulates the transition from Tex^prog or effector-like cells to more differentiated states (Extended Data Fig. 4a). Following secondary transfer of either subpopulation, ZMYND8 deficiency led to accumulation of effector-like cells and reduction of Tex^term cells (Fig. 2e,f and Extended Data Fig. 4b,c). Then, we tested the effects of ZMYND8 deficiency on the differentiation of Tex^int into Tex^KLR and Tex^term cells (Extended Data Fig. 4d). ZMYND8 deficiency enhanced Tex^KLR cell accumulation and impaired Tex^term cell formation (Extended Data Fig. 4e,f). Therefore, targeting ZMYND8 increases accumulation of effector-like cells by enhancing the differentiation of Tex^prog cells into effector-like states and restricting their transition into Tex^term cells; moreover, these effects are associated with enhanced differentiation of Tex^int cells into Tex^KLR over Tex^term cells.

These findings raised the possibility that targeting ZMYND8 enhances CD8⁺ T cell effector function and antiviral capacity. Indeed, Gzma and Gzmb were upregulated in ZMYND8-deficient effector-like and Tex^term states (Extended Data Fig. 4g,h). Accordingly, ZMYND8-deficient P14 cells contained increased GZMA⁺ and GZMB⁺ frequencies in multiple tissues (Extended Data Fig. 4i–k). To directly assess the cytotoxic capacity of ZMYND8-deficient P14 cells, we used an established ex vivo killing assay¹¹ (Extended Data Fig. 4l) and found that ZMYND8-deficient T_ex subpopulations all showed increased target cell killing capacity (Fig. 2g and Extended Data Fig. 4m).

In an in vivo killing assay²², mice receiving ZMYND8-deficient P14 cells showed more efficient clearance of splenocytes pulsed with LCMV epitope glycoprotein_33–41 (gp33) peptide (Fig. 2h), suggesting improved target cell killing in vivo upon ZMYND8 deficiency. Additionally, there was an increased ratio of adoptively transferred ZMYND8-deficient P14 cells compared with endogenous gp33-specific CD8⁺ T cells in these mice (Extended Data Fig. 4n). Furthermore, ZMYND8-deficient P14 cells exhibited increased effector-like and reduced Tex^term cell frequencies compared to either endogenous gp33-specific CD8⁺ T cells or sgNTC-expressing P14 cells (Extended Data Fig. 4o). Concurrently, Tex^KLR cells were increased (Extended Data Fig. 4p,q), supporting a critical role for ZMYND8 in dictating Tex^KLR versus Tex^term fate choices. Finally, in viral load measurements¹⁰, mice receiving ZMYND8-deficient P14 cells showed a modest reduction of viral burdens (Extended Data Fig. 4r). Thus, ZMYND8 deletion improves antiviral immunity during LCMV Cl13 infection.

IL-2-based therapies reinvigorate T_ex antiviral responses^23,24. Accordingly, the STAT5CA Up signature was enriched in effector-like cells (Fig. 2i and Extended Data Fig. 5a), with ZMYND8-deficient Tex^prog and effector-like cells showing elevated STAT5CA Up and decreased STAT5CA Down activities (Extended Data Fig. 5b). Moreover, ZMYND8 deficiency increased STAT5 activity (Extended Data Fig. 5c,d), and this was associated with altered STAT5CA-mediated signatures (Extended Data Fig. 5e–g and Supplementary Table 8). Furthermore, in CD8⁺ T cells from a public dataset²⁴, ZMYND8 activity was negatively associated with IL-2 responses (Extended Data Fig. 5h). We therefore tested whether ZMYND8 deficiency enhances IL-2 therapeutic effects on antiviral control (Extended Data Fig. 5i). Mice receiving the ZMYND8-deficient P14 cells plus IL-2 combination exhibited lower viral loads compared to those receiving either monotherapy (Extended Data Fig. 5j). Mice receiving this combination therapy also showed further clearance of gp33 peptide-pulsed splenocytes (Fig. 2j), indicative of improved in vivo killing. These results reveal a cooperative effect on antiviral immunity by combining Zmynd8 targeting and IL-2 treatment.

We next explored the underlying cellular mechanisms. IL-2 treatment markedly reduced the accumulation of Tex^term cells, whereas effector-like cells were increased (Fig. 2k,l and Extended Data Fig. 5k), largely mirroring effects of ZMYND8 loss. The combination of ZMYND8 deletion and IL-2 treatment resulted in a more pronounced increase in effector-like cells, including Tex^KLR cells (Fig. 2l,m). Together, ZMYND8 restricts IL-2 responses, and combinatorial targeting of ZMYND8 in CD8⁺ T cells together with IL-2 treatment boosts antiviral immunity.

We next explored the mechanistic basis of how ZMYND8 restrains IL-2 responses. Based on assay for transposase-accessible chromatin using sequencing (ATAC–seq) analysis (Extended Data Fig. 5l), chromatin accessibility was increased at multiple effector-associated gene loci in ZMYND8-deficient effector-like cells (Extended Data Fig. 5m–o). Transcription factor activity inference revealed increased STAT5A and STAT5B activities in ZMYND8-deficient Tex^prog and effector-like cells (Fig. 3a, Extended Data Fig. 5p and Supplementary Table 9). Furthermore, genes with enhanced chromatin accessibility in both ZMYND8-deficient subpopulations were enriched for IL-2–STAT5 signalling signatures (Extended Data Fig. 5q), suggesting that ZMYND8 deficiency alters the epigenetic landscape of IL-2 target genes.

To further test this, we performed a single-cell multiome analysis. ZMYND8-deficient cells exhibited increased effector-like cells and reduced Tex^prog and Tex^term cells (Extended Data Fig. 6a–c), and increased chromatin accessibility at Gzma, Gzmb and Klre1 but reduced accessibility at Cd101 (Extended Data Fig. 6d). STAT5 motifs were enriched in ZMYND8-deficient total P14 cells and their subpopulations (Extended Data Fig. 6e). Next, we directly assessed STAT5 phosphorylation following IL-2 stimulation, and observed increased phosphorylated STAT5 (p-STAT5) levels in ZMYND8-deficient cells (Fig. 3b). ZMYND8-deficient T_ex subpopulations also showed reduced TOX levels across multiple time points and tissues (Fig. 3c and Extended Data Fig. 6f,g). The expression of the STAT5 targets Il2ra and Il2rb mediates sensitivity to IL-2². Il2ra expression was highest in Tex^KLR cells, while Il2rb expression was largely comparable in Tex^int, Tex^KLR and Tex^term cells (Extended Data Fig. 6h). Similar patterns were observed for CD25 and CD122 surface expression (Extended Data Fig. 6i,j), suggesting that effector-like populations are marked by selective CD25 upregulation. Furthermore, ZMYND8 deletion enhanced CD25 and CD122 expression, including in effector-like cells (Fig. 3d,e and Extended Data Fig. 6k–o). Together, these data show that ZMYND8 functions as an epigenetic brake on IL-2R–STAT5 signalling in T_ex states.

To establish conserved effects, we generated ZMYND8-deficient human CD8⁺ T cells and assessed IL-2–STAT5 signalling. p-STAT5 levels were increased upon ZMYND8 deficiency (Fig. 3f). Next, we utilized an established in vitro model²⁵ to test the effects of ZMYND8 deletion on human T cell states (Extended Data Fig. 7a). As expected²⁵, compared to acutely stimulated cells, cells receiving chronic T cell receptor (TCR) stimulation showed hallmark features of exhaustion (Extended Data Fig. 7b,c). Of note, ZMYND8-deficient human CD8⁺ T cells displayed higher expression of CD25 and CD122 under these conditions (Fig. 3g and Extended Data Fig. 7d). Furthermore, ZMYND8 deficiency decreased the expression of PD-1 and TOX but augmented IFNγ, TNF and GZMB production, with stronger effects observed under the chronic stimulation condition (Fig. 3h,i and Extended Data Fig. 7e–h). These results were validated using independent sgRNAs targeting ZMYND8 (Extended Data Fig. 7i–n). Moreover, during chronic stimulation, ZMYND8 deficiency increased and decreased the frequencies of Ki-67⁺ cells and active caspase-3⁺ cells, respectively (Extended Data Fig. 7o,p). Furthermore, loss of ZMYND8 improved the long-term persistence of human CD8⁺ T cells during chronic stimulation (Extended Data Fig. 7q). These results reveal conserved effects for ZMYND8 in mouse and human CD8⁺ T cells.

We next tested whether the effects of ZMYND8 deficiency on T_ex cells are mediated by IL-2–STAT5 signalling. We first generated P14 cells lacking ZMYND8 and STAT5B alone or in combination, followed by adoptive transfer into mice and LCMV Cl13 infection (Extended Data Fig. 8a,b). Co-deletion of STAT5B reversed the increases in effector-like and Tex^KLR cell accumulation in ZMYND8-deficient cells (Fig. 3j,k), as well as the enhanced expression of GZMA, GZMB and IFNγ (Extended Data Fig. 8c). STAT5B co-deficiency also largely reversed the phenotypes of reduced TOX expression (Fig. 3l) and Tex^prog or Tex^term accumulation (Extended Data Fig. 8d,e). Similar effects were largely observed upon co-deletion of CD25 or CD122 in ZMYND8-deficient cells (Extended Data Fig. 8f–o). Collectively, these results show that ZMYND8 restrains IL-2R expression and STAT5 signalling, thereby contributing to increased TOX expression and CD8⁺ T cell exhaustion.

We mapped chromatin occupancy of ZMYND8 by performing cleavage under targets and release using nuclease (CUT&RUN) assay. Among core ZMYND8 binding sites (2,863 peaks in total), IL-2–STAT5 signalling was the most enriched pathway (Fig. 4a), which included Il2ra and Il2rb (Supplementary Table 10). Accordingly, ZMYND8 mainly bound to the gene body of Il2ra and more upstream region of Il2rb (Extended Data Fig. 9a), suggesting that ZMYND8 may directly modulate IL-2R expression in CD8⁺ T cells during chronic infection.

ZMYND8 is an epigenetic reader that recognizes various histone modifications²⁶, so we performed CUT&RUN for these known histone marks²⁶ and assessed their overlaps with ZMYND8-bound genomic locations. The majority of core ZMYND8 binding sites were in promoter, intronic and distal regions (Extended Data Fig. 9b), resembling binding patterns of H3K4me1 and H3K27Ac marks (Fig. 4b and Extended Data Fig. 9c). ZMYND8 binding on Il2ra and Il2rb extensively overlapped with H3K4me1 and H3K27Ac-bound regions (Fig. 4c), whereas there was limited overlap between ZMYND8 binding sites with those of H3K14Ac, H3K36me2, H4K16Ac and H3K4me3, especially on Il2ra relative to Il2rb (Extended Data Fig. 9d). Since gene loci marked by both H3K4me1 and H3K27Ac often define active enhancer elements²⁷, these findings suggest that ZMYND8 preferentially binds to active enhancer elements. Next, we performed CUT&RUN for H3K27Ac using wild-type T_ex subpopulations and superimposed the results with ZMYND8-bound regions on Il2ra. H3K27Ac deposition at ZMYND8-bound regions on Il2ra was highest in effector-like and lowest in Tex^term cells (Extended Data Fig. 9e,f), in line with increased Il2ra expression in effector-like cells (Extended Data Fig. 9g). Therefore, H3K27Ac deposition at ZMYND8-bound active enhancer regions of Il2ra is associated with Il2ra expression.

H3K27Ac levels are regulated by the histone acetyltransferase p300²⁷. The occupancies of H3K27Ac and p300 were enriched in ZMYND8-bound regions in CD8⁺ T cells (Extended Data Fig. 9h,i). CUT&RUN assays revealed that p300 and ZMYND8 co-occupied enhancer regions on Il2ra and Il2rb, and p300 occupancy and H3K27Ac levels were increased at these loci in ZMYND8-deficient cells (Fig. 4d,e and Extended Data Fig. 9j–l). Furthermore, co-immunoprecipitation assays revealed the interaction of ZMYND8 and p300 in CD8⁺ T cells (Extended Data Fig. 9m). Finally, in ATAC–seq profiling, chromatin regions with increased accessibility in ZMYND8-deficient cells were enriched for p300 occupancy sites (Fig. 3a). Together, ZMYND8 deficiency enhances p300 chromatin occupancy, H3K27Ac deposition and p300-associated chromatin accessibility at selective loci in CD8⁺ T cells.

To test whether ZMYND8 functions as a transcriptional repressor for Il2ra transcription by restricting p300-mediated enhancer activity, we performed a luciferase reporter assay in Jurkat cells nucleofected with reporter constructs containing the IL2RA enhancer (CaRE4)²⁸. Upon TCR stimulation, co-expression of ZMYND8 markedly suppressed p300-mediated transactivation of IL2RA (Fig. 4f), supporting a direct inhibitory effect of ZMYND8 on p300-driven enhancer activity on the IL2RA locus. To dissect the functional domains required for ZMYND8 to repress p300-driven transcriptional activity, we constructed ZMYND8 truncation mutants lacking the individual PHD (ΔPHD), BRD (ΔBRD), PWWP (ΔPWWP), triple-reader cassette (ΔPHD–BRD–PWWP) or MYND (ΔMYND) domain (Extended Data Fig. 9n). Notably, the ΔPHD, ΔBRD, ΔPWWP (to a lesser extent) and ΔPHD–BRD–PWWP mutants did not suppress p300-driven IL2RA transcription (Extended Data Fig. 9o). These findings suggest that the histone-reader domains²⁹ of ZMYND8 contribute to repressing p300-dependent transcriptional regulation. Moreover, co-immunoprecipitation assays in HEK293T cells revealed reduced interactions between p300 and ZMYND8 truncation mutants lacking the PHD, PWWP (to a lesser extent) or PHD–BRD–PWWP domain (Extended Data Fig. 9p). Collectively, the histone-reader domains of ZMYND8 contribute to both its interaction with and repression of p300.

Targeting Ep300 reduced effector-like cell frequency but increased the proportions of Tex^prog and Tex^term cells (Extended Data Fig. 10a–c and Supplementary Table 11), which were validated by flow cytometry analysis (Extended Data Fig. 10d–g). Therefore, p300 deficiency produces phenotypic effects that oppose those of ZMYND8 deficiency. To directly test whether the effect of ZMYND8 deletion is dependent upon p300, we co-targeted ZMYND8 and p300 in Cas9⁺ P14 cells (Extended Data Fig. 10h), followed by adoptive transfer and LCMV Cl13 infection. Co-deletion of p300 blocked the increased effector-like states induced by ZMYND8 loss (Fig. 4g,h), and rectified the reductions in Tex^prog or Tex^term cell accumulation (Extended Data Fig. 10i,j). Co-deletion of p300 also reversed the increased CD25 and CD122 expression on ZMYND8-deficient P14 cells (Fig. 4i). Together, these findings define a ZMYND8–p300 axis that reciprocally regulates IL-2R expression and effector-like cell differentiation.

Zmynd8 expression was higher in CD8⁺ T cells at late stage of chronic infection versus early stage or acute infection (Fig. 5a). During chronic infection, Zmynd8 or Tox expression progressively increased and was highest in Tex^term and lowest in effector-like cells, especially in Tex^KLR cells (Fig. 5b,c and Extended Data Fig. 11a). Furthermore, Zmynd8 expression or ZMYND8-activated signature was enriched in the Tex^term state, whereas ZMYND8-suppressed signature was most prevalent in the Tex^KLR state (Extended Data Fig. 11b–e). In human CD8⁺ T cells from HIV-infected patients³⁰, ZMYND8 expression was positively correlated with exhaustion-related genes (Extended Data Fig. 11f), and the ZMYND8-activated signature activity was highest in individuals with high viral titres (Extended Data Fig. 11f), suggesting the association of ZMYND8 with human CD8⁺ T cell exhaustion.

Beyond chronic infection, the ZMYND8-activated signature and Zmynd8 expression were also increased in Tex^term compared with Tex^prog or effector-like cells in mouse tumour models^31,32, and the ZMYND8-suppressed signature was increased in effector-like cells (Extended Data Fig. 11g–i). In a mouse liver cancer model³³, ZMYND8-activated and suppressed signatures were respectively upregulated and downregulated as T cells became progressively dysfunctional (Extended Data Fig. 11j). Similarly, in intratumoural CD8⁺ T cells from patients with melanoma³⁴, ZMYND8-activated signature and Zmynd8 expression were highest in the dysfunctional state (Extended Data Fig. 11k). Together, ZMYND8 expression and/or activity are conserved features of terminal exhaustion across chronic viral infections and tumours.

Prolonged TCR stimulation in vitro recapitulates exhaustion-like features in T cells³⁵. Following sustained TCR stimulation in vitro, ZMYND8 expression gradually increased (Extended Data Fig. 11l). Furthermore, ZMYND8 expression was higher in human CD8⁺ T cells under the chronic versus acute stimulation condition²⁵ (Fig. 5d), with progressively increased ZMYND8 expression observed over time (Fig. 5e). Therefore, chronic TCR stimulation promotes ZMYND8 expression in both mouse and human CD8⁺ T cells.

Defects in IL-2 signalling in CD8⁺ T cells are a major barrier for cancer immunotherapy², so we tested whether ZMYND8 deletion unleashes IL-2 signalling in tumours. ZMYND8-deficient P14 cells exhibited increased levels of CD25, CD122 and p-STAT5, along with increased NKG2D and reduced TOX expression, in B16F10 tumours expressing gp33 (B16-gp33) (Fig. 5f). In a human CD19 chimeric antigen receptor (CAR) T cell model^19,31,36, ZMYND8-deficient CAR T cells also showed increased CD25, CD122 and p-STAT5 levels and reduced TOX expression (Extended Data Fig. 11m,n). NKG2D expression was also upregulated on ZMYND8-deficient CAR T cell populations (Extended Data Fig. 11o). Therefore, ZMYND8 deficiency enhances IL-2R–STAT5 signalling and promotes NK-like features in tumour-reactive CD8⁺ T cells, including CAR T cells.

Next, we assessed the antitumour effects of targeting ZMYND8. In B16-gp33 tumour-bearing mice, adoptive transfer of ZMYND8-deficient P14 cells reduced tumour burden and prolonged the survival of tumour-bearing mice (Fig. 5g and Extended Data Fig. 12a–d). In public datasets from melanoma or sarcoma-bearing mice treated with or without IL-2^37,38, ZMYND8 activity in CD8⁺ T cells was inversely associated with IL-2 responses (Extended Data Fig. 12e,f). Moreover, combining IL-2 treatment with adoptive cell therapy of ZMYND8-deficient P14 cells further enhanced tumour control and prolonged the survival of B16-gp33 tumour-bearing mice relative to either monotherapy (Fig. 5h and Extended Data Fig. 12g–i). ZMYND8-deficient pmel cells (which recognize an endogenous B16F10 melanoma antigen) also exhibited improved control of B16F10 tumour growth, with IL-2 combination therapy further enhancing tumour control and extending the survival of tumour-bearing mice (Fig. 5i and Extended Data Fig. 12j). Collectively, ZMYND8 deficiency in tumour-reactive CD8⁺ T cells acts cooperatively with IL-2 therapy, resulting in potent antitumour effects.

Next, we tested whether ZMYND8 deletion regulates ICB responses. In public datasets from patients with advanced basal cell carcinoma and squamous cell carcinoma³⁹, an activated CD8⁺ T cell population induced by anti-PD-1 treatment showed reduced ZMYND8-activated signature activity, as well as lower ZMYND8 expression, compared with the exhausted population (Fig. 5j and Extended Data Fig. 12k,l). The reciprocal pattern was observed for the ZMYND8-suppressed signature (Fig. 5j and Extended Data Fig. 12l). Moreover, ICB responders in colorectal cancer⁴⁰ and melanoma⁴¹ showed negative correlation with ZMYND8 activity or expression in intratumoural CD8⁺ T cells (Extended Data Fig. 12m–o). Therefore, ZMYND8 activity and expression are negatively associated with ICB responses in human CD8⁺ T cells from patients with cancer.

Increased IL-2 engagement in CD8⁺ T cells enhances the efficacy of ICB^23,24. In B16-gp33 tumours treated with anti-PD-L1, ZMYND8 deficiency still increased the frequencies of CD25⁺ cells among intratumoural T_ex populations (Fig. 5k and Extended Data Fig. 12p,q). ZMYND8-deficient P14 cells also exhibited increased p-STAT5 (Fig. 5l) and decreased TOX levels (Extended Data Fig. 12r) under both isotype and anti-PD-L1 treatment conditions. Moreover, whereas TOX expression was increased in intratumoural control P14 cells after cessation of anti-PD-L1 treatment, such upregulation was dampened by ZMYND8 deficiency (Extended Data Fig. 12s), suggesting that ZMYND8 deficiency partly overcomes maladaptive reprogramming of T_ex cells following ICB⁴². Accordingly, the combination of ZMYND8-deficient P14 cells with PD-L1 blockade improved tumour control, mouse survival and intratumoural P14 cell accumulation compared with either treatment alone (Fig. 5m,n and Extended Data Fig. 12t,u).

Similarly, during LCMV Cl13 infection, TOX expression was substantially suppressed upon ZMYND8 deletion under both isotype and anti-PD-L1 treatment conditions (Fig. 5o). Furthermore, mice receiving adoptive transfer of ZMYND8-deficient P14 cells plus anti-PD-L1 showed greater in vivo killing capacity (Fig. 5p), associated with reduced viral loads (Extended Data Fig. 12v). Collectively, ZMYND8 is a key driver of TOX expression and CD8⁺ T cell exhaustion in chronic infection and tumours and represents a therapeutic target to enhance protective immunity and boost the efficacies of IL-2 and ICB therapies (Extended Data Fig. 12w).

T cell exhaustion remains a major barrier for effective immunity against chronic infection and cancer^1,3, yet mechanisms governing the divergence of effector-like and Tex^term cell fate choices are not well understood. Here we identified that the chromatin reader ZMYND8 impeded the formation of effector-like states but promoted Tex^term cell generation, thereby limiting responses to immunotherapies. ZMYND8 deficiency improved cytolytic function of T_ex populations, including the Tex^term subpopulation whose functional reinvigoration remains challenging^1,3. Loss of ZMYND8 also enhanced IL-2 signalling but lowered TOX expression, suggesting the beneficial effects of targeting ZMYND8 to drive expansion of effector-like (especially Tex^KLR) cells and counteract maladaptive T_ex reprogramming. Elevated ZMYND8 expression and activity in T_ex cells were a shared feature across chronic infection and cancer. These results suggest that chronic TCR stimulation elicits ZMYND8-mediated feedback inhibition of IL-2 signalling, thereby highlighting an epigenetic rheostat imposing signal 1 (chronic antigen stimulation)-induced suppression of signal 3 cytokine signalling to enforce T cell exhaustion.

During chronic stimulation, T_ex cells receive suboptimal IL-2 signals and acquire a hypofunctional state², so strategies that engage high-affinity IL-2R signalling are likely to induce robust therapeutic effects. We found that ZMYND8 directly bound to Il2ra gene enhancers to suppress their transcriptional activation, revealing an epigenetic surveillance mechanism that accounts for the progressive loss of IL-2Rα expression and impaired IL-2R–STAT5 signalling in T cell exhaustion². Moreover, increased IL-2R signalling altered T_ex heterogeneous states and especially enhanced effector-like cell generation in the context of ZMYND8 deficiency. Given the dose-limiting toxicities of systemic IL-2 therapy², these findings have implications for emerging IL-2 and IL-2R engineering strategies to enhance IL-2 signalling selectively in therapeutic CD8⁺ T cells². Moreover, the observed negative association between ZMYND8 and ICB response suggests that targeting ZMYND8 may complement ICB-based immunotherapies by enhancing IL-2–IL-2Rα engagement²⁴. Nonetheless, the broader applicability of our findings beyond chronic LCMV infection and melanoma warrants further investigation.

Despite limited studies in immune cells, ZMYND8 is a chromatin reader that recognizes multiple histone modifications in tumour cells^29,43. We revealed that ZMYND8 and the histone acetyltransferase p300 selectively co-occupied active enhancers in IL-2 signalling genes, and ZMYND8 suppressed p300 activity and H3K27Ac modification at these gene loci (Extended Data Fig. 12w). Targeting ZMYND8 relieved this epigenetic ‘brake’ and increased IL-2Rα expression and downstream STAT5 activation, with p300 co-deletion reversing these effects. Therefore, ZMYND8 is a potent repressor of p300 activity that is required for IL-2R–STAT5 signalling, highlighting an epigenetic–cytokine signalling axis underlying CD8⁺ T cell hypofunctional adaptation. As p300 recruitment and chromatin engagement are facilitated by STAT5 in helper CD4⁺ T cells⁴⁴, ZMYND8-dependent suppression of p300-mediated H3K27Ac deposition may also require STAT5 in CD8⁺ T cells. Collectively, we have identified an epigenetic rheostat ZMYND8 mediating the reciprocal generation of Tex^KLR versus Tex^term cells and the marked downregulation of the high-affinity IL-2R subunit during T cell exhaustion. These findings uncover that epigenetic rewiring connects immune signals 1 and 3 to shape T_ex heterogeneity and function.

Mice

C57BL/6, P14 TCR-transgenic (Tg)⁴⁵, pmel TCR-Tg⁴⁶ and Rosa26-Cas9 knock-in⁴⁷ mice were purchased from the Jackson Laboratory. Human CD19 CAR-Tg mice were provided by T. Geiger³⁶. We crossed Rosa26-Cas9 knock-in mice with P14, pmel or human CD19 (hCD19) CAR-Tg mice to express Cas9 in antigen-specific CD8⁺ T cells. The Cas9-expressing mice were fully backcrossed to the C57BL/6 background. Sex-matched (male or female) mice were used at 7–12 weeks of age unless otherwise noted and assigned randomly to control and experimental groups. All mice were maintained in a specific pathogen-free facility in the Animal Resource Center at St. Jude Children’s Research Hospital. Mice were kept with 12 h–12 h light–dark cycles that coincide with daylight in Memphis, TN, USA. The St. Jude Children’s Research Hospital Animal Resource Center housing facility was maintained at 30–70 % humidity and 20–25 °C. Animal protocols were approved by and performed in accordance with the Institutional Animal Care and Use Committee (IACUC) of St. Jude Children’s Research Hospital.

Cell lines

The Plat-E cell line was provided by Y.-C. Liu and cultured in Dulbecco’s modified essential medium (DMEM) (Gibco) supplemented with 10% (v/v) FBS and 1% (v/v) penicillin–streptomycin. B16-gp33 cell line was provided by B. Youngblood and the B16-hCD19 cell line was generated as described³¹. B16F10, BHK (CL-10), Vero (CCL-81), Jurkat (E6-1) and HEK293T cell lines were purchased from the American Type Culture Collection (ATCC). All tumour cell lines were cultured in RPMI 1640 medium (Gibco) supplemented with 10% (v/v) FBS and 1% (v/v) penicillin–streptomycin. No commonly misidentified cell lines were used in this study (International Cell Line Authentication Committee). Cell lines used in this study were not independently authenticated or tested for Mycoplasma contamination.

Naive T cell isolation and viral transduction

Naive Cas9-expressing P14, pmel or hCD19 CAR-Tg cells were isolated from the spleen and peripheral lymph nodes of P14-Cas9, pmel-Cas9 and CAR-Tg-Cas9 mice using a naive CD8α⁺ T cell isolation kit (Miltenyi Biotec) according to the manufacturer’s instructions. Purified naive P14, pmel and hCD19 CAR-Tg T cells were activated in vitro for 20 h with 10 μg ml^–1 anti-CD3 (145-2C11; Bio-X-Cell) and 5 μg ml^–1 anti-CD28 (37.51; Bio-X-Cell) antibodies before viral transduction. Viral transduction was performed by spin-infection at 900g at 25 °C for 3 h with 10 μg ml^–1 polybrene (Sigma-Aldrich), followed by 2 h rest at 37 °C and 5% CO₂. After transduction, cells were cultured in T cell medium (Click’s medium (IrvineScientific) supplemented with 10% FBS, 55 μM 2-mercaptoethanol and 1× penicillin–streptomycin–l-glutamine (Gibco)) with recombinant human IL-2 (20 IU ml^–1; National Cancer Institute or PeproTech), mouse IL-7 (12.5 ng ml^–1; PeproTech) and IL-15 (25 ng ml^–1; PeproTech) for 3 days. Transduced cells were sorted based on the expression of Ametrine or GFP (as indicated in the figure legends) using a Reflection cell sorter (iCyt) before adoptive transfer into recipient mice. sgRNAs were designed using an online tool (https://portals.broadinstitute.org/gppx/crispick/public), and the sgRNAs used in this study are listed in Supplementary Table 11. Unless otherwise indicated in the text, the guides used for targeting Zmynd8, ZMYND8 and Ep300 in all experiments were sgZmynd8.g2, sgZMYND8.g1 and sgEp300.g2, respectively. The retroviral sgRNA vector was previously described^19,22. Retrovirus was produced by co-transfecting Plat-E cells with the core plasmid (sgRNA plasmid) and the packaging plasmid pCL-Eco (Addgene, #12371) and was collected 48 h after transfection. Co-deletion experiments were performed through double transduction of Cas9-expressing CD8⁺ T cells with Ametrine- and GFP-expressing retroviral vectors (Ametrine and GFP co-expression marked cells that had undergone successful dual transduction), followed by flow cytometry analysis.

Adoptive T cell transfer

For chronic LCMV infection, a total of 1 × 10⁴ retrovirus-transduced Cas9⁺ P14 CD8⁺ T cells were adoptively transferred intravenously into sex-matched Cas9-expressing recipient mice before infection to minimize the potential rejection. For tumour models, retrovirus-transduced Cas9⁺ P14, pmel or hCD19 CAR-Tg cells were adoptively transferred intravenously to the tumour-bearing mice on either day –1 (1 × 10⁶ cells per mouse) or day 7 (4 × 10⁶ per mouse) after tumour inoculation (as indicated in the figure legends). In the single-colour transfer system to assess antitumour effects, antigen-specific T cells transduced with sgNTC or the indicated sgRNAs (expressing the same fluorescent reporter protein) were transferred to separate hosts. In the dual-colour transfer system (for assessment of phenotypic alterations), cells transduced with the indicated sgRNAs marked by the expression of Ametrine, were mixed at a 1:1 ratio with those transduced with sgNTC labelled with GFP (spike), followed by adoptive transfer to the same host. To calculate fold changes in the dual-colour transfer system, the frequency of indicated population or geometric mean fluorescence intensity (gMFI) of indicated protein was calculated relative to spike (sgNTC) cells from the same host. Specifically, the proportion or gMFI of sgRNA-transduced cells was divided by the proportion or gMFI of spike cells and further normalized to the ratio of pre-transfer input samples. For the double knockout experiments, Cas9-expressing P14 cells co-transduced with sgNTC or sgZmynd8 (both GFP⁺)-expressing retrovirus together with sgNTC, sgStat5b, sgIl2ra, sgIl2rb or sgEp300 (all Ametrine⁺)-expressing retrovirus. Then, co-transduced Cas9⁺ P14 cells (all GFP⁺Ametrine⁺) were mixed at a 1:1 ratio with cells transduced with sgNTC (spike cells; Ametrine⁺) and co-transferred into the same Cas9⁺ mice, followed by LCMV Cl13 infection (dual-colour transfer system). The quantification of cell number was performed by calculating the numbers of indicated sgRNA-transduced cells and the sgNTC-transduced spike cells from the same LCMV Cl13-infected or tumour-bearing host, followed by normalization to the tumour weight in the tumour model.

LCMV Cl13 infection

LCMV Cl13 virus was grown in BHK-21 cells (ATCC, CCL-10), and viral titres were determined by plaque formation assay on Vero cells (ATCC, CCL-81). To induce chronic LCMV Cl13 infection, 2 × 10⁶ plaque-forming units (PFU) of LCMV Cl13 were injected intravenously post-adoptive transfer of P14 cells where noted. To quantify LCMV viral loads in serum and tissues of chronically infected mice, LCMV quantitative PCR (qPCR) assay was applied as previously described⁴⁸. In brief, 10 mg liver and kidney samples were homogenized in the presence of lysis buffer (Buffer RLT (QIAGEN) supplemented with 1% 2-mercaptoethanol), and total RNA was isolated using the RNeasy Mini Kit (QIAGEN) according to the manufacturer’s instructions. For serum samples, RNA was extracted from 100 μl serum using RNeasy Micro Kit (QIAGEN). Then, 15 μl RNA was used in a 20 μl cDNA reverse transcription reaction with High-Capacity cDNA Reverse Transcription Kit (Thermo) and the GP-R primer shown below. cDNA from liver and kidney samples was diluted 200-fold, and cDNA from serum samples was diluted 5-fold before adding to Power SYBR Green Master Mix (Applied Biosystems) containing LCMV-specific new GP primers: GP-R (S pos. 970-991), GCAACTGCTGTGTTCCCGAAAC; and GP-F (S pos. 877-901): CATTCACCTGGACTTTGTCAGACTC.

Amplification was assessed using QuantStudio7 Flex Real-Time PCR System (Applied Biosystems). The samples from uninfected mice were used as negative control. The viral load was calculated using a known LCMV Cl13 stock (quantified by plaque formation assay) as a standard.

Secondary adoptive cell transfer assays

Cas9-expressing P14 cells transduced with sgNTC (Ametrine⁺) or sgZmynd8 (GFP⁺) were mixed at a 1:1 ratio and co-transferred into C57BL/6 mice, followed by infection with LCMV Cl13 on the same day. At 10 dpi, Tex^prog (Ly108⁺TIM-3⁻CX3CR1⁻), effector-like (Ly108⁻TIM-3⁺CD101⁻) or Tex^int (Ly108⁻CX3CR1⁺KLRG1⁻) populations among control (sgNTC) and ZMYND8-deficient (sgZmynd8) P14 cells in the spleen were purified by sorting. LCMV Cl13 infection-matched mice received co-transfer of a 1:1 mixture of control and ZMYND8-deficient Tex^prog cells (1 × 10⁵ each), effector-like cells (2.5 × 10⁵ each) or Tex^int cells (2.5 × 10⁵ each). The differentiation of each T_ex subpopulation into various T_ex subpopulations (as indicated in the figures and their legends) was examined on day 9 after secondary transfer (that is, at 19 dpi) by flow cytometry analysis.

In vivo administration of biologics in LCMV Cl13-infected mice

For PD-L1 blockade, IgG isotype control antibody (200 μg per mouse; clone LTF-2, Bio-X-Cell) or anti-mouse PD-L1 antibody (200 μg per mouse; clone 10 F.9G2, Bio-X-Cell) was administered intraperitoneally into LCMV Cl13-infected mice at 13, 16, 19, 22 and 25 dpi. For IL-2 therapy, PBS or recombinant human IL-2 (rhIL-2) (1.5 × 10⁴ IU per mouse; National Cancer Institute) was injected intraperitoneally into LCMV Cl13-infected mice once daily from 30–35 dpi for a total of 6 daily injections.

Tumour models and immunotherapeutic treatments

In the single-colour transfer system for tumour therapy assays, 5 × 10⁵ B16-gp33 or B16F10 melanoma cells were subcutaneously injected into the right flank of C57BL/6 mice. On days 7–10 after tumour inoculation (as indicated in the figures or their legends), mice bearing tumours of a similar size were randomly divided into treatment groups (5–8 mice per group). Then, Cas9-expressing P14 (for the treatment of B16-gp33 melanoma) or pmel (for the treatment of B16F10 melanoma) CD8⁺ T cells transduced with sgNTC or the indicated sgRNAs (with the same fluorescent reporter protein) were adoptively transferred (4 × 10⁶ cells per mouse) individually to tumour-bearing mice. For anti-PD-L1 treatment, the B16-gp33 tumour-bearing mice were treated with two intraperitoneal injections of anti-PD-L1 (200 μg per mouse; clone 10 F.9G2, Bio-X-Cell) or IgG isotype control antibody (200 μg per mouse; clone LTF-2, Bio-X-Cell) on days 15 and 18 after tumour inoculation. For IL-2 treatment, tumour-bearing mice were randomly divided into two groups on day 10 after tumour inoculation and intraperitoneally injected with either PBS or rhIL-2 (1 × 10⁵ IU per mouse, dissolved in PBS) every day and 5 times in total. For the tumour model shown in Extended Data Fig. 12a–c, retrovirus-transduced Cas9⁺ P14 cells (1 × 10⁶ cells per mouse) were adoptively transferred intravenously to the C57BL/6 mice one day before subcutaneous inoculation of B16-gp33 tumour cells (1 × 10⁶ per mouse). Mice were monitored for tumour growth; tumours were measured every 2 days with digital callipers and tumour volumes were calculated using the following formula: length × width × width × π/6. For all tumour growth curves, the graphs depict the growth curves of individual tumours (dotted lines) and the average tumour volumes (bold lines). To isolate intratumoural lymphocytes, tumours were collected on the indicated days after inoculation, excised, minced and digested with 1 mg ml⁻¹ collagenase IV (LS004188, Worthington Biochemicals) and 200 U ml⁻¹ DNase I (DN25-1G, Sigma-Aldrich) for 1 h at 37 °C and passed through 70-μm filters to remove undigested tumour tissues. Tumour-infiltrating lymphocytes (TILs) from B16-gp33 or B16-hCD19 melanoma were further isolated by density-gradient centrifugation over Percoll (17089101, Cytiva) at 2,500 rpm for 20 min at room temperature. Tumour size limits were approved to reach a maximum of 3,000 mm³ or ≤20% of body weight (whichever was lower) by the IACUC of St. Jude Children’s Research Hospital.

Flow cytometry

For analysis of surface markers, cells were stained in PBS (Gibco) containing 2% FBS. Surface proteins were stained for 30 min at 4 °C. For transcription factor staining, cells were stained for surface molecules and then fixed with 2% paraformaldehyde (Thermo Fisher Scientific) for 30 min at room temperature, followed by permeabilization with FOXP3/transcription factor staining buffer set, according to the manufacturer’s instructions (00-5523-00, eBioscience). Intracellular staining for cytokines was performed using BD CytoFix/CytoPerm fixation/permeabilization kit (554774, BD Biosciences) after stimulation with phorbol 12-myristate 13-acetate (PMA; Sigma-Aldrich) and ionomycin (Sigma-Aldrich) in the presence of monensin (GolgiStop, 554724, BD Bioscience) for 4 h. 7-AAD (A9400, 1:200, Sigma-Aldrich) or fixable viability dye (65-0865-14; 1:2,000, eBioscience) was used for dead-cell exclusion. The following antibodies from eBioscience were used: APC–anti-CD25 (PC61.5, 17-0251-82, 1:200); APC–anti-NKG2D (1D11, 17-5878-42, 1:200); APC–anti-perforin (eBioOMAK-D, 17-9392-80, 1:200); PE–anti-TOX (TXRX10, 12-6502-82, 1:100); PE–anti-CD101 (Moushi101, 12-1011-82, 1:200); PE/Cyanine7–anti-NKG2D (CX5, 25-5882-82, 1:200); PE/Cyanine7–anti-TIM-3 (RMT3-23, 25-5870-82, 1:400); PE/Cyanine7–anti-CD101 (Moushi101, 25-1011-82, 1:200); PE/Cyanine7–anti-KLRG1 (13F12F2, 25-9488-42, 1:200); PerCP-eFluor 710–anti-CD39 (24DMS1, 46-0391-82, 1:400); PerCP-eFluor 710–anti-GZMA (GzA-3G8.5, 46-5831-82, 1:200). The following antibodies from BioLegend were used: PE–anti-CD122 (5H4, 105906, 1:200); Alexa Fluor 700–anti-CD8α (53-6.7, 100730, 1:400); Brilliant Violet 785–anti-TCRβ (H57-597, 109249, 1:400); APC–anti-Ly108 (330-AJ, 134610, 1:400); APC–anti-CD122 (5H4, 105912, 1:200); Brilliant Violet 421–anti-TNF (MP6-XT22, 506328, 1:200); Brilliant Violet 421–anti-mouse/human KLRG1 (2F1/KLRG1, 138414, 1:200); APC/Cyanine7–anti-NK1.1 (PK136, 108724, 1:200); Brilliant Violet 711–anti-CD8 (53-6.7, 100748, 1:400); Brilliant Violet 711–anti-CD366 (TIM-3) (RMT3-23, 119727, 1:400); Brilliant Violet 421–anti-CX3CR1 (SA011F11, 149023, 1:400); Brilliant Violet 650–anti-CX3CR1 (SA011F11, 149033, 1:400); PE/Cyanine7–anti-mouse Ki-67 (16A8, 652426, 1:200); Alexa Fluor 647–anti-human/mouse GZMB (GB11, 515405, 1:100); PE–anti-IFNγ (XMG1.2, 505808, 1:200). FITC–anti-human CD279 (PD-1) (A17188B, 621612, 1:200); Alexa Fluor 700–anti-human CD8 (SK1, 344724, 1:200); PerCP/Cyanine5.5–anti-human IFNγ (B27, 506528, 1:200); Brilliant Violet 711–anti-human CD366 (TIM-3) (F38-2E2, 345024, 1:400); Brilliant Violet 605–anti-human CD25 (BC96, 302632, 1:200); PE/Cyanine7–anti-human CD122 (IL-2Rβ) (TU27, 339014, 1:200); PE–anti-IFNγ (XMG1.2, 505808, 1:200). The following antibodies from BD Biosciences were used: PE–anti-STAT5 (pY694) (47, 612567, 1:20); PE–anti-NKG2A (16A11, 568952, 1:200); Brilliant Violet 605–anti-Ly108 (13G3, 745250, 1:400); Brilliant Violet 711–anti-CD94 (18d3, 740760, 1:400); Brilliant Violet 421–anti-active caspase-3 (C92-605.rMAb, 570863, 1:200); PerCP/Cyanine5.5–anti-human Ki-67 (B56, 561284, 1:200); Alexa Fluor 488–anti-human TNF (Infliximab297.rMAb, 570954, 1:100). Alexa Fluor 700–anti-TIM-3 (FAB1529RN, 1:400) was from R&D Systems, and APC–anti-human/mouse TOX (REA473, 130-118-335, 1:200) was from Miltenyi Biotec.

For p-STAT5 staining, splenocytes from LCMV Cl13-infected mice or TILs from B16-gp33 or B16-hCD19 melanoma were restimulated with 20 IU ml⁻¹ rhIL-2 at 37 °C for 30 min in T cell medium containing antibodies for surface molecules, followed by fixation with 2% paraformaldehyde for 30 min at room temperature and permeabilization using 90% ice-cold methanol for 30 min on ice. Then, cells were stained for anti-STAT5 (pY694) (1:20) for 30 min at room temperature. Flow cytometry data were acquired using BD FACSDiva software (v8.0.1) on Fortessa or Symphony A3 cytometers (BD Biosciences) and were analysed using FlowJo software (v10.10.0; TreeStar). Representative gating strategies for all flow cytometry results are provided under Supplementary Fig. 1.

For in vivo labelling of circulatory or tissue-resident P14 cells in the spleen, PE–anti-CD8α (S18018E; 2 μg) prepared in PBS was injected intravascularly into LCMV Cl13-infected mice containing control and ZMYND8-deficient P14 cells at 21 dpi, and the recipient mice were euthanized 5 min after injection as previously described¹¹. Splenocytes were prepared for flow cytometry analysis with surface staining using Alexa Fluor 700–anti-CD8α antibody (53-6.7, 1:400). The proportions of control and ZMYND8-deficient total P14 cells and their T_ex subpopulations in the white pulp (PE^–) and red pulp (PE⁺) were analysed by flow cytometry.

Naive CD8⁺ T cell RNP electroporation

For CRISPR–Cas9 editing of naive mouse CD8⁺ T cells, RNP complexes were prepared by mixing 2 µl of 100 µM sgRNA (Synthego) and 2 µl of 40 µM Cas9 (Alt-R S.p. Cas9 Nuclease), and incubated for 10 min at room temperature. Electroporation was performed on 1–3 × 10⁶ naive CD8⁺ T cells using the 4D-Nucleofector X Unit (with program DN100) and P3 primary cell 4D-Nucleofector X Kit (V4XP-3032, Lonza). Pre-warmed complete T cell medium (100 μl) was immediately added to cells, followed by incubation at 37 °C for 10 min to allow cells to recover. Cells were washed extensively before transfer into recipient mice.

In vitro acute and chronic stimulation of human CD8⁺ T cells

Peripheral blood mononuclear cells were isolated from consented healthy donors by density-gradient separation using Lymphoprep (GE Healthcare). Naive CD8⁺ T cells (CCR7⁺CD45RO^–) were sorted and activated with ImmunoCult Human CD3/CD28 T Cell Activator (STEMCELL Technologies) at a density of 1 × 10⁶ cells ml^–1 in X-VIVO 15 medium (Lonza) supplemented with 5% human serum (Sigma-Aldrich), 50 µM 2-mercaptoethanol and 1× penicillin–streptomycin–l-glutamine (Gibco). Two days after activation, RNP-Cas9 complexes were prepared as described above, and 1–2 × 10⁶ activated CD8⁺ T cells were electroporated using the 4D-Nucleofector X Unit (using program EH-115) and P3 primary cell 4D-Nucleofector X Kit (Lonza). Pre-warmed X-VIVO 15 medium (100 μl) was immediately added to the cells, followed by incubation at 37 °C for 15 min to allow cells to recover. Cells were then transferred to a 48-well plate (Corning) and cultured for 2 additional days in complete X-VIVO 15 medium supplemented with 50 IU ml⁻¹ rhIL-2. Afterwards, for acute stimulation, edited human CD8⁺ T cells were maintained under the above culture conditions, with fresh medium and cytokines replaced every 2 days. For chronic stimulation, edited human CD8⁺ T cells were further activated with 2 μg ml^–1 plate-bound anti-human CD3 antibody (OKT3, BioLegend), and cells were replated onto fresh anti-human CD3 antibody-coated plates and replaced with fresh X-VIVO 15 medium containing 50 IU ml^–1 rhIL-2 every 2 days. Cells were collected and analysed by flow cytometry and immunoblot analyses as indicated in the figure legends. Intracellular staining for cytokines (for example, IFNγ) or GZMB in human CD8⁺ T cells cultured under both acute and chronic stimulation conditions was performed after stimulation with ionomycin and phorbol 12-myristate 13-acetate (PMA) in the presence of GolgiStop (monensin) for 4 h.

Fresh human blood leukopaks from healthy donors were obtained from the Blood Donor Center at St. Jude Children’s Research Hospital. Ethical oversight for studies involving human specimens was provided by the Institutional Review Board (IRB) of St. Jude Children’s Research Hospital. As all leukapheresis products were de-identified, the isolation of naive CD8⁺ T cells and related experiments did not constitute human-subjects research.

Ex vivo and in vivo killing assays

For ex vivo killing assay, splenocytes from naive CD45.2⁺ C57BL/6 mice were pulsed with 0.2 µM gp_33–41 peptide (target) or the irrelevant OVA_257–264 SIINFEKL peptide (non-target) at 37 °C for 1 h. Target and non-target cells were labelled with 0.5 μM Cell Trace Violet (CTV; Thermo Fisher Scientific) or 5 μM CTV, respectively, per the manufacturer’s instructions. Target (CTV^low) and non-target (CTV^hi) cells were mixed at a 1:1 ratio and incubated with or without sgNTC and sgZmynd8-expressing Tex^prog (Ly108⁺CX3CR1^–), effector-like (Ly108^–CX3CR1⁺) and Tex^term (Ly108^–CX3CR1^–) subpopulations (sorted from the spleen of the same host at 21 dpi (from the dual-colour transfer system)) at a 1:3 effector-to-target ratio for 16 h in T cell medium as described¹¹. Percentage of specific lysis was calculated as follows: specific lysis (%) = 100 – (100 × percentage of gp33-pulsed targets remaining after co-culture with effectors)/(percentage of gp33-pulsed targets after culture without effectors).

In vivo killing assays were performed as previously described²². In brief, CD45.2⁺ splenocytes were pulsed with 0.2 µM gp_33–41 peptide or PBS at 37 °C for 1 h. These antigen- or PBS-pulsed splenocytes were then labelled with 0.5 μM CTV (CTV^low) or 5 μM CTV (CTV^hi), respectively, at 37 °C for 15 min. At 21 (for Fig. 2h), 36 (for Fig. 2j) or 28 (for Fig. 5p) dpi, the CTV^low and CTV^hi splenocytes were mixed at a 1:1 ratio, and a total of 2 × 10⁷ cells were adoptively transferred to LCMV Cl13-infected mice that received sgNTC or sgZmynd8 (Ametrine⁺)-expressing cells (single-colour transfer system), followed by analysis of in vivo cytotoxicity against these splenocytes after 3 h.

Plasmid generation

To generate ZMYND8-V5 truncation constructs, the wild-type ZMYND8-V5 plasmid (Addgene #65401) was used as the template. All truncations were generated using the Q5 Site-Directed Mutagenesis Kit (E0554S, NEB) according to the manufacturer’s instructions. The primers used for each truncation were as follows: ΔPHD-F: ATTCCGTCCATCCTG, ΔPHD-R: ACAGTAGCAGAATGCATC; ΔBRD-F: TTTACTCTGGGTCTCG, ΔBRD-R: GAAGTATGTCCAGAATGTTATC; ΔPWWP-F: ATTGCTACAAGGCTCAC, ΔPWWP-R: ATTCCTTTTTCTGTGAAAAAG; ΔPHD–BRD–PWWP-F: ATTCCGTCCATCCTG, ΔPBP-R: ATTCCTTTTTCTGTGAAAAAG; ΔMYND-F: CCACTGCTTCTTCTTG, ΔMYND-R: ACCCAGTCAGCTACTG.

Immunoprecipitation and immunoblot analyses

For immunoprecipitation in Extended Data Fig. 9m, 2 × 10⁷ activated CD8⁺ T cells were lysed in ice-cold Pierce IP Lysis Buffer (87787, Thermo Fisher Scientific) supplemented with protease and phosphatase inhibitor cocktail (78442, Thermo Fisher Scientific). The cell lysates were then incubated with 2 µg of anti-ZMYND8 (11633-1-AP, Proteintech) or normal rabbit IgG (2729, Cell Signalling Technology) antibody, with rotation overnight at 4 °C, followed by incubation with Protein A/G magnetic beads (88802, Thermo Fisher Scientific) for 3 h. The beads were washed 3 times with ice-cold IP lysis buffer, resuspended with 1× Laemmli sample buffer (1610747, Bio-Rad), and then boiled at 95 °C for 10 min, followed by immunoblot analysis.

For immunoprecipitation in Extended Data Fig. 9p, HEK293T cells (4 × 10⁵) were seeded in each well of a 6-well plate, and plasmids were transfected when the cells reached approximately 80% confluency. In brief, wild-type or truncated ZMYND8-V5 constructs, together with 5 μg pcDNA3.1-p300-6×His, were mixed with TransIT-293 (MIR 2706, Mirus Bio) in 200 μl Opti-MEM medium (31985, Gibco) for each well. The total amount of each ZMYND8-V5 truncation construct was optimized to achieve comparable expression (0.5 μg for wild-type and ΔMYND, 0.6 μg for ΔPHD, ΔBRD and ΔPWWP, and 0.8 μg for ΔPHD–BRD–PWWP). At 36 h after transfection, HEK293T cells were lysed in ice-cold Pierce IP Lysis Buffer containing protease and phosphatase inhibitor cocktail with rotation at 4 °C for 30 min. The cell lysate was centrifuged at 13,000g for 10 min at 4 °C, and the supernatant was incubated with anti-V5 magnetic beads (SAE0203, Sigma-Aldrich) at 4 °C for 2 h. Beads were washed 3 times with ice-cold IP Lysis Buffer and resuspended with 1× complete Laemmli sample buffer. All the protein samples were boiled at 95 °C for 10 min, and then separated for immunoblot analysis.

For immunoblot analysis, cells were lysed in RIPA buffer (Thermo Fisher Scientific) supplemented with protease and phosphatase inhibitor cocktail. The protein concentrations of cell lysates were determined by BCA protein assay kit (Thermo Fisher Scientific). Equivalent amounts of total protein were denatured by adding Laemmli protein sample buffer (Bio-Rad) and boiling at 95 °C for 5 min, and then separated using 4–12% Criterion XT Bis-Tris protein gel (3450125, Bio-Rad) and transferred to polyvinylidene fluoride (PVDF) membranes (1620177, Bio-Rad). The membranes were blocked using 3% Bovine Serum Albumin (BSA) (A9418, Sigma-Aldrich) for 1 h at room temperature and then incubated overnight with primary antibodies at 4 °C. The membranes were washed three times with Tris-buffered saline containing 0.1% Tween-20 (TBST) and then incubated with secondary antibodies for 2 h at room temperature. After washing three times with TBST, HRP was activated with SuperSignal West Dura Extended Duration Substrate (34075, Thermo Fisher Scientific) and visualized with chemiluminscent detection system using Amersham Imager 600 (GE Healthcare Life Sciences). The blots were then processed and quantified using Image J software.

Primary antibodies and dilutions were as follows: anti-ZMYND8 (1:1,000, 11633-1-AP) and anti-V5 tag (1:3,000, 14440-1-AP) were from Proteintech; anti-6× His tag (1:1,000, ab18184) was from Abcam; anti-phospho-STAT5 (Y694) (1:1,000, 9351S), anti-STAT5 (1:1,000, 94205S), anti-STAT5B (1:1,000, 34662S), anti-IL-2Rα/CD25 (1:1,000, 36128S), anti-p300 (1:1,000, 57625S), anti-GAPDH (1:5,000, 2118S) and anti-β-Actin (called ACTB;1:5,000, 4970) were from Cell Signalling Technology. Secondary antibodies and dilutions were as follows: HRP-conjugated anti-rabbit IgG (H + L) (1:3,000, W4011) and HRP-conjugated anti-mouse IgG (H + L) (1:3,000; W4021) were from Promega, and HRP-conjugated anti-rabbit IgG, light chain specific (1:3,000; SA00001-7L) was from Proteintech. Densitometric quantification of phosphorylated protein levels was normalized relative to the corresponding total protein, and densitometric quantification of total protein expression was normalized relative to the loading control ACTB. All densitometric quantifications depict the fold changes compared with the relative control (set equal to 1.0) and are shown below the immunoblot image.

RNA isolation and gene expression profiling

Cells were lysed with Buffer RLT in the RNeasy Mini Kit (QIAGEN), and total RNA was extracted according to the manufacturer’s instructions. Then, 1 μg total RNA was reverse transcribed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). The cDNA was then diluted and added to Power SYBR Green Master Mix (Applied Biosystems) containing gene-specific PCR primers. Gene amplification was assessed using QuantStudio7 Flex Real-Time PCR System (Applied Biosystems). Actb was used as the housekeeping control. A list of the primers is provided in Supplementary Table 12.

Luciferase assay

pGL4.23-IL2RA CaRE4 scramble (#91852), pGL4.23-IL2RA CaRE4 (#91850), GFP-ZMYND8 (#65401) and pcDNA3.1-p300 (#23252) were obtained from Addgene, and pTK-Green Renilla Luc Vector (16154) was from Thermo Fisher Scientific. Among them, pGL4.23-IL2RA CaRE4 scramble and pGL4.23-IL2RA CaRE4 plasmids contain a scrambled sequence and the IL2RA enhancer CaRE4 sequence, respectively, both upstream of a generic minimal promoter²⁸. For the experiments shown in Fig. 4f and Extended Data Fig. 9o, each Firefly luciferase (Luc) IL2RA CaRE4 or scramble construct (500 ng) and Renilla Luc plasmid (70 ng) were electroporated with or without p300-expressing plasmid (500 ng) alone or together with wild-type or indicated mutant ZMYND8-expressing plasmid (500 ng) into 5 × 10⁵ Jurkat cells using the 4-D Nucleofector (with program CL-120) and 20 μl Nucleofection Buffer SE (V4XC-1032, Lonza). Jurkat cells were rested for 18 h and then split into either an untreated control plate or stimulation plate pre-coated with anti-human CD3 (clone OKT3, 10 μg ml⁻¹, BioLegend) and anti-human CD28 (clone CD28.2, 10 μg ml⁻¹, BioLegend) antibodies. Luciferase expression was assessed using the Dual-Glo Luciferase Assay (E2920, Promega) on a 96-well plate luminometer after 24 h of stimulation. Firefly Luc activity was normalized to Renilla Luc activity for each well and was reported as fold induction over pGL4.23-IL2RA scramble vector under unstimulated condition.

CUT&RUN assay

Sample preparation and sequencing

Total Cas9⁺ P14 cells transduced with sgNTC (Ametrine⁺) or sgZmynd8 (GFP⁺) were sorted from the spleen of LCMV Cl13-infected recipient mice at 7 dpi (due to abundant expression of IL-2R at this time point (Extended Data Fig. 1k,l)), or 3 T_ex subpopulations (Ly108⁺CX3CR1⁻, Ly108⁻CX3CR1⁺ and Ly108⁻CX3CR1⁻) among control (sgNTC-expressing) P14 cells were sorted at 21 dpi. A total of 0.5–1 × 10⁵ cells were used for CUT&RUN using the CUTANA ChIC/CUT&RUN Kit (EpiCypher) according to the manufacturer’s instructions. The antibodies (0.5 μg per reaction) used for CUT&RUN were as follows: anti-ZMYND8 (11633-1-AP) was from Proteintech; Rabbit IgG negative control antibody (13-0042) and anti-H3K4me3 (13-0041) were from EpiCypher; anti-H3K4me1 (ab176877), anti-H3K36me2 (ab9049), anti-H3K14Ac (ab52946), anti-H3K27Ac (ab4729) and anti-H4K16Ac (ab109463) were from Abcam. anti-p300 (57625S) was from Cell Signalling Technology. Here is the detailed information about the biological replicates for CUT&RUN samples: for Fig. 4d and Extended Data Fig. 9a, IgG (n = 2), anti-ZMYND8 (n = 3) and anti-p300 (n = 2); for Fig. 4c and Extended Data Fig. 9d (n = 2 for all histone modification samples). Representative genome tracks from one biological replicate are shown; for Fig. 4e and Extended Data Fig. 9j, n = 2 for each antibody (anti-H3K27Ac, anti-ZMYND8 and anti-p300) in either sgNTC or sgZmynd8-expressing cells. Representative genome tracks from one biological replicate are shown; for Extended Data Fig. 9e, Tex^prog (n = 1), effector-like (n = 2) and Tex^term (n = 2). CUT&RUN-enriched DNA from H3K27Ac, ZMYND8 or p300 CUT&RUN assays was quantified by qPCR in Extended Data Fig. 9k,l using specific primers, as previously described⁴⁹ (Supplementary Table 12). The Actb gene sequence was used as control genomic locus. The CUT&RUN–qPCR data was normalized and analysed using the comparative C_T method (also known as 2^–ΔΔCT method), as previously described⁴⁹. Primer sequences used in CUT&RUN–qPCR are provided in Supplementary Table 12. Alternatively, CUT&RUN libraries were prepared using CUTANA CUT&RUN Library Prep Kit (EpiCypher) according to the manufacturer’s protocol. Barcoded libraries were quantified on an Agilent TapeStation. The final libraries were sequenced on the Illumina NovaSeq platform using paired-end (2× 50 bp) with target reads of 20 million per sample. Representative in-house generated raw and processed CUT&RUN assay data shown in Fig. 4c–e and Extended Data Fig. 9a,d,e,j have been deposited into the GEO database with the Superseries identifier GSE342764.

Data analysis

CUT&RUN raw reads were processed using nf-core/cutandrun pipeline⁵⁰. The peaks for the histone marks, ZMYND8 and p300 binding peaks were called by seacr⁵¹ (for histone marks) or MACS2⁵² (for ZMYND8 and p300) algorithms. Genome annotation of binding sites was performed by plotAnnoBar function in ChIPseeker R package⁵³ (v1.40.0). The numbers of intersected peak regions were calculated by enrichPeakOverlap function in ChIPseeker. ZMYND8 binding sites shown in Extended Data Fig. 9a were identified in at least two of three CUT&RUN samples using anti-ZMYND8 in P14 cells compared with isotype control. In addition, a two-tailed Fisher’s exact test was used to assess whether an MSigDB gene set was significantly enriched (P < 0.05) among genes nearest to ZMYND8 binding sites. Tracks were visualized using Integrative Genomics Viewer (IGV, v2.4.13). For Cistrome transcriptional regulator binding or histone mark analysis⁵⁴, ZMYND8-bound regions were uploaded to http://dbtoolkit.cistrome.org/ to analyse transcriptional regulators (including transcription factors and chromatin regulators) or histone marks with significant binding overlaps using default parameters. For comparing H3K27Ac level in Tex^prog, effector-like and Tex^term cells, consensus peak regions were generated by concatenating the H3K27Ac peak files from all samples, sorting the resulting genomic intervals, and merging overlapping or adjacent peaks using bedtools software (v2.25.0). Read counts over the consensus peak regions were then quantified using bedtools ‘multicov’ command, which calculated the number of sequencing reads overlapping each peak from the aligned BAM file of each sample. The resulting peak-by-sample count matrix was used for downstream quantitative and differential analysis using DESeq2 R package⁵⁵ (v1.32.0). Normalized average H3K27Ac read counts at the ZMYND8-bound active enhancer region of the Il2ra locus were visualized by ComplexHeatmap R package (v2.20.0). The code for analysis of CUT&RUN data in the relevant figure panels has been deposited to Zenodo (https://doi.org/10.5281/zenodo.21924005 (ref. ⁵⁶)).

scRNA-seq

Sample preparation

Cas9-expressing P14 cells transduced with either sgNTC (Ametrine⁺) or sgZmynd8 (GFP⁺) were mixed at a 1:1 ratio and then injected intravenously into the same Cas9⁺ recipient mice, followed by LCMV Cl13 infection (n = 2 per group). sgNTC and sgZmynd8-transduced Cas9⁺ P14 cells were sorted from the same host at 21 dpi. After cell counting and centrifugation at 2,000 rpm for 5 min, the supernatant was removed, and cells were resuspended and diluted in 1× PBS containing 0.04% BSA at concentration of 1 × 10⁶ cells per ml. Single-cell libraries were prepared using a Chromium Single Cell 3′ Library and Gel Bead kit (v3.1; 10X Genomics). In brief, the single-cell suspensions were loaded onto the Chromium Controller according to their respective cell counts to generate 9,000 single-cell gel beads in emulsion per sample. Each sample was loaded into a separate channel. The cDNA content of each sample after cDNA amplification of 12 cycles was quantified and quality checked using a High Sensitivity D5000 chip in a TapeStation (Agilent Technologies) to determine the number of PCR amplification cycles to produce a sufficient library for sequencing. After library quantification and quality checking using a D5000 chip (Agilent Technologies), samples were diluted to 3.5 nM for loading onto a HiSeq 4000 (Illumina) with a 2× 100-bp paired-end kit using the following cycles: 28 cycles read 1, 10 cycles i7 index, 10 cycles i5 index and 90 cycles read 2. An average of 300 million reads per sample was obtained (approximately 20,000 reads per cell). In-house generated raw and processed scRNA-seq data have been in the NCBI GEO database and are accessible through the GEO SuperSeries access number GSE342764.

Data analysis

FASTQ files were processed using Cell Ranger⁵⁷ (v7.1.0) using the Cell Ranger count command with default settings. Gene expression reads were aligned to the mouse genome (mm10 from ENSEMBL GRCm38 loaded from 10X Genomics). Underlying cell variations derived in P14 cells in the single-cell gene expression data were visualized using a two-dimensional projection by UMAP with the Seurat R package⁵⁸ (v5.1.0). sgNTC and sgZmynd8-transduced Cas9⁺ P14 cells were further clustered and annotated as four cellular states (Tex^prog, Tex^int, Tex^KLR and Tex^term): Tcf7⁺Slamf6⁺ (Tex^prog), Cx3cr1⁺Mki67⁺Klrg1^low (Tex^int), Cx3cr1⁺Klrg1^hiKlre1^hiKlrc1^hiKlrd1^hi (Tex^KLR) and Cx3cr1^–Cd101⁺ (Tex^term). Violin and dot plots that depict the expression levels of selective genes were generated using the VlnPlot and DotPlot function, respectively, in the Seurat R package. Activity scores of gene signatures curated from public gene sets or data generated in-house (see below for details) were calculated using the AddModuleScore function in Seurat. Ahmed CXCR5⁺ stem-like CD8⁺ T cell⁵⁹ and Ahmed CXCR5^– terminally differentiated CD8⁺ T cell⁵⁹ signatures were curated by comparing CXCR5⁺CD8⁺ T cells with TIM-3⁺CD8⁺ T cells from LCMV Cl13-infected mice (GSE84105); the gene signatures were generated using the top 200 upregulated genes and top 200 downregulated genes (ranked by log₂FCs), respectively. The Kaech effector and memory signatures⁶⁰ were curated by comparing IL-7R^low with IL-7R^hi effector CD8⁺ T cells from microarray analysis (GSE8678); the Kaech effector gene signature was generated using the top 200 upregulated genes (ranked by log₂FCs) followed by removal of probes without gene names, while the Kaech memory gene signature was generated using the top 200 downregulated genes (ranked by log₂FCs) followed by removal of probes without gene names. The Regev cell cycle signature was previously described⁶¹. The core tissue-resident memory T cell signature was previously described⁶². The STAT5CA upregulated (Up) and downregulated (Down) gene signatures were curated from a public scRNA-seq dataset of STAT5CA-expressing versus control CD8⁺ T cells (GSE214116)²¹ using genes with log₂FC (STAT5CA versus control) >1 (Up) or <–1 (Down) and Bonferroni corrected P < 0.05. The in-house ZMYND8-activated signature was generated using downregulated genes (log₂FC <–0.5; Bonferroni corrected P < 0.05) in sgZmynd8-expressing Havcr2 (encodes TIM-3)⁺ P14 cells versus sgNTC-expressing Havcr2⁺ P14 cells from scRNA-seq profiling at 21 dpi. The ZMYND8-suppressed signature was generated using upregulated genes (log₂FC > 0.5; Bonferroni corrected P < 0.05) in sgZmynd8-expressing Havcr2⁺ P14 cells versus sgNTC-expressing Havcr2⁺ P14 cells from scRNA-seq profiling at 21 dpi. Epigenetic landscape in silico deletion analysis (LISA)⁶³ (https://lisa.cistrome.org/) was performed using the top 250 upregulated genes (ranked by log₂FC) in sgZmynd8-expressing compared with sgNTC-expressing total P14 cells to infer the transcription factors with upregulated activity upon ZMYND8 deficiency. Pseudotime trajectory analysis was performed to infer the differentiation trajectory of T_ex cells in sgNTC and sgZmynd8-expressing P14 cells. The analysis was conducted (using the default parameters in the Monocle3 R package⁶⁴ (v1.3.7)) on the four T_ex states, with the Tex^prog state set as the starting point. Density plots were used to compare pseudotime distribution between genotypes or T_ex differentiation states. In Extended Data Fig. 3f, the region between the dashed lines denotes the interval in which sgZmynd8-expressing P14 cells accumulated most prominently relative to sgNTC controls. Consistent with the inferred trajectory, sgZmynd8-expressing cells were enriched primarily in the Tex^KLR state, and to a lesser extent the Tex^int state. The code for analysis of scRNA-seq data in the relevant figure panels has been deposited to Zenodo (https://doi.org/10.5281/zenodo.21924005 (ref. ⁵⁶)).

Pre-ranked GSEA

For scRNA-seq analysis, to compare gene expression between genotypes (sgZmynd8-expressing compared with sgNTC-expressing), a non-parametric two-tailed Wilcoxon rank-sum test was applied, and genes were ranked by log₂FC values. GSEA (v4.2.3) analysis of the STAT5CA Up signature in sgZmynd8-expressing P14 compared with sgNTC cells was used to derive leading-edge genes⁶⁵, and those showing significant upregulation (log₂FC > 0.5, adjusted P < 0.05) in sgZmynd8-expressing cells are shown in the heatmap. For pathway enrichment analysis, pre-ranked GSEA (an analysis of GSEA against a user-supplied, ranked list of genes) was then performed using the MSigDB collection (v7.4) for each comparison⁶⁵.

Public dataset analysis

Multiple public scRNA-seq datasets from the GEO databases were re-analysed to examine the correlation of Zmynd8 (mouse) or ZMYND8 (human) expression and ZMYND8-activated and ZMYND8-suppressed activity scores (described above) with CD8⁺ T cell exhaustion. Seurat R package (v5.1.0) was used for preprocessing and visualization, similar to that used for the in-house generated data. Zmynd8 expression in endogenous H2D^b-gp33-specific CD8⁺ T cells from early and late stages of acute and chronic infection⁶⁶ (Fig. 5a) and T_ex states from chronic infection¹⁵ (GSE122712; Extended Data Fig. 11c) were visualized using the DotPlot function in Seurat. ZMYND8-activated and ZMYND8-suppressed activity scores in T_ex states from chronic infection¹⁵ (GSE122712) were visualized using the VlnPlot function in Seurat. Zmynd8 expression in OT-I cells from B16-OVA melanoma³¹ (GSE216909) was visualized using the DotPlot function in Seurat. ZMYND8-activated and ZMYND8-suppressed activity scores in T_ex states from B16-OVA melanoma³¹ (GSE216909) were visualized using the VlnPlot function in Seurat. For scRNA-seq datasets from a genetically engineered mouse model (GEMM) for lung cancer³² (GSE164177), ZMYND8-activated and ZMYND8-suppressed activity scores in CD8⁺ T cells were visualized using the VlnPlot function in Seurat. For the RNA-seq dataset from a GEMM for liver cancer³³ (GSE89307), the relative expression of Tcf7, Pdcd1, Tox and the average expression of ZMYND8-activated or ZMYND8-suppressed genes of tumour-specific CD8⁺ T cells were visualized using the Heatmap function in the ComplexHeatmap R package (v2.20.0). For the dataset from human patients infected with HIV³⁰ (GSE157829), ZMYND8 expression and ZMYND8-activated and ZMYND8-suppressed activity scores were examined in CD8⁺ T cells from the peripheral blood of healthy donor and HIV-infected individuals with low and high viral titre (as previously defined³⁰). In the human melanoma dataset³⁴ (GSE123139), the naive-like, transitional and dysfunctional CD8⁺ T cell annotations were based on the reported markers in each subpopulation³¹. ZMYND8 expression and the activity scores for ZMYND8-activated and ZMYND8-suppressed signatures were examined in these subpopulations.

To assess the correlation of the responsiveness to ICB with ZMYND8 activity in CD8⁺ T cells from human melanoma⁴¹ (GSE120575), the Kendall rank order correlation between ZMYND8 expression or the average expression of ZMYND8-activated genes and other reported markers that are positively associated (IL7R, SELL, TCF7) or negatively associated (HOPX, LGALS1, VCAM1, SNAP47, CCL3, FASLG, MT2A, EPSTI1, GBP1, PSMB2, NDUFB3, CD38, GBP4, WARS, PRDX3) with ICB response was performed in CD8⁺ T cells in patients with melanoma treated with ICB. For other skin cancers, such as advanced basal cell carcinoma (BCC)³⁹ (GSE123813) and squamous cell carcinoma (SCC)³⁹ (GSE123813), ZMYND8 expression and the activity scores of ZMYND8-activated and ZMYND8-suppressed signatures were compared using violin plots in the CD8⁺ T cell clusters (originally annotated in the literature³⁹) after ICB treatment. For colorectal cancer (GSE205506)⁴⁰, the ZMYND8-activated and ZMYND8-suppressed activity scores were compared in intratumoural CD8⁺ T cells from patients who achieved pathological complete response (pCR) or non-pCR after ICB treatment, as well as untreated patients.

To assess the correlation of the responsiveness to IL-2 therapy with ZMYND8 activity in CD8⁺ T cells during chronic infection, the activity scores of the ZMYND8-activated and ZMYND8-suppressed signatures were compared in P14 cells from LCMV Cl13-infected mice treated with or without IL-2 (GSE206732)²⁴. For tumour models, the activity scores of ZMYND8-activated and ZMYND8-suppressed signatures were compared in CD8⁺ T cells from B16F10-B2m^–/– melanoma treated with vehicle and mRNA-encoded IL-2³⁷ or from T3 sarcoma tumours at day 2 after treatment with vehicle or IL-2 (GSE252650)³⁸.

Single-cell multiome profiling

For multiome analysis, sgNTC and sgZmynd8-expressing Cas9⁺ P14 cells were sort-purified from the spleens of recipient mice at day 21 post-LCMV Cl13 infection (n =  2 per group). All samples were centrifuged at 2,000 rpm for 5 min at 4 °C. The supernatant was removed, and the pellets were resuspended in 50 μl 1× PBS plus 0.04% BSA. Nuclei for Single Cell Multiome ATAC + Gene Expression Sequencing were isolated, washed, and counted per manufacturer’s instructions (10X Genomics). Single nuclei suspensions were loaded onto the Chromium Controller according to their respective cell counts to generate single nuclei GEMs per the manufacturer’s protocol. Each sample was loaded into a separate channel. Libraries were prepared using the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Bundle (10X Genomics). The cDNA content of each sample was amplified (7 cycles for ATAC library, 12 cycles for Gene Expression library), and then quantified and quality checked using a High Sensitivity DNA chip with a TapeStation 4200 analyser (Agilent Technologies). Samples for ATAC library sequencing were diluted to 150 pM for loading onto the NovaSeq (Illumina) with a 2× 100 paired-end kit using the following read cycles: 50 cycles Read 1 N, 8 cycles i7 Index, 24 cycles i5 Index, and 49 cycles Read 2 N. An average of 343,111,424 reads per sample was obtained (~45,285 read pairs per nucleus). Samples for Gene Expression library sequencing were diluted to 150 pM for loading onto the NovaSeq (Illumina) with a 2× 100 paired-end kit using the following read cycles: 28 cycles Read 1, 10 cycles i7 Index, 10 cycles i5 Index, and 90 cycles Read 2 N. An average of 439,208,964 reads per sample was obtained (~57,979 read pairs per nucleus).

For single-nuclear RNA-seq (snRNA-seq) analysis from multiome profiling, the Cell Ranger ARC v2.0.2 Single-Cell Software Suite (10X Genomics) was implemented to process the raw sequencing data from the Illumina HiSeq run. This pipeline performed de-multiplexing, alignment (mm10), and barcode processing to generate gene-cell and peak-cell matrices used for downstream gene expression and ATAC analysis, respectively. Cells with low or high unique molecular identifier (UMI) counts were filtered. A total of 24,160 cells (sgNTC replicate 1: 4,764; sgNTC replicate 2: 7,142; sgZmynd8 replicate 1: 5,241; and sgZmynd8 replicate 2: 7,013) were captured with an average of 363 mRNA molecules (UMIs, median: 327, range: 196–1,770). The expression level of each gene was normalized using the NormalizeData function with a scale factor of 10⁶. In-house generated processed multiomics data have been deposited in the NCBI GEO database and are accessible through the GEO SuperSeries access number GSE342764.

Sample merging and processing in scATAC–seq

ATAC peaks from each sample were merged to create a common peak set and quantified by Signac R package (v1.14.0). Specifically, peak coordinates for each sample were loaded and converted to genomic ranges. The ‘reduce’ function in GenomicRanges R package (v1.55.4) was used to first create a common set of peaks. Fragment objects were created using the CreateFragmentObject function for each sample to quantify peaks using the FeatureMatrix function. Quantified matrices were used to create a Seurat object for each sample, and the standard merge function was then used to merge the objects. Normalization and linear dimensional reduction were performed on the merged single-cell ATAC–seq (scATAC–seq) dataset through latent semantic indexing using RUNTFIDF, FindTopFeatures, and RunSVD functions with default parameters. The scATAC–seq dataset was then integrated with snRNA-seq data through the corresponding cell barcode for each cell. Using the weighted nearest-neighbour methods in the Seurat R package, a joint neighbour graph that represents both the gene expression and DNA accessibility measurements was computed by FindMultiModalNeighbors function. CD8⁺ T cell states were annotated by their markers similar as the scRNA-seq analysis.

Differential accessibility, motif enrichment analysis

Differentially accessible peaks between cell types or genotypes were calculated by the FindMarkers function with min.pct = 0.1. Transcription factor motifs were identified using the FindMotifs function based on the differentially upregulated and downregulated peaks. Additionally, we also computed a per-cell motif activity score by running ChromVAR (runChromVAR function) in the Signac R package (v1.14.0) and directly tested for differential activity scores between genotypes in P14 cells or the indicated T_ex cell states using the FindMarkers function. The code for analysis of multiomics data in the relevant figure panels has been deposited to Zenodo (https://doi.org/10.5281/zenodo.21924005 (ref. ⁵⁶)).

ATAC–seq

Sample preparation

Cas9-expressing P14 cells transduced with either sgNTC (Ametrine⁺) or sgZmynd8 (GFP⁺) were mixed at a 1:1 ratio and then injected intravenously into the same Cas9⁺ recipient mice, followed by LCMV Cl13 infection. At 14 dpi, Tex^prog (Ly108⁺TIM-3^–) and effector-like (Ly108^–TIM-3⁺) cells were sorted from the same host for ATAC–seq analysis (n = 4 biological replicates per group). Sorted cells were incubated in 50 µl ATAC–seq lysis buffer (10 mM Tris-HCl, pH 7.4, 10 mM NaCl, 3 mM MgCl₂, 0.1% IGEPAL CA-630) on ice for 10 min. The resulting nuclei were pelleted at 500g for 10 min at 4 °C. The supernatant was carefully removed with a pipette and discarded. The pellet was resuspended in 50 µl transposase reaction mix (25 µl 2× TD buffer, 22.5 µl nuclease-free water and 2.5 µl transposase) and incubated for 30 min at 37 °C. After the reaction, the DNA was cleaned up using the QIAGEN MinElute kit. The barcoding reaction was run using the NEBNext HiFi kit based on manufacturer’s instructions and amplified for five cycles as described^19,22 using the same primers. The optimal cycle numbers were assessed from 5 µl (of 50 µl) from the previous reaction mix using KAPA SYBRFast (Kapa Biosystems) and a 20-cycle amplification on an Applied Biosystems 7900HT. Optimal cycles were determined from the linear part of the amplification curve, and the remaining 45 µl of PCR reaction was amplified in the same reaction mix using the optimal cycle number.

Data analysis

ATAC–seq analysis was performed as previously described^19,22. In brief, 2× 50 bp paired-end reads from NovaSeq were trimmed for Nextera adapters using Trimmomatic (v0.36; paired-end mode, parameters: LEADING:10 TRAILING:10 SLIDINGWINDOW:4:18 MINLEN:25) and aligned to the mouse genome mm10 with BWA (v0.7.16, default parameters). Duplicates were marked using Picard (v2.9.4), and only non-duplicated, properly paired reads were retained with SAMtools (-q 1 -F 1804, v1.9). To account for Tn5 transposase shift, reads were offset by +4 bp (sense strand) and –5 bp (antisense strand), then separated into nucleosome-free, mono-, di-, and tri-nucleosome fractions by fragment size as described⁶⁷. BigWig files were generated using the central 80 bp of fragments, scaled to 3 × 10⁷ nucleosome-free reads. Clear nucleosome-free peaks and mono-, di-, and tri-nucleosome patterns were observed in IGV (v2.4.13). All samples yielded ~2 × 10⁸ nucleosome-free reads, indicating high data quality. Next, peaks were called on nucleosome-free reads using MACS2 (v2.1.1.20160309, parameters: --extsize 200 --nomodel). To ensure reproducibility, peaks were retained only if called with a stricter cutoff (MACS2 -q 0.05). Consensus peaks were defined as those present in ≥50% of replicates, with non-reproducible peaks discarded. Between sgNTC and sgZmynd8-transduced Tex^prog and effector-like cells, reproducible peaks overlapping by ≥100 bp were merged. Read counts for each region were obtained with bedtools (v2.25.0).

For differential accessibility analysis, nucleosome-free reads were normalized to counts per million, and differences were assessed using a negative binomial model implemented in DESeq2 R package (v1.32.0)⁵⁵. Differentially accessible open chromatin regions (OCRs) were defined as those with P < 0.05 and |log₂FC|> 0.5. PCA was performed with the prcomp function in R. Differential accessibility regions were assigned to the nearest genes using HOMER (v4.9.1) software⁶⁸, and functional enrichment was assessed with MSigDB signatures⁶⁹ by a two-tailed Fisher’s exact test. For motif analysis, 1,000 unchanged regions (P > 0.5) were used as background. Motifs were scanned using FIMO from the MEME suite⁷⁰ (v4.11.3; parameters: --thresh 1e-4 --motif-pseudo 0.0001) against the TRANSFAC 2019 Vertebrata database. Enrichment in differential accessible OCRs versus background was evaluated with two-tailed Fisher’s exact tests. For transcription factor footprinting, binding activity was inferred and visualized using RGT-HINT⁷¹ (v0.13.2). For Cistrome transcriptional regulator binding analysis, OCRs with increased accessibility (P < 0.05 and log₂FC > 0.5) in sgZmynd8-expressing Tex^prog and effector-like cells compared to sgNTC counterparts were uploaded to http://dbtoolkit.cistrome.org/ to analyse transcriptional regulators (including transcription factors and chromatin regulators) with significant binding overlaps using default parameters. Genome tracks of average ATAC–seq signals (for each genotype) in indicated gene loci were presented using Integrative Genomics Viewer (IGV, v2.4.13). In-house generated raw and processed ATAC–seq data have been deposited in the NCBI GEO database and are accessible through the GEO SuperSeries access number GSE342764. The code for analysis of ATAC–seq data in the relevant figure panels has been deposited to Zenodo (https://doi.org/10.5281/zenodo.21924005 (ref. ⁵⁶)).

In vivo single-cell and bulk CRISPR screening

Selection of epigenetic factors for library design

To select the epigenetic factors that were potentially involved in CD8⁺ T cell exhaustion in the chronic infection, we performed differential gene expression analyses from public scRNA-seq^11,15 (GSE149876 and GSE122712) and RNA-seq¹⁵ (GSE123235) datasets and differential accessibility analysis of OCRs in a scATAC–seq dataset¹⁶ (GSE230363), to nominate genes with differential expression/accessibility (threshold: log₂FC > 0.3, P < 0.05 for Up; log₂FC < –0.3, P < 0.05 for Down) between different T_ex states (Tex^term versus Tex^prog, Tex^term versus effector-like or Tex^prog versus effector-like). Those genes with differential expression or accessibility were overlapped with two published gene lists^16,72 of epigenetic factors to select a total of 91 epigenetic factors. Nine transcription factors (Bach2⁷³, Batf^17,18,19, Myb⁷⁴, Prdm1⁷⁵, Tbx21⁷⁶, Tcf7^77,78, Tox^4,5,6,7,8, Tox2⁸ and Zeb2^79,80) known to regulate T_ex cell differentiation were also included in the custom library for a total of 100 genes targeted (Supplementary Table 1).

Sample preparation

The in vivo screening approach was modified based on previous studies^19,31. In brief, retrovirus was produced by co-transfecting the retroviral library with pCL-Eco (Addgene, #12371) in Plat-E cells. 48 h after transfection, the supernatant containing viral particles was collected and frozen at −80 °C. Cas9-expressing P14 cells were transduced to achieve 20–30 % transduction efficiency. Transduced cells were sorted based on the expression of Ametrine, and an aliquot of 1 × 10⁶ transduced Cas9⁺ P14 cells was saved as ‘input’. Transduced Cas9⁺ P14 cells (1 × 10⁴) were then transferred intravenously to Cas9⁺ mice followed by LCMV Cl13 infection (2 × 10⁶ PFU) one day later. A total of 60 recipient mice were used in 2 screens combined (in Fig. 1a). For scCRISPR screening, donor-derived total P14 cells from a total of 15 Cas9⁺ mice were sorted and pooled for scCRISPR analysis at 28 dpi. In total, five 10X reactions (Chromium Next GEM Single Cell 5’ Kit v2, PN-1000263 and 5’ CRISPR Kit, PN-1000451) were used. In-house generated raw and processed scCRISPR screening data have been deposited in the NCBI GEO database and are accessible through the GEO SuperSeries access number GSE342764. For bulk CRISPR screening, donor-derived Tex^prog (Ly108⁺CX3CR1^–), effector-like (Ly108^–CX3CR1⁺) and Tex^term (Ly108^–CX3CR1^–) P14 subpopulations were sorted at 28 dpi and frozen at −80 °C prior to genomic DNA extraction. At least 5 × 10⁴ sorted T_ex subpopulations (> 150× cell coverage per sgRNA) were used for deep sequencing analysis. For scCRISPR screening of IL-2 signalling regulators (Extended Data Fig. 1m), an in vivo scCRISPR screen targeting known positive regulators (Jak3, Stat5a and Stat5b) and negative regulators (Socs1 and Socs3) of IL-2 signalling (along with sgNTCs and additional validation controls for the screening phenotypes) was performed, using a similar strategy as described in Fig. 1a. The sgRNA library contained 3 sgRNAs targeting each of the IL-2 signalling genes. In-house generated processed scCRISPR screening data in Extended Data Fig. 1m have been deposited into Zenodo (https://doi.org/10.5281/zenodo.21909366 (ref. ⁸¹)).

Library preparation for scCRISPR screening

Sorted P14 cells were resuspended and diluted in 1× PBS (Thermo Fisher Scientific) containing 0.04% BSA (Amresco) at concentration of 1 × 10⁶ cells ml^–1. Both the gene expression library and the CRISPR screening library were prepared using the Chromium Next GEM Single Cell 5’ v2 kit and 5’ CRISPR Kit for CRISPR Screening (10X Genomics). In brief, the single-cell suspensions were loaded onto the Chromium Controller according to their respective cell counts to generate 10,000 single-cell gel beads in emulsion (GEMs) per sample. Each sample was loaded into four separate channels. The resulting libraries were quantified, and quality checked by TapeStation (Agilent). Samples were diluted and loaded onto the NovaSeq (Illumina) to a sequencing depth of 300 million reads per channel for gene expression libraries and 200 million reads per channel for CRISPR screening libraries.

Data preprocessing

Alignments and count aggregation of gene expression and sgRNA reads were completed using Cell Ranger⁵⁷ (v7.1.0). Gene expression and sgRNA reads were aligned using the cellranger count command with default settings. Gene expression reads were aligned to the mouse genome (mm10 from ENSEMBL GRCm38 loaded from 10X Genomics). sgRNA reads were aligned to sgRNA library using the pattern TATTTCTAGCTCTAAAAC(BC) to capture sgRNA sequences. Only droplets with ≥2 sgRNA UMIs were used in further analyses. The filtered feature matrices were imported to create assays for a Seurat object containing both gene expression and CRISPR guide capture matrices. A third assay summarizing the total gene-level counts of all four sgRNAs for each target gene was also created, followed by pooling of 5 samples using the ‘merge’ function. Cells were initially quality filtered based on the percentage of mitochondrial reads <10% (to remove dead cells) and the number of detected RNA features <7,500 and UMI feature <50,000 (to remove doublets for downstream gene expression analysis). Cells detected with sgRNAs targeting two or more genes were then removed to avoid interference from multi-sgRNA-transduced cells. A total of 19,032 P14 cells passed quality filtering and were used for downstream analysis. For the scCRISPR screening of IL-2 signalling regulators along with additional perturbations (Extended Data Fig. 1m), similar preprocessing was conducted as above. In brief, cells were first filtered based on the percentage of mitochondrial reads <10% (to remove dead cells) and the number of detected RNA features <6,000 and UMI feature <40,000 (to remove doublets for downstream gene expression analysis). Cells detected with sgRNAs targeting two or more genes were then removed to avoid interference from multi-sgRNA-transduced cells. A total of 2,825 P14 cells containing sgRNAs targeting IL-2 signalling genes (Jak3, Stat5a, Stat5b, Socs1 and Socs3) and sgNTC passed quality filtering and were used for downstream analysis. To evaluate the enrichment or depletion of each perturbation relative to sgNTC (Extended Data Fig. 1c), cells expressing sgRNAs targeting the same gene (3 sgRNAs per gene) were pooled. We calculated the log₂FC in cell number by comparing the number of cells expressing each perturbation with the number of sgNTC-expressing cells, followed by normalization using the ratio of the number of sgRNAs targeting each perturbation to the number of sgNTC sgRNAs in the sgRNA library. To calculate replicate-level statistics for the relative enrichment and depletion of total P14 cells in Supplementary Table 7 (specifically, five 10X Genomics reactions (technical replicates) for the scCRISPR screen in Fig. 1a and eight technical replicates for the scCRISPR screen in Extended Data Fig. 1m), the average gene-level log₂FC was calculated as the mean of gene-level log₂FCs across all technical replicates. To assess the reproducibility of gene-level effects across technical replicates, two-sided, one-sample t-tests were performed on the replicate-specific, gene-level log₂FC values against a null hypothesis of zero mean effect. Resulting P values were adjusted for multiple hypothesis testing using the Benjamini–Hochberg procedure to control the false discovery rate (FDR).

For cell clustering, the FindClusters function of the Seurat R package (v5.1.0) was used to identify the clusters in P14 cells in an unbiased manner. To determine the molecular determinants for P14 cell developmental trajectory, clusters were annotated as four cellular states (Tex^prog, Tex^int, Tex^KLR and Tex^term) based on expression of indicated markers: Tcf7⁺Slamf6⁺ (Tex^prog), Cx3cr1⁺Mki67⁺Klrg1^low (Tex^int), Cx3cr1⁺Klrg1^hiKlre1^hiKlrc1^hiKlrd1^hi (Tex^KLR) and Cx3cr1^–Cd101⁺ (Tex^term). Dot plots showing the relative average expression (after scaled normalization) of marker genes or effector molecules and cytokines in different clusters were visualized using the DotPlot function in the Seurat R package. Pseudotime trajectory analysis was performed (using default parameters in the Monocle3 R package⁶⁴ (v1.3.7)) on the four T_ex states with the Tex^prog state set as the starting point. Activity scores of gene signatures (such as the Tex^prog, Tex^int, Tex^KLR and Tex^term signatures in Extended Data Fig. 1d curated from literature¹³) were calculated using AddModuleScore function of the Seurat package for the four cellular states and visualized by violin plots. To visualize the distribution of cells with a specific perturbation (at the gene level) on the UMAP, contour density plots were generated using the ggplot2 (v3.5.1) R package.

Network analysis

For network analysis, to mitigate the potential for observed phenotypes being driven by an outlier single sgRNA (which can be contributed by low cell recovery and/or sparse nature of scRNA-seq datasets), sgRNAs recovered in fewer than ten cells were excluded, similar as described^31,82. Perturbations for which two sgRNAs failed to meet this minimum recovery threshold were excluded from downstream gene program analyses, resulting in the exclusion of a total of six genes (Batf, Ezh2, Nat10, Prmt1, Ssrp1 and Trim28). Along with 1,092 cells expressing sgNTC controls (representing 30 individual sgRNAs), we recovered an average of 189 cells per gene and 65 cells per sgRNA for gene-specific perturbations (Supplementary Table 6). log₂FC (perturbation versus sgNTC) values were used to quantify the regulatory effect of each perturbation on target genes. The following bioinformatic analytical steps were applied to define regulomes associated with P14 cell differentiation. Step 1, similar to our previous publication³¹, our initial computational analysis calculated the individual gene regulatory effect of a specific genetic perturbation on target genes. Specifically, differential gene expression analysis was performed using the FindMarkers function in Seurat R package⁵⁸, by comparing each perturbation to the sgNTC control. log₂FC (perturbation versus sgNTC) values were used to quantify the regulatory effects of each perturbation on target genes. Step 2, to define co-regulated gene programs associated with T_ex cell differentiation, differential expression analysis was performed by comparing each of the four T_ex states with all other states combined. Differentially expressed genes (|log₂FC (each state versus the other 3 states)| > 0.5) were combined, and redundant genes across states were removed. A gene × perturbation matrix (279 × 94, with log₂FC of perturbation versus sgNTC values) was then constructed to generate a perturbation map (Fig. 1c, top left). Step 3, co-regulated gene programs (of the aforementioned 279 genes) were identified by Spearman correlation hierarchical clustering, with four major programs annotated, namely stemness (G1), proliferation (G2), effector function (G3) and exhaustion (G4) (Fig. 1c, bottom left), based on enrichment of pathways (Extended Data Fig. 1h). Meanwhile, co-functional modules were defined by Spearman correlation hierarchical clustering, resulting in nine modules labelled as M1–M9 (Fig. 1c, top right).

Step 4, following these initial steps to define gene regulatory effects, co-regulated gene programs, and co-functional modules, we then computed the average gene regulatory effect of a specific perturbation on a given gene program, defined as the average log₂FC across all genes within that co-regulated gene program. As a final step, the average gene regulatory effect of a co-functional module on a given gene program was calculated as the mean of the gene regulatory effects across all genetic perturbations within the co-functional module (Fig. 1e and Extended Data Fig. 1i). The strength of regulation between perturbation modules and gene programs was visualized using the ggalluvial R package (v0.12.5), where the line width represents regulatory strength, colour (red versus blue) indicates positive or negative effects, and module height reflects the overall magnitude of regulation. The regulatory effect of perturbations on Tox expression was visualized by Cytoscape software (v3.10.3).

Gene-level perturbation ranking analysis

For gene-level perturbation ranking between T_ex states (for example, Fig. 1f,g), cells expressing sgRNAs targeting the same gene (3 sgRNAs per gene) were pooled. For each gene perturbation, an enrichment score was calculated as the ratio of cells in a given T_ex state relative to all other T_ex states. To account for differences in the baseline abundance of individual T_ex states, the enrichment score was normalized by the corresponding global cell ratio of that T_ex state relative to all other T_ex states across the entire dataset, and is reported as a normalized log₂FC. To calculate replicate-level statistics for enrichment or depletion of T_ex states (Supplementary Table 7), the average gene-level log₂FC was calculated as the mean of gene-level log₂FCs across all technical replicates. To assess the reproducibility of gene-level effects across technical replicates, the P values and FDR were subsequently calculated across the technical replicates as described above. GSEA was performed on the scCRISPR dataset by comparing sgZmynd8-expressing P14 cells with sgNTC-expressing P14 cells, using STAT5CA Up and Down gene signatures. The code for analysis of scCRISPR data in the relevant figure panels has been deposited to Zenodo (https://doi.org/10.5281/zenodo.21924005 (ref. ⁵⁶)).

Library preparation for bulk CRISPR screening

Genomic DNA was extracted by using the DNeasy Blood & Tissue Kits (69506, QIAGEN). Primary PCR was performed by using the KOD Hot Start DNA Polymerase (71086, Millipore) and the following pair of Nextera next-generation sequencing (NGS) primers: (Nextera NGS forward (-F): TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGttgtggaaaggacgaaacaccg; Nextera NGS reverse (-R): GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGccactttttcaagttgataacgg). Primary PCR products were purified using the AMPure XP beads (A63881, Beckman). A second PCR reaction was performed to attach Illumina adapters and indexes to barcode each sample. Hi-Seq 50-bp single-end sequencing (Illumina) was performed for library sequencing.

Data analysis for bulk CRISPR screening

FASTQ read files obtained after sequencing were demultiplexed using Hi-Seq analysis software (Illumina) and processed using mageck (v0.5.9.4) software⁸³. Raw count tables were generated using the mageck count command by matching the sgRNA sequence of the aforementioned. Read counts for sgRNAs were normalized against median read counts across all samples for each screening. For each gene or sgRNA in the sgRNA library, the log₂FC for enrichment or depletion was calculated using the mageck test command, with the gene-lfc-method parameter as the mean and control-sgrna parameter using the list of sgNTCs. The sgRNA count table and analysis of bulk CRISPR screening data have been deposited to Zenodo (https://doi.org/10.5281/zenodo.21909366 (ref. ⁸¹)). The code for analysis of bulk CRISPR data in the relevant figure panels has been deposited to Zenodo (https://doi.org/10.5281/zenodo.21924005 (ref. ⁵⁶)).

Statistical analysis for biological experiments

For biological experiment (non-omics) analyses, data were analysed by Prism software (v11 for tumour growth curve experiments and v9 for other experiments; GraphPad) using two-tailed unpaired Student’s t-tests, one-way analysis of variance (ANOVA) or two-way ANOVA, as indicated in the figure legends. A two-tailed Student’s t-test was used to compare the two groups. For multiple comparisons, a one-way or two-way ANOVA with Tukey’s or Sidak’s multiple-comparisons test was applied. For comparing mouse tumour growth curves, two-way ANOVA with Geisser–Greenhouse correction was applied. The log-rank (Mantel–Cox) test was performed to compare mouse survival curves. Two-tailed Wilcoxon rank-sum tests were applied for activity score (without Bonferroni correction adjustment) or expression (with Bonferroni correction adjustment) analysis of scRNA-seq data. P < 0.05 was considered statistically significant with P value annotations shown as *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. In all bar plots, data are presented as the mean ± s.e.m. The exact P values are shown in the source data files that accompany this manuscript.

Reporting summary

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

The authors declare that the data supporting the findings of this study are available within the manuscript and its Supplementary Information. All bulk CRISPR and scCRISPR screening, ATAC–seq, CUT&RUN, scRNA-seq and single-cell multiomics data described in the manuscript have been deposited into Gene Expression Omnibus (GEO) Superseries access number GSE342764 and Zenodo (https://doi.org/10.5281/zenodo.21909366 (ref. ⁸¹)). Public scRNA-seq, bulk RNA-seq or scATAC–seq datasets are available through GSE122712, GSE164177, GSE89307, GSE157829, GSE123139, GSE120575, GSE123813, GSE205506, GSE206732, GSE252650, GSE149876, GSE123235 and GSE230363, Zenodo (https://zenodo.org/records/10604293 (ref. ⁸⁴)). Hallmark, C2, C5 and C7 collections were from the MSigDB (https://www.broadinstitute.org/gsea/msigdb/). The TRANSFAC 2019 Vertebrata database was commercially available at GeneXplain GmbH (https://genexplain.com/transfac-database/). Source data are provided with this paper.

The original code for bioinformatics analyses has been deposited at Zenodo (https://doi.org/10.5281/zenodo.21924005 (ref. ⁵⁶)).

1. Sun, Q. & Dong, C. Regulators of CD8⁺ T cell exhaustion. Nat. Rev. Immunol. 26, 129–151 (2025).

   Article  PubMed  Google Scholar 

2. Shouse, A. N., LaPorte, K. M. & Malek, T. R. Interleukin-2 signaling in the regulation of T cell biology in autoimmunity and cancer. Immunity 57, 414–428 (2024).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

3. Giles, J. R., Globig, A. M., Kaech, S. M. & Wherry, E. J. CD8⁺ T cells in the cancer-immunity cycle. Immunity 56, 2231–2253 (2023).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

4. Khan, O. et al. TOX transcriptionally and epigenetically programs CD8⁺ T cell exhaustion. Nature 571, 211–218 (2019).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

5. Scott, A. C. et al. TOX is a critical regulator of tumour-specific T cell differentiation. Nature 571, 270–274 (2019).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

6. Alfei, F. et al. TOX reinforces the phenotype and longevity of exhausted T cells in chronic viral infection. Nature 571, 265–269 (2019).

   Article  ADS  CAS  PubMed  Google Scholar 

7. Yao, C. et al. Single-cell RNA-seq reveals TOX as a key regulator of CD8⁺ T cell persistence in chronic infection. Nat. Immunol. 20, 890–901 (2019).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

8. Seo, H. et al. TOX and TOX2 transcription factors cooperate with NR4A transcription factors to impose CD8⁺ T cell exhaustion. Proc. Natl Acad. Sci. USA 116, 12410–12415 (2019).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

9. Hudson, W. H. et al. Proliferating transitory T cells with an effector-like transcriptional signature emerge from PD-1⁺ stem-like CD8⁺ T cells during chronic infection. Immunity 51, 1043–1058.e1044 (2019).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

10. Zander, R. et al. CD4⁺ T cell help is required for the formation of a cytolytic CD8⁺ T cell subset that protects against chronic infection and cancer. Immunity 51, 1028–1042.e4 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

11. Beltra, J. C. et al. Developmental relationships of four exhausted CD8⁺ T cell subsets reveals underlying transcriptional and epigenetic landscape control mechanisms. Immunity 52, 825–841.e8 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

12. Giles, J. R. et al. Shared and distinct biological circuits in effector, memory and exhausted CD8⁺ T cells revealed by temporal single-cell transcriptomics and epigenetics. Nat. Immunol. 23, 1600–1613 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

13. Daniel, B. et al. Divergent clonal differentiation trajectories of T cell exhaustion. Nat. Immunol. 23, 1614–1627 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

14. Shi, H., Doench, J. G. & Chi, H. CRISPR screens for functional interrogation of immunity. Nat. Rev. Immunol. 23, 363–380 (2023).

    Article  CAS  PubMed  Google Scholar 

15. Miller, B. C. et al. Subsets of exhausted CD8⁺ T cells differentially mediate tumor control and respond to checkpoint blockade. Nat. Immunol. 20, 326–336 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

16. Baxter, A. E. et al. The SWI/SNF chromatin remodeling complexes BAF and PBAF differentially regulate epigenetic transitions in exhausted CD8⁺ T cells. Immunity 56, 1320–1340.e10 (2023).

    Article  CAS  PubMed  Google Scholar 

17. Chen, Y. et al. BATF regulates progenitor to cytolytic effector CD8⁺ T cell transition during chronic viral infection. Nat. Immunol. 22, 996–1007 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

18. Seo, H. et al. BATF and IRF4 cooperate to counter exhaustion in tumor-infiltrating CAR T cells. Nat. Immunol. 22, 983–995 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

19. Wei, J. et al. Targeting REGNASE-1 programs long-lived effector T cells for cancer therapy. Nature 576, 471–476 (2019).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

20. Kumar, S. et al. CARM1 inhibition enables immunotherapy of resistant tumors by dual action on tumor cells and T cells. Cancer Discov. 11, 2050–2071 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

21. Beltra, J. C. et al. Stat5 opposes the transcription factor Tox and rewires exhausted CD8⁺ T cells toward durable effector-like states during chronic antigen exposure. Immunity 56, 2699–2718.e11 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

22. Huang, H. et al. In vivo CRISPR screening reveals nutrient signaling processes underpinning CD8⁺ T cell fate decisions. Cell 184, 1245–1261.e21 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

23. West, E. E. et al. PD-L1 blockade synergizes with IL-2 therapy in reinvigorating exhausted T cells. J. Clin. Invest. 123, 2604–2615 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

24. Hashimoto, M. et al. PD-1 combination therapy with IL-2 modifies CD8⁺ T cell exhaustion program. Nature 610, 173–181 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

25. Belk, J. A. et al. Genome-wide CRISPR screens of T cell exhaustion identify chromatin remodeling factors that limit T cell persistence. Cancer Cell 40, 768–786.e7 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

26. Chen, Y., Tsai, Y. H. & Tseng, S. H. Regulation of ZMYND8 to treat cancer. Molecules 26, 1083 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

27. Creyghton, M. P. et al. Histone H3K27ac separates active from poised enhancers and predicts developmental state. Proc. Natl Acad. Sci. USA 107, 21931–21936 (2010).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

28. Simeonov, D. R. et al. Discovery of stimulation-responsive immune enhancers with CRISPR activation. Nature 549, 111–115 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

29. Cao, Z. et al. ZMYND8-regulated IRF8 transcription axis is an acute myeloid leukemia dependency. Mol. Cell 81, 3604–3622.e10 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

30. Wang, S. et al. An atlas of immune cell exhaustion in HIV-infected individuals revealed by single-cell transcriptomics. Emerg. Microbes Infect. 9, 2333–2347 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

31. Zhou, P. et al. Single-cell CRISPR screens in vivo map T cell fate regulomes in cancer. Nature 624, 154–163 (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

32. Burger, M. L. et al. Antigen dominance hierarchies shape TCF1⁺ progenitor CD8 T cell phenotypes in tumors. Cell 184, 4996–5014.e26 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

33. Philip, M. et al. Chromatin states define tumour-specific T cell dysfunction and reprogramming. Nature 545, 452–456 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

34. Li, H. et al. Dysfunctional CD8 T cells form a proliferative, dynamically regulated compartment within human melanoma. Cell 176, 775–789.e18 (2019).

    Article  CAS  PubMed  Google Scholar 

35. Zhao, M. et al. Rapid in vitro generation of bona fide exhausted CD8⁺ T cells is accompanied by Tcf7 promotor methylation. PLoS Pathog. 16, e1008555 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

36. Zheng, W. et al. Regnase-1 suppresses TCF-1⁺ precursor exhausted T-cell formation to limit CAR-T-cell responses against ALL. Blood 138, 122–135 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

37. Beck, J. D. et al. Long-lasting mRNA-encoded interleukin-2 restores CD8⁺ T cell neoantigen immunity in MHC class I-deficient cancers. Cancer Cell 42, 568–582.e11 (2024).

    Article  CAS  PubMed  Google Scholar 

38. Moynihan, K. D. et al. IL2 targeted to CD8⁺ T cells promotes robust effector T-cell responses and potent antitumor immunity. Cancer Discov. 14, 1206–1225 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

39. Yost, K. E. et al. Clonal replacement of tumor-specific T cells following PD-1 blockade. Nat. Med. 25, 1251–1259 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

40. Li, J. et al. Remodeling of the immune and stromal cell compartment by PD-1 blockade in mismatch repair-deficient colorectal cancer. Cancer Cell 41, 1152–1169.e7 (2023).

    Article  PubMed  Google Scholar 

41. Sade-Feldman, M. et al. Defining T cell states associated with response to checkpoint immunotherapy in melanoma. Cell 175, 998–1013.e20 (2018).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

42. Pauken, K. E. et al. Epigenetic stability of exhausted T cells limits durability of reinvigoration by PD-1 blockade. Science 354, 1160–1165 (2016).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

43. Shen, H. et al. Suppression of enhancer overactivation by a RACK7–histone demethylase complex. Cell 165, 331–342 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

44. Villarino, A. V. et al. A central role for STAT5 in the transcriptional programing of T helper cell metabolism. Sci. Immunol. 7, eabl9467 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

45. Pircher, H., Burki, K., Lang, R., Hengartner, H. & Zinkernagel, R. M. Tolerance induction in double specific T-cell receptor transgenic mice varies with antigen. Nature 342, 559–561 (1989).

    Article  ADS  CAS  PubMed  Google Scholar 

46. Overwijk, W. W. et al. Tumor regression and autoimmunity after reversal of a functionally tolerant state of self-reactive CD8⁺ T cells. J. Exp. Med. 198, 569–580 (2003).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

47. Platt, R. J. et al. CRISPR-Cas9 knockin mice for genome editing and cancer modeling. Cell 159, 440–455 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

48. McCausland, M. M. & Crotty, S. Quantitative PCR technique for detecting lymphocytic choriomeningitis virus in vivo. J. Virol. Methods 147, 167–176 (2008).

    Article  CAS  PubMed  Google Scholar 

49. Panday, A., Elango, R., Willis, N. A. & Scully, R. A modified CUT&RUN-seq technique for qPCR analysis of chromatin-protein interactions. STAR Protoc. 3, 101529 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

50. Ewels, P. A. et al. The nf-core framework for community-curated bioinformatics pipelines. Nat. Biotechnol. 38, 276–278 (2020).

    Article  CAS  PubMed  Google Scholar 

51. Meers, M. P., Tenenbaum, D. & Henikoff, S. Peak calling by sparse enrichment analysis for CUT&RUN chromatin profiling. Epigenetics Chromatin 12, 42 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

52. Zhang, Y. et al. Model-based analysis of ChIP–seq (MACS). Genome Biol. 9, R137 (2008).

    Article  PubMed  PubMed Central  Google Scholar 

53. Yu, G., Wang, L. G. & He, Q. Y. ChIPseeker: an R/Bioconductor package for ChIP peak annotation, comparison and visualization. Bioinformatics 31, 2382–2383 (2015).

    Article  CAS  PubMed  Google Scholar 

54. Zheng, R. et al. Cistrome Data Browser: expanded datasets and new tools for gene regulatory analysis. Nucleic Acids Res. 47, D729–D735 (2019).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

55. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

56. Shi, H. & Chi, H. Zmynd8 manuscript code [Computer software]. Zenodo https://doi.org/10.5281/zenodo.21924005 (2026).

57. Zheng, G. X. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

58. Hao, Y. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol. 42, 293–304 (2024).

    Article  ADS  CAS  PubMed  Google Scholar 

59. Im, S. J. et al. Defining CD8⁺ T cells that provide the proliferative burst after PD-1 therapy. Nature 537, 417–421 (2016).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

60. Joshi, N. S. et al. Inflammation directs memory precursor and short-lived effector CD8⁺ T cell fates via the graded expression of T-bet transcription factor. Immunity 27, 281–295 (2007).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

61. Tirosh, I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352, 189–196 (2016).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

62. Milner, J. J. et al. Runx3 programs CD8⁺ T cell residency in non-lymphoid tissues and tumours. Nature 552, 253–257 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

63. Qin, Q. et al. Lisa: inferring transcriptional regulators through integrative modeling of public chromatin accessibility and ChIP–seq data. Genome Biol. 21, 32 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

64. Trapnell, C. et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat. Biotechnol. 32, 381–386 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

65. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

66. Fagerberg, E. et al. KLF2 maintains lineage fidelity and suppresses CD8 T cell exhaustion during acute LCMV infection. Science 387, eadn2337 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

67. Buenrostro, J. D., Giresi, P. G., Zaba, L. C., Chang, H. Y. & Greenleaf, W. J. Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position. Nat. Methods 10, 1213–1218 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

68. Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38, 576–589 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

69. Liberzon, A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst. 1, 417–425 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

70. Grant, C. E., Bailey, T. L. & Noble, W. S. FIMO: scanning for occurrences of a given motif. Bioinformatics 27, 1017–1018 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

71. Li, Z. et al. Identification of transcription factor binding sites using ATAC-seq. Genome Biol. 20, 45 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

72. Guo, A. et al. cBAF complex components and MYC cooperate early in CD8⁺ T cell fate. Nature 607, 135–141 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

73. Yao, C. et al. BACH2 enforces the transcriptional and epigenetic programs of stem-like CD8⁺ T cells. Nat. Immunol. 22, 370–380 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

74. Tsui, C. et al. MYB orchestrates T cell exhaustion and response to checkpoint inhibition. Nature 609, 354–360 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

75. Shin, H. et al. A role for the transcriptional repressor Blimp-1 in CD8⁺ T cell exhaustion during chronic viral infection. Immunity 31, 309–320 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

76. Kao, C. et al. Transcription factor T-bet represses expression of the inhibitory receptor PD-1 and sustains virus-specific CD8⁺ T cell responses during chronic infection. Nat. Immunol. 12, 663–671 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

77. Utzschneider, D. T. et al. T cell factor 1-expressing memory-like CD8⁺ T cells sustain the immune response to chronic viral infections. Immunity 45, 415–427 (2016).

    Article  CAS  PubMed  Google Scholar 

78. Chen, Z. et al. TCF-1-centered transcriptional network drives an effector versus exhausted CD8 T cell-fate decision. Immunity 51, 840–855.e5 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

79. Dominguez, C. X. et al. The transcription factors ZEB2 and T-bet cooperate to program cytotoxic T cell terminal differentiation in response to LCMV viral infection. J. Exp. Med. 212, 2041–2056 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

80. Omilusik, K. D. et al. Transcriptional repressor ZEB2 promotes terminal differentiation of CD8⁺ effector and memory T cell populations during infection. J. Exp. Med. 212, 2027–2039 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

81. Shi, H. & Chi, H. Bulk CRISPR and IL2 signalling scCRISPR screen [Dataset]. Zenodo https://doi.org/10.5281/zenodo.21909366 (2026).

82. Song, B. et al. Decoding heterogeneous single-cell perturbation responses. Nat. Cell Biol. 27, 493–504 (2025).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

83. Li, W. et al. MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens. Genome Biol. 15, 554 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

84. Streuber, M. An mRNA-encoded, long-lasting Interleukin-2 restores CD8⁺ T cell neoantigen immunity in MHC class I-deficient cancers [Data set]. Zenodo https://doi.org/10.5281/zenodo.10604293 (2024).

Download references

Author notes

1. These authors contributed equally: Yan Wang, Hao Shi

Authors and Affiliations

1. Department of Immunology, St. Jude Children’s Research Hospital, Memphis, TN, USA

   Yan Wang, Hao Shi, Nicole M. Chapman, Anil KC, Renqiang Sun, Hao Song, Xiaoxi Meng, Xiang Sun & Hongbo Chi

Contributions

Y.W. conceived, designed and performed cellular and molecular experiments, analysed data and wrote the manuscript. H. Shi performed bioinformatic analyses, wrote the manuscript and helped to conceive and design cellular and molecular experiments. N.M.C. co-wrote the manuscript. A.K. prepared scRNA-seq and ATAC–seq libraries. R.S. generated ZMYND8 truncation constructs and performed co-immunoprecipitation experiments. H. Song performed part of the immunoblot analysis. X.M. performed bioinformatic analysis. X.S. helped to perform scCRISPR screening library preparation (first two steps). H.C. helped to conceive and design experiments, co-wrote the manuscript and provided overall direction.

Corresponding author

Correspondence to Hongbo Chi.

Competing interests

H.C. consults for Kumquat Biosciences Inc. and TCura Bioscience and is a co-inventor on patents/patent applications in the field of immunotherapy. The other authors declare no competing interests.

Peer review information

Nature thanks David Barbie, Tuoqi Wu, Keji Zhao and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions