Ezh2 phosphorylation state determines its capacity to maintain CD8+ T memory precursors for antitumor immunity

Memory T cells sustain effector T-cell production while self-renewing in reaction to persistent antigen; yet, excessive expansion reduces memory potential and impairs antitumor immunity. Epigenetic mechanisms are thought to be important for balancing effector and memory differentiation; however, the epigenetic regulator(s) underpinning this process remains unknown. Herein, we show that the histone methyltransferase Ezh2 controls CD8+ T memory precursor formation and antitumor activity. Ezh2 activates Id3 while silencing Id2, Prdm1 and Eomes, promoting the expansion of memory precursor cells and their differentiation into functional memory cells. Akt activation phosphorylates Ezh2 and decreases its control of these transcriptional programs, causing enhanced effector differentiation at the expense of T memory precursors. Engineering T cells with an Akt-insensitive Ezh2 mutant markedly improves their memory potential and capability of controlling tumor growth compared to transiently inhibiting Akt. These findings establish Akt-mediated phosphorylation of Ezh2 as a critical target to potentiate antitumor immunotherapeutic strategies.

D uring an immune response, naive CD8 + T cells (T N ) differentiate into short-lived effector cells, which eliminate the immediate threat, and long-term memory T cells, which are able to rapidly expand and elaborate effector function upon reencountering antigen [1][2][3] . T cells with impaired memory potential are unable to clear persistently expressed antigens, including those associated with chronic infections or tumors [2][3][4][5][6][7][8][9] . One important clinical scenario requiring optimal memory potential is adoptive cell therapy (ACT), where complete destruction of malignant cells requires maintenance of effector T cells (T EFF ) over weeks or months 7,8,[10][11][12] . Akt activation drives CD8 + T cells towards terminal differentiation and diminishes their memory potential 13 . Major transcription factors (TFs), including Id3, Id2, T-bet, Eomes and Blimp-1 control the differentiation and function of effector and memory cells 1,3 . Thus far, the mechanism by which CD8 + T cells epigenetically integrate Akt signaling and these major TFs to regulate the memory potential for controlling tumor growth remains poorly defined.
Histone methylation regulates gene transcription patterns involved in multiple cellular processes [14][15][16] . In T cells, trimethylation of histone H3 at lysine 4 (H3K4me3) is enriched within gene promoters associated with active transcription. In contrast, H3K27me3 is deposited within gene loci correlated with repression of genes important for T-cell proliferation, differentiation and survival. Upon T-cell activation, the majority of these gene loci lose repressive H3K27me3 modifications, which is accompanied with initiating gene transcription [17][18][19] . Ezh2 together with Suz12 and Eed form core components of polycomb repressive complex-2 (PRC2) that catalyzes H3K27me3 and acts primarily as a gene silencer 16,[20][21][22] . Intriguingly, Ezh2 has non-canonical activity in the regulation of signaling proteins and gene activation [23][24][25][26][27] . However, whether Ezh2 may regulate T-cell-fate decisions, and if this Ezh2 activity can be modified during T-cell responses, have not previously been determined.
Here we demonstrate that Ezh2 is essential for the development and maintenance of T memory precursors and associated antitumor immunity. Ezh2 activates Id3 while silencing Id2, Prdm1, and Eomes, promoting the expansion of memory precursors and their differentiation into functional memory cells. Akt mediates phosphorylation of Ezh2, which in turn eases Ezh2 transcriptional control, causing enhanced effector differentiation at the expense of memory precursors. Thus, Akt-mediated phosphorylation of Ezh2 may serve as a critical target to potentiate antitumor immunotherapeutic strategies.

Results
Ezh2 regulates effector and memory T-cell persistence. To evaluate the impact of Ezh2 in CD8 + T-cell responses, we used an experimental ACT model, of which transferred T cells were more effective in controlling tumor growth in lymphopenic mice compared with lymphoreplete hosts 4,11,28 . We deleted Ezh2 in melanoma-associated antigen gp100-specific CD8 + T-cell receptor (TCR)-transgenic Pmel-1 cells by crossing Ezh2 fl/fl CD4-Cre B6 mice 29 to Pmel-1 mice 30 , which produced T-cell-specific Ezh2-knockout Pmel-1 mice (Ezh2 −/− Pmel-1). Transfer of wildtype (WT) but not Ezh2 −/− Pmel-1 T N repressed the growth of pre-established B16 melanoma in lymphodepleted mice (Fig. 1a). Using lymphopenic recipients without B16 tumor, we observed that the frequency and number of WT and Ezh2 −/− Pmel-1 cells was similar in the spleen 4 days after transfer and immunization, while Ezh2 −/− Pmel-1 cell numbers were decreased by 7 days and 35 days (Fig. 1b). The discrepancy was not due to differences in organ tropism, as fewer Ezh2 −/− Pmel-1 cells were detected in peripheral blood (PB) and lymph node (LN) ( Supplementary  Fig. 1). Further, although Ezh2 deficiency did not affect the capacity of Pmel-1 cells to produce IFN-γ during the effector phase (7 days), it caused a reduction of IFN-γ-producing cells by 35 days (Fig. 1c), suggesting an impaired persistence of tumorreactive T cells.
To test if impaired persistence of Ezh2-deficient T cells might be associated with their decreased capability to expand and elaborate effector function upon antigen rechallenge, we transferred equal numbers of WT and Ezh2 −/− memory Pmel-1 cells, which were recovered from primary recipients 42 days after gp100-immunization, into secondary lymphopenic recipients. Seven days after gp100-rechallenge, Ezh2 −/− memory Pmel-1 cells produced approximately 2-and 6-fold fewer total CD8 + T cells and IFN-γ-secreting effectors, respectively, compared to their WT counterparts (Fig. 1d, e). In addition, when donor T cells derived from these secondary recipients were cultured with gp100 for 5 days, Ezh2 −/− memory progenies were unable to expand compared to their WT counterparts (Fig. 1f). This suggests that decreased persistence of activated Ezh2 −/− Pmel-1 cells may result from their impaired memory potential.
We used lymphoreplete mice to validate our conclusion. Equal number of congenically labeled WT (CD45.2 + , Thy1.1 + ) and Ezh2 −/− (CD45.2 + , Thy1.2 + ) Pmel-1 cells were co-injected into non-irradiated B6/SJL mice (CD45.1 + , Thy1.2 + ), followed by infection with vaccinia virus encoding gp100 (VVA-gp100). This also allowed us to test a cell-autonomous effect of Ezh2 deficiency. Again, although loss of Ezh2 did not impair the initial proliferation in the spleen 3 days after immunization, it caused approximately 2-fold reduction in number by 5 days and 35 days (Supplementary Fig. 2a). Impaired expansion of Ezh2 −/− Pmel-1 cells was not the result of decreased proliferation but dependent on increased apoptosis ( Supplementary Fig. 2b, c). Ezh2 deficiency decreased the frequency of Pmel-1 cells producing high levels of IFN-γ 5 days and 35 days after immunization, without influencing IL-2 and granzyme B (GzmB) ( Supplementary Fig. 2d). Ex vivo culture of 35 day-memory T cells revealed that Ezh2 deficiency resulted in 5-fold less expansion of Pmel-1 cells upon rechallenging with gp100 ( Supplementary Fig. 2e). Furthermore, while gp100restimulation induced a 2-fold higher frequency of IFN-γproducing cells in WT 35 day-memory T cells compared to non-stimulation controls, it did not increase IFN-γ production by Ezh2 −/− memory T cells ( Supplementary Fig. 2f). Thus, Ezh2 is crucial for persistence of effector and memory T cells under either lymphopenic or lymphoreplete conditions. Ezh2 promotes memory precursor formation. A typical T-cell response contains three characteristic phases: clonal expansion, apoptotic contraction, and memory phase (Fig. 2a). We therefore used three different strategies to determine at which T-cell response phase(s) Ezh2 is required to support effector and memory T-cell persistence. During the initial phase, antigenactivated CD8 + T cells generate two subsets of memory precursors: CD44 hi CD62L hi central memory precursor CD8 + T cells (T CMP ) and CD44 hi CD62L lo T EFF . T CMP are less differentiated and have greater capability than T EFF to replicate and generate memory cells 2,3,5-7 . We characterized T CMP and T EFF from WT and Ezh2 −/− Pmel-1 cell recipients 4 days and 7 days after transfer into lymphodepleted mice. CD8 + T cells expressing high levels of KLRG1 (KLRG1 hi ) represent a terminally differentiated proliferating cells 3 . Ezh2 deficiency caused higher frequency of KLRG-1-expressing CD62L hi -and CD62L lo -T cells, which occurred 4 days after immunization and increased by 7 days (Fig.2b). This difference was even greater in PB (Fig. 2c). Ezh2 deficiency also induced a skewed differentiation of activated CD8 + T cells towards T EFF , as evidenced by increased fraction of T EFF and a corresponding reduction of both T CMP frequency and numbers throughout the expansion phase (Fig. 2d). Similar results were observed when responding cells were segregated into memory potential cells (MPC) and short-lived effector cells (SLECs) based on KLRG-1 and IL-7Rα expression, featured by increased percentage and number of SLECs at 4 days (Fig. 2d). Notably, while loss of Ezh2 did not affect the survival of T CMP at 4 days and 7 days, it caused enhanced apoptosis of T EFF early during expansion (4 days) (Fig. 2e). Since KLRG-1 normally is not expressed on the surface of WT CD62L hi CD8 + T cells 3,31 , ectopic expression of KLRG-1 on the surface of Ezh2-deficient CD62L hi CD8 + T cells probably indicates their precocious differentiation. Thus, Ezh2 is important for preserving the pool size of T CMP , primarily through a mechanism of restraining differentiation into SLECs. This was supported by the observation that both Ezh2 −/− T CMP and T EFF expressed 370-and 500-fold more transcripts, respectively, of the senescence gene p19 Arf than their WT counterparts (Fig. 2f)   Ezh2 is required for CD8 + T cells to control tumor growth and promote memory precursor formation. a WT and Ezh2 −/− naive Pmel-1 cells (1 × 10 6 , Thy1.1 + ) were transferred into sublethally irradiated (5 Gy) B6 mice (Thy1.2 + ) that had pre-established B16 melanoma, followed by treatment with IL-2 (1 × 10 5 IU per injection, i.p., twice a day) and gp100-pulsed DCs (gp100-DCs, 1 × 10 6 per mouse, i.p.) for 3 days. Tumor size was monitored over time. b-f WT and Ezh2 −/− T N Pmel-1 cells (1 × 10 6 , Thy1.1 + ) were transferred into sublethally irradiated non-tumor-bearing B6 mice, followed by immunization with IL-2 and gp100-DCs for 3d. b Donor T cells were collected from the spleen 4 days, 7 days, and 35 days after adoptive transfer. Plots and graphs show the frequency and numbers of donor T cells. c Percentage of IFN-γ-producing donor T cells in the spleen. d Donor T cells were collected from the spleen of WT and Ezh2 −/− Pmel-1 cell primary recipients 42 days after transfer, and separately transferred into sublethally irradiated secondary non-tumor-bearing B6 mice (4 × 10 4 cells per mouse), followed by treatment with IL-2 and gp100-DCs at 42 days, 43 days, and 44 days. By 49 days, donor T cells were collected from the spleen of the secondary mice. Plots and graph show the percentage of donor Pmel-1 cells. e Donor T cells derived from these secondary mice were activated with anti-CD3 Ab for 5 hrs to measure their production of IFN-γ. Graph shows the number of IFN-γ + Pmel-1 cells in the spleen. f Donor T cells collected at 49d from the secondary mice were cultured ex vivo with IL-7 + IL-15 in the presence or absence of gp100 for additional 5 days. Plots and graphs show the percentage of donor T cells in cultures. *p < 0.05, **p < 0.01, and ***p < 0.001 (two-tailed unpaired t test). The data are representatives of three independent experiments with n = 5 mice per group in each (a; mean ± SD), or two experiments (b-f, n = 3-6 mice per group in each, mean ± SD) NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02187-8 ARTICLE Ezh2-deficiency-mediated change in surface phenotype was largely associated with T-cell differentiation rather than activation and exhaustion markers (e.g., CD69, CD25, CD122, and PD-1) in these lymphodepleted hosts (Fig. 2g). Using lymphoreplete mice immunized by VVA-gp100, we validated the impact of Ezh2 deficiency on reducing T CMP and MPCs and increasing SLECs without changing the expression of CD122 and PD-1 during expansion phase ( Supplementary Fig. 3). To examine the cell autonomous effect of Ezh2 on endogenous CD8 + T-cell responses, we co-transferred equal amount of WT (Thy1.1 + CD45.2 + ) and Ezh2 −/− (Thy1.1 − CD45.2 + ) splenocytes into lymphoreplete B6/SJL mice (Thy1.1 − CD45.1 + ), followed by infection with vaccinia virus encoding OVA (VVA-OVA). As compared to WT CD8 + T cells, Ezh2 −/− CD8 + T cells produced 1.5-to 2-fold more OVA 257-264 -specific T cells 3 days after infection, maintained at 4 days, and dramatically declined by 5 days (Supplementary Fig. 4a, b). Notably, the increase of OVA 257-264 -specific Ezh2 −/− CD8 T cells 3 days after infection was associated with enhanced proliferation rates ( Supplementary  Fig. 4c), increases of KLRG1 hi cells ( Supplementary Fig. 4d, e) and decreased ratio of T CMP vs. T EFF (Supplementary Fig. 4f). Thus, loss of Ezh2 leads to an earlier occurrence of T-cell response peak, exaggerated terminal differentiation and loss of memory precursors.   were transferred into sublethally irradiated non-tumor-bearing B6 mice (Thy1.2 + ), followed by immediate treatment with IL-2 and gp100-DCs for 3 days. By 7 days after transfer, donor T CMP and T EFF were highly purified using FACS sorter, and transferred into sublethally irradiated secondary recipients that had been immunized with gp100-DCs 7 days earlier (described in Supplementary Fig.4). Forty-two days later, donor T cells were collected from the spleen of these secondary recipients (h), and further cultured ex vivo for additional 5 days (i). Plots and graphs show the frequency of donor T cells derived from the secondary recipients of WT and Ezh2 −/− T CMP and T EFF . *p < 0.05, **p < 0.01, and ***p < 0.01 (two-tailed unpaired t test). The data are representative of four independent experiments with n = 3 mice per group in each (b-g; mean ± SD) or two experiments with n = 4 mice per group in each (h, i, mean ± SD) The second phase is featured by massive apoptosis of effector cells and the transition of memory precursors into mature memory cells. To evaluate if Ezh2 deficiency affected this transition, we purified T CMP and T EFF from primary lymphodepleted recipients 7 days after transfer of Pmel-1 cells and immunization, and separately transferred them into gp100 immunization-matched secondary recipients ( Supplementary  Fig. 5). WT T CMP produced 6-fold more memory cells than WT T EFF 42 days after transfer (Fig. 2h), confirming that T CMP have greater ability than T EFF to produce memory cells 3,32 . Ezh2 −/− T EFF failed to produce detectable memory T cells. However, Ezh2 −/− T CMP produced similar percentages of memory-phenotype cells as did WT T CMP (Fig. 2h), but their progenies were unable to expand upon gp100 rechallenge, unlike those of WT T CMP (Fig. 2i). Thus, Ezh2 is dispensable for the homeostatic survival of T CMP during contraction phase, but is important for the transition of both T CMP and T EFF into mature memory T cells.
Increased apoptosis of Ezh2 −/− T cells during the effector phase might result in long-term consequences on recall response capacity during the final memory phase. To test this, we immunized lymphoreplete Ezh2 fl/fl mice with VVA-gp100 ( Supplementary Fig. 6a). Forty-two days later when gp100specific memory CD8 + T cells were formed (Supplementary Fig. 6b), we isolated splenocytes from these mice and treated them with TAT-Cre to delete Ezh2 ( Supplementary Fig. 6c), followed by culturing them ex vivo for 5 days, with or without gp100 addition. Deletion of Ezh2 by TAT-Cre dramatically decreased the capacity of these Ezh2 fl/fl memory CD8 + T cells to expand and produce IFN-γ upon gp100 rechallenge ( Supplementary Fig. 6d, e). These results indicate that Ezh2 is important for maintaining the recall response capacity of developed memory T cells.
Ezh2 orchestrates gene programs of memory properties. To identify the Ezh2-targeted genes associated with effector and memory differentiation, we performed RNA sequencing of WT and Ezh2 −/− Pmel-1 cells after TCR activation for 3 days. Ezh2 deficiency had minimal effect on gene expression in T N ( Fig. 3a; Supplementary Data 1). In contrast, TCR activation of Ezh2 −/− Pmel-1 cells led to the up-regulation of 279 genes and downregulation of 168 genes compared to their WT counterparts ( Fig. 3a; Supplementary Data 2). Genes altered in activated Ezh2 −/− Pmel-1 cells were associated with cellular proliferation, cell death and survival, as well as cell function and maintenance (Fig. 3b). Ezh2 deficiency upregulated Id2, Prdm1, and Eomes ( Fig. 3c), all critical for effector differentiation and functionality 1,3 , and decreased Id3, which controls memory formation [33][34][35] . RT-PCR analysis validated these changes (Fig. 3d, e). RNA-seq gene profiling confirmed the role of Ezh2 in restraining effector differentiation, as evidenced by upregulated effector molecules and chemokine receptors in TCR-activated Ezh2 −/− CD8 + T cells ( Supplementary Fig. 7a).
To assess the impact of Ezh2 deficiency on TF expression in memory precursors, we isolated T CMP and T EFF from WT and Ezh2 −/− Pmel-1 cell recipients 4 days after activation. As compared to WT T CMP , Ezh2 −/− T CMP had lower levels of Id3, but higher expression of Id2 and Eomes ( Supplementary Fig. 7b). In T EFF , Ezh2 deficiency led to increased expression of Id2, Eomes, and Prdm1 ( Supplementary Fig. 7b). We performed chromatin immunoprecipitation (ChIP) analysis and observed that Ezh2 bound to the promoter regions of these gene loci ( Supplementary Fig. 7c). This was confirmed using chromatin from activated Ezh2 −/− Pmel-1 cells 3 days after activation as evidenced by decreased amount of Ezh2 and H3K27me3 at the promoter region of these gene loci ( Supplementary Fig. 7d). Ezh2 appeared to have differential effects on Tbx21 (which encodes Tbet) expression between T CMP vs. T EFF via an unknown mechanism ( Supplementary Fig. 7b, c). Since CD8 + T cells that express high levels of Prdm1, Id2 and Tbx21 but low levels of Id3 are reported to undergo accelerated and enhanced terminal differentiation; 1,3,33,34 we propose that Ezh2 orchestrates the expression of these TFs for controlling stepwise effector differentiation and memory formation of CD8 + T cells.
Phosphorylation of Ezh2 impairs the maintenance of memory precursors. During the immune response, normal CD8 + T cells express high levels of Ezh2 upon antigen activation, however, they still undergo a "programmed" differentiation into terminal T EFF . This points to a mechanism that modifies Ezh2 function in T cells, reducing Ezh2 regulation of memory T cells. We found that TCR-activated Pmel-1 cells expressed 42-and 23-fold higher Ezh2 protein 3 days and 7 days after culture, respectively, than T N (Fig. 4a). However, as compared to T N , activated CD8 + T cells showed a profound decrease in Ezh2 function, as evidenced by the fact that 3 day-and 7 day-Pmel-1 cells contained 2-and 5fold less H3K27me3, respectively (Fig. 4b), increase of Ezh2silenced genes Id2, Eomes and Prdm1 while decreasing Ezh2activated gene Id3 (Fig. 4c), and impaired antitumor activity (Fig. 4d). ChIP analysis showed that in CD8 + T N , Ezh2 bound to the promoter region of Id3, Id2, Prdm1, and Eomes (Fig. 4e). However, the amount of Ezh2 within these promoter regions was dramatically decreased in proliferating CD8 + T cells, which occurred 3 days after activation and persisted throughout 7 days (Fig. 4f). This decreased presence of Ezh2 was paralleled by a reduction of H3K27me3 at the Prdm1 and Eomes loci (Fig. 4g). Thus, Ezh2 is dissociated from the promoter regions of these TFs as early as 3 days after activation.
To determine if T CMP and T EFF differentially modify Ezh2 function, we recovered them from WT Pmel-1-cell recipients 7 days after activation. As compared to T N , T EFF expressed higher levels of Ezh2 ( Supplementary Fig. 8a), upregulated Id2, Prdm1 and Eomes, and decreased Id3 ( Supplementary Fig. 8b), and reduced binding of Ezh2 at the regulatory regions of Id3, Prdm1 and Eomes ( Supplementary Fig. 8c). T CMP decreased the amount of Ezh2 within the promoters of Id3, Eomes and Prdm1 compared to T N , but they retained more Ezh2 at the promoter regions of these gene loci than T EFF (Supplementary Fig. 8c). This correlated with higher levels of Id3 transcripts but lower transcription of Prdm1 and Eomes in T CMP than T EFF (Supplementary Fig. 8b). Thus, T EFF have more dramatic reduction of Ezh2 function compared to T CMP .
Phosphatidylinositol-3-kinase (PI3K)/Akt is important for Tcell proliferation and differentiation 13 . In cancer cells, Akt phosphorylates Ezh2 at serine 21 (pEzh2 S21 ), thereby suppressing Ezh2 enzymatic activity 36 . We assessed if Akt phosphorylation of Ezh2 might induce the dissociation of Ezh2 from these TF loci in proliferating T cells. As compared to T N , T CMP and T EFF from B6 mice that received Pmel-1 cells 7 days earlier had 8.3-fold and 3.7-fold more pEzh2 S21 , respectively ( Supplementary Fig. 8d). This correlated with decreased H3K27me3 and increased pAkt ( Supplementary Fig. 8a, d). pEzh2 S21 occurred in TCR-activated CD8 + T cells 3 days after culture. By 5 days and 7 days, activated cells expressed 2.3-fold and 8.4-fold more pEzh2 S21 than T N (Fig. 5a). Treatment with MK2206, an allosteric inhibitor of Akt, led to a marked decrease of pEzh2 S21 and increase of H3K27me3 in proliferating CD8 + T cells (Fig. 5b-d). Inhibiting Akt or its upstream activator PI3K by PI103 likewise upregulated Id3 expression while reducing Id2, Eomes and Prdm1 (Fig. 5e), affirming an orchestrated regulatory pathway across these TFs. Inhibiting Akt activity also increased the amount of Ezh2 and H3K27me3 at the promoter regions of Id3, Id2, Eomes, and Prdm1 loci (Fig. 5f, g). Rapamycin inhibition of mTOR, a downstream effector of Akt pathway 13,37 , increased Eomes expression and reduced Id3 (Fig. 5e), suggesting that the mechanism of Ezh2 inhibition by Akt differs from mTORdependent effects.
To establish a specific effect of Akt-mediated phosphorylation on Ezh2, we infected Pmel-1 cells with retrovirus encoding Aktphosphorylation-resistant Ezh2 in which the serine at amino acid 21 was replaced by alanine (named Ezh2 S21A ). Pmel-1 cells expressing Ezh2 S21A increased H3K27me3 compared to Ezh2 control (Fig. 5h), upregulated Id3 transcript and repressed Eomes and Prdm1 expression (Fig. 5i), and increased Ezh2 binding at Id3, Id2, Eomes and Prdm1 loci (Fig. 5j). Notably, introduction of Ezh2 S21A increased the H3K27me3 level within the Eomes and Prdm1 loci but not the Id3 and Id2 loci (Fig. 5k), suggesting that epigenetic regulation of these loci occurs via a different mechanism. Using lymporeplete hosts infected by VVA-gp100, we confirmed that WT Pmel-1 cells induced Ezh2 and pEzh2 S21 3 days after activation and markedly increased by 5 days (Supplementary Fig. 9a, b), which was inversely correlated to the decreased cellular H3K27me3 (Supplementary Fig. 9c). In aggregate, Akt activation profoundly reduces Ezh2 function in activated CD8 + T cells.
To define the impact of Ezh2 phosphorylation by Akt in CD8 + T cells, we made a retrovirus encoding a phosphomimetic Ezh2, in which serine 21 was replaced by aspartate (named Ezh2 S21D ) 36 . We separately introduced Ezh2 S21A and Ezh2 S21D into activated Ezh2 −/− Pmel-1 cells to assess their specific effects in T cells without endogenous Ezh2 (Fig. 6a). As compared to Ezh2 S21D , Ezh2 S21A increased H3K27me3 in activated CD8 + T cells (Fig. 6b), and induced high levels of Id3 but reduced Eomes and Prdm1 transcripts in these T cells 7 days after activation (Fig. 6c). Upon transfer into lymphopenic B6 mice and treatment with IL-2 and gp100-DCs, Ezh2 −/− Pmel-1 cells overexpressing Ezh2 S21A produced more donor T cells (Figs. 6d) and 6-fold more Id3 hi donor cells, compared to Ezh2 S21D 6d after transfer (Fig. 6e). Overexpressing Ezh2 S21A caused 3-fold smaller in frequency of KLRG1 hi T cells than Ezh2 S21D , without changing the fraction of CD62L hi cells and IFN-γ-producing cells (Fig. 6e). As compared to Ezh2 S21A , overexpressing WT Ezh2 in Ezh2 −/− Pmel-1 cells was able to rescue their survival in vivo (Fig.6d), but less effective in modifying the expression of Id3, Prdm1 and Eomes (Fig.6c) and unable to sustain Id3 hi T cells (Fig.6e). These investigations   Fig.3 Ezh2 orchestrates the expression of genes critical for effector differentiation and memory formation. a-c WT and Ezh2 −/− Pmel-1 cells were cultured in the presence of anti-CD3/CD28 Abs and IL-2 for 3 days. Cells were collected to extract RNA for sequencing. Using one-way ANOVA analysis, we selected transcripts with p < 0.01 and q < 0.01 for comparing paired groups and at least a 1.5-fold difference from the means for the paired groups. a Venn diagram shows the number of differentially expressed genes. b Table shows the signaling pathways most regulated by Ezh2 in activated Pmel-1 cells, which was identified by INGENUITY pathway analysis. c Heat map shows the cluster of genes associated with Cell Function and Maintenance in activated WT and Ezh2 −/− Pmel-1 cells. d, e Gene expression was assessed by real-time RT-PCR. *p < 0.05, and ***p < 0.001 (two-tailed unpaired t test). The data are representative of four independent experiments (d, e; mean ± SD) identify Akt-mediated phosphorylation of Ezh2 as a novel and important mechanism regulating effector differentiation and memory formation, beyond T-cell survival.
Akt-insensitive Ezh2 augments antigen-specific T-cell antitumor efficacy. ACT for cancer requires sufficient amplification and persistence of tumor-specific T cells to eradicate the tumor 4,6,7,11 . Current in vitro methods to expand cells to sufficient numbers impair the maintenance of memory properties in cultured T cells 6,7 . Ex vivo treatment of expanding CD8 + T cells with an Akt inhibitor led to increase of memory-phenotype cells and antitumor immunity 38,39 . Indeed, transfer of Pmel-1 cells treated by MK2206 during ex vivo culture induced greater antitumor activity than untreated Pmel-1 cells, but was less potent than Ezh2 S21A -transduced Pmel-1 cells (Fig. 7a). As compared to MK2206-treated Pmel-1 cell recipients, Ezh2 S21A -Pmel-1 cell recipients generated higher frequency of total Pmel-1 cells and Id3 hi Pmel-1 cells, but a lower frequency of KLRG1 hi cells in the spleen 9 days after transfer (Fig. 7b).
Enhanced tumor immunity by Ezh2 S21A -Pmel-1 cells could potentially result from increased expression of overall Ezh2 protein. To test it, we assessed whether overexpressing normal Ezh2 in Pmel-1 cells, which is susceptible to Akt phosphorylation, may influence their antitumor activity. Transfer of Pmel-1 cells expressing Ezh2 S21A had dramatically enhanced capacity to inhibit tumor growth compared to either Ezh2 or GFP control (Fig. 7c). Ezh2 S21A -, Ezh2-and GFP-Pmel-1 cell recipients of B16 melanoma had similar amounts of donor T cells in PB within 7d after transfer. Over time, Ezh2 S21A -Pmel-1 cell recipients produced approximately 10-fold more in frequency of donor T cells than either Ezh2-or GFP-Pmel-1 cell recipients between 10 and 14 days (Fig. 7d). Upon rechallenge with gp100-DCs 18 days after transfer, Ezh2 S21A -Pmel-1 cell recipients had~4-fold more circulating Pmel-1 cells than Ezh2-and GFP-Pmel-1 cell recipients by 24 days (Fig. 7d), demonstrating an enhanced recall response. We also found that ectopic expression of Ezh2 S21A induced a lower frequency of KLRG1 hi cells in vivo 10 days after transfer compared to Ezh2 and GFP controls (Fig. 7e). Analysis of tumor infiltrating lymphocytes (TILs) showed that as compared to GFP-and Ezh2-Pmel-1 cell recipients, Ezh2 S21A -Pmel-1 cell recipients had 1.5-fold higher frequency of TILs (Figs. 7f), 5-fold more total GFP + TILs and 4-fold more IFN-γ-producing GFP + TILs (Fig. 7g). Overexpression of Ezh2 S21A induced 2-fold and 4fold more CD62L hi CD8 + T cells in the spleen and tumor, respectively, than GFP-and Ezh2-Pmel-1 cell recipients, but did not prevent Pmel-1 cells from differentiating into CD62L lo T EFF in vivo (Fig. 7h). Thus, overexpressing Ezh2 S21A dramatically improves the persistence of tumor-reactive CD8 + T cells in vivo upon chronic exposure to tumor antigen.
However, neither Id3-nor Ezh2-overexpression completely restored the capability of Ezh2 −/− Pmel-1 cells to produce memory cells like GFP-transduced WT Pmel-1 controls (Fig. 8a-c). Since Ezh2 was critical for generating memory precursors within 5 days of activation ( Fig. 2 and Supplementary  Fig. 3), and since retrovirally introduced genes did not reach peak expression by 5 days after T-cell activation 23 , one plausible explanation for the partial rescue effect of Id3 is its delayed expression in Ezh2 −/− Pmel-1 cells.
To assess the direct impact of Ezh2 on Id3 activation, we cloned the promoter region of the Id3-gene (ranging from −1.5 kb upstream of the transcription start site (TSS) to + 0.5Kb downstream of the TSS) into the pGL3 luciferase reporter vector and constructed an Id3-specific pGL3 reporter (Id3-pGL). Overexpressing Ezh2 S21A , but not Ezh2 S21D , activated Id3 transcription (Fig. 8d). To evaluate the effect of Ezh2 enzymatic activity on Id3 transcription, we deleted endogenous Ezh2 in 3T3 cells using CRISPR (Fig. 8e), followed by viral transduction of various Ezh2 mutants. Mutant Ezh2 S21A had a greater capacity than Ezh2 to   Fig. 5 Phosphorylation of Ezh2 by Akt dissociates Ezh2 from the promoter regions of major TF loci. a WT Pmel-1 cells were stimulated with anti-CD3/ CD28 Ab + IL-2. Immunoblot analysis of Pmel-1 cells stimulated in vitro for 3 days, 5 days, and 7 days, probed with Abs against Ezh2 and phosphorylated Ezh2 S21. Unstimulated T N were used as control. b-d Immunoblot analysis of Pmel-1 CD8 + T cells stimulated in vitro for 7 days, with or without treatment of MK2206, probed with indicated Abs. e-g WT Pmel-1 cells were cultured with anti-CD3/CD28 Ab + IL-2, with or without treatment of PI103, MK2206 or rapamycin for 7 days. Real-time RT-PCR analysis of Ezh2-targeted genes (e). ChIP analysis of cultured Pmel-1 cells treated with PI103, or MK2206. Graphs show the deposition of Ezh2 (f) and H3K27me3 (g) at the promoter regions of Id3, Id2, Prdm1, and Eomes. h-k WT Pmel-1 cells were stimulated with anti-CD3/CD28 Ab for 36 h, followed by infection with MigR1 retrovirus (GFP) or MigR1 retrovirus encoding flag-tagged Ezh2, or Ezh2 S21A . Pmel-1 cells were collected at 7 days after culture. Immunoblot analysis of GFP and Ezh2 S21A Pmel-1 cells, probed with indicated Abs (h). Real-time RT-PCR analysis of their expression of major TFs (i). ChIP analysis shows the deposition of Ezh2 (j) and H3K27me3 (k) at the promoter regions of these major TFs loci. *p < 0.05, **p < 0.01, and ***p < 0.001 (two-tailed unpaired t test). The data are representatives of three independent experiments (a-d), or two experiments (e-k; mean ± SD) activate Id3, whereas the loss-of-function mutant Ezh2 H689A 40 failed to activate Id3 (Fig. 8f), confirming that Ezh2 requires its methyltranferase activity to activate Id3 transcription. To understand the mechanism by which Ezh2 stimulated Id3 transcription, we assessed if Ezh2 influenced histone modification states at the promoter region of Id3. Recent studies demonstrate that TFs (e.g., Tbx21, Prdm1, Eomes and Irf4) are bivalent for both H3K27me3 + and H3K4me3 + histone methylation under naive state 18 . We observed that the Id3 locus was marked by both H3K27me3 and H3K4me3 in CD8 + T N (Fig. 8g). TCR-activated CD8 + T cells reduced H3K4me3 and Pol II at Id3 locus upon differentiation (Fig. 8g). Loss of Ezh2 led to decreased deposition of H3K4me3 and Pol II at Id3 locus, but not Id2 locus (Fig. 8h). Thus, Ezh2 activation of Id3 transcription in T cells likely involves the enzyme(s) that catalyzes H3K4me3. Akt phosphorylation of Ezh2 leads to dissociation of Ezh2 from the Id3 locus and subsequent resolution into non-permissive state. This is in contrast to a recent study showing that in cancer cells, Aktinduced pEzh2 S21 activates those genes lacking H3K27me3 modifications 24 .

Discussion
Understanding the epigenetic mechanisms that mediate continual replication of antigen-driven T cells without senescence during reaction to chronic infection and cancer may lead to new strategies to improve the efficacy of ACT 4,7,8,11,32,[41][42][43] . Minimally differentiated CD8 + T cells, including T N and T CMP , exhibit superior persistence and enhanced antitumor activity compared with differentiated T EFF 4,6,10,44 . The goal therefore is to generate cellular products rich in less-differentiated memory precursors, however, the mechanisms that restrain effector differentiation and maintain memory potential remains poorly understood. Tcell signal strength, which determines the quantity and quality of T-cell responses and includes signals from the TCR, costimulatory molecules and cytokines, is believed to be epigenetically controlled. However, the epigenetic regulator(s), responsible for converting these signals into the transcriptional programs that generate and maintain memory precursors, has not been fully identified 2,45-50 . Thus, our observations that Ezh2 regulates effector, memory precursor and antigen-specific recall response have profound implications towards the development of new   Fig. 6 Inhibiting Akt-mediated phosphorylation of Ezh2 enhances the generation of T CMP . Ezh2 −/− Pmel-1 cells (Thy1.1 + ) were stimulated with anti-CD3/ CD28 Ab + IL-2 for 36 h, followed by infection with MigR1 retrovirus encoding GFP, Ezh2, Ezh2 S21D and Ezh2 S21A , respectively. At 7 days, after stimulation, these infected T cells were collected. a Experiment scheme. b Immunoblot analysis of Pmel-1 cells transduced with GFP, Ezh2, Ezh2 S21D and Ezh2 S21A , probed with Flag, H3K27me3 and H3. c RT-PCR analysis of gene expression in Pmel-1 cells transduced by indicated genes. d, e These 7 day-Pmel-1 cells (Thy1.1 + ) were transferred into sublethally irradiated B6 mice (Thy1.2 + ), followed by treatment with IL-2 and gp100-DCs for 3d after transfer. d Donor T cells were collected from the spleen, PB and LN at 6d after adoptive transfer, and measured by flow cytometric analysis. e Plots and graphs show the percentage of of Id3 hi , KLRG1 hi , CD62L + and IFN-γ + within donor T cells from the spleen. *p < 0.05, **p < 0.01, and ***p < 0.001 (two-tailed unpaired t test). The data are representatives of three experiments (b, c; mean ± SD) or two experiments with n = 4 mice per group in each (d, e; mean ± SD) NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02187-8 ARTICLE strategies to optimize the expansion and quality of therapeutic T cells for ACT. Histone modifications occur in T cells early after antigen priming [17][18][19] . Recent studies suggested that CD8 + T-cells decreased the amount of H3K27me3 at the promoter regions of genes associated with cell proliferation and differentiation within 24 h 18 . We found that 3 days after TCR engagement Ezh2 was highly induced in CD8 + T cells, however, its function on activating Id3 and repressing Id2, Eomes and Prdm1 was decreased. It has been shown that low expression of Id3 and high levels of Id2, Blimp-1 and Eomes are associated with enhanced effector differentiation but decreased memory potential 1,3,5,33,34,51,52 . In line with these observations, our data suggested that through coordinating the expression of these major TFs in activated CD8 + T cells, Ezh2 played essential roles in preserving T CMP and preventing precocious effector differentiation. We also observed that Ezh2 directly bound to the promoter region of Id3 and activated Id3 transcription in a dose-dependent manner. Introduction of Id3 into activated CD8 + T cells lacking Ezh2 rescued their recall response capacity for producing effector cells upon antigen reencounter. These data identify the importance of Ezh2 in regulating Id3-mediated memory T-cell development and function.
The results from our studies establish that Akt-mediated Ezh2 phosphorylation as a critical target for optimizing the expansion and quality of therapeutic T cells used for treating cancer and chronic infections. The capability of antigen-specific T cells to replicate and persist in vivo is crucial for controlling tumor growth and eliminating virus-infected cells 4,7,8,11,32,[41][42][43] . In support of this, we have identified that Akt-unphosphorylated Ezh2-mediated epigenetic effects are involved in regulating the quantity and quality of memory precursors. These data support the observation that ex vivo treatment of expanding CD8 + T cells   with certain Akt inhibitors led to increased frequency of memory precursors and improved antitumor immunity in vivo 38,39 . However, ex vivo Akt inhibitor treatment resulted in a transient effect and did not prevent reactivation of Akt in CD8 + T cells in vivo upon antigen rechallenge. When signals activating Akt persist, the prolonged presence of Akt-insensitive Ezh2 in tumorreactive CD8 + T cells is required for them to preserve their memory properties in vivo. Continual treatment with Akt inhibitors in vivo might extend its effect on promoting memory T cells. However, effective cancer immunotherapy requires collective efforts of effector and memory T cells 4,6,11 . Systemic administration of Akt inhibitors may potentially inhibit effector responses and cause adverse effects. Future investigations will be identifying new pharmacological approaches that can selectively block the Ezh2 and Akt interaction in T cells for improving antitumor efficacy.
Our findings together with Kakaradov's studies 53 also indicate that the impact of Ezh2 on antigen-driven CD8 + T cells may vary at different activation state and differentiation stage. Ezh2 promoted the survival and expansion of activated CD8 + T cells later after antigen priming. By 5 days after activation, Ezh2 deficiency selectively increased apoptosis in T EFF rather than T CM -like cells 53 . However, early after antigen activation (e.g., by 3 days) Ezh2 was dispensable for survival of antigen-primed CD8 + T cells 53 . We also identified that loss of Ezh2 caused preferential decrease of T CMP pool independent of cell apoptosis early during expansion, while markedly increasing the fraction of terminal T EFF . Since both the first division of activated CD8 + T cells and their subsequent differentiation may influence the ratio of T CMP and T EFF 46,47,53,54 , we propose that Ezh2 is important for preserving the T CMP pool and restraining precocious terminal differentiation. Interestingly, Kakaradov's studies showed that despite the presence of Ezh2, about 75% of P14 CD8 + T cells had undergone terminal differentiation (e.g., expression of KLRG1 hi ) at 4 days after LCMV infection and only 1% of activated P14 CD8 + T cells were T CMP -like phenotype 7 days after infection 53 . In contrast, we found that WT Pmel-1 cells consisted of <10% KLRG1 hi cells and~50% T CMP at the peak of effector response. Loss of Ezh2 led to production of 4-fold more terminal KLRG1 hi cells, accompanied with increased expression of 370-fold more p19 Arf transcript. These differences between our studies and Kakaradov's are likely explained by different experimental systems and the possibility that LCMV-induced excessive terminal differentiation could mask certain impact of Ezh2 in CD8 + T cells. However, both studies support previous observations that T N are able to develop into memory cells without transitioning through an effector stage [54][55][56][57][58][59][60][61][62][63] , in which Ezh2 plays a crucial role.
In conclusion, Ezh2 plays multiple roles in antigen-driven CD8 + T-cell responses (Fig. 8i). Ezh2 promotes the survival and expansion of proliferating effector cells. Ezh2 also functions as a key molecular gatekeeper for generating CD8 + T memory precursors, restraining terminal differentiation and maintaining antigen-specific recall response capability. Ezh2 activates Id3 to regulate the persistence and function of effector and memory CD8 + T cells. Ezh2 also silences Id2, Prdm1 and Eomes to temper effector differentiation. Importantly, Ezh2 itself is functionally modified by Akt phosphorylation, which diminishes the capacity for Ezh2 to stimulate Id3 and silence Id2, Prdm1 and Eomes, thereby driving effector differentiation and reducing memory potential. Our findings also open new perspective to understand how extrinsic signals from environmental cues may influence T-cell immunity via modulating Ezh2 function.
Ezh2 was deleted in gp100-specific CD8 + T-cell receptor -transgenic Pmel-1 cells by backcrossing Ezh2 fl/fl CD4-Cre B6 mice to Pmel-1 mice to produce T-cellspecific Ezh2-knockout Pmel-1 mice (Ezh2 −/− Pmel-1). Both female and male mice (6-12 weeks of age) were used for experiment. Experimental protocols were approved by the Temple University's Committee on Use and Care of Animals, and the University of Michigan Committee on Use and Care of Animals.
Cell preparation and culture. CD44 lo CD8 + T N cells were isolated from spleen and LN (purity usually 90-95%). T cells were cultured in the presence of anti-CD3 (2 μg/ml; BD bioscience) and anti-CD28 (2 μg/ml; BD bioscience) Abs together with recombinant human IL-2 (10 ng/ml; R&D Systems). In vivo sorted T cells and cultured T cells were restimulated with anti-CD3 Ab (1 μg/ml) or human gp100 (10 −6 M) for 5 h before intracellular staining. BM-derived c-kit + cells were used to generate DCs. For TAT-Cre experiments, 10 6 purified CD8 + T cells or splenocytes were incubated in 100 μl serum-free RPMI containing 90-100 μg/ml TAT-Cre fusion protein (a gift from Dr. Warren Pear, University of Pennsylvania) for 16 h at 37°C. After extensive washing, treated cells were kept in vitro culture for another 5 days.
Retroviral construction and T-cell infection. MigR1 retroviral vector was provided by Dr. Warren Pear (University of Pennsylvania). Ezh2, Ezh2 mutants, and Id3 cDNA was cloned into MigR1 (GFP) vector. For retroviral infection, CD8 + T cells were prestimulated with anti-CD3/CD28 Ab for 24 h, and then the retrovirus supernatant was added in the presence of 8 µg/ml polybrene (Sigma). Cells were spinoculated at 3000 r.p.m, 32°C for 3 h. The same retroviral infection procedure was repeated 24 h later 23 .
Counting of adoptively transferred cells. Mice were killed after infection. Samples were enriched for mononuclear cells or CD8 + T cells (Miltenyi), and cells were counted by trypan blue exclusion. The frequency of transferred T cells was determined by measurement of the expression of CD8 and Thy1.1 by flow cytometry. The absolute number of Pmel-1 cells was calculated by multiplication of the total cell count with the percentage of CD8 + Thy1.1.
RNA-Seq analysis. RNA sequencing was carried out at the sequencing core at University of Michigan Sequencing Core (Ann Arbor, Michigan). Transcriptome analysis was performed on RNA isolated from fresh naive and cultured CD8 + T cells. Briefly, total RNA was isolated from T cells using an RNAeasy kit (QIA-GEN) and RNA-seq libraries were prepared using SureSelect RNA Library Preparation kits (Agilent Technologies). Samples were run on a HiSeq 2000 sequencing system (Illumina), and at least 37.5 × 10 6 single-end reads were obtained per sample. Expression was evaluated by determining the fragment per kilobase per million reads values. Using one-way ANOVA analysis, we selected transcripts with p < 0.01 and q < 0.01 for comparing paired groups and at least a 1.5-fold difference from the means for the paired groups. RNAseq data were deposited in the NCBI's Gene Expression Omnibus database (accession no. GSE76755).
Reverse transcription polymerase chain reaction (RT-PCR). RNA was isolated with an RNeasy Mini kit (Qiagen) and cDNA was generated by reverse transcription (Applied Biosystems). Real-time RT-PCR was performed with a SYBR green PCR mix (ABI Biosystems) in the Realplex Eppendorf Real-time PCR instrument (Eppendorf AG). Gene expression levels were calculated relative to the 18s gene. Data were collected and quantitatively analyzed on a Realplex sequence detection system (Eppendorf AG), and Applied Biosystems StepOne Plus Real-time PCR systems (Applied Biosystems). The primer sequences are listed in Supplementary Table 1.
ChIP. A Millipore ChIP kit and Diagenode ChIP kit was used for ChIP assay. DNA-protein complexes were crosslinked with formaldehyde at a final concentration of 1%. Sonicated extracts were precleared and incubated with Abs specific to Ezh2, H3K27me3, or nonspecific anti-IgG. The immunoprecipitated DNA was quantified by real-time quantitative PCR. The primer sequences are listed in Supplementary Table 2. The Ab information is listed in Supplementary  Table 3.
Western blot analysis. Cell lysates were prepared in lysis buffer (150 mM NaCl, 50 mM Tris-Cl, 1% Triton X-100, 0.1% SDS, 0.5% sodium deoxycholate, 0.02% sodium azide, 1 mM sodium vanadate, protease inhibitors (10 μg/ml leupeptin, 10 μg/ml aprotinin, and 1 mM phenylmethylsulfonyl fluoride)). Protein concentrations were determined using a Bio-Rad protein assay. Ten to 50 μg protein was loaded in 4-12% Mini-PrROTEAN TGX Precast protein Gels (Bio-rad), and transferred to PVDF membrane. Membranes were blocked in 3% BSA for 1 h, incubated with the primary antibody overnight at 4°C, washed, and incubated with the appropriate horseradish-conjugated secondary antibody for 1 h at room temperature. An enhanced chemiluminescent (ECL) kit was used to visualize the signal (Thermo Fisher). The Abs used for western blot analysis are listed in Supplementary Table 3. Uncropped blots for western blot analysis are shown in Supplementary Fig. 11.
Flow cytometric analysis and cell lines. The Abs used for flow cytometric analyses were purchased from eBioscience, BioLegend, or BD Biosciences. The Ab information is listed in Supplementary Table 3. Flow cytometric analyses were performed with FACS LSRII (BD Biosciences) as described 64,65 . B16 mouse melanoma cell line was purchased from American Type Culture Collection, and were cultured in RPMI-1640 medium (Gibco) supplemented with 10% FBS. No mycoplasma contamination was detected in any of the cultures using a mycoplasma detection kit.
Statistical analyses. Unless otherwise specified, statistical tests were performed using unpaired two-tailed Student's t test. Where necessary, the Shapiro-Wilk test was used to test for normality of the underlying sample distribution. No blinding was done, as objective quantitative assays such as flow cytometry, were used. Experimental sample sizes were chosen using power calculations with preliminary experiments and/or were based on previously described variability in similar experiments 30,45,50,[65][66][67] . Samples that had undergone technical failure during processing were excluded from analyses. Where relevant, recipient mice were randomized before adoptive transfer. p values of 0.05 or less were considered significant.
Data availability. The RNA-seq data has been deposited in National Center for Biotechnology Information Gene Expression Omnibus (GEO)database (https:// www.ncbi.nlm.nih.gov/geo) under the accession code GSE76755. All other materials are available from the corresponding authors upon reasonable request.