Polycomb Repressive Complex 2-Mediated Chromatin Repression Guides Effector CD8 + T Cell Terminal Differentiation and Loss of Multipotency

Transcription

1 Article Polycomb Repressive Complex 2-Mediated Chromatin Repression Guides Effector CD8 + T Cell Terminal Differentiation and Loss of Multipotency Graphical Abstract Authors Simon M. Gray, Robert A. Amezquita, Tianxia Guan, Steven H. Kleinstein, Susan M. Kaech Correspondence simon.gray@yale.edu (S.M.G.), susan.kaech@yale.edu (S.M.K.) In Brief Cytotoxic CD8 + T cells either terminally differentiate and die or form a rapidly responding population of memory T cells after pathogen clearance. Gray et al. define a temporal model for how effector T cells lose memory cell potential through selective epigenetic silencing of promemory genes. Highlights d d H3K27me3 is more abundant at certain pro-memory genes in TE CD8 + T cells PRC2 is required for CD8 + T cell clonal expansion and TE differentiation Accession Numbers GSE72408 SRA SRP d d H3K27me3 is deposited during late effector CD8 + T cell differentiation FOXO1 regulates H3K27me3 deposition at certain promemory loci in TE cells Gray et al., 2017, Immunity 46, April 18, 2017 ª 2017 Elsevier Inc.

2 Immunity Article Polycomb Repressive Complex 2-Mediated Chromatin Repression Guides Effector CD8 + T Cell Terminal Differentiation and Loss of Multipotency Simon M. Gray, 1,4, * Robert A. Amezquita, 1,4 Tianxia Guan, 1 Steven H. Kleinstein, 1,2,3 and Susan M. Kaech 1,5, * 1 Department of Immunobiology, Yale University School of Medicine, New Haven, CT 06510, USA 2 Interdepartmental Program in Computational Biology and Bioinformatics, Yale University School of Medicine, New Haven, CT 06510, USA 3 Department of Pathology, Yale University School of Medicine, New Haven, CT 06510, USA 4 These authors contributed equally 5 Lead Contact *Correspondence: simon.gray@yale.edu (S.M.G.), susan.kaech@yale.edu (S.M.K.) SUMMARY Understanding immunological memory formation depends on elucidating how multipotent memory precursor (MP) cells maintain developmental plasticity and longevity to provide long-term immunity while other effector cells develop into terminally differentiated effector (TE) cells with limited survival. Profiling active (H3K27ac) and repressed (H3K27me3) chromatin in naive, MP, and TE CD8 + T cells during viral infection revealed increased H3K27me3 deposition at numerous pro-memory and pro-survival genes in TE relative to MP cells, indicative of fate restriction, but permissive chromatin at both pro-memory and pro-effector genes in MP cells, indicative of multipotency. Polycomb repressive complex 2 deficiency impaired clonal expansion and TE cell differentiation, but minimally impacted CD8 + memory T cell maturation. Abundant H3K27me3 deposition at pro-memory genes occurred late during TE cell development, probably from diminished transcription factor FOXO1 expression. These results outline a temporal model for loss of memory cell potential through selective epigenetic silencing of pro-memory genes in effector T cells. INTRODUCTION Cytotoxic CD8 + T cells help clear intracellular bacterial and viral infections and protect against future infections by forming a long-lived, rapidly responding population of memory T cells. During an acute viral infection, the pathogen-specific CD8 + T cells develop into a heterogeneous population of cytotoxic effector and memory T cells, comprised of subsets discernable by their phenotypes, functions, anatomical locations, and longterm fates (Kaech and Cui, 2012). Many anti-viral effector CD8 + T cells differentiate into terminal effector (TE) cells (distinguished by stable expression of killer cell lectin like receptor G1 [KLRG1] and repression of interleukin 7 receptor a [IL-7Ra] [KLRG1 hi IL- 7R lo ]) that migrate into the periphery as interferon-g (IFN-g)-producing cytotoxic T lymphocytes (CTLs) (Kaech and Cui, 2012). However, the TE cells have the least memory cell potential and display the greatest rates of contraction after infection. TE cells are considered terminally differentiated because they proliferate poorly in response to homeostatic cytokines (e.g., IL-15) or antigen and maintain their phenotypic and functional properties upon restimulation (i.e., they do not adopt properties of other effector subsets) (Joshi et al., 2007). In contrast, subsets of memory precursor (MP) effector cells (distinguished by higher expression of IL-7Ra [IL-7R hi ]) are intrinsically more fit to persist and self-renew (Joshi et al., 2007; Best et al., 2013; Sarkar et al., 2008). The MP cells are also multipotent, developing into diverse types of memory CD8 + T cells (e.g., central, effector, and resident memory T cells) that form recalled effector cells upon secondary infection (Sarkar et al., 2008; Mackay et al., 2013; Joshi et al., 2007). Given that these CD8 + T cell fate decisions determine the quantity and quality of immunological memory that forms after vaccination and infection, it is critical to understand how and when they are specified. Temporal regulation of effector CD8 + T cell differentiation and the factors involved have begun to be delineated. Even as early as the first cell division, asymmetric partitioning of the Mechanistic Target of Rapamycin Complex 1 (mtorc1) and 2 and the transcription factors (TFs) c-myc and T-bet biases the daughters cells toward effector or memory cell fates (Chang et al., 2011; Verbist et al., 2016; Pollizzi et al., 2016). Other studies have shown that commitment to a TE cell fate can be visualized in early effector CD8 + T cells 3.5 to 4.5 days postinfection (p.i.) by increased expression of KLRG1 or the interleukin 2 receptor a (IL-2Ra) chain or reduced expression of the transcription factor ID3 (see references within Kaech and Cui, 2012; Best et al., 2013; Kalia et al., 2010). The KLRG1 lo early effector cells remain uncommitted at this stage and continue to give rise to KLRG1 hi IL-7R lo TE cells as well as KLRG1 hi IL-7R hi and KLRG1 lo IL-7R hi cells that differ in their long-term fates and the types of memory cells they form (Kaech and Cui, 2012). CD62L hi central memory (Tcm) and CD103 hi tissue-resident memory (Trm) cells progressively form several weeks after infection, demonstrating that effector and memory CD8 + T cell specification is dynamic (Mackay et al., 2013; Kaech and Cui, 2012). Further, recent work has shown that these memory cells have distinct roles in immune surveillance 596 Immunity 46, , April 18, 2017 ª 2017 Elsevier Inc.

3 and homeostasis, distinguished by expression of the chemokine receptor CX3CR1 (Gerlach et al., 2016). TE versus MP cell differentiation is influenced by cytokines such as type 1 interferons, IL-12, and IL-2, which are transduced through transcription factors STAT4, STAT5, and the AKT and mtor pathways to induce effector gene expression and repress quiescence and pro-memory genes (see references within Kaech and Cui, 2012). Increasing levels of inflammation induce graded expression of TFs, including T-bet, ZEB2, and Blimp-1, that cooperatively promote TE-signature gene expression and terminal differentiation (Kaech and Cui, 2012; Dominguez et al., 2015; Shin et al., 2013; Omilusik et al., 2015). Simultaneously, the AKT-mediated inactivation of the TF FOXO1 represses expression of MP-signature genes, including Il7r, Sell, Tcf7, Lef1, Bach2, and other memory promoting genes (see references within Kaech and Cui, 2012; Kim et al., 2013). While the above studies offer insight into the molecular control of differentiation of diverse effector T cell subsets during infection, they do not explain how the changes in gene expression are stably inherited in daughter cells to generate terminally differentiated TE cells fated to die or multipotent MP cells fated to persist and generate memory cells. Dynamic regulation of epigenetic and chromatin states will influence how T cells acquire or lose plasticity and/or how particular T cell fates are determined. To better elucidate the epigenetic mechanisms by which TE cells become committed to a terminal fate and MP cells remain multipotent in the context of changing environments during acute lymphocytic choriomeningitis virus (LCMV) infection, we profiled active chromatin associated histone 3 lysine 27 acetyl (H3K27ac) and repressed chromatin associated histone 3 lysine 27 trimethyl (H3K27me3) genome-wide in MP and TE effector CD8 + T cells. This demonstrated biased deposition of repressive H3K27me3 at MP-signature genes in TE cells indicating preferential repression of pro-memory genes as TE cells terminally differentiate. Conversely, MP cells did not contain greater amounts of H3K27me3 at TE-signature genes despite lower transcriptional activity, illuminating their epigenetic plasticity. Additionally, we found that inactivity of the Polycomb Repressive Complex 2 (PRC2), which catalyzes de novo H3K27 trimethylation, via deletion of the methyltransferase Enhancer of Zeste Homolog 2 (EZH2) or its cofactor Embryonic Ectoderm Development (EED) (Margueron and Reinberg, 2011) in virusspecific CD8 + T cells, impaired the formation of terminally differentiated TE cells. While having minimal impact on memory CD8 + T cell maturation, Ezh2 deficiency impaired secondary responses of memory cells to reinfection. Finally, we found that abundant deposition of H3K27me3 occurred relatively late during effector development to stably silence pro-memory genes specifically in TE cells, probably in a FOXO1-regulated manner. These results outline a sophisticated model for how memory cell potential is lost as effector CD8 + T cells terminally differentiate through epigenetic silencing of pro-memory genes. RESULTS Epigenetic Repression of Pro-memory Genes in Terminally Differentiated Effector CD8 + T Cells TE and MP cells differ in their developmental potential (i.e., multipotency) and long-term fates after viral infection, and while they express several genes in common, they also have distinct gene expression signatures (referred to as MP- and TE-signature genes throughout) (Dominguez et al., 2015; Joshi et al., 2007). To understand how these developmental differences arise, we characterized the epigenetic states of these virus-specific effector CD8 + T cell subsets via genome-wide profiling of the histone modifications H3K27ac or H3K27me3 to infer regions of active or repressed chromatin, respectively. To this end, naive Thy1.1 + P14 (LCMV GP specific) T cell receptor (TCR) transgenic CD8 + T cells were transferred into naive wild-type (WT) C57BL/6 recipient mice that were subsequently infected with lymphocytic choriomeningitis virus (LCMV)-Armstrong strain, which causes an acute systemic infection. Ten days after infection (d10 p.i.), pure populations of KLRG1 hi IL-7R lo (TE) and KLRG1 lo IL- 7R hi (MP) CD8 + T cells were sorted and the chromatin was immunoprecipitated using antibodies against H3K27ac and H3K27me3 followed by high-throughput sequencing (ChIP-seq). A total of 8,296 H3K27me3 and 9,076 H3K27ac high-quality peaks (p value < 10 5 ) were identified across MP and TE ChIP-seq samples (Figures S1A S1C). These peaks were annotated to the nearest gene transcriptional start site (TSS) (Figure S1D). From these annotations, H3K27ac deposition exhibited a positive correlation with gene expression across MP and TE cells (Figure S1G, left), while H3K27me3 deposition correlated negatively with gene expression (Figure S1G, right). Next, we identified consensus peaks that contained significant differences in the amount of H3K27me3 or H3K27ac deposition (referred to as differentially modified regions [DMRs]) between MP and TE cells. DMRs were defined as having a fold-change greater than 1.2 with an FDR less than 0.1. Regions failing to meet these criteria were labeled as common regions between MP and TE cells. As above, these DMRs and common regions were annotated to the nearest TSS (Figures S1E and S1F) to identify related patterns of mrna expression in MP and TE cells. Volcano plots of the log 2 (fold-change) of H3K27ac deposition showed that, as expected, DMRs with increased H3K27ac in MP cells (cluster 1) or TE cells (cluster 2) were associated with increased mrna expression in each respective cell population (MP- and TE-signature genes are labeled red and blue, respectively) (Figure 1A). In contrast, differential deposition of H3K27me3 showed that MP cells possessed few differentially methylated loci (cluster 3), whereas TE cells had an abundance of highly methylated loci relative to MP cells (cluster 4), many of which were associated with MP-signature genes (Figure 1B). We next plotted a 20 kb window centered on each significant H3K27ac or H3K27me3 DMR and clustered the loci based on the four quadrants defined by the volcano plots (Figure 1C). Clusters 1 and 2 show the significant H3K27ac DMRs in MP and TE cells, respectively, and the mrna expression at many of these loci correlated with the relative greater abundance of H3K27ac within each cell population (Figure 1C, right). However, note that the relative ratio of H3K27ac between MP and TE cells at the DMRs in cluster 2 was more similar than those in cluster 1 (Figure 1D), signifying that MP cells had comparable amounts of histone acetylation at many TE-signature gene loci despite the lowered mrna expression (possibly indicative of transcriptional poising). Clusters 3 and 4 show the significant H3K27me3 DMRs (Figure 1C), wherein one can see that many H3K27me3 DMRs in TE cells correlated with transcriptional repression of Immunity 46, , April 18,

4 Figure 1. Higher Levels of H3K27me3 Deposition Are a Defining Feature of TE Cells Compared to MP Cells LCMV-specific KLRG1hiIL-7Rlo (TE) and KLRG1loIL-7Rhi (MP) P14 CD8+ T cells were purified at d10 after LCMV-Armstrong infection and ChIP-seq was performed for H3K27me3 and H3K27ac. See Supplemental Experimental Procedures for details, but briefly, consensus peaks from replicate TE and MP samples were compared to identify common regions or those that contain significantly differentially modified regions (DMRs) of H3K27me3 or H3K27ac (FDR [BenjaminiHochberg] < 0.1 and fold-change > 1.2). (A and B) Volcano plots comparing differential deposition of (A) H3K27ac and (B) H3K27me3 between MP and TE cells were used to identify DMRs. Differential abundance of H3K27ac or H3K27me3 deposition [log2(fold-change)] is plotted by [ log10(fdr)], where positive fold-change values represent higher deposition of the histone modification in MP cells and negative values represent higher deposition in TE cells. Horizontal dashed line denotes log10(fdr) of 0.1, while vertical (legend continued on next page) 598 Immunity 46, , April 18, 2017

5 Figure 2. TE Cells Restrict Memory Cell Potential by Epigenetically Repressing MP-Signature Genes Alignment tracks of H3K27ac and H3K27me3 deposition across MP (red) and TE (blue) cells at (A) MP-signature and (B) TE-signature genes. MP- and TE-signature genes were defined based on differential mrna expression (>1.5 fold-change, FDR < 0.1). Statistically significant differentially modified regions (DMRs) are marked by rectangles below tracks, with red bars representing DMRs where the modification is higher in MP cells and blue bars representing DMRs where the modification is higher in TE cells. Black bars demarcate common consensus peaks that are not differentially modified in one cell population over the other. Data shown contain the union of significant consensus peaks identified across two independent biological replicates of ChIP-seq experiments for H3K27ac and H3K27me3 (n = mice/group/replicate). See also Figure S2. MP-signature genes (cluster 4). Similarly, comparing the ratio of H3K27ac to H3K27me3 in TE and MP cells showed that TE cells generally had a bias toward a more repressed state (lower ac/me3 ratio) than MP cells (Figure 1E). In summary, these data demonstrated that the most significant, numerous, and substantial changes in H3K27me3 and H3K27ac deposition preferentially occurred at pro-memory, MP-signature genes that were epigenetically and transcriptionally repressed in the TE cells (Figure 1). To understand how the epigenetic profiles of individual genes differed between MP and TE cells, we examined several important and well-studied MP- and TE-signature genes. At many MP-signature genes (such as Id3, Tcf7, and Bach2) and pro-survival genes (such as Bcl2) the TE cells contained substantially more H3K27me3 and less H3K27ac deposition than MP cells (Figures 2A and S2). A similar pattern was also observed at Trm-signature genes (Mackay et al., 2013) including Sik1, Skil, and Cdh1 (Figure S2), suggesting that epigenetic silencing of these loci in TE cells accounted for their decline in plasticity and longevity. In contrast, there was relatively little H3K27me3 deposition at most pro-effector, TE-signature genes, such as Klrg1, Tbx21 (T-bet), and Prdm1 (Blimp-1), in either TE or MP cells (Figure 2B). Altogether, these data outlined an epigenetic dichotomy between TE and MP cells and illustrated that as effector CD8 + dashed lines denote a ± log 2 transformed fold-change of 1.2. DMRs associated to MP and TE gene expression signatures are labeled as blue and red dots, respectively. All remaining consensus peaks are referred to as common regions between MP and TE cells (labeled as light gray dots). (C) Deposition of H3K27me3 and H3K27ac in MP and TE cells centered on DMRs ± 10 kb as identified in volcano plots (A and B). Cluster 1 (dark blue) = H3K27ac deposition higher in MP than TE, cluster 2 (light blue) = H3K27ac deposition higher in TE than MP, cluster 3 (green) = H3K27me3 deposition higher in MP than TE, cluster 4 (orange) = H3K27me3 deposition higher in TE than MP. Line plots at top show the summary distributions across each cluster for each H3K27ac and H3K27me3 in MP and TE cells, respectively. Scatterplots on far right show Log 2 (fold-change) of mrna expression between MP and TE cells for the DMRassociated genes and summaries of entire gene expression distributions across clusters 1 4 or common consensus peaks are shown in boxplots. (D and E) Line plots show the ratios of H3K27ac or H3K27me3 deposition (normalized to common consensus peaks, see Figure S1I) between MP and TE cells or within MP or TE cells for each cluster. Data shown contain the union of significant consensus peaks identified across two independent biological replicates of ChIP-seq experiments for H3K27ac and H3K27me3 (n = mice/group/replicate). See also Figure S1. Immunity 46, , April 18,

6 Figure 3. EZH2 Is Required for H3K27me3 Deposition and Antiviral CD8 + T Cell Clonal Expansion (A) Naive CD8 + T cells from Ezh2 f/f mice and CD8 + T cells from Ezh2 f/f and Ezh2 f/f CD4Cre + mice activated in vitro with acd3 and acd28 for 3 days were purified using FACS and the amounts of EZH2, b-actin, H3K27me3, and total H3 were measured by western blot (data from two different experiments). Note, H3K27me3 is virtually undetectable in activated Ezh2 f/f CD4Cre + CD8 + T cells. (B) Ezh2 f/f and Ezh2 f/f GzmBCre + mice were infected with LCMV Armstrong and the number of splenic D b GP and D b NP MHC class I tetramer + CD8 + T cells combined were enumerated at d8 p.i. (C) Bar graph shows viral titer in the serum of Ezh2 f/f and Ezh2 f/f GzmBCre + mice at d8 p.i. (D) Naive P14 Ezh2 f/f (red) and Ezh2 f/f CD4Cre + (black line) CD8 + T cells were labeled with CellTrace Violet and stimulated for 72 hr in vitro with GP peptide. Unstimulated naive P14 CD8 T cells are shown in gray. (E) Congenically mismatched naive P14 + Thy1.1 + Ly5.2 + Ezh2 f/f (red) and Thy1.2 + Ly5.2 + Ezh2 f/f CD4Cre + (black line) CD8 + T cells were pulsed with CellTrace Violet and adoptively co-transferred into Thy1.2 + Ly5.1 + WT recipient mice that were subsequently infected with LCMV-Armstrong and analyzed for cell division 60 hr later. P14 + CD8 + T cells from an uninfected recipient are shown in gray. (F) Bar graphs show IFN-g, TNFa, and IL-2 production in GP peptide-stimulated Ezh2 f/f and Ezh2 f/f GzmBCre + CD8 + T cells at d8 p.i. Data shown are representative of two (C E), three (A), or five (F) independent experiments (n = 4 10 mice/group/experiment for C and F) or cumulative of five independent experiments (n = 21 mice/group) (B). Data are expressed as mean ± SD. *p = 0.02, ****p < T cells terminally differentiated into TE cells, many pro-memory genes were selectively remodeled into a repressive state by the accumulation of H3K27me3. This provided a genomic understanding for how memory cell potential was lost as effector CD8 + T cells terminally differentiated through epigenetic silencing of pro-memory genes. The reciprocal process did not appear to occur in MP cells as they maintained permissive or active chromatin states at both MP- and TE-signature genes. Given that memory cells derived from the MP subset will need to express pro-effector, TE-signature genes quickly upon antigen re-exposure, our data support a view where the TE fate is not epigenetically repressed in MP cells, but rather remains open or poised. These data argue that the MP versus TE cell fate decision process differs from a conventional binary cell fate choice where each cell type represses the fate-determining genes of the alternative fate. Rather, MP cells maintain multipotency for both memory and effector fates, while TE cells restrict memory fates. EZH2 Is Required for the Anti-viral CD8 + T Cell Response To better understand how H3K27me3 deposition was regulated in CD8 + T cells, we generated mice that conditionally deleted Ezh2 (the de novo methyltransferase of PRC2) in thymocytes by breeding mice containing floxed Ezh2 alleles to mice expressing Cre-recombinase under the control of the CD4 promoter, enhancer, and silencer, generating Ezh2 f/f CD4Cre + mice. Credeficient littermate Ezh2 f/f mice were used as controls. Western blotting of EZH2 and H3K27me3 in naive and activated CD8 + T cells from Ezh2 f/f CD4Cre + mice or littermate controls showed that (1) EZH2 protein is substantially induced in activated T cells and (2) H3K27me3 was virtually undetectable in activated Ezh2 f/f CD4Cre + CD8 + T cells (Figure 3A). The latter point demonstrated that EZH2-containing PRC2 was the exclusive writer of de novo H3K27me3 in activated CD8 + T cells. To examine the role of EZH2 in the anti-viral CD8 + T cell response, we generated mice that conditionally deleted Ezh2 600 Immunity 46, , April 18, 2017

7 Figure 4. EZH2 Is Required for KLRG1 hi CD27 lo TE CD8 + T Cell Differentiation (A) Ezh2 f/f (solid) and Ezh2 f/f GzmBCre + (open) mice were infected with LCMV-Armstrong, and at d4.5 p.i. the percentage of KLRG1 hi virus-specific CD8 + T cells was determined in the peripheral blood using D b GP and D b NP MHC class I tetramer + staining and flow cytometry. (B D) The same mice as in (A) were examined at d8 p.i. for expression of pro-effector and pro-memory receptors and TFs on splenic D b GP tetramer + CD8 + T cells. (B) Histograms show the relative surface expression of the indicated receptors in Ezh2 f/f (black) and Ezh2 f/f GzmBCre + (gray) CD8 + T cells. (C) Bar graph shows KLRG1/IL-7R subsets as a percentage of D b GP Ezh2 f/f (solid) and Ezh2 f/f GzmBCre + (open) CD8 + T cells. (D) Bar graphs show intracellular mean fluorescence intensity (MFI) of the indicated TFs in Ezh2 f/f (solid) and Ezh2 f/f GzmBCre + (open) CD8 + T cells. Data shown are representative of three (A) or five (B, D) or cumulative of five (C) independent experiments (n = 4 10 mice/group/experiment). Data are expressed as mean ± SD. **p = 0.006, ***p = , ****p < See also Figures S3 and S4. in activated CD8 + T cells by crossing Ezh2 f/f mice to those expressing Cre under the control of the Granzyme B promoter, generating Ezh2 f/f GzmBCre + mice. In this system, Ezh2 was not deleted in naive CD8 + T cells, but was deleted within hr after activation or infection (Figures S3A and S3B; Dominguez et al., 2015). The Ezh2 f/f and Ezh2 f/f GzmBCre + LCMVspecific CD8 + T cells were then analyzed during infection and this showed that Ezh2 was necessary for normal effector CD8 + T cell expansion and viral control. Ezh2 f/f GzmBCre + mice formed 10-fold fewer LCMV-specific anti-viral CD8 + T cells at d8 p.i. and had delayed viral clearance compared to Cre (Ezh2 f/f ) littermate controls (Figures 3B and 3C). The defect in effector CD8 + T cell expansion in the absence of Ezh2 was probably due to increased apoptosis as opposed to defects in cell division since Ezh2 f/f CD4Cre + CD8 + T cells divided at similar rates to Ezh2 f/f controls early after activation both in vitro and in vivo during LCMV infection (Figures 3D and 3E). Examination of cytokine production in a 5 hr ex vivo peptide stimulation assay revealed that Ezh2 f/f GzmBCre + CD8 + T cells were still polyfunctional, with equivalent percentages of IFN-g + and IL-2 + cells, but modestly fewer tumor necrosis factor-a + (TNFa + ) cells, as compared to the Ezh2 f/f controls (Figure 3F). Therefore, EZH2 was required for efficient clonal expansion of activated CD8 + T cells during an acute viral infection, similar to what has been observed in CD4 + T cells after Toxoplasma gondii infection (Yang et al., 2015). EZH2-Containing PRC2 Is Required for Differentiation of Terminal Effector CD8 + T Cells To investigate how EZH2 controls effector CD8 + T cell differentiation, we assessed the phenotype of LCMV-specific Ezh2 f/f GzmBCre + CD8 + T cells at days 4 8 p.i. During the early effector phase (d4.5 p.i.), despite reduced expansion, a similar percentage of KLRG1 hi effector CD8 + T cells formed between Ezh2 f/f and Ezh2 f/f GzmBCre + cells (Figure 4A). However, by d8 p.i., very few KLRG1 hi IL-7R lo TE-like cells were observed in the virus-specific Ezh2 f/f GzmBCre + CD8 + T cells relative to Ezh2 f/f littermate control cells (Figures 4B and 4C). Rather, the CD8 + T cells lacking Ezh2 generated effector cells that expressed lower amounts of KLRG1 and IL-7R and increased amounts of CD27 and CD62L, compared to the littermate controls (Figures 4B and 4C). In addition, the Ezh2-deficient anti-viral CD8 + T cells expressed higher amounts of pro-memory TFs including TCF1 (TCF7), FOXO1, and Eomes and lower amounts of the pro-effector TF T-bet (Figure 4D). Similarly, in different genetic models, Ezh2 f/f CD4Cre + mice formed few virus-specific Immunity 46, , April 18,

8 KLRG1 hi IL-7R lo TE-like cells and many more KLRG1 lo IL- 7R hi CD27 hi CD62L hi cells compared to Ezh2 f/f controls after infection (Figure S3C), and replication of this phenotype in mixed Ezh2 f/f : Ezh2 f/f GzmBCre + bone marrow chimeras at d8 p.i. confirmed that EZH2 functions in a CD8 + T cell-intrinsic manner to generate TE-like cells (Figure S3D). Lastly, because EZH2 is known to have non-histone targets in cells (Gunawan et al., 2015), we examined the phenotypes of virus-specific CD8 + T cells lacking the PRC2 subunit EED by generating Eed f/f CD4Cre + mice that were infected with LCMV-Armstrong. This showed that the phenotypes of the CD8 + T cells lacking Eed were very similar to those lacking Ezh2 at day 8 p.i. (Figure S4). Together, these data suggested that PRC2 (containing both EZH2 and EED) was intrinsically required for the expansion and terminal differentiation of effector CD8 + T cells and that in the absence of PRC2, effector CD8 + T cells acquired more MP-like qualities. EZH2 Is Not Required for Memory CD8 + T Cell Formation, but Is Required for Protective Immunity We next examined whether EZH2 was required for memory CD8 + T cell formation by two approaches. First, we created mice in which we could inducibly delete Ezh2 conditionally in CD8 + T cells using the GranzymeB-ER T2 Cre system (which regulates expression of the estrogen-receptor (ER)-Cre fusion protein by the GzmB promoter and ER-Cre nuclear localization via tamoxifen treatment) (Bannard et al., 2009). These Ezh2 f/f GzmB-ER T2 Cre mice also contained a ROSA26 flox STOP flox YFP cassette to report ER-Cre activity in the virus-specific CD8 + T cells. In brief, Ezh2 f/f GzmB-ER T2 Cre mice and Ezh2 f/+ GzmB-ER T2 Cre littermate controls were infected with LCMV- Armstrong and 8 days later, after effector CD8 + T cell formation, the mice were gavaged with 2 mg of tamoxifen every 3 days from d8 to 23 p.i., then sacrificed at d30 p.i. to measure the numbers and phenotypes of LCMV-specific (YFP + ) memory CD8 + T cells. Ezh2 deletion in this system was confirmed after completion of tamoxifen treatment by genomic DNA PCR analysis in virusspecific CD8 + T cells, which showed near complete deletion of Ezh2 (Figure S5). This experiment showed that loss of Ezh2 in anti-viral CD8 + T cells after d8 p.i. did not have a major impact on the quantity or quality of memory CD8 + T cells that formed (Figures 5A 5C). In agreement, similar numbers and types of memory CD8 + T cells formed in Ezh2 f/f GzmBCre + mice (the non-inducible Cre system described in Figure 3) as in the littermate controls at d30 p.i. (Figures 5D 5F). Thus, EZH2 was critical for naive/effector CD8 + T cell differentiation, but not effector/memory differentiation, in line with the expression pattern of EZH2 in antiviral CD8 + T cells (data not shown). Second, to examine the protection provided by Ezh2-deficient memory CD8 + T cells upon secondary infection, small numbers of Ezh2-deficient or Ezh2-sufficient memory CD8 + T cells specific for the LCMV-epitope GP (isolated from Ezh2 f/f GzmBCre + or Ezh2 f/f mice, respectively) were transferred into naive B6 recipients that were subsequently infected with a recombinant strain of Listeria monocytogenes that expresses the GP epitope (LM-33). Similar to what was observed during the primary infection, the Ezh2-deficient memory CD8 + T cells did not expand or clear bacteria as effectively as the control memory CD8 + Tcells(Figures 5G and 5H). These data demonstrated that EZH2 was not required for the formation of memory CD8 + T cells in general but was required for their protective recall responses and generation of secondary effector cells. TE Cell Fates Are Determined during Late Effector Cell Development Prior work has shown that effector CD8 + T cells transcriptionally repress pro-memory genes early after activation and that commitment to effector or memory cell fates can be observed prior to d4.5 p.i. (Kalia et al., 2010; Best et al., 2013; Sarkar et al., 2008), but when are these cell fates precisely determined? We reasoned that delineating when the pro-memory loci are selectively remodeled by PRC2 would help to better temporally define TE cell determination. Therefore, we first examined whether EZH2 was required for the early transcriptional repression of MP-signature genes, such as Tcf7, Foxo1, Bach2, and Id3, in CD8 + T cells during acute infection (Figure S6). This showed that at day 4.5 p.i., the mrna expression was reduced equivalently in both the Ezh2-deficient and Ezh2-sufficient CD8 + T cells relative to naive CD8 + T cells (Figure 6A), indicating that H3K27me3 deposition was not required for the initial transcriptional repression of pro-memory genes during the early effector phase. Next, we monitored when H3K27me3 deposition increased at MP-signature loci (Tcf7 and Bach2) in effector CD8 + T cells and found that it was lower in naive CD8 + T cells and early effector cells at days 1.5 and 4.5 p.i. compared to day 10 TE cells (Figure 6B). There was no difference observed between KLRG1 hi and KLRG1 lo effector cells at day 4.5 p.i.; this is when KLRG1 hi cells begin to appear during infection and demonstrate commitment to KLRG1 hi IL-7R lo TE cell fates (Figure 6B; Joshi et al., 2007). However, by day 10 p.i., H3K27me3 dramatically increased preferentially in KLRG1 hi IL-7R lo TE cells, but not KLRG1 lo IL-7R hi MP cells (Figure 6B). This result suggested that increased H3K27me3 occurred during a late stage of effector T cell differentiation to stabilize repression of promemory genes whose transcription was already shut down in developing TE cells. Moreover, d30 p.i. memory CD8 + Tcells displayed low levels of H3K27me3 deposition, comparable to the MP cell population from which most memory cells descended (Figure 6B). Altogether, these results delineated a molecular timeline between commitment (early effector phase) and determination (late effector phase) of TE cell fates by revealing that EZH2-epigenetic remodeling was not required for early transcriptional repression of pro-memory genes but was required later to sustain repression. This significant increase in H3K27me3 observed in the late effector phase may represent the quintessential step in TE cell fate determination. FOXO1 Regulates Deposition of H3K27me3 in CD8 + T Cells What controls PRC2-mediated H3K27me3 deposition at promemory genes in TE cells? We reasoned that key transcription factors involved in MP and TE cell differentiation, such as T-bet, STAT4, and FOXO1, may be involved (Joshi et al., 2007; Kim et al., 2013). To examine this idea, we performed H3K27me3 ChIP-qPCR at the pro-memory genes Tcf7 and Bach2 from P14 + KLRG1 hi IL-7R lo CD8 + T cells from d10 p.i. that were lacking 602 Immunity 46, , April 18, 2017

9 Figure 5. EZH2 Is Not Required for Memory Formation, but Is Required for Secondary Responses (A C) Ezh2 f/+ GzmBER T2 Cre + and Ezh2 f/f GzmBER T2 Cre + mice were infected with LCMV-Armstrong, then starting at d8 p.i., EZH2 was inducibly deleted in virusspecific CD8 + T cells by tamoxifen (2 mg) administration every 3 days for 3 weeks. (A) The splenic YFP + D b GP CD8 + T cells were examined at d30 p.i. by flow cytometry and contour plots (left) show expression of KLRG1 and IL-7R. Bar graphs (right) show the numbers of Ezh2 f/+ GzmBER T2 Cre + (solid) and Ezh2 f/f GzmBER T2 Cre + (open) virus-specific YFP + CD8 + T cells in each of the four KLRG1/ IL-7R subsets. (B) Bar graph shows percentage of YFP + D b GP CD8 + T cells expressing CD62L. (C) Bar graph shows FOXO1 protein levels in YFP + D b GP CD8 + T cells. (D H) Ezh2 f/f and Ezh2 f/f GzmBCre + mice (the non-inducible Cre system described in Figure 3) were infected with LCMV-Armstrong and at d30 p.i. the phenotype (D F) and protective capacity (G and H) of the GP specific memory CD8 + T cells were examined. (D) Contour plots (left) show expression of KLRG1 and IL-7R and bar graphs (right) show the numbers of Ezh2 f/f (solid) and Ezh2 f/f GzmBCre + (open) virus-specific CD8 + T cells in each of the four KLRG1/IL-7R subsets. (E) Bar graph shows percentage of D b GP CD8 + T cells expressing CD62L. (F) Bar graph shows FOXO1 protein levels in D b GP CD8 + T cells. (G and H) 50,000 GP specific Ezh2 f/f or Ezh2 f/f GzmBCre + memory CD8 + T cells (from D) were transferred into naive B6 mice that were then infected with recombinant Listeria monocytogenes expressing the GP epitope (LM-33). (G) At d4.5 post-challenge, the numbers of recalled GP specific Ezh2 f/f and Ezh2 f/f GzmBCre + CD8 + T cells were enumerated in the spleen. (H) LM-33 bacterial titers in the liver were determined at d3 post-challenge. Data shown are representative of two (B, C, E, F, H) or cumulative of two (G) or three (A, D) independent experiments; n = 9/group (A), n = 9 13/group (D, G), n = 3 5 mice/group/experiment (B, C, E, F, H). Data are expressed as mean ± SD. *p = 0.02, **p = 0.01, ***p = See also Figure S5. (1) Foxo1 or (2) Stat4 or (3) retrovirally overexpressing (OE) T-bet. Control (WT) P14 + KLRG1 hi IL-7R lo CD8 + T cells were infected with empty-vector-gfp retroviruses. Despite the profound effects of T-bet OE and Stat4 deficiency on promoting and repressing TE cell development, respectively, neither of these genetic alterations affected abundance of H3K27me3 at these loci relative to control (WT) cells. In contrast, KLRG1 hi IL-7R lo cells lacking Foxo1 had significantly higher amounts (>2-fold) of H3K27me3 at the pro-memory genes Tcf7 and Bach2 (Figure 7A). This result suggested that FOXO1 restrained PRC2 activity at such pro-memory loci. One may predict that if FOXO1 affects H3K27me3 at MPsignature genes, such as Tcf7 and Bach2, its expression may be higher in MP cells and it may preferentially bind to such loci. Immunity 46, , April 18,

10 Figure 6. Increased H3K27me3 Deposition Occurs Specifically in Late-Stage TE Cells (A) Tbx21, Tcf7, Foxo1, Bach2, and Id3 mrna were measured in purified naive (day 0) Ezh2 f/f (solid) P14 + CD8 + T cells, and in Ezh2 f/f (solid) and Ezh2 f/f GzmBCre + (open) P14 + CD8 + T cells from day 4.5 p.i. using qrt-pcr. Data are representative of three independent experiments (n = 2 3 mice/group/experiment). (B) WT P14 + CD8 + T cells were purified from naive (day 0) and LCMV-Armstrong-infected mice at days 1.5, 4.5 (KLRG1 hi and KLRG1 lo populations), 10 (KLRG1 hi IL-7R lo TE and KLRG1 lo IL-7R hi MP populations), and 30 p.i., and then ChIP-qPCR was performed for H3K27me3 using primers to the Tcf7 TSS and Bach2 intron 1 (black). Region within Ttn served as a negative control (white) for H3K27me3 based on genome-wide H3K27me3 ChIP-seq analysis. Data are cumulative of four independent experiments (n = 2 4 mice/group/experiment) with percent of input for each sample normalized internally to the percent of input for the day 30 sample, thereby calculating fold enrichment relative to the d30 sample as plotted. Data are expressed as mean ± SD. *p < 0.05, **p < See also Figure S6. Using flow cytometry, we observed that KLRG1 lo IL-7R hi MP cells contained more FOXO1 protein than KLRG1 hi IL-7R lo TE cells (Figure 7B). We then examined putative FOXO1 binding sites in the four different clusters of DMRs identified in Figure 1 using previously generated FOXO1 ChIP-seq data (Kim et al., 2013) (GEO: GSE46943) from naive CD8 + T cells. The density of FOXO1 peaks surrounding common consensus peaks (i.e., non-dmrs), H3K27ac DMRs, or H3K27me3 DMRs in MP and TE cells (i.e., clusters 1 4 defined in Figure 1) was measured and plotted (Figure 7C). FOXO1 binding was densely enriched near the H3K27ac common peaks or DMRs in cluster 1, suggesting that FOXO1 generally bound to active genes and open chromatin regions (note, FOXO1 binding was not concentrated proximal to H3K27ac DMRs in TE cells [cluster 2]) (Figure 7C, top). There were no FOXO1 binding sites in H3K27me3 DMRs in MP cells (cluster 3), but interestingly, FOXO1 binding sites were preferentially and densely enriched near H3K27me3 DMRs found in TE cells (cluster 4) (Figure 7C, bottom). This analysis predicted that FOXO1 preferentially bound to gene loci that are transcriptionally active in MP cells and repressed in TE cells. Pathway analysis of the FOXO1-bound sites within 2 kb of consensus peaks in CD8 + T cells showed that chromatin organization and histone modifications were the most highly enriched biological processes, suggesting a potentially important role for FOXO1 in regulating the chromatin state of CD8 + T cells (Figure S7). Collectively, these results provide a deeper mechanistic understanding of how and when memory cell potential and longevity is gained or lost in effector CD8 + T cells during viral infection decreased FOXO1 expression and binding to pro-memory genes in TE cells may allow for increased PRC2 activity and stable epigenetic repression, leading to the determination of a TE cell fate. Concurrently, increased FOXO1 expression in developing MP cells may shield pro-memory genes from H3K27me3 deposition, allowing for sustained or even increased expression as memory CD8 + T cells mature (Best et al., 2013; Kaech et al., 2002; Sarkar et al., 2008). DISCUSSION The mechanisms underlying differentiation and commitment of effector and memory CD8 + T cells is an active area of investigation. While the asymmetric cell division, signal-strength, and decreasing-potential models seek to explain how the TE and MP cell populations become distinct (Kaech and Cui, 2012), they do not explain how or when TE cells become committed to their identity and lose memory cell potential and longevity, nor how MP cells remain plastic to form diverse memory cell populations that can regenerate effector cells. Our study illuminated mechanisms for how this occurred by identifying that several pro-memory and pro-survival genes were selectively remodeled in developing KLRG1 hi IL-7R lo TE cells to contain greater H3K27me3 and lesser H3K27ac deposition, which more stably restricted transcription and consequently memory cell potential and lifespan. KLRG1 lo IL-7R hi MP cells, on the other hand, maintained pro-memory, pro-survival, and TE-signature gene loci in active or permissive epigenetic states, allowing for the present expression of memory genes and the future expression of effector genes. Further, we showed that PRC2 complex members EZH2 and EED were required for effector cell expansion and the differentiation of KLRG1 hi IL-7R lo TE cells and that PRC2 activity may be restrained at MP-signature genes by FOXO1. These data provided mechanistic insight for how developmental plasticity was epigenetically wired in subsets of effector CD8 + T cells or lost in others as they terminally differentiated. In many respects, the epigenetic changes in TE cells resembled those in stem cells differentiating into specialized cell types that is, the cells turned on lineage-determining genes (i.e., TE-signature genes) and silenced stemness genes (i.e., MPsignature genes). However, an important distinction in MP cells compared to embryonic stem cells is that EZH2 helps to preserve embryonic stem cell identity by depositing H3K27me3 and repressing lineage-determining genes (Chen and Dent, 604 Immunity 46, , April 18, 2017

11 Figure 7. FOXO1 Regulates H3K27me3 Deposition (A) WT (control cells infected with empty-vector- GFP retrovirus), Foxo1 f/f CD4Cre +, Tbet-overexpressing (Tbet OE ), and Stat4 / KLRG1 hi IL-7R lo (TE) Thy1.1 P14 + CD8 + T cells were purified by FACS from infected mice from day 10 p.i. and ChIP-qPCR was performed for H3K27me3 using primers to the Tcf7 TSS and Bach2 intron 1 (black). A region within Ttn served as a negative control (white) for H3K27me3 based on genome-wide H3K27me3 ChIP-seq analysis. Data are cumulative from two (Stat4 / ), three (Tbet OE ), or four (Foxo1 f/f CD4Cre + ) independent experiments (n = 2 [Stat4 / ], 7 [Tbet OE ], or 11 [Foxo1 f/f CD4Cre + ] mice/per group) with percentage of input for each sample normalized internally to the percentage of input of WT, thereby calculating fold enrichment relative to WT as plotted. (B) Bar graph shows FOXO1 protein levels in KLRG1 hi and KLRG1 lo WT P14 + CD8 + T cells and Foxo1 f/f CD4Cre + P14 + CD8 + T cells at d10 p.i. with LCMV-Armstrong in vivo. Data are expressed as mean ± SD. *p < 0.05, **p < 0.01, ****p < (C) Significant FOXO1 binding sites (p value < 10 5 ) identified by ChIP-seq from naive CD8 + T cells (GEO: GSE46943) were annotated to the nearest consensus peak. Density plots were calculated from the distribution of distances from a FOXO1 binding site to the nearest consensus peak and visualized separately for DMRs versus common regions for H3K27ac (top) and H3K27me3 (bottom). Select genes associated with FOXO1 binding are noted below the density plots and colored according to cluster 1 4 or common. See also Figure S ); contrastingly, our data showed that MP cells maintained TE-signature genes in a permissive state rather than a repressed state. Our data more clearly delineated a stepwise process from specification to determination of TE fates during viral infection. While transcriptional repression of pro-memory genes occurs rapidly after CD8 + T cell activation (Best et al., 2013), our results showed that this early transcriptional repression occurred independent of EZH2. Moreover, the pro-memory loci Tcf7 and Bach2 were not as heavily H3K27me3 methylated in KLRG1 hi cells at day 4.5 p.i. as compared to day 10 p.i. Thus, despite being committed to a TE cell fate at day 4.5, evidence of more stable epigenetic repression of pro-memory genes in TE cells did not occur until several days later; this suggested that determination of TE cell fates occurred at later stages of effector cell differentiation and H3K27me3 stabilized already repressed gene loci, similar to what has been observed in thymocyte lineage commitment (Zhang et al., 2012). Epigenetic silencing by DNA methylation also regulates CD8 + T cell effector formation (Scharer et al., 2013; Ladle et al., 2016) and therefore it will be important to study the temporal and spatial relationship between H3K27me3 and DNA methylation in effector CD8 + T cells. Prior work mapping genome-wide changes in histone acetylation and methylation in virus-specific effector and memory cells after murine influenza infection (Russ et al., 2014) demonstrated that many genes were epigenetically remodeled during the effector to memory transition. Although epigenetic differences between effector T cell subsets were not compared, as done here, it is possible that such changes occurred over time because of selective maintenance of IL-7R hi MP cells. It is also possible that these changes occurred due to active removal of H3K27me3 in the memory cells and their progenitors by the H3K27me3-specific demethylases UTX and JMJD3 (Manna et al., 2015). Of interest, the mrna expression of Utx and Jmjd3 is increased between the effector to memory cell transition (Kaech et al., 2002). What drives this pro-memory gene-specific, late deposition of H3K27me3 in TE cells? Given that sustained mtor activity was found to drive TE cell differentiation and impair memory cell development (Kaech and Cui, 2012; Pollizzi et al., 2015), in part due to impaired FOXO1 activity (Kim et al., 2013), we reasoned that FOXO1 may be an interesting candidate to consider in regulating PRC2 activity. Moreover, FOXP3, another fork-head box TF family member, was found to bind to EZH2 and regulate PRC2 activity in Treg cells (DuPage et al., 2015; Arvey et al., 2014). Indeed, our data suggested that FOXO1 prevented deposition of H3K27me3 on certain pro-memory genes in CD8 + T cells. Although we were not able to demonstrate biochemical interaction between FOXO1 and PRC2 subunits in co-immunoprecipitation experiments in effector CD8 + T cells (data not shown), we found that Foxo1 deficiency, but not T-bet OE or Stat4 deficiency, led to increased H3K27me3 levels at certain pro-memory genes in KLRG1 hi IL-7R lo LCMV-specific effector Immunity 46, , April 18,

12 CD8 + T cells. Given that FOXO1 was more highly expressed in MP than TE cells, we postulated that pro-memory genes may be shielded from H3K27me3 deposition in MP cells by FOXO1. Additionally, FOXO1 binding sites were enriched for genes involved in chromatin remodeling and organization in CD8 + T cells, suggesting that FOXO1 may actively remodel chromatin to retain a more pluripotent state in MP cells. Furthermore, increased AKT-mediated phosphorylation of FOXO1 in developing KLRG1 hi, IRF4 hi cells may lead to nuclear exclusion followed by ubiquitination, acetylation, and proteosomal degradation of FOXO1, which commits the effector cells to terminal differentiation (Chang et al., 2011; Staron et al., 2014; Lin et al., 2015). Altogether, our data suggested that FOXO1 may regulate PRC2-mediated deposition of H3K27me3 in CD8 + T cells as they terminally differentiated; however, future studies are required to clarify the precise mechanism of FOXO1 regulation. Our data also highlighted the role of epigenetic bivalency, where a locus contains both active- and repressive-associated histone modifications, in cellular differentiation. Similar to previous studies in CD8 + T cells (Russ et al., 2014; Araki et al., 2009), we found that many genes were bivalently modified during CD8 + T cell differentiation. Further, we found that the relative ratio of H3K27ac/H3K27me3 differed between MP and TE cells at numerous differentially expressed genes. TE cells had a lower relative ratio of H3K27ac/H3K27me3 than MP cells at many pro-memory and pro-survival genes, suggesting that the relative level, not simple presence or absence, of repressive to activating epigenetic modifications determined whether repressed promemory genes became stably epigenetically silenced. An alternative explanation of the observed bivalency is undistinguished heterogeneity in the TE and MP populations, wherein cells may differ epigenetically despite sharing the same population identifying surface markers. While 90% of TE cells appear stably terminally differentiated, we have previously shown that 10% of these cells transform into IL-7R hi cells (Joshi et al., 2007). Thus, there may be some convertibility within this subset and it could involve epigenetic remodeling and erasing of H3K27me3 from the pro-memory gene loci in response to different stresses or tissue environments. In summary, this work identified a model for how memory cell potential was lost as effector CD8 + T cells terminally differentiated through epigenetic silencing of genes controlling T cell survival and memory cell fates and adds to our growing understanding of the profound importance of epigenetic regulation in T cell plasticity and developmental potential (Vahedi et al., 2012). EXPERIMENTAL PROCEDURES Mice, Infections, and Treatments C57BL/6 (B6) mice were purchased from Charles River Laboratories. P14 (LCMV H-2D b GP specific) T cell receptor (TCR) transgenic mice were obtained from R. Ahmed (Pircher et al., 1989). Ezh2 flox/flox (Shen et al., 2008), Eed flox/flox (Xie et al., 2014), and CD4Cre (Lee et al., 2001) mice were purchased from Jackson Laboratories. GranzymeB-Cre (GzmBCre) mice (Jacob and Baltimore, 1999) were provided by J. Jacob (Emory University, Atlanta, GA). GranzymeB-ER T2 Cre ROSA26 flox STOP flox YFP mice (Bannard et al., 2009) were provided by Doug Fearon (Cold Spring Harbor Laboratory, Cold Spring Harbor, NY). Foxo1 flox/flox mice (Ouyang et al., 2009) were provided by Ming Li (Memorial Sloan Kettering Cancer Center, New York, NY). Stat4 / mice (Kaplan et al., 1996) were provided by M. Grusby (Harvard Medical School, Boston, MA). P14 chimeric mice were generated by transferring Thy1.1 + P14 CD8 + T cells to naive Thy1.2 + B6 recipients. Ezh2D/flox PCR was previously described (Shen et al., 2008). All animal experiments were completed according to approved Institutional Animal Care and Use Committee protocols. Details of infections and treatments are found in the Supplemental Experimental Procedures. Gene Expression by qrt-pcr and Immunoblotting Immunoblotting was performed as previously described using rabbit monoclonal antibodies from Cell Signaling: EZH2 (D2C9), b-actin (13E5), H3K27me3 (C62B11), and H3 (D1H2) (Hand et al., 2007). Details of these procedures and list of qpcr primers and antibodies are found in the Supplemental Experimental Procedures. Chromatin Immunoprecipitation and ChIP-Sequencing ChIP was performed with anti-h3k27me3 (Abcam cat# ab6002, RRID: AB_305237), anti-h3k27ac (Abcam cat# ab4729; RRID: AB_ ), and anti-mouse IgG antibodies. Sequencing was performed on an Illumina HiSeq 2500 with four samples per lane (170M reads are distributed at 42.5M reads per sample with 75 bp reads in single-end mode). Details of these procedures and list of qpcr primers are found in the Supplemental Experimental Procedures. Flow Cytometry, Surface and Intracellular Staining, Peptide Stimulations, and Antibodies Flow cytometry and sorting were performed on an LSRII and BD FACS Aria II (Becton Dickinson), respectively, and analyzed with FlowJo software (FlowJo, LLC). Details of these procedures and list of antibodies are found in the Supplemental Experimental Procedures. Mixed Bone Marrow Chimeras and Retroviral Everexpression MSCV-empty-vector-GFP and MSCV-T-bet-GFP (Tbet-overexpression) (Szabo etal., 2000) wereobtainedfrom L. Glimcher (Dana Farber, Boston, MA). Detailsof these procedures are found in the Supplemental Experimental Procedures. Statistical Analysis Results from flow cytometry and qpcr data were presented as mean ± standard deviation. Statistical significance was computed with Prism 7 (GraphPad Software) by paired or unpaired Student s t test with a p value of < 0.05 considered as significant. Details of statistical analysis for ChIP-seq data are found in the Supplemental Experimental Procedures. ACCESSION NUMBERS Data are available online at the Gene Expression Omnibus (GEO: GSE72408) and the Sequence Read Archive (SRA: SRA and SRP101899). SUPPLEMENTAL INFORMATION Supplemental Information includes seven figures, one table, and Supplemental Experimental Procedures (including analysis of ChIP-seq data, differential modification analysis, in vitro culture, and CellTrace Violet proliferation assay) and can be found with this article online at j.immuni AUTHOR CONTRIBUTIONS S.M.G., R.A.A., T.G., and S.M.K. designed the research. S.M.G. and T.G performed the experiments. R.A.A. performed the bioinformatics analyses under the advisement of S.H.K. and S.M.K. S.M.G., R.A.A., T.G., S.H.K., and S.M.K. analyzed the results. S.M.G., R.A.A., and S.M.K. wrote the manuscript. All authors read and approved the final version of this manuscript. ACKNOWLEDGMENTS We thank members of the Kaech and Kleinstein laboratories for helpful comments and suggestions. This study was supported in part by the Howard Hughes Medical Institute (HHMI) (S.M.K.), HHMI Gilliam Fellowship (R.A.A.), 606 Immunity 46, , April 18, 2017

13 the HHMI International Student Research Fellowship (T.G.), the NIH grants R37 AI (S.M.K.), F30 AI (S.M.G.), 1S10OD (S.M. Mane), and NIH/NIGMS T32 GM (S.M.G.). This work was supported by the HPC facilities operated by the Yale Center for Research Computing and the Yale Center for Genome Analysis. Received: August 30, 2016 Revised: January 23, 2017 Accepted: February 14, 2017 Published: April 11, 2017 REFERENCES Araki, Y., Wang, Z., Zang, C., Wood, W.H., 3rd, Schones, D., Cui, K., Roh, T.Y., Lhotsky, B., Wersto, R.P., Peng, W., et al. (2009). Genome-wide analysis of histone methylation reveals chromatin state-based regulation of gene transcription and function of memory CD8+ T cells. Immunity 30, Arvey, A., van der Veeken, J., Samstein, R.M., Feng, Y., Stamatoyannopoulos, J.A., and Rudensky, A.Y. (2014). Inflammation-induced repression of chromatin bound by the transcription factor Foxp3 in regulatory T cells. Nat. Immunol. 15, Bannard, O., Kraman, M., and Fearon, D.T. (2009). Secondary replicative function of CD8+ T cells that had developed an effector phenotype. Science 323, Best, J.A., Blair, D.A., Knell, J., Yang, E., Mayya, V., Doedens, A., Dustin, M.L., and Goldrath, A.W.; Immunological Genome Project Consortium (2013). Transcriptional insights into the CD8(+) T cell response to infection and memory T cell formation. Nat. Immunol. 14, Chang, J.T., Ciocca, M.L., Kinjyo, I., Palanivel, V.R., McClurkin, C.E., Dejong, C.S., Mooney, E.C., Kim, J.S., Steinel, N.C., Oliaro, J., et al. (2011). Asymmetric proteasome segregation as a mechanism for unequal partitioning of the transcription factor T-bet during T lymphocyte division. Immunity 34, Chen, T., and Dent, S.Y. (2014). Chromatin modifiers and remodellers: regulators of cellular differentiation. Nat. Rev. Genet. 15, Dominguez, C.X., Amezquita, R.A., Guan, T., Marshall, H.D., Joshi, N.S., Kleinstein, S.H., and Kaech, S.M. (2015). 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, DuPage, M., Chopra, G., Quiros, J., Rosenthal, W.L., Morar, M.M., Holohan, D., Zhang, R., Turka, L., Marson, A., and Bluestone, J.A. (2015). The chromatin-modifying enzyme Ezh2 is critical for the maintenance of regulatory T cell identity after activation. Immunity 42, Gerlach, C., Moseman, A., Loughhead, S., Alvarez, D., and Von Andrian, U.H. (2016). CX3CR1 distinguishes three antigen-experienced CD8 T cell subsets with distinct migratory, functional and homeostatic properties. J. Immunol. 196, Gunawan, M., Venkatesan, N., Loh, J.T., Wong, J.F., Berger, H., Neo, W.H., Li, L.Y.J., La Win, M.K., Yau, Y.H., Guo, T., et al. (2015). The methyltransferase Ezh2 controls cell adhesion and migration through direct methylation of the extranuclear regulatory protein talin. Nat. Immunol. 16, Hand, T.W., Morre, M., and Kaech, S.M. (2007). Expression of IL-7 receptor a is necessary but not sufficient for the formation of memory CD8 T cells during viral infection. Proc. Natl. Acad. Sci. USA 104, Jacob, J., and Baltimore, D. (1999). Modelling T-cell memory by genetic marking of memory T cells in vivo. Nature 399, Joshi, N.S., Cui, W., Chandele, A., Lee, H.K., Urso, D.R., Hagman, J., Gapin, L., and Kaech, S.M. (2007). Inflammation directs memory precursor and shortlived effector CD8(+) T cell fates via the graded expression of T-bet transcription factor. Immunity 27, Kaech, S.M., and Cui, W. (2012). Transcriptional control of effector and memory CD8+ T cell differentiation. Nat. Rev. Immunol. 12, Kaech, S.M., Hemby, S., Kersh, E., and Ahmed, R. (2002). Molecular and functional profiling of memory CD8 T cell differentiation. Cell 111, Kalia, V., Sarkar, S., Subramaniam, S., Haining, W.N., Smith, K.A., and Ahmed, R. (2010). Prolonged interleukin-2ralpha expression on virus-specific CD8+ T cells favors terminal-effector differentiation in vivo. Immunity 32, Kaplan, M.H., Sun, Y.L., Hoey, T., and Grusby, M.J. (1996). Impaired IL-12 responses and enhanced development of Th2 cells in Stat4-deficient mice. Nature 382, Kim, M.V., Ouyang, W., Liao, W., Zhang, M.Q., and Li, M.O. (2013). The transcription factor Foxo1 controls central-memory CD8+ T cell responses to infection. Immunity 39, Ladle, B.H., Li, K.P., Phillips, M.J., Pucsek, A.B., Haile, A., Powell, J.D., Jaffee, E.M., Hildeman, D.A., and Gamper, C.J. (2016). De novo DNA methylation by DNA methyltransferase 3a controls early effector CD8+ T-cell fate decisions following activation. Proc. Natl. Acad. Sci. USA 113, Lee, P.P., Fitzpatrick, D.R., Beard, C., Jessup, H.K., Lehar, S., Makar, K.W., Pérez-Melgosa, M., Sweetser, M.T., Schlissel, M.S., Nguyen, S., et al. (2001). A critical role for Dnmt1 and DNA methylation in T cell development, function, and survival. Immunity 15, Lin, W.H., Adams, W.C., Nish, S.A., Chen, Y.H., Yen, B., Rothman, N.J., Kratchmarov, R., Okada, T., Klein, U., and Reiner, S.L. (2015). Asymmetric PI3K signaling driving developmental and regenerative cell fate bifurcation. Cell Rep. 13, Mackay, L.K., Rahimpour, A., Ma, J.Z., Collins, N., Stock, A.T., Hafon, M.L., Vega-Ramos, J., Lauzurica, P., Mueller, S.N., Stefanovic, T., et al. (2013). The developmental pathway for CD103(+)CD8+ tissue-resident memory T cells of skin. Nat. Immunol. 14, Manna, S., Kim, J.K., Baugé, C., Cam, M., Zhao, Y., Shetty, J., Vacchio, M.S., Castro, E., Tran, B., Tessarollo, L., and Bosselut, R. (2015). Histone H3 Lysine 27 demethylases Jmjd3 and Utx are required for T-cell differentiation. Nat. Commun. 6, Margueron, R., and Reinberg, D. (2011). The Polycomb complex PRC2 and its mark in life. Nature 469, Omilusik, K.D., Best, J.A., Yu, B., Goossens, S., Weidemann, A., Nguyen, J.V., Seuntjens, E., Stryjewska, A., Zweier, C., Roychoudhuri, R., et al. (2015). Transcriptional repressor ZEB2 promotes terminal differentiation of CD8+ effector and memory T cell populations during infection. J. Exp. Med. 212, Ouyang, W., Beckett, O., Flavell, R.A., and Li, M.O. (2009). An essential role of the Forkhead-box transcription factor Foxo1 in control of T cell homeostasis and tolerance. Immunity 30, Pircher, H., B urki, K., Lang, R., Hengartner, H., and Zinkernagel, R.M. (1989). Tolerance induction in double specific T-cell receptor transgenic mice varies with antigen. Nature 342, Pollizzi, K.N., Patel, C.H., Sun, I.H., Oh, M.H., Waickman, A.T., Wen, J., Delgoffe, G.M., and Powell, J.D. (2015). mtorc1 and mtorc2 selectively regulate CD8 + T cell differentiation. J. Clin. Invest. 125, Pollizzi, K.N., Sun, I.H., Patel, C.H., Lo, Y.C., Oh, M.H., Waickman, A.T., Tam, A.J., Blosser, R.L., Wen, J., Delgoffe, G.M., and Powell, J.D. (2016). Asymmetric inheritance of mtorc1 kinase activity during division dictates CD8(+) T cell differentiation. Nat. Immunol. 17, Russ, B.E., Olshanksy, M., Smallwood, H.S., Li, J., Denton, A.E., Prier, J.E., Stock, A.T., Croom, H.A., Cullen, J.G., Nguyen, M.L., et al. (2014). Distinct epigenetic signatures delineate transcriptional programs during virus-specific CD8(+) T cell differentiation. Immunity 41, Sarkar, S., Kalia, V., Haining, W.N., Konieczny, B.T., Subramaniam, S., and Ahmed, R. (2008). Functional and genomic profiling of effector CD8 T cell subsets with distinct memory fates. J. Exp. Med. 205, Scharer, C.D., Barwick, B.G., Youngblood, B.A., Ahmed, R., and Boss, J.M. (2013). Global DNA methylation remodeling accompanies CD8 T cell effector function. J. Immunol. 191, Shen, X., Liu, Y., Hsu, Y.J., Fujiwara, Y., Kim, J., Mao, X., Yuan, G.C., and Orkin, S.H. (2008). EZH1 mediates methylation on histone H3 lysine 27 and complements EZH2 in maintaining stem cell identity and executing pluripotency. Mol. Cell 32, Immunity 46, , April 18,

14 Shin, H.M., Kapoor, V.N., Guan, T., Kaech, S.M., Welsh, R.M., and Berg, L.J. (2013). Epigenetic modifications induced by Blimp-1 regulate CD8 + T cell memory progression during acute virus infection. Immunity 39, Staron, M.M., Gray, S.M., Marshall, H.D., Parish, I.A., Chen, J.H., Perry, C.J., Cui, G., Li, M.O., and Kaech, S.M. (2014). The transcription factor FoxO1 sustains expression of the inhibitory receptor PD-1 and survival of antiviral CD8(+) T cells during chronic infection. Immunity 41, Szabo, S.J., Kim, S.T., Costa, G.L., Zhang, X., Fathman, C.G., and Glimcher, L.H. (2000). A novel transcription factor, T-bet, directs Th1 lineage commitment. Cell 100, Vahedi, G., Takahashi, H., Nakayamada, S., Sun, H.-W., Sartorelli, V., Kanno, Y., and O Shea, J.J. (2012). STATs shape the active enhancer landscape of T cell populations. Cell 151, Verbist, K.C., Guy, C.S., Milasta, S., Liedmann, S., Kaminski, M.M., Wang, R., and Green, D.R. (2016). Metabolic maintenance of cell asymmetry following division in activated T lymphocytes. Nature 532, Xie, H., Xu, J., Hsu, J.H., Nguyen, M., Fujiwara, Y., Peng, C., and Orkin, S.H. (2014). Polycomb repressive complex 2 regulates normal hematopoietic stem cell function in a developmental-stage-specific manner. Cell Stem Cell 14, Yang, X.P., Jiang, K., Hirahara, K., Vahedi, G., Afzali, B., Sciume, G., Bonelli, M., Sun, H.W., Jankovic, D., Kanno, Y., et al. (2015). EZH2 is crucial for both differentiation of regulatory T cells and T effector cell expansion. Sci. Rep. 5, Zhang, J.A., Mortazavi, A., Williams, B.A., Wold, B.J., and Rothenberg, E.V. (2012). Dynamic transformations of genome-wide epigenetic marking and transcriptional control establish T cell identity. Cell 149, Immunity 46, , April 18, 2017

15 Immunity, Volume 46 Supplemental Information Polycomb Repressive Complex 2-Mediated Chromatin Repression Guides Effector CD8 + T Cell Terminal Differentiation and Loss of Multipotency Simon M. Gray, Robert A. Amezquita, Tianxia Guan, Steven H. Kleinstein, and Susan M. Kaech

16 Supplemental Data Polycomb repressive complex 2-mediated chromatin repression guides effector CD8 + T cell terminal differentiation and loss of multipotency Simon M. Gray 1,*,, Robert A. Amezquita 1,*, Tianxia Guan 1, Steven H. Kleinstein 1,2,3, and Susan M. Kaech 1,,4 Supplemental Experimental Procedures Infections and treatments For primary infection, 5-6 week old mice were infected with 2x10 5 PFU of the LCMV Armstrong strain by intra-peritoneal (i.p.) injection. Viral titers were measured by plaque assay as previously described (Ahmed et al., 1984). For challenge with secondary infection, mice were infected intravenously (i.v.) with 1x10 5 CFU/mouse of recombinant Listeria monocytogenes strain XFL203 which expresses the GP33-41 epitope of LCMV, called LM-33 (Kaech and Ahmed, 2001). For LM-33 titers, organs were harvested and processed in antibiotic free media, then incubated with an equal volume of 1% Triton X-100 made in H2O, and plated in dilutions on Difko Brain Heart Infusion Agar plates (Becton Dickinson). Tamoxifen purchased from Cayman Chemical (Ann Arbor, MI) was dissolved in peanut oil at 20 mg/ml and mice were orally gavaged with 100 µl (2mg) per treatment. Gene expression by qrt-pcr and immunoblotting RNA was isolated from 100,000 sorted cells by TRIzol extraction (Life Technologies) followed by ethanol precipitation. SSRTII (Life Technologies) was used for CDNA preparation. qrt-pcr analysis was performed with an Agilent Mx3000P qpcr system using itaq Universal SYBR Green super mix (Bio-Rad Laboratories). Relative expression was calculated as fold change over the ribosomal gene Rpl9 (L9). The following primers were used: Bach2 forward: 5 -TGAGGTACCCACAGACACCA-3 Bach2 reverse: 5 - TGCCAGGACTGTCTTCACTG-3 ; Foxo1 forward: 5 -TGTCAGGCTAAGAGTTAGTGAGCA-3 Foxo1 reverse: 5 - GGGTGAAGGGCATCTTTG-3 ; Id3 forward: 5 -GACTCTGGGACCCTCTCTC-3 Id3 reverse: 5 - ACCCAAGTTCAGTCCTTCTC-3 ; Tbet forward: 5 -AGCAAGGACGGCGAATGTT-3 Tbet reverse: 5 - GTGGACATATAAGCGGTTCCC-3 ; Rpl9 forward: 5 -TGAAGAAATCTGTGGGTCG-3 Rpl9 reverse: 5 - GCACTACGGACATAGGAACTC-3 ; Tcf7 forward 5 -AGAAGCCAGTCATCAAGAAA-3 Tcf7 reverse 5 - CATTTCTTTTTCCTCCTGTG-3. Immunoblotting was performed as previously described using rabbit monoclonal antibodies from Cell Signaling: EZH2 (D2C9), b-actin (13E5), H3K27me3 (C62B11), H3 (D1H2)(Hand et al., 2007).

17 Chromatin immunoprecipitation (ChIP) and ChIP-Sequencing (ChIP-Seq) ChIP experiments were performed on FACS purified in vivo adoptively transferred P14 + CD8 + T cells. CD8a + CD44 + Thy1.1 + cells were sorted based on expression of KLRG1 and IL7Ra to purified populations of TE (KLRG1 Hi IL7Ra Lo ) and MP (KLRG1 Lo IL7Ra Hi ) cells. Cells were crosslinked with 1% formaldehyde in 10% fetal bovine serum containing RPMI medium for 10 minutes at 37 C. Crosslinking was stopped by addition of 2.5M glycine at 1:20 dilution for 5 minutes at room temperature. Washed cells were lysed and sonicated to obtain chromatin fragments of 150 to 500 base pairs. ChIP was performed on sonicated chromatin from 1-10 million cells with anti-h3k27me3 (Abcam, ab6002) and anti-h3k27ac (Abcam, ab4729) antibodies. Immunoprecipitated DNA was purified, amplified, processed into a library, and sequenced on an Illumina HiSeq 2500 with 4 samples per lane (170M reads are distributed at 42.5M reads per sample with 75bp reads in single-end mode). The resulting fastq file from each sample was first trimmed for TruSeq adapters, and those with 5 /3 end qualities lower than 30/20 were excluded from further analysis. Reads were then aligned to the mm10 (GRCm38) reference genome using bowtie2. Alignments were then filtered to remove duplicate reads and blacklisted regions as defined by ENCODE, and visualized as bigwig files using the DeepTools bamcoverage utility. FastQC was run on both raw and processed data to assess sequencing quality. Conventional ChIP was performed with anti-mouse IgG as negative control and immunoprecipitated DNA was analyzed by qpcr as above with the following primers: Tcf7 TSS forward: 5 -CCTTCGGACTCATTCACCAG-3 Tcf7 TSS reverse:5 -GCGAGGAACAGGACGATAAG-3 ; Id3 TSS forward: 5 - ACTCAGCTTAGCCAGGTGGA-3 Id3 TSS reverse: 5 - CACCTGAAGGTCGAGGATGT-3 ; Bach2 intron 1 forward: 5 - TCAGCCTTTAAGAGCCCAAA Bach2 intron 1 reverse: 5 - AAAGGGGGACCCCTCTAAAT; Ttn forward: 5 - CCGCATCTTTGACACTGAGA Ttn reverse: 5 - AAAGGGTGACCAGGAGCTTT. Analysis of ChIP-seq Data. MACS2 Peak Calling. Peaks were first called for each sample using MACS2 v2.1.0, normalizing relative to input in each case. For H3K27ac, default macs2 settings were used, and for H3K27me3, the --broad setting was invoked. Replication of Peaks. Given that ChIP-seq samples were acquired in separate batches, and additionally since each sample possessed different IP efficiencies, a single significance threshold cutoff for peak calling proved insufficient to filter out low-quality peaks. To address this, q-value thresholds were defined for each individual sample by fitting the distribution of all candidate peak q-values to a mixture of two Gaussian distributions. The top 75% of peaks from the distribution with lower q-values were retained for further analysis. Consensus Peaksets. A consensus peakset was defined first for each condition based on the replicates. To accomplish this, bedtools intersect was used to capture consensus regions, followed by defining a peakset with respect to the given mark by using bedtools merge to combine these consensus

18 peaksets. Annotation of Peaksets. Regions were annotated to the nearest gene TSS using the bedtools v2.26 closest utility for regions with one-to-many mappings for a given transcript, the most highly expressed transcript across MP and TE CD8 + T cells was used to reduce annotation to be one-to-one. Region Functional Annotation. Gene ontology analysis of genomic regions was performed using the Stanford GREAT online tool(mclean et al., 2010). Differential Modification Analysis Read counts were first tallied over the consensus peaksets for each mark per sample in R using the GenomicAlignments v1.10 package s summarizeoverlaps function, ignoring strand and in union mode. Subsequently, analysis was performed DESeq2 v1.14.1, with the read counts normalized by fitting the data to a local smoothed dispersion fit to better capture the observed dispersion-mean relationship. Differential analysis was then performed to determine statistically significant differentially modified regions, defined as those with FDR (Benjamini-Hochberg) less than 0.1 and fold-change greater than 1.2. Statistical analysis and visualization of sequencing data was performed using custom R scripts. Flow cytometry, surface and intracellular staining, peptide stimulations, and antibodies Lymphocyte isolation, LCMV peptide stimulations, MHC class I tetramer production, and surface and intracellular staining were performed as previously described(murali-krishna et al., 1998). Conjugated antibodies were purchased from Biolegend: KLRG1 (2F1), IFNg (XMG1.2), CD62L (MEL-14), CD8a (53-6.7), CD44 (IM7), Thy1.1 (OX-7), T-bet (4B10), TNFa (MP6-XT22); and ebioscience: IL7Ra (A7R34), CD27 (LG.7F9), IL-2 (JES6-5H4), Eomes (Dan11mag). Primary unconjugated rabbit monoclonal antibodies were purchased from Cell Signaling: TCF1 (C63D9), FOXO1 (C29H4), EZH2 (D2C9), and detected by anti-rabbit IgG 647 or anti-rabbit IgG 488 antibody (ThermoFisher Scientific). Foxp3/Transcription Factor Staining Buffer Set (ebioscience) was used for intracellular permeabilization. Flow cytometry was performed on an LSRII (Becton Dickinson) and analyzed with FlowJo software (FlowJo, LLC). Sorting was performed with a BD FACS Aria II (Becton Dickinson). In vitro culture and CellTrace Violet proliferation assay For in vitro cultures, 1x10 6 naïve splenocytes were cultured with GP33-41 peptide, or acd3 (1µg/ml) (Biolegend) and acd28 (0.5 µg/ml) (Biolegend), and IL-2 (10ng/ml) (R&D Systems) for the specified time. For proliferation assays, naïve splenocytes were labeled with CellTrace Violet (ThermoFisher Scientific) according to the vendor protocol, then either cultured in vitro as above or adoptively transferred to recipients and infected with LCMV Armstrong as above.

19 Mixed bone marrow chimeras To generate mixed bone marrow chimeras, bone marrow from Thy1.1 + Ly5.2 + EZH2 +/+ mice was mixed in a 80:20 ratio with bone marrow from Thy1.2 + Ly5.2 + Ezh2 f/f GzmBCre + mice and used to reconstitute naïve wildtype lethally irradiated Thy1.2 + Ly5.1 + recipient mice. Two months post-reconstitution, mixed bone marrow chimeric mice were infected with LCMV-Armstrong and lymphocytes were isolated and analyzed as described above. Retroviral overexpression Retrovirus supernatant was produced in HEK293T cells. MSCV-empty-vector-GFP and MSCV-T-bet-GFP (Tbet-overexpression)(Szabo et al., 2000) were obtained from L. Glimcher (Dana Farber, Boston, MA). P14 + TCR transgenic mice were superinfected i.v. with 2x10 6 PFU of the LCMV-Armstrong strain. 24 hours later, P14 + splenocytes were spin-transduced at 6 million cells per 24 well-plate well for 90 minutes at 37 C with viral supernatants from 293T cells supplemented with 8μg/mL of hexadimethrine bromide (Sigma, H9268), then 1x10 5 P14 + CD8 + T cells were adoptively transferred to naïve B6 recipient mice that were subsequently infected with 2x10 5 PFU of LCMV-Armstrong strain.

20 SUPPLEMENTAL FIGURES and SUPPLEMENTAL FIGURE LEGENDS Figure S1. (related to Figure 1)

21 Figure S1. (related to Figure 1) Quality control and summary statistics of H3K27ac and H3K27me3 ChIP-seq in Memory Precursors and Terminal Effector cells. A) Table summarizing quality control metrics for replicates of H3K27ac and H3K27me3 ChIP-Seq datasets following data processing. Dup% = % duplicate reads, ReadL = read length, FragL = fragment length, RelCC = relative cross-coverage score, SSD = squared sum of deviations, RiP% = % reads in peaks. B) Stacked bar graph of quality control (QC) pass-fail rate of reads from replicates of H3K27ac and H3K27me3 ChIP-Seq datasets. Reads passing QC are in blue and reads failing QC are in gray. The percentage of reads passing QC is shown below each bar. C) Bar graph showing the number of consensus peaks (blue) shared between replicates (light gray and dark gray) of H3K27ac and H3K27me3 ChIP-Seq datasets. D-F) Consensus peaks were annotated to the TSS of the nearest gene. Shown are bar plots pertaining to the number of regions falling within a given distance for all H3K27ac and H3K27me3 regions (D), then separately for DMRs versus Common regions (E), and finally only for DMRs (F). G) Scatter plot of the normalized mean deposition of H3K27ac (left) and H3K27me3 (right) across MP and TE cells versus the normalized mean gene expression in MP and TE cells, where each region was associated with the gene with the nearest TSS as shown in D-F. H) Deposition of H3K27me3 and H3K27ac in MP and TE cells centered on Common regions +/-10kb. Common regions are defined as regions with FDR > 0.1 and/or fold-change < 1.2 in volcano plots in Fig 1A (Cluster 5) and B (Cluster 6). Line plots at top show the summary distributions across each cluster for each H3K27ac and H3K27me3 in MP and TE cells, respectively. I) Line plots show the ratio of H3K27ac or H3K27me3 comparing MP versus TE cells for each set of Common regions in clusters 5 and 6 (top), and similarly, the ratio of H3K27ac to H3K27me3 within both MP and TE cells (bottom). Data shown contain the union of significant consensus peaks identified across two independent biological replicates of ChIP-Seq experiments for H3K27ac and H3K27me3 (A-I; n=10-20 mice/group/replicate).

22 Figure S2. (related to Figure 2)

23 Figure S2. (related to Figure 2) T resident memory genes and Pro-survival genes exhibit substantially more H3K27me3 and less H3K27ac deposition in TE versus MP cells. Alignment tracks of H3K27ac and H3K27me3 deposition across MP and TE cells at tissue resident memory (TRM) and pro-survival genes (TRM signature genes were defined by (Mackay et al., 2013)). Statistically significant differentially modified regions (DMRs) are marked by rectangles below tracks with red bars representing enrichment in MP cells and blue bars representing enrichment in TE cells. Black bars demarcate called peaks that are not enriched in one population over the other. Data shown contain the union of significant consensus peaks identified across two independent biological replicates of ChIP-Seq experiments for H3K27ac and H3K27me3 (A-B; n=10-20 mice/group/replicate).

24 Figure S3. (related to Figure 4)

25 Figure S3. (related to Figure 4) Validation of Ezh2 deletion characteristics on CD8+ T cell effector development and protein expression. A) MFI of EZH2 protein level at d4.5 p.i. in virus-specific (D b GP33-41 and D b NP MHC class I tetramer + ) Ezh2 f/f and Ezh2 f/f GzmBCre + CD8 + T cells from the peripheral blood. B) Congenically mismatched naïve P14 + Ezh2 f/f (red) and Ezh2 f/f GzmBCre + (blue) CD8 + T Cells were pulsed with CellTrace Violet and adoptively co-transferred to the same congenically mismatched WT recipient mouse, which was infected with LCMV Armstrong. Plot shows in vivo proliferation of splenic P14 + CD8 + T cells at 60 hrs p.i. P14 + cells from an uninfected recipient are shown as non-divided control. Bar graph shows MFI of EZH2 protein level in splenic P14 + Ezh2 f/f (solid) and Ezh2 f/f GzmBCre + (open) CD8 + T cells at day 0, 2, 2.5, and 4 p.i. C) Ezh2 f/f (solid) and Ezh2 f/f CD4Cre + (open) mice were infected with LCMV-Armstrong and splenic GP33-41-specific CD8 + T cells were enumerated at d8 p.i; KLRG1, CD27, and CD62L expression were determined at d8 p.i. on splenic GP33-41-specific CD8 + T cells (Ezh2 f/f is black line, Ezh2 f/f CD4Cre + is gray line); the percentage of splenic GP33-41-specific CD8 + T cells in KLRG1/IL7R subsets was determined at d8 p.i.; and the intracellular mean fluorescence intensity (MFI) of the indicated TFs was measured by flow cytometry. D) Mixed 80% Ezh2 f/f GzmBCre + to 20% Ezh2 f/f bone marrow chimeras (BMC) were infected with LCMV- Armstrong and splenic virus-specific (D b GP33-41 and D b NP MHC class I tetramer + ) CD8 + T cells were examined at d8 p.i. Plots show number of LCMV-specific CD8 + T cells, number of LCMV-specific KLRG1 + CD8 + T cells, percentage of D b GP33-41 tetramer + CD8 + T cells expressing KLRG1, and MFI of TCF1, FOXO1, and Tbet in splenic D b GP33-41 tetramer + Ezh2 f/f and Ezh2 f/f GzmBCre + CD8 + T cells paired from the same BMC mouse. Data shown are representative of two (B) or five (A) independent experiments, or cumulative of three (D) or five (C) independent experiments. Data are expressed as mean ± SD. *p<0.02, ***p<0.001, ****p< n=3-5 mice/group/experiment (C-D) or n=4-10 mice/group/experiment (A).

26 Figure S4. (related to Figure 4)

27 Figure S4. (related to Figure 4) EED deletion is functionally similar to EZH2 deletion day 8 postinfection. Eed f/f and Eed f/f CD4Cre + mice were infected with LCMV Armstrong and the number of splenic virus-specific (D b GP33-41 and D b NP MHC class I tetramer + ) CD8 + T cells were enumerated at d8 p.i. B) Contour plots (left) show surface expression of KLRG1 and IL7R on splenic D b GP33-41 tetramer + Eed f/f (solid) and Eed f/f CD4Cre + (open) CD8 + T cells from d8 p.i. Bar graph (right) shows average percentages for each subset. C) Bar graph shows number of splenic virus-specific (D b GP33-41 and D b NP MHC class I tetramer + ) Eed f/f (solid) and Eed f/f CD4Cre + (open) CD8 + T cells in KLRG1/IL7R subsets at d8 p.i. D) Ezh2 f/f, Ezh2 f/f CD4Cre +, Eed f/+ CD4Cre + and Eed f/f CD4Cre + CD8 + T cells were activated in vitro with acd3 and acd28 for 3 days, sort purified, and probed for H3K27me3 by western blot. Data shown are representative of two (D) or cumulative of two (A-C) independent experiments (n=3-5 mice/group/experiment). Data are expressed as mean ± SD. *p<0.05, ***p<0.001, ****p<0.0001

28 Figure S5. (related to Figure 5)

29 Figure S5. (related to Figure 5) Assessment of EZH2 deletion after completion of tamoxifen treatment in Ezh2 f/f GzmB-ER T2 Cre. EZH2 deletion was assessed by genomic DNA PCR on Ezh2 f/+ GzmBER T2 Cre + and Ezh2 f/f GzmBER T2 Cre + CD8 + T cells purified by FACS at d28 p.i. (20 days after tamoxifen treatment). Purified Ezh2 f/f and Ezh2 f/f CD4Cre + CD8 + T cells activated in vitro for 1.5 days served as a positive control.

30 Figure S6. (related to Figure 6)

31 Figure S6. (related to Figure 6) Profiling of mrna expression in WT CD8 + T cells at early timepoints. A) Tbx21, Tcf7, Foxo1, Bach2, and Id3 mrna were measured using qrt-pcr in sort purified WT P14 + CD8 + T cells from days 0 (naïve), 1.5, and 4.5 p.i. with LCMV-Armstrong. Data from days 0 and 4.5 p.i. is the same as Fig 6A. Data representative of 3 independent experiments. n=2-3/group/experiment. Data are expressed as mean ± SD. *p<0.05, **p<0.01, ***p<0.001

32 Figure S7. (related to Figure 7)