The Transcription factor Runx3 establishes chromatin accessibility of cis-regulatory landscapes that drive memory cytotoxic T lymphocyte formation

Summary

T cell receptor (TCR) stimulation of naïve CD8⁺ T cells initiates reprogramming of cis-regulatory landscapes that specify effector and memory cytotoxic T lymphocyte (CTL) differentiation. We mapped regions of hyper-accessible chromatin in naïve cells during TCR stimulation and discovered that the transcription factor (TF) Runx3 promoted accessibility to memory CTL-specific cis-regulatory regions prior to the first cell division, and was essential for memory CTL differentiation. Runx3 was specifically required for accessibility to regions highly enriched with IRF, bZIP and Prdm1-like TF motifs, upregulation of TFs Irf4 and Blimp1, and activation of fundamental CTL attributes in early effector and memory precursor cells. Runx3 ensured nascent CTLs differentiated into memory CTLs by preventing high expression of the TF T-bet, slowing effector cell proliferation, and repressing terminal CTL differentiation. Runx3 overexpression enhanced memory CTL differentiation during iterative infections. Thus, Runx3 governs chromatin accessibility during TCR stimulation and enforces the memory CTL developmental program.

Graphical Abstract

Chromatin accessibility in naïve CD8⁺ T cells is globally reprogrammed during infection. Wang et al. show that the transcription factor Runx3 programs chromatin accessibility during stimulation of naïve CD8⁺ T cells, which establishes early transcriptional circuits that drive differentiation of nascent cytotoxic T lymphocytes (CTLs), and also represses terminal differentiation, to ensure they develop into long-lived memory CTLs.

Introduction

During acute infections, stimulation of T cell antigen receptors (TCRs) and co-stimulatory receptors on antigen specific naïve CD8⁺ T cells initiates their exponential accumulation and differentiation into a population of nascent CTLs that harbors the precursors of long-lived memory CTLs (CTLmem) (Chang et al., 2014). Near the peak response, the population of differentiated cells is heterogeneous and comprises both terminal effector (TE) CTLs, which have short half-lives after the infection resolves, and memory precursor (MP) effector CTLs, which efficiently generate CTLmem. Individual naïve CD8⁺ T cells have the potential to differentiate into both MP and TE CTLs, and ultimately CTLmem cells (Buchholz et al., 2016), but the transcriptional mechanisms that establish nascent CTLs, and then ensure developing MP CTLs remain differentiated from TE CTLs are still unclear.

Modulation of chromatin structure underlies the identity of effector CD4⁺ and CD8⁺ T cells (Agarwal and Rao, 1998; Pipkin et al., 2007). The most extensive chromatin remodeling occurs as naïve CD8⁺ T cells mature into each CTL subset (Araki et al., 2009; He et al., 2016; Russ et al., 2014; Scharer et al., 2017; Scott-Browne et al., 2016; Yu et al., 2017). Although differences exist in which cis-regulatory regions are accessible in CTLmem, as compared to TE and MP cells in the effector phase, there is extensive overlap in the regions that are accessible in all subsets. Thus, distinct CTL “lineages” utilize discrete suites of cis-regulatory regions, but also rely an extensive common cis-regulatory network that is established after naïve T cell activation (He et al., 2016; Scharer et al., 2017; Scott-Browne et al., 2016; Yu et al., 2017). While multiple TFs are known to be essential for CTL differentiation, those that pioneer chromatin accessibility to establish CTL differentiation are still unclear (Chang et al., 2014).

In this study, we found that cis-regulatory regions which are opened upon TCR stimulation, and that correspond to those which are accessible in fully developed circulating CTLmem, were enriched with motifs recognized by the Runx-family of TFs. We present multiple loss-of-function and gain-of-function studies which have delineated that Runx3 was essential for programming chromatin accessibility characteristic of CTLmem, inducing transcription typical of normal MP CTL, repressing essential features of terminal CTL and ensuring genuine CTLmem develop during viral infection. These results suggest that Runx3 programs CTLmem formation by regulating chromatin accessibility during TCR stimulation, and curtailing terminal differentiation.

TCR stimulation rapidly induces chromatin accessibility of memory CTL-specific cis-regulatory regions that are enriched with Runx-family TF motifs

Differential TCR signals “program” memory CTL development (Teixeiro et al., 2009), and a single brief encounter with peptide-MHC by the TCR is sufficient to induce naïve CD8⁺ T cells to differentiate into memory CTL (Kaech and Ahmed, 2001; van Stipdonk et al., 2001). To explore this process at the chromatin level, we used ATAC-seq to map regions of hyper-accessible chromatin genome-wide in naïve CD8⁺ T cells during TCR stimulation (Figure S1A and B) (Buenrostro et al., 2013). Prior to the first cell division, TCR signals induced remodeling of putative cis-acting sequences that significantly overlapped biologically meaningful regions that have been found to be “opened” or “closed” in differentiated CTL subsets after Lymphocytic choriomeningitis virus (LCMV) infection, as compared to naïve cells (Fig. 1A and 1B and Figure S1C) (Scharer et al., 2017; Scott-Browne et al., 2016; Yu et al., 2017). Thus, TCR stimulation induces de novo chromatin remodeling prior to cell division that opens a substantial fraction of the cis-regulatory landscape that is accessible in fully developed CTLmem.

Figure 1. Runx-factor motifs are enriched within accessible cis-acting regions of CTLmem that are opened upon TCR stimulation.

Naïve CD8⁺ T cells (CD44^lo CD62L^hi) purified from C57BL/6 mice were analyzed by ATAC-seq before and after TCR stimulation. Data are from two biological replicates.

(A) The absolute number and overlap of all ATAC-seq regions with significantly increased (top) and decreased (bottom) accessibility in TE CTL (red), CTLmem (green), and TCR-stimulated cells (no fill) relative to naïve CD8⁺ T cells (DEseq2 P<0.05, log₂ fc >1) are represented in the Venn diagrams. The overlap of TCR-regulated ATAC-seq regions and those found in each CTL subset is summarized (right).

(B) Scatterplots depict changes in accessibility (fold-change in Tn5 insertions) within ATAC-seq regions upon TCR stimulation (y-axis), and in each differentiated CTL (x-axis), compared to in naïve cells. All ATAC-seq regions (dots) that exhibited significant changes in accessibility (both x-axis and y-axis, P<0.05) are plotted. Peaks near representative genes are highlighted (blue).

(C) The heatmap plots the frequencies of the most enriched TF motifs in differentially accessible regions. Plotted motifs have on-target frequency >15%, on-target frequency/background frequency ratio ≥1.5, and P≤0.05. For all enriched motifs, see Table S1.

(D) The heatmap shows fold changes in PageRank scores for TF motifs in ATAC-seq regions of differentiated CTL subsets versus naïve cells, ordered based on highest PageRank CTLmem/naïve cell ratio, after selecting TFs with RNA-seq TPM in naïve cells ≥1, and a PageRank score in CTLmem ≥0.003.

See also Figure S1.

To discover potential transcription factors (TFs) important for this process, we identified TF motifs that were both statistically enriched and highly frequent within all regions that underwent differential remodeling (Figure 1C and Tables S1 and S2). Motifs recognized by TFs from the Runx, Ets, IRF and bZIP families were the most highly enriched in regions that were opened upon TCR stimulation, and of these, Runx and Ets motifs were the most strongly enriched in regions that were also accessible in CTLmem (Figure 1C). We stratified the potential operational activity of TFs in the different mature CTL lineages using PageRank (Figure 1D), an algorithm that prioritizes TFs whose cognate motifs are enriched within accessible ATAC-seq regions that are annotated to genes whose expression correlates with the specific CTL phenotypes (Yu et al., 2017). Among all of the prioritized TFs, the PageRank scores for Runx3 and Runx2 were strongly enhanced in CTLmem compared to naïve cells. The enrichment of Runx-motifs after TCR stimulation and in developed CTLmem suggested that Runx TFs could be early drivers of CTLmem differentiation.

Runx3-deficiency impairs differentiation of MP CTL and increases TE CTL differentiation

Runx1, Runx2, and Runx3 encode the Runt-family TFs of mammalians (Levanon et al., 1994). All three Runx TFs were substantially expressed in differentiated CTL (Figure S1D–S1F). Each Runx TF can bind the consensus Runx-motif with high affinity when associated with their heterodimeric partner Cbfb (Bartfeld et al., 2002; Tahirov et al., 2001). To define roles for each factor, we retrovirally transduced TCR-transgenic P14 CD8⁺ T cells that are specific for the LCMV GP33 epitope with distinct short hairpin RNAs embedded in a microRNA context (shRNAmirs) that targeted each of the Runx TFs, and interrogated their effects on formation of differentiated CTL subsets in wildtype (WT) hosts after LCMV infection (Figure 2A, experimental schemes and S2A), as judged by surface KLRG1 and CD127 (IL-7Rα) expression (Chen et al., 2014).

Figure 2. Depletion of Runx3 in adoptively transferred P14 cell ablates their development into MP and DP cells during LMCV infection.

Naïve P14 cells were activated in vitro, transduced with retroviral constructs and analyzed after adoptive transfer to WT hosts subjected to LCMV infection.

(A) Schematic depicts the approach using shRNAmirs.

(B) Charts show the absolute numbers of shRNAimr-transduced P14 cells after LCMV infection.

(C-D) Flow cytometry plots show representative surface staining after gating transduced P14 cells. Data are pooled from two independent experiments (B-D).

(E) The adoptive transfer approach to analyzing conditional Runx3-disruption (left). Flow cytometry plots show representative surface staining after gating P14 cells that activated the Cre-dependent YFP reporter (top). Data are one representative experiment of at least ten replicates.

Bar charts show sample means. Symbols are values from individual mice. *P<0.05, **P<0.01 and ***P<0.001 by two-tailed unpaired Student’s t test.

See also Figure S2.

Previous studies have demonstrated that KLRG1^lo CD127^lo early effector (EE) cells can differentiate into TE (KLRG1^hi CD127^lo) cells, double positive (DP) effector (KLRG1^hi CD127^hi) cells and MP effector (KLRG1^lo CD127^hi) cells (Plumlee et al., 2015). Multiple distinct shRNAmirs specific for Runx3 or Cbfb modestly reduced overall P14 cell numbers (Figure S2A and 2B). Runx3 shRNAmirs increased the percentage of EE cells, suggesting that reduced Runx3 expression impairs differentiation of EE into more mature CTL subsets (Figure 2C), and consistent with this, depletion of either Runx3 or Cbfb decreased the percentages of both DP and MP cells. However, depletion of either TF also increased the percentages of TE cells (Figure 2C and D). Thus, reduced Runx3 expression impairs DP and MP CTL development, and skews differentiation toward a TE CTL phenotype. In contrast, Runx1 or Runx2 depletion did not consistently produce phenotypes (Figure S2A–S2C). In line with this, Runx3 was the most highly expressed Runx-TF in naïve, memory precursor and memory CTL subsets (Figure S1D–F).

To extend the RNAi studies, we used retrovirally-delivered Cre-recombinase to disrupt the Runx3 locus in P14 cells concurrent with TCR activation, and analyzed the effect of its loss after adoptive transfer of these cells into WT hosts infected with LCMV (Figure 2E and S2D–E). Runx3-disruption reduced the accumulation of P14 cells after LCMV infection (data not shown), blocked differentiation of both DP and MP phenotype cells and accentuated the frequencies of TE-phenotypic cells, similar to the RNAi results (Figure 2E and S2E). Thus, Runx3 is required in a cell-intrinsic fashion to promote DP and MP CTL differentiation, and restrain TE CTL differentiation during LCMV infection.

To examine the role of Runx3 in endogenous antigen-specific cells, we crossed Runx3^fl/fl mice to dLck-Cre and ERT2-Cre mice to facilitate conditional Runx3 disruption in post-thymic CD8⁺ T cells (Figure S3) (Naoe et al., 2007; Ruzankina et al., 2007; Zhang et al., 2005). To monitor Cre-activity in cells from these mice, they were also crossed to Rosa26 EYFP reporter mice (abbreviated sYFP), which carry a LoxP-flanked transcriptional terminator upstream of EYFP (Srinivas et al., 2001). Inactivation of one Runx3 allele in Runx3^+/fl sYFP dLck-Cre mice marginally reduced overall effector cell accumulation during LCMV infection (Figure 3A and S3A–S3C). However, this ablated most LCMV-tetramer-specific cells with DP and MP CTL phenotypes and increased the fraction of cells with a TE CTL phenotype (Figure 3B and S3D), reaffirming the RNAi phenotypes.

Figure 3. Conditional Runx3-disruption in post-thymic CD8⁺ T cells ablates development of endogenous LCMV-specific MP cells and CTLmem.

Runx3^+/+, Runx3^+/fl, and Runx3^fl/fl sYFP mice transgenic for either dLck-Cre, or ERT2-Cre transgenes were analyzed after LCMV-Armstrong (LCMV-Arm) infection.

(A and C) The absolute numbers of YFP⁺ CD44^hi CD8⁺ cells are summarized 8 days p.i. in the spleen.

(B and D) Flow cytometry plots show representative staining of LCMV-specific (GP33 tet⁺) CD8⁺ T cells in the spleen. The frequencies of CD8⁺ YFP⁺ GP33-tet⁺ cells bearing the indicated CTL phenotypes are summarized.

(E and F) Flow cytometry plots show representative GP33 tetramer staining, and the bar charts summarize the frequencies and numbers of GP33 tet⁺ cells in the spleen.

(G) Thy1-disparate P14 cells transduced with either control or Cbfb-specific shRNAmirs were mixed and co-transferred to naïve hosts. Representative flow cytometry plots show frequencies transduced P14 cells 30 days after LCMV infection, and the bar charts summarize the absolute numbers of transduced P14 cells in the spleen for all mice.

(H) The flow cytometry plots show representative phenotypes of gated GP33 tet⁺ cells in the spleen, and the bar charts summarize the frequencies of GP33 tet⁺ cells with the indicated CTL phenotypes.

(I) Viral titers were determined by plaque assay for individual mice.

(J) Naïve host mice received no transfer (No P14) or 2.5×10⁵ Cre-retrovirus-transduced memory-like P14 cells that were Runx3 WT (+/+) or Runx3-deficient (fl/fl) (see methods) before challenge with 50,000 CFU of LM-GP33. LM-GP33 colony forming units (CFU) in the spleen were quantified by limiting dilution 4 days after challenge.

(K-L) YFP⁺ CD8⁺ CD44^hi cells were FACS-isolated from the spleens of Runx3^+/+ and Runx3^fl/fl sYFP dLck-Cre mice 8 days after LCMV-Arm infection. Whole cell lysates were analyzed by immunoblotting (K), and RNA was extracted and analyzed by qRT-PCR (L).

Data are from at least two biological replicates. Bar charts indicate sample means. Symbols indicate values from individual mice. *P<0.05, **P<0.01, ***P<0.001, two-tailed unpaired Student’s t test. See also Figure S3.

Disruption of both Runx3 alleles in Runx3^fl/fl sYFP dLck-Cre mice virtually eliminated Runx3 protein expression in FACS-purified YFP⁺ CD8⁺ cells (Figure S3A and S3B), strongly impaired accumulation of LCMV-specific cells (Figure 3A and S3C), and decreased the frequencies of NP396-tetramer⁺ cells (Figure S3C), which normally dominate the response. This correlated with increased fractions of EE cells (Figure 3B and Figure S3D), and decreased frequencies of DP cells, beyond that caused by loss of only one Runx3 allele (Figure 3B and S3D). In addition, although disruption of both Runx3 alleles also increased the relative frequencies of TE cells, and decreased the frequencies of MP cells compared to WT controls, these effects were comparable to those observed upon loss of only one Runx3 allele (Figures 3B and S3D).

To confirm these results in a setting wherein only a minority of all endogenous T cells inactivated Runx3, we treated sYFP ERT2-Cre mice carrying WT or conditional Runx3 alleles with tamoxifen, to induce Cre activity, and subsequently infected the mice with LCMV (Figure S3E and S3F). In this setting, 10–40% of CD8⁺ T cells activated YFP expression (data not shown), which correlated with gene-dose dependent reduction in Runx3 protein in FACS-purified YFP⁺ CD8⁺ T cells (Figure S3F). Reduced Runx3 expression in Runx3^+/fl sYFP ERT2-Cre mice did not impair accumulation of LCMV-specific cells in the spleen (Figure 3C), but impaired the fractions of MP and DP phenotype cells and increased the fraction of TE phenotypic cells (Figure 3D and S3H). Further reduction of Runx3 expression in cells from Runx3^fl/fl sYFP ERT2-Cre mice impaired cell accumulation (Figure 3C), and generation of both DP and MP phenotype cells, while also increasing the fractions of TE phenotype cells.

Taken altogether, these results indicate that partial Runx3 deficiency impairs the differentiation of both DP and MP phenotype cells, and accentuates the development of cells with a TE phenotype, without strongly impairing effector cell accumulation. In contrast, complete Runx3-deficiency severely reduces effector cell accumulation and arrests differentiation near the EE cell stage, and strongly blocks differentiation of both DP and MP cells, but permits some cells to acquire the TE CTL phenotype, although their numbers are reduced.

Runx3 is essential for memory CTL formation

At memory time-points after LCMV infection, there were many fewer LCMV-specific CD8⁺ T cells in both Runx3^fl/fl sYFP dLck-Cre (Figure 3E) and Runx3^fl/fl sYFP ERT2-Cre mice (Figure 3F) compared to control mice. Adoptively transferred P14 cells carrying Cbfb shRNAmirs also generated at most 10-fold fewer CTLmem than control shRNAmir transduced cells in WT hosts infected with LCMV, indicating that reduced CTLmem numbers in the absence of Runx-activity is cell-intrinsic (Figure 3G). The residual Runx3-deficient cells that persisted at memory time points in dLck-Cre mice were more effector-like phenotypically (Figure 3H). Delayed viral clearance in these mice could have promoted this phenotype, but YFP⁺ GP33 tet⁺ cells in Runx3^fl/fl sYFP ERT2-Cre mice (only a minority of cells delete Runx3) also appeared more effector-like beyond day 50 p.i. (data not shown). Regardless, Runx3-deficient CD8⁺ T cells were not protective in effector or memory-like settings (Figure 3I and 3J), and exhibited defective cytolytic effector gene expression (Figure 3K and 3L), as expected (Cruz-Guilloty et al., 2009; Shan et al., 2017). Thus, Runx3 is essential for formation of bona fide CTLmem.

Runx3 is essential during TCR stimulation for chromatin accessibility of the memory CTL-specific cis-regulatory landscape

To explore early roles of Runx3 within the cis-regulatory landscapes of CTL, we analyzed chromatin accessibility during TCR stimulation of naïve CD8⁺ T cells sorted from Runx3^+/+ and Runx3^fl/fl sYFP dLck-Cre mice. Accessible chromatin regions in unstimulated WT and Runx3-deficient naïve cells exhibited minimal differential accessibility (Figure 4A, top). However, most regions that remodeled upon TCR stimulation in WT cells were not remodeled in Runx3-deficient cells (Figure 4A and 4B). Although Runx3 was required during TCR stimulation for the accessibility of regions that are also accessible in all mature CTL subsets (i.e., TE, MP and CTLmem) (Figures 4C–4D and S4A–S4B), it preferentially affected opening of regions that are more accessible in mature CTLmem, when compared to either naïve cells (Figure 4D, top panel, upper right quadrant), or to TE CTL (Figure 4D, bottom plot, upper right quadrant). Those that were specifically accessible in CTLmem compared to TE CTL were annotated near genes enriched for immunological biological processes (Figure S4C, upper panel). Conversely, regions whose accessibility increased in the absence of Runx3 during TCR stimulation were preferentially accessible in naïve cells (Figure 4D, top plot, lower left quadrant), or in TE CTL as compared to CTLmem (Figure 4D, bottom plot, lower left quadrant), and the latter were annotated to genes enriched for non-immunological processes (Figure S4C, lower panel). Together, these results suggest that Runx3 promotes accessibility to CTLmem-specific cis-acting regions during TCR stimulation. Consistent with this interpretation, Runx3 binding sites overlapped a significantly greater fraction of CTLmem-associated accessible regions, than TE CTL-associated accessible regions (20.6% of CTLmem versus 13.1% of TE CTL sites, P<0.0001, Chi-square, Figure S4D) (Lotem et al., 2013; Scott-Browne et al., 2016).

Figure 4. Runx3 establishes CTLmem-associated chromatin accessibility during TCR stimulation.

Naïve, CD44^lo CD62L^hi YFP⁺ CD8⁺ T cells from Runx3^+/+ (WT) or Runx3^fl/fl (KO) sYFP dLck-Cre mice were purified and analyzed before and after TCR stimulation. Data from two biological replicates were used for all analyses.

(A) The Venn diagram depicts overlapping ATAC-seq regions (≥15 TPM) in naïve WT and KO cells (top); only 161 peaks differed significantly (DEseq2 P<0.05, log₂ fc >1, not shown). ATAC-seq regions that increased (opened) or decreased (closed) in accessibility upon TCR stimulation compared to naïve cells were identified (DEseq2 P<0.05, log₂ fc >1). Venn diagrams depict overlap between differential ATAC-seq regions of WT and KO cells.

(B) Differentially accessible ATAC-seq regions 24 hours after TCR stimulation relative to naïve cells were identified (DEseq2 P<0.01). The change in accessibility within these regions in KO versus WT cells (left side) upon TCR stimulation, and in naïve versus TCR-stimulated WT cells was plotted in the heatmap (log2 fold change in Tn5 insertion sites), ordered by increasing accessibility in WT cells 24 hours after stimulation. Representative genes near ATAC-seq regions are indicated to left.

(C) Charts summarize the percentages of ATAC-seq regions that became accessible relative to naïve cells upon TCR stimulation (DEseq2 P<0.05, log₂ fc >1) of those that are also more accessible in each CTL subset relative to naïve cells (DEseq2 P<0.05, log₂ fc >1).

(D) The change in accessibility of ATAC-seq regions (gray dots) that differ significantly in WT versus KO cells 24 hours post TCR stimulation (DEseq2 P<0.05) and in CTLmem versus naïve CD8⁺ T cells (top plot), or TE CTL (bottom plot) (DEseq2 P<0.05) are plotted. Total (black) and Runx3-occupied (red) regions are indicated. Selected genes proximal to ATAC-seq regions are highlighted (blue).

(E) The change in TF motif frequencies in accessible regions of KO versus WT cells identified in (A) during TCR stimulation are expressed as ratios in the heatmap (Table S2, and methods for ranking motifs). Plotted TF motifs selected as in Figure 1C.

(F) Charts summarize absolute numbers of TF motifs in accessible regions of WT and KO cells as identified in (A) after TCR stimulation.

(G) The in vivo TF footprints derived from Tn5 insertion frequency (ATAC-seq reads) over representative TF motifs within accessible ATAC-seq regions identified 24 hours after TCR stimulation as in (A).

(H) Purified P14 CD8⁺ T cell nuclei were extracted with the indicated NaCl concentrations. Supernatants were cleared by centrifugation. Pellets were completely solubilized. Equivalent pellet and supernatants were analyzed by immunoblotting.

See also Figure S4.

To define the cis-regulatory sequences controlled by Runx3, we analyzed the enrichment of TF motifs within Runx3-dependent accessible regions. The frequencies of motifs recognized by Runx, bZIP, RHD, IRF and Prdm1-like factors were decreased in accessible regions of Runx3-deficient cells after TCR stimulation (Figures 4E and Table S2, all enriched motifs), which resulted from reduced numbers of these motifs in accessible regions (Figure 4F, left panels). Conversely, frequencies of ETS, KLF, NRF and bHLH motifs were increased in accessible regions of Runx3-deficient cells after TCR stimulation (Figures 4E and Table S2, all enriched motifs), due to increased numbers of these motifs in accessible regions (Figure 4F, right panels). Similar effects were observed when CTL subset-specific cis-regions were analyzed (Figure S4E). Accordingly, in vivo TF footprints at representative Runx, IRF, bZIP, and Prdm1 motifs were evident in WT cells but not in Runx3-deficient cells, indicating cognate TFs inefficiently bind these motifs in the absence of Runx3 (Figure 4G, top panels), whereas footprints over ETS, NRF1, bHLH, or KLF motifs were present in both WT and Runx3-deficient cells, indicating cognate TFs bind these sites in the absence of Runx3 (Figure 4G, bottom panels). Thus, Runx3 is essential during TCR stimulation for the accessibility of cis-regulatory regions and binding of other key TFs that drive CTL differentiation.

To determine whether Runx3 associates with chromatin prior to TCR stimulation, we isolated nuclei from purified naïve P14 CD8⁺ T cells and analyzed its resistance to extraction with salt. Most Runx3 and histone H1 was not extracted in the presence of 150mM NaCl (Figure 4H), whereas histone H1 was extracted completely with 400 mM NaCl, as expected. In contrast, Runx3 was not extracted entirely until most of the nucleosome core protein histone H2 had also been extracted. Thus, a substantial amount of Runx3 associates with chromatin in naïve cells before TCR stimulation.

Runx3 drives transcription that induces differentiation of authentic MP CTL and represses differentiation of TE CTL

To define how Runx3 controls gene expression in developing CTL we used RNA-seq to analyze phenotypically defined CTL subsets during LCMV infection (Figure 5A). Gene expression in the different phenotypic CTL subsets of both WT and Runx3-deficient cells diverged progressively from naïve cells, confirming that Runx3-deficient cells are activated in vivo, and undergo gene expression changes throughout the response to LCMV infection (Figure 5B, left PCA plot). However, gene expression in Runx3-deficient cells was distinct from WT cells, especially when comparing EE and MP phenotypic subsets isolated on day 8 post infection (Figure 5B, right PCA plot).

Figure 5. Runx3 is required for transcription that drives normal MP CTL differentiation.

(A) Scheme for RNA-seq analysis of mRNA in FACS-purified P14 CTL subsets. Two biological replicates were performed.

(B) PCA analysis of all WT and KO Runx3 CTL subsets and naïve CD8⁺ T cells purified from P14 Tcra^−/− mice (left plot). All ex vivo subsets from day 8 were analyzed separately (right plot).

(C) The heatmaps depict enrichment of genes in each CTL subset that were upregulated (Up) or downregulated (Down) in Runx3-deficient cells for previously defined gene clusters (Best et al., 2013) (pAdj ≤0.05; DEseq2, see Methods). Gene clusters were colored based on similarity.

(D) The change in expression of all genes (dots) in Runx3-deficient cells is shown in the Volcano plots. Genes within gene-cluster families are colored, and exemplar TF gene symbols are indicated. Colored values denote the number of differentially expressed genes in each gene cluster family (pAdj <0.05). See also Figure S5.

(E and F) Congenic P14 Runx3^+/+ and Runx3^fl/fl sYFP cells were transduced with Cre-retrovirus (Cre-RV), mixed 1:1 and transferred to naïve hosts that were infected with LCMV-Arm. On day 8 p.i., mice were injected intravenously with anti-CD8a antibodies to label intravascular cells. (E) Flow cytometry plots show representative in vivo and ex vivo CD8 staining. Bar charts summarize the sample means for individual mice in one representative experiments. P-value was calculated using a two tailed, paired Student’s t test. (F) Plots show representative intracellular TCF1 staining, and denote TCF1 MFI, in phenotypic CTL subsets.

(G and H) Chromatin-associated RNA expression was quantified by RNA-seq analysis of two biological replicates. Purified CD8⁺ T cells from naïve Runx3^+/+ (WT) and Runx3^fl/fl (KO) mice were transduced with Cre-cDNA or Runx3-cDNA retroviruses (OE), followed by FACS-purification and culture with 100 U/mL rhIL-2 for 6 total days (Pipkin et al., 2010). (G) Differentially expressed genes in Runx3 KO and OE cells were identified relative to WT cells (DEseq2, pAdj ≤0.05). The heatmap depicts the change in their expression in KO or OE cells relative to WT cells, and as comparison, in polyclonal TE versus MP phenotypic cells sorted from spleens of WT C57BL/6 mice 8 days after LCMV-Arm infection. (H) The change in expression of all genes in Runx3 KO and OE cells compared to WT cells is plotted. Exemplar TF gene symbols are denoted.

We identified Runx3-dependent genes at all times and quantified their association among previously defined gene-clusters that correlate with different stages of CTL development (Best et al., 2013); and grouped related clusters (Roman numerals) into “families” (Figure 5C, colored tomato, blue, and gold, Table S3). Runx3-deficient cells exhibited altered gene expression in both KLRG1^lo and KLRG1^hi cells on day 5 p.i. (Figure 5C, day 5 left panels), and in definitive MP-phenotypic cells on day 8 (Figure 5C, day 8 right panels). Specifically, Runx3-deficient KLRG1^lo cells on day 5 inefficiently upregulated genes underlying early T cell activation (Figure 5D, left plot, tomato) and late effector and memory CTL properties (Figure 5D, left plot, blue). This phenotype was more extreme for effector and memory associated genes in MP phenotypic cells on day 8 (Figure 5D, right plot, blue). The expression of key TFs that drive effector and memory CTL attributes was reduced in KLRG1^lo cells (Figure 5D, Irf4, Prdm1, Id2) (Chang et al., 2014; Man et al., 2013; Rutishauser et al., 2009). Concomitantly, Runx3-deficient cells also failed to repress genes expressed in naïve cells that are normally downregulated upon activation, and that are re-expressed in MP cells (Figure 5D, gold), especially those encoding TFs that promote T cell quiescence and lymphoid retention (Figure 5D, Tcf7, Id3, Bach2, Foxo3) (Chang et al., 2014; Shan et al., 2017). Moreover, in addition to the well-studied exemplar TF genes highlighted above, Runx3-deficient cells exhibited dysregulated expression of many genes encoding TFs that recognize motifs that were enriched in cis-regions that required Runx3 for chromatin accessibility, although their functions in CD8⁺ T cells are currently unappreciated (Figure S5). Thus, Runx3-deficient MP-phenotypic cells are not normal MP CTL, as they lack key functional attributes of CTL, and appear to have not appropriately repressed genes typical of naïve cells.

Paradoxically, Runx3-deficient KLRG1^hi and KLRG1^lo cells on day 5 p.i, overexpressed Tbx21 and Zeb2, which encode TFs that fundamentally drive TE CTL differentiation (Figure 5D, left panel and data not shown) (Dominguez et al., 2015; Joshi et al., 2007; Omilusik et al., 2015) (Figure 5C, right panel). Nevertheless, Runx3-deficient cells did not develop into genuine TE CTL as they overexpressed Sell (encoding CD62L) (not depicted) (Youngblood et al., 2017). In addition, Runx3-deficient cells accumulated in the splenic white pulp aberrantly, and overexpressed the TF TCF1, consistent with previous studies (Pipkin et al., 2010; Shan et al., 2017)(Figure 5E and 5F). Thus, Runx3 is necessary for repressing gene expression that promotes aspects of T cell homeostasis, lymphoid homing as well as terminal CTL differentiation.

The obvious mislocalization of Runx3-deficient cells in vivo prompted using a primary cell culture model to study Runx3-dependent gene-expression in a setting that obviates potential effects from these differential locales (Seo et al., 2016). To define effects that were transcriptional, we used RNA-seq analysis of chromatin-associated RNAs, which primarily comprise nascent RNAs undergoing transcription (Bhatt et al., 2012). Runx3 was both necessary and sufficient in vitro for upregulating Irf4, Prdm1 and Id2 transcription, and downregulating Bach2 and Tcf7 transcription (Figure 5G and 5H). Therefore, Runx3 controls transcription of TF genes that program CTL, independently of locales that Runx3-deficient cells aberrantly experience in vivo.

Runx3 initiates CTL differentiation by promoting expression of Irf4 and Blimp1 and driving chromatin accessibility of their binding sites

To define Runx3-dependent transcriptional circuits, we used the PageRank algorithm (Yu et al., 2017) in a personalized approach that integrated both the RNA-seq and ATAC-seq data, which prioritized TFs based on motifs enriched within Runx3-dependent ATAC-seq regions annotated to genes whose transcription was also dependent on Runx3. This implicated Irf4, Blimp1, JunD, BATF, and Bach2 as candidate TFs that required Runx3 for chromatin accessibility and transcriptional regulation (Figure 6A). We found significant overlap of Runx3, Irf4, BATF, Bach2 and to lesser extent Jun and JunD ChIP-seq binding sites, which confirmed the PageRank predictions (Figure 6B) (Kurachi et al., 2014; Roychoudhuri et al., 2016). Moreover, Runx3 was required during TCR activation for chromatin accessibility at multiple putative cis-regulatory regions in genes encoding key positive regulators of CTL differentiation (e.g., Irf4, Prdm1, Id2, Eomes, and Il2ra) that bind Runx3, Irf4, and Batf, or all three factors in conjunction (Figure 6C and S6). Thus, within hours of TCR stimulation Runx3 directly promotes chromatin accessibility of cis-regulatory regions that are bound by multiple key TFs that drive CTL differentiation.

Figure 6. Runx3 establishes Irf4 and Blimp1 transcriptional circuits by promoting chromatin accessibility and transcription of Irf4 and Prdm1.

(A) Personalized Page Rank scores were calculated and expressed as ratios, for top scoring TFs. Scores were calculated in Runx3 WT and KO cells from ATAC-seq data (Figure 4A) after 24 hours of TCR stimulation and RNA-seq data (Figure 5G).

(B) The extent of pairwise overlap between TF ChIP-seq peaks was calculated using Jaccard’s index (yellow heatmap) and values were clustered by Euclidean distance. The pairwise overlap between Runx3 and the other indicated TFs are shown as ratios of Runx3 sites and sites for each other TF (blue heatmap).

(C) Genome browser tracks depict normalized TF occupancy and ATAC-seq chromatin accessibility. Asterisks signify Runx3 ChIP-seq peaks (MACS, P<1e-5).

(D) Purified naïve (CD44^lo CD62L^hi YFP⁺) CD8⁺ T cells from Runx3^+/+ or Runx3^fl/fl sYFP dLck-Cre mice were stimulated using plate-bound anti-CD3 and anti-CD28. Histograms show representative intracellular Irf4 staining (top). The percentages of Irf4^hi cells after TCR stimulation (symbols) are plotted from two independent biological replicates.

(E) P14 Runx3^+/+ and Runx3^fl/fl sYFP cells transduced with Cre-retrovirus (Cre-RV) were analyzed in the spleen. Transduced KLRG1^lo cells were gated (left), and the mean fluorescence intensity (MFI) values (symbols) of Irf4 staining in all mice are plotted (right). Horizontal bars indicate the means. Data are from at least two biological replicates.

(F) 50,000 Blimp1-YFP reporter P14 cells transduced with shRNAmirs were transferred to WT hosts. Blimp1-YFP reporter fluorescence was analyzed in transduced cells after gating TE and MP phenotypic cells in the spleen after LCMV-Arm infection (top). Representative Blimp1-YFP reporter expression is shown (middle). The frequencies of gated Blimp1-YFP⁺ cells (symbols) were plotted (bottom). Horizontal bars indicate the means. Data are from at least two biological replicates.

P-values calculated by two-tailed unpaired (D and F) or paired (E) Student’s t test. See also Figure S6.

TCR stimulation induces Batf and Irf4, which both bind to the Prdm1 locus and promote Blimp1 expression, and all three factors promote effector and memory CTL differentiation (Chang et al., 2014; Man et al., 2013; Raczkowski et al., 2013; Rutishauser et al., 2009). Consistent with decreased chromatin accessibility at Runx3 binding sites in the Irf4 and Prdm1 loci in the absence of Runx3 (Figure 6C), naïve Runx3-deficient cells inefficiently induced Irf4 upon early TCR stimulation (Figure 6D), and expressed less Irf4 in KLRG1^lo cells on day 6 of LCMV infection (Figure 6E). In addition, RNAi of Cbfb in P14 cells reduced Blimp1-YFP reporter expression specifically in MP-phenotypic cells (Figure 6F). These results imply that Runx3 is required during TCR stimulation to establish feedforward and feedback transcriptional circuits during TCR stimulation by promoting accessibility to Irf4 and Batf binding sites in cis-regulatory regions of Irf4, and Prdm1 and other downstream regulatory genes that promote CTL differentiation.

Enhanced Runx3 expression increases memory CTL differentiation and negatively regulates proliferation of TE CTL

Runx3 haploinsufficient cells inefficiently formed MP CTL (refer to Figure 2), suggesting that Runx3 dosage influences the development of MP versus TE CTLs. Therefore, we tested whether enhanced Runx3 expression was sufficient to increase CTLmem. P14 cells overexpressing Runx3 cDNA, compared to control-transduced cells, rapidly manifested a MP phenotype and repressed the TE phenotype at early times (Figure 7A and S7A), and uniformly exhibited a central memory CD8⁺ T cell phenotype (T_cm), and harbored a greater number of T_cm in the spleen at memory times during LCMV infection (Figure 7B and S7A–S7B). Moreover, Runx3-transduced T_cm cells isolated on day 90–200 p.i. regenerated significantly more secondary DP and MP CTL upon rechallenge, compared to control T_cm cells after mixed transfers into new naïve hosts (Figure S7C) and infection with either LCMV (Figure 7C) or LM-GP33 (Figure S7D). Thus, increased Runx3 concentrations enhance differentiation of MP CTL in primary and secondary responses.

Figure 7. Runx3 drives differentiation of CTLmem and restrains terminal differentiation.

Congenically mismatched P14 Runx3^+/+ cells transduced with empty (mock) or Runx3-cDNA (OE) retrovirus were mixed and transferred to naïve hosts subjected to LCMV-Arm infection.

(A-B) Flow cytometry plots show representative staining of transduced cells in the spleen on days 8 (A) and 90 (B) following primary (1°) LCMV i nfection.

(C) On day 90 after 1° LCMV infection, mock and Runx3 OE (CD44^hi CD62L^hi) P14 cells were isolated by FACS, mixed 1:1 and 10,000 total cells were transferred to naïve recipients. Flow cytometry plots (top) show representative surface phenotype in the spleen after these hosts were infected with LMCV-Arm (2°).

(D-E) Flow cytometry plots show representative 5-ethynyl-2′-deoxyuridine (EdU) labeling of transduced P14 cells from the spleen after gating on the indicated CTL phenotypes. P14 Runx3^+/+ or Runx3^fl/fl sYFP cells were transduced with Cre-retrovirus (D), or P14 Runx3^+/+ cells were transduced with empty (mock) or Runx3-cDNA (OE) retrovirus (E).

(F) Naïve P14 cells were transduced individually or together with the indicated retroviral constructs (upper panels; negative controls, mock) to overexpress Runx3 (Ametrine) and/or T-bet (Tbx21, GFP) (lower panels), and were analyzed after gating on cells expressing one or both fluorescent reporters (left). Flow cytometry plots show representative staining of KLRG1 and CD127 in the spleen. P14 cell numbers were normalized to mock-transduced cells based on the frequencies of each transduced subset upon adoptive transfer.

(G) P14 cells transduced with the indicated constructs (as in Figure 2E) were re-isolated from the spleen by FACS on day 5 p.i. LCMV-Arm. Whole cell lysates were analyzed by immunoblotting.

(H) T-bet expression in P14 cells transduced with empty (Mock) or Runx3 cDNA (OE) constructs was analyzed after gating on transduced cells with the indicated phenotypes.

(I) Model showing the role of Runx3 during memory CTL differentiation.

All bar charts summarize the mean values of transduced P14 cells of different CTL phenotypes in the spleens of individual mice (symbols). Data are one representative of (A-E, G-H), or pooled from (F), at least two independent experiments. *P<0.05, **P<0.01, ***P<0.001. P-values computed using two-tailed paired Student’s t test. See also Figure S7.

Lineage-tracing studies support a model wherein CTLmem develop linearly from individual naïve cells via activated precursors that divide slowly, while a small number of precursors that proliferate rapidly differentiate into TE CTL (Buchholz et al., 2016). Indeed, WT polyclonal EE and TE phenotypic cells incorporated much more 5-ethynyl-2′-deoxyuridine (Edu) than DP and MP cells (Figure S7E). We found that Runx3-deficient KLRG1^lo P14 cells incorporated much less Edu, whereas Runx3-deficient KLRG1^hi P14 cells incorporated similar amounts of Edu as WT P14 cells (Figure 7D). Thus, Runx3 is dispensable for proliferation of cells that upregulate KLRG1, consistent with the enhanced relative frequencies of TE-phenotype cells that develop in the absence of Runx3. In contrast, Runx3 is required for proliferation of nascent CTL, consistent with reduced overall numbers of Runx3-deficient CTL (refer to Figure 2). Reciprocally, Runx3 overexpression in P14 cells reduced Edu-incorporation in both KLRG1^lo and KLRG1^hi cells (Figure 7E), reduced Ki-67 staining in total P14 cells in vivo (Figure S7F), and also reduced cell accumulation in cell culture (Figure S7G), indicating that increased Runx3 expression slows proliferation of effector cells. Therefore, Runx3 may favor CTLmem formation by retarding proliferation of cells that differentiate toward TE cells.

Runx3 represses terminal differentiation by restraining high expression of T-bet

We hypothesized that the predilection of Runx3-deficient cells to ultimately manifest a TE phenotype was associated with overexpression Tbx21 and Zeb2 mRNAs. In line with this notion, T-bet protein was overexpressed in both EE and TE CTL that were Runx3-deficient or Cbfb-depleted (Figure S7H) and, Tbx21 RNAi rescued the propensity of Runx3-deficient cells to preferentially acquire the TE-like phenotype (Figure S7I). Moreover, Runx3 was dominant over T-bet because its overexpression restored the differentiation of DP and MP CTL in cells overexpressing T-bet in vivo (Figure 7F), prevented high T-bet expression in WT cells prior to emergence of definitive TE cells on day (Figure 7G), and downregulated the T-bet^hi state specifically in definitive EE, TE and DP cells on day 8 after LCMV infection (Figure 7H). In contrast, T-bet overexpression did not repress Runx3 (Figure 7G). Therefore, Runx3 functions upstream of T-bet and prevents acquisition of the T-bet^hi state in vivo. These results suggest that Runx3 ensures CTLmem form by repressing T-bet-driven terminal CTL differentiation (Figure 7I).

Discussion

A central unanswered problem in immunology centers on how memory T cells develop from the overall burst of effector cells that are elicited upon primary infection. A very brief period of antigenic stimulation is sufficient to initiate a complete program of memory CTL differentiation (Kaech and Ahmed, 2001; van Stipdonk et al., 2001). Our results indicate that TCR and minimal co-stimulatory signals rapidly establishes this program in naïve cells by driving chromatin accessibility to a large number of cis-regulatory regions that are also accessible in CTL subsets from the effector and memory phases after infection. Approximately 15% of the CTLmem cis-regulatory landscape became accessible prior to the first cell division. The most frequent TF motifs encoded by these regions were recognized by TFs from the Runx, ETS, IRF, RHD, bZIP, and ZF-KLF family of TFs, suggesting that their integrated functions drive the transcriptional reprogramming of naïve CD8⁺ T cells that leads to CTLmem formation.

Runx3 was required during TCR stimulation for establishing chromatin accessibility of cis-regulatory regions that are associated with all CTL subsets. However, it was almost universally required for opening those that were CTLmem-specific, and that contained motifs recognized by IRF, RHD, bZIP and Prdm1-like family TFs. This suggests that TFs in these families promote CTLmem differentiation. Consistent with this, Runx3 was necessary for normal expression of Irf4 and Blimp1 (and other TFs) in EE and MP CTL, and for chromatin accessibility in the Irf4 and Prdm1 loci at regions that also bound Irf4 and Batf. Thus, TCR signals appear to collaborate with Runx3 to establish Irf4-dependent feedforward and feedback transcriptional circuits, given that Irf4 promotes Blimp1 expression (Man et al., 2013). Therefore, Irf4 and Blimp1 might promote CTLmem differentiation in a Runx3-dependent fashion, even though Irf4 and Blimp1 are also essential for TE CTL differentiation (Kurachi et al., 2014; Man et al., 2013; Rutishauser et al., 2009; Xin et al., 2016).

Runx3 appears to enforce development of CTLmem by ensuring that nascent CTL remain differentiated from TE CTL, and this likely involves multiple mechanisms. The accessible chromatin landscape that was established upon TCR stimulation of Runx3-deficient cells was dominated by Ets, ZF-KLF and bHLH family TF motifs, and this correlated with generation of aberrant MP and TE-like cells. This might have been potentiated by Runx1 and Ets1, and other TFs such as Tcf1, which were overexpressed in Runx3-deficient cells (Shan et al., 2017). Thus, Runx3 is important for ensuring that certain cis-regions become inaccessible during differentiation. It is possible this altered regime contributed to overexpression of genes encoding the TFs T-bet and Zeb2, which drive terminal differentiation. In addition, Runx3 might ensure CTLmem cells develop by promoting accessibility to cis-acting regions controlled by TFs that deactivate TE CTL differentiation. For example, Runx3-deficient cells overexpressed Bach2, which normally restrains TE CTL differentiation, but did not establish access to Bach2 motifs, indicating that Bach2’s normal function depends on Runx3 (Kurachi et al., 2014; Roychoudhuri et al., 2016).

In addition, our results are consistent with spatially integrated functions of Runx3 and T-box TFs, which likely influence the differentiation of CTL at distinct times, in different tissue environments (Cruz-Guilloty et al., 2009; Intlekofer et al., 2005; Joshi et al., 2007; Milner et al., 2017; Reis et al., 2014). T-box motifs were statistically enriched in cis-regions that required Runx3 for accessibility upon TCR stimulation, although they were not prioritized because they also occurred frequently in background sequences. However, Runx3 and T-bet ChIP-seq binding sites overlapped substantially, which supports the notion that Runx3 might directly modulate the activity of T-bet (and or Eomes) at regions co-occupied by Runx3 and either T-box protein, and supports a role for both TF families in promoting differentiation of CTLmem. Thus, Runx3-driven chromatin accessibility probably governs T-box-TF mediated control of effector and memory CTL differentiation.

In conclusion, Runx3 appears to be an apical TF in the hierarchy controlling memory CTL differentiation, and exhibits several characteristics of validated pioneer TFs (Iwafuchi-Doi and Zaret, 2014), which suggests it might function to instigate chromatin accessibility during TCR stimulation. We anticipate that Runx3-dependent chromatin remodeling during TCR stimulation involves the concerted action of multiple conventional TFs, chromatin binding proteins and remodeling enzymes, which could vary upon context. Given that Runx3 is expressed prior to a number of other key TFs that are activated by TCR stimulation, it stands to reason that Runx3 is one of the earliest factors to initiate programming of CTL. The enhanced differentiation of CTLmem upon Runx3 overexpression suggests that manipulating its activity might have utility for reprogramming CD8⁺ T cells in order to manufacture durable immunity (Milner et al., 2017). The notion of reprogramming non-cytolytic cells into CTL is very likely possible, as heritably silenced human PRF1 loci from fibroblasts can be reprogrammed into an active format directly when transferred to a cytolytic cell environment (Pipkin et al., 2007). Collectively, these studies establish the paradigm for chromatin-level reprogramming of T cells to harness immunity.

STAR Methods

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Matthew Pipkin (mpipkin@scripps.edu).

Experimental Model and Subject Details

Mice

C57BL/6 mice, Runx3^fl/fl mice, ERT2-Cre and Rosa26 stop-flox EYFP reporter mice (abbreviated sYFP) were from the Jackson Laboratory. P14 TCR transgenic and Blimp1-YFP BAC transgenic reporter mice have been previously described (Rutishauser et al., 2009). P14 bearing disparate Thy1 alleles were intercrossed with the lines described above to generate Runx3 conditional lines. Only heterozygous dLck-Cre mice were maintained and used in this study. For inducing Cre-mediated recombination in ERT2-Cre, mice were injected intraperitoneally (IP) daily with 1.5mg of tamoxifen in corn oil, for three consecutive days. For analysis of endogenous T cell responses, 6–12 week-old male or female mice were used. For adoptive transfer experiments, 6–8 week old male or female recipient mice were used. All mice were maintained in specific-pathogen free facilities and used according to protocols approved by the Institutional Animal Care and Use Committee of TSRI-FL.

Cell lines

Plat-E (female), Vero (female), Vero-E6 (female) and BHK-21 (male) cells were grown in DMEM (Gibco-Invitrogen) with 4.5 g/L of D-glucose, FBS (Gibco, 10%) and penicillin/streptomycin (100 U/mL, Invitrogen) at 37°C in a humidified atmospher e of 5% CO2/95% air.

Virus

LCMV-Armstrong and LCMV-Clone13 stocks were produced by plating 7.5 million BHK21 cells and infecting with 400,000 PFU of either virus, 24 hours after plating cells in T175 flasks. Viral supernatants were collected 48 hours after infection, similar to as described (Ahmed et al., 1984).

Method Details

Infections, plaque and colony forming assay

In experiments to examine endogenous CD8⁺ T cell responses mice were infected (IP) with 2×10⁵ PFU of LCMV-Armstrong strain (Arm). For experiments involving unmanipulated naïve P14 cells, 5,000 to 100,000 cells were adoptively transferred to naïve mice that were infected (IP) the following morning with 2×10⁵ PFU of LCMV-Arm. For experiments using in vitro transduced P14 cells, either 50,000 activated cells were adoptively transferred and mice were infected with 2×10⁵ PFU of LCMV-Arm, 1 hour after adoptive transfer, or up to 500,000 activated cells were transferred and mice were infected with 1.5×10⁵ PFU LCMV-Clone 13, 1 hour after transfer; both settings result in acute infections (Chen et al., 2014). Homogenates for plaque assays were generated by disrupting spleens in 1x PBS/2.0% FBS, straining and storing 100uL aliquots at −80°C, which were thawed and disr upted 20 times with a tight Dounce homogenizer. Listeria monocytogenes-GP33 (LM-GP33) was grown on brain heart infusion media. To quantify CFU in mice, tissues were disrupted in 1x PBS/2.0% FBS, strained and solubilized in a final concentration of 0.5% Triton X-100, and serial dilutions were plated. To immunize mice carrying adoptively transferred P14 cells, mice were inoculated with 10,000 CFU. For LM-GP33 protection assays, P14 cells were activated by TCR stimulation for 48 hours, transduced with Cre-expressing retroviruses, and cultured for one day in 10U/ml of rhIL2, sorted to homogeneity based on YFP expression and then cultured for an additional 3 days in 10U/ml of rhIL2, which induces memory-CD8⁺ T cell-like cells (Pipkin et al., 2010). Cells were collected after culture and 25×10⁴ of YFP⁺ cells were transferred to naïve C57BL/6 mice. One day later, mice were challenged with 5×10⁴ of CFU of LM-GP33 via retro-orbital injection. Colony formation assay of spleens were performed on day 4 post-infection.

shRNAmir and cDNA retroviral packaging

The MSCV-based retroviral vectors MigR1 and LMPd were used to transduce activated CD8⁺ T cells (Chen et al., 2014). MigR1 encoding IRES-GFP, was used to deliver Runx3 (long form) and Tbx21 cDNAs. LMPd, encoding Pgk-Ametrine1.1, was used to express shRNAmirs specific for Runx1, Runx2, Runx3, Cbfb, Cd4 and Tbx21 derived mRNAs. Retroviral supernatants were generated in Platinum-E packaging cells as previously described (Chen et al., 2014).

Flow cytometry

Single cell suspensions prepared from spleens or heparinized blood were treated with RBC lysis buffer, washed and incubated with antibody cocktails for surface staining. All of the antibodies used in this study, including CD4 (RM4–5), CD8 (53–6.7), CD44 (IM7), CD62L (MEL-14), CXCR3 (CXCR3–173), CD25 (PC61), CD69 (H1.2F3), CD127 (A7R34), KLRG1 (2F1/KLRG1), CD27 (LG.3A10), CD43 (1B11), T-bet were purchased from Biolegend, except BUV395 conjugated CD8 antibody from BD Biosciences. LCMV GP33 (KAVYNFATC), GP276(SGVENPGGYCL) and NP396 (FQPQNGQFI) H-2Db tetramers conjugated to either BV421 or APC were obtained from NIH Tetramer Facility.

CD8⁺ T cell transduction

CD8⁺ T cells were isolated using magnetic beads and negative selection, as well as by FACS to isolate YFP⁺ cells, or naïve cells based on CD44 and CD62L expression. For primary TCR-mediated activation, purified CD8⁺ T cells were plated at a density of 4×10⁵ cells/cm² at a concentration of 1×10⁶/mL and on plates pre-coated with goat anti-hamster capture antibody (50μg/mL) and media supplemented with hamster anti-CD3 (2C11) and anti-CD28 (37.51) (1μg/mL each). For retroviral transductions, 16 hours after initial plating, the T cell conditioned media was taken off, saved and replaced with retroviral supernatants containing 10μg/mL polybrene, and the cultures were immediately centrifuged at ~ 500×g for 90 min at ~ 37 degrees C, followed by incubation for an additional 3 hours at 37 degrees C in the CO₂ incubator. For adoptive transfer, 50–500×10³ activated cells were transferred to naïve congenically mismatched C57BL/6 mice (see supplementary figure legends for details). After adoptive transfer of activated T cells, the cells were allowed to equilibrate in the mice for 1 hour prior to administering infections. For in vitro culturing, after 4 hours of transduction retroviral supernatants were aspirated and replaced with the original T cell conditioned media. After 48 hours of total TCR stimulation, the activated cells were directly resuspended, counted and recultured by diluting the cells into fresh media containing 10 or 100 U/mL rhIL-2 (NCI). Cells were recultured as 5×10⁵ cells/mL, every 24 hours following removal from the TCR stimulus.

Ex vivo effector CD8⁺ T cell sorting

Thy1.1 positive selection was used to enrich P14 cells in splenocytes prepared from 5 days- or 8 days-infected host mice. Thy1.1 negative selection was used to enrich P14 cells in splenocytes prepared from 90 days-infected host mice. After staining enriched cells with CD8-BV510, Thy1.1-PerCP/Cy5.5, ThCD127-PE and KLRG1-APC, different effector subsets were sorted on BD FACSAria™ Fusion flow cytometer.

qPCR analysis

Cells were used to extract RNA by Trizol, followed by cDNA synthesis using GoScript™ Reverse Transcriptase (from Promega) and qPCR using Taqman probes (from Life Technology) as indicated in Key Resource Table.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-mouse CD8a - BUV395	eBioscience	Cat#563786
Anti-mouse TCF1/TCF7 Alexa Fluor - 647	Cell Signaling	Cat#6709
Anti-mouse Ki67 - PE/Cy7	eBioscience	Cat#25–5698-82, RRID:AB_11220070
Anti-mouse Granzyme B - PE/Cy7	eBioscience	Cat#25–8898-82, RRID:AB_10853339
Anti-mouse T-bet Alexa Fluor - 647	BioLegend	Cat#644804, RRID:AB_1595466
Anti-mouse IRF4 Alexa Fluor - 647	BioLegend	Cat#646407, RRID:AB_2564047
Anti-mouse CD90.1 - PerCP/Cy5.5	BioLegend	Cat#202516, RRID:AB_961437
Anti-mouse CD27 - PerCP/Cy5.5	BioLegend	Cat#124213, RRID:AB_2073424
Anti-mouse CD62L - PerCP/Cy5.5	BioLegend	Cat#104431, RRID:AB_2187123
Anti-mouse KLRG1 (MAFA) antibody - PerCP/Cy5.5	BioLegend	Cat#138417, RRID:AB_2563014
Anti-mouse CD103 - PerCP/Cy5.5	BioLegend	Cat#121415, RRID:AB_1574957
Anti-mouse CD44 - PE/Cy7	BioLegend	Cat#103029, RRID:AB_830786
Anti-mouse CD69 - PE/Cy7	BioLegend	Cat#104512, RRID:AB_493564
Anti-mouse CD90.1 - PE/Cy7	BioLegend	Cat#202518, RRID:AB_1659223
Anti-mouse CXCR3 - PE/Cy7	BioLegend	Cat#126516, RRID:AB_2245493
Anti-mouse CD90.2 - PE/Cy7	BioLegend	Cat#140309, RRID:AB_10645336
Anti-mouse CD4 - PE/Cy7	BioLegend	Cat#100528, RRID:AB_312729
Anti-mouse KLRG1 (MAFA) - APC	BioLegend	Cat#138412, RRID:AB_10641560
Anti-mouse CD103 - APC	BioLegend	Cat#121410, RRID:AB_535952
Anti-mouse CD90.2 - APC	BioLegend	Cat#105312, RRID:AB_313183
Anti-mouse CD90.1 - APC	BioLegend	Cat#202526, RRID:AB_1595470
Anti-mouse CD8 - APC	BioLegend	Cat#100712, RRID:AB_312751
Anti-mouse CD44 - Alexa Fluor - 700	BioLegend	Cat#103026, RRID:AB_493713
Anti-mouse CD44 - FITC	BioLegend	Cat#103006, RRID:AB_312957
Anti-mouse CD90.2 - FITC	BioLegend	Cat#105306, RRID:AB_313177
Anti-mouse CD90.1 - PE	BioLegend	Cat#202523, RRID:AB_1595635
Anti-mouse CD44 - PE	BioLegend	Cat#103007, RRID:AB_312958
Anti-mouse CD127 - PE	BioLegend	Cat#135010, RRID:AB_1937251
Anti-mouse CD90.1 - BV421	BioLegend	Cat#202529, RRID:AB_10899572
Anti-mouse CD4 - BV421	BioLegend	Cat#100544, RRID:AB_11219790
Anti-mouse CD90.2 - BV421	BioLegend	Cat#105341, RRID:AB_2632888
Anti-mouse KLRG1 (MAFA) - BV421	BioLegend	Cat#138414, RRID:AB_2565613
Anti-mouse CD90.1 - BV510	BioLegend	Cat#202535, RRID:AB_2562643
Anti-mouse CD90.2 - BV510	BioLegend	Cat#105335, RRID:AB_2566587
Anti-mouse CD4 - BV510	BioLegend	Cat#100559, RRID:AB_2562608
Anti-mouse CD8 - BV510	BioLegend	Cat#100752, RRID:AB_2563057
Anti-mouse CD90.1 - BV605	BioLegend	Cat#202537, RRID:AB_2562644
Anti-mouse CD62L - BV605	BioLegend	Cat#104438, RRID:AB_2563058
Anti-mouse PD-1 - BV605	BioLegend	Cat#135220, RRID:AB_2562616
Anti-mouse CD90.1 - BV605	BioLegend	Cat#202537, RRID:AB_2562644
Anti-mouse CD4 - Alexa Fluro 594	BioLegend	Cat#100446, RRID:AB_2563182
Anti-mouse CD25 - PerCP/Cy5.5	BioLegend	Cat#102030, RRID:AB_893288
Anti-mouse CD25 - BV421	BioLegend	Cat#102033, RRID:AB_10895908
Anti-mouse CD43 - APC	BioLegend	Cat#143208, RRID:AB_11149685
Anti-mouse Runx3	Homemade
Anti-mouse CBFb	Thermo Fisher Scientific	Cat#PA5–35322, RRID:AB_2552632
Anti-mouse Runx2	SCBT	Cat#sc-390715, RRID:AB_2637033
Anti-mouse T-bet	BioLegend	Cat#644801, RRID:AB_1595608
Anti-mouse Runx1	Proteintech	Cat#25315–1-AP, RRID:AB_
Anti-mouse Histone H2B	BioLegend	Cat#688702, RRID:AB_2629641
Anti-mouse Histone H1.2	Proteintech	Cat#15446–1-AP, RRID:AB_10858631
Anti-mouse Beta-Actin	BioLegend	Cat#622102, RRID:AB_315946
Anti-mouse Perforin	Enzo	Cat#ALX-804–057-C100, RRID:AB_2052202
Anti-mouse Granzyme B	BioLegend	Cat#662801, RRID:AB_10583068
Bacterial and Virus Strains
Listeria monocytogenes-GP33	A gift from Anjana Rao Lab	N/A
LCMV Armstrong	A gift from Shane Crotty Lab	N/A
LCMV clone 13	A gift from Shane Crotty Lab	N/A
Chemicals, Peptides, and Recombinant Proteins
LCMV GP33 peptide tetramer APC	NIH Tetramer Core Facility	N/A
LCMV GP276 peptide tetramer APC	NIH Tetramer Core Facility	N/A
LCMV NP306 peptide tetramer APC	NIH Tetramer Core Facility	N/A
Tamoxifen	SIGMA-ALDRICH	Cat#T5648–1G
4-Hydroxytamoxifen	SIGMA-ALDRICH	Cat#H7904–5MG
AGENCOURT AMPURE XP, 60 ML	BECKMAN COULTER	Cat#A63881
Hexadimethrine bromide	SIGMA-ALDRICH	Cat#H9268–5G
Critical Commercial Assays
eBioscience™ Foxp3 / Transcription Factor Staining Buffer Set	Thermo Fisher Scientific	Cat#00–5523-00
Click-iT™ EdU Alexa Fluor™ 647 Flow Cytometry Assay Kit	Thermo Fisher Scientific	Cat#C10419
NucleoSpin® Gel and PCR Clean-up	MACHEREY-NAGEL	Cat#740609.50
NucleoSpin® Plasmid (NoLid)	MACHEREY-NAGEL	Cat#740499.250
NucleoBond® Xtra Midi	MACHEREY-NAGEL	Cat#740410.50
Nextera DNA Library Prep Kit (24 samples)	Illumina	Cat#FC-121–1030
Qubit dsDNA HS Assay Kit	Thermo Fisher Scientific	Q32854
EasySep™ Mouse CD8+ T Cell Isolation Kit	STEMCELL Technologies	Cat#19853a
EasySep™ Mouse CD90.1 Positive Selection Kit	STEMCELL Technologies	Cat#18958
Pierce™ ECL Plus Western Blotting Substrate	Thermo Fisher Scientific	Cat#32132
SMART-Seq® v4 Ultra® Low Input RNA Kit for Sequencing	Takara Clontech	Cat#634888
TruSeq Stranded Total RNA	illumina	Cat#20020596
SuperScript™ III First-Strand Synthesis System	Thermo Fisher Scientific	Cat#18080051
Phusion High-Fidelity DNA Polymerase	Thermo Fisher Scientific	Cat#F530L
Qiagen MinElute PCR Purification Kit	Qiagen	Cat#28006
NEBNext High-Fidelity 2x PCR Master Mix	New England Labs	Cat #M0541
TaqMan™ Universal Master Mix II, no UNG	Thermo Fisher Scientific	Cat#4440040
CD90.1 MicroBeads, mouse and rat	Miltenyl Biotech	Cat#130–094-523
Deposited Data
GEO super series accession GSE111149	This paper	N/A
Experimental Models: Cell Lines
Plat-E	Cell Biolabs	Cat#RV-101
Vero	ATCC	ATCC# CCL-81
Vero C1008	ATCC	ATCC# CCL-1586
BHK-21	ATCC	ATCC# CCL-10
Experimental Models: Organisms/Strains
C57BL6/J	Jackson Laboratories	Cat#000664
Runx3	Jackson Laboratories	Cat#008773
P14 Thy1.1	A gift from Rafi Ahmed Lab	N/A
Ert2-Cre	Jackson Laboratories	Cat#008085
dLCK	Jackson Laboratories	Cat#012837
YFP	Jackson Laboratories	Cat# 006148
Oligonucleotides
Mouse Runx3 Taqman probe	Thermo Fisher Scientific	Mm00490666_m1
Mouse Runx2 Taqman probe	Thermo Fisher Scientific	Mm00501584_m1
Mouse Runx1 Taqman probe	Thermo Fisher Scientific	Mm01213404_m1
Mouse Tbx21 Taqman probe	Thermo Fisher Scientific	Mm00450960_m1
Mouse Gzmb Taqman probe	Thermo Fisher Scientific	Mm00442834_m1
Mouse Prf1 Taqman probe	Thermo Fisher Scientific	Mm00812512_m1
Mouse Actb Taqman probe	Thermo Fisher Scientific	Mm00607939_s1
Recombinant DNA
pMIG	Addgene	Cat#9044
pMIG-Myc-Runx3	Homemade	N/A
pLMPd-Ametrine	Homemade	Chen et al., 2014
See Table S4 for sequences used in short hairpin RNA interference	This paper	N/A
pCL-Eco packaging vector	Addgene	Cat#21371
Software and Algorithms
Flowjo X	Tree Star	N/A
Prism 7	GraphPad	N/A
Adobe Illustrator	Adobe	N/A
Rstudio	Rstudio	N/A

Immunoblot analysis

Equivalent cell numbers were lysed in 1X Laemmli sample buffer and boiled for 10 min, resolved by 8% SDS-PAGE, the proteins were blotted onto nitrocellulose membranes in Towbin transfer buffer. Membranes were incubated with primary antibodies, such as anti-Runx1(Proteintech, 25315–1-AP), anti-Runx2 (Santa Cruz Biotechnology, sc-390715), anti-Runx3 (produced custom at Covance, by immunizing rabbits with human Runx3 (220–429) peptide produced at TSRI), anti-T-bet (Biolegend, 644801), anti-Cbfb (Santa Cruz Biotechnology, sc-56751) and anti-β-actin. After incubation with HRP-conjugated secondary antibody, target proteins were detected by Pierce ECL Plus Western Blotting Substrate.

RNA-seq

For RNA-seq analysis of in vivo CTL, distinct phenotypic subsets were isolated by FACS and sorted into DMEM containing 10% FBS, and then were washed in 1x PBS. Full-length cDNAs were generated directly from ~50,000 cells per sample using oligo(dT) priming and the SMART-Seq v4 Ultra Low Input RNA Kit (Clontech), and 150 pg of cDNA was used to prepare strand-specific paired-end sequencing libraries (Nextera XT, Illumina). Libraries were quantified using Qubit HS DNA assay (Invitrogen) and their quality was assessed on Agilent Technology 2100 Bioanalyzer using a High Sensitivity DNA chip (Agilent) and dual-indexed pooled libraries were sequenced. For RNA-seq analysis of in vitro differentiated CTL, and polyclonal ex vivo CTL subsets, either total RNA or chromatin-associated (nascent) RNA was extracted in TRIZoL reagent (Bhatt et al., 2012), and was converted into strand-specific paired-end sequencing libraries using the TruSeq stranded total RNA kit (Illumina). Samples were DNase I treated. 500ng of total RNA was rRNA-depleted. Chromatin-associated RNA contained less than 5% of rRNA based on bioanalyzer analysis and was not rRNA-depleted. 50ng of chromatin-associated or rRNA-depleted total RNA was fragmented, converted to cDNA, adapter-ligated with unique indices and PCR amplified. The final libraries were quantified and sequenced (NextSeq 500, Illumina) using paired-end 75bp chemistry.

ATAC-seq

For ATAC-seq, 5×10⁴ nuclei were isolated from naïve and stimulated CD8⁺ T cells as described for lymphocytes (Pipkin and Lichtenheld, 2006) and treated in 50 μl of Tn5 transposase containing transposition reaction mixture (Nextera DNA Sample Prep Kit, Illumina) at 37°C for 30 min. Genomic DNA was extracted by MinElute PCR Purification Kit (from QIAGEN) and subjected to preliminary PCR using indexed primers (Buenrostro et al., 2013) and NEBNext High-Fidelity 2X PCR master mix. Aliquots from each reaction were removed and amplified using quantitative PCR to estimate the number of additional cycles for half-maximal amplification and additional cycles were applied to the remaining pre-amplified PCR reactions accordingly. The PCR reactions were purified and fragment size-distributions were quantified using a Bioanalyzer to confirm that independent samples exhibited similar fragment distributions.

Bioinformatic analysis

For analysis of ChIP-seq data, reads were aligned to UCSC mm10 with Bowtie2 using base settings (Langmead and Salzberg, 2012) and peaks were called using MACS using base settings (Zhang et al., 2008). Raw sequencing reads from Runx3 (GSE50131) (Lotem et al., 2013); Tbx21 (GSE72408) (Dominguez et al., 2015); Bach2 (GSE77857) (Roychoudhuri et al., 2016); BATF, cJun, Irf4, JunB, and JunD (GSE54191) (Kurachi et al., 2014) were downloaded from the SRA database. Peaks were filtered to remove those associated with alternative annotations, mitochondrial DNA, and blacklist sites (Carroll et al., 2014), in R using GenomeInfoDb, and the ChIP blacklist GenomicRanges packge, and the commands: keepStandardChromosomes(a,pruning.mode = ‘coarse’),dropSeqlevels(a,’chrM’,pruning.mode = ‘coarse’), and subsetByOverlaps(a,black_list,invert=TRUE). To quantify the extent of pairwise overlap between different TFs, the Jaccard’s index was calculated according to (sum(width(intersect(a,b)))/sum(width(union(a,b)))). Overlapping ChIP-seq peaks were computed using Homer mergePeaks function and the fractions of sites overlapping Runx3 ChIP-seq peaks were calculated.

RNA-seq derived reads from CTL subsets (Figure 5) were aligned to UCSC mm10 with Star and reads counted with HTseq count (Anders and Huber, 2010; Dobin et al., 2013). Differential expression was determined using DESeq2. PCA plots were generated from log normalized count data generated through DESeq2 and PCAExplorer (Love et al., 2014). TPMs were calculated from raw count data by dividing by the number of counts by exon length in kilobases (RPK), dividing the total number of counts by 1 million (CM) and then dividing the RPK by CM. Genes with raw counts less than 10 were excluded from the TPM calculation. RNA-seq reads from in vitro CTL samples were aligned to UCSC mm9 with Bowtie2 and reads were counted with HTseq count (Anders and Huber, 2010; Dobin et al., 2013), and differential expression was calculated with DESeq2. Enrichment in Figure 5C was calculated by the proportion of overlap of significant Runx3-dependent genes with the corresponding IMGEN cluster (Best et al., 2013). P-values associated with differences between the number of Runx3-dependent genes within specific gene-clusters (Table S3) were calculated using the binomial distribution in R.

FASTQ files from ATAC-seq reads were aligned to UCSC mm10 with Bowtie (-p 15 -m 1 –best - strata -X 2000 -S --fr --chunkmbs 1024). ATAC-seq data from TE, MP and memory CTL subsets were acquired from the GEO database (GSE88987) (Scott-Browne et al., 2016). PCR duplicates, mitochondria DNA, Y chromosome DNA and fragments larger than 100bp were removed by samtools. Coordinates of reads were then shifted (plus strand +4, minus strain −5) to represent real Tn5 binding sites. Peaks were called using MACS2 (macs2 callpeak –t inputfile –f BED –g mm –n outputfile –nomodel −1 0.01 –keep-dup all –call-sumits -B). Peaks from all samples were then merged using Homer (mergePeaks -d 200). Counts of Tn5 insertion site numbers were generated by HTseq-count. Comparisons of differential peaks were performed by DESeq2 after filtering out low read peaks (count number <5 in all samples in comparison). Genes within 100kb up or down stream of the peaks were annotated to peaks. Motif enrichment was calculated using Homer (findMotifsGenome.pl -size given -mis 3 -mask). Go-term analysis of genes identified around peaks were performed with ClusterProfiler (Yu et al., 2012). Transcription factor footprints were identified by RGT-hint (Gusmao et al., 2016). PageRank analysis was performed with Taiji (Yu et al., 2017). Plots were generated by R ggplot2, python Matplotlib_venn and Seaborn. All bedgraph files of ATAC-seq or ChIP-seq are normalized by multiplying counts to a scaling factor calculated by 10⁹/ accumulation-factor (accumulation-factor calculation: awk ‘{sum+=(($3-$2)*$4)} END {print sum}’). Bedgraph files were transformed to bigwig by bedgraphtobigwig, and uploaded as UCSC custom tracks.

Quantification and Statistical Analysis

Data are presented as mean ± SEM. Sample numbers and experimental repetitions are indicated in the figures, or methods section above. For analysis of endogenous T cell responses, statistical significance was computed using independent Student’s t-test. For adoptive transfer experiments in which controls were transferred to separate mice, statistical significance was computed using the unpaired Student’s t-test. For adoptive transfer experiments in which control and experimental T cells were mixed and co-transferred into the same host mice, statistical significance was computed using the paired Student’s t-test. Unless specifically denoted in the Figures, * P < 0.05, ** P < 0.01 or *** P <0.001 were considered statistically significant. The number of sample replicates and statistical cut-offs used in the analysis of genomics data are indicated in the figure legends and specific methods.

Supplementary Material

1

Table S1. Enriched TF motifs within differentially remodeled ATAC-seq regions after naïve cell stimulation, Related to Figure 1.

2

Table S2. Enriched TF motifs within Runx3-dependent ATAC-seq regions after TCR stimulation, Related to Figure 4.

3

Table S3. The enrichment of Runx3-dependent genes within gene clusters that define different aspects of CTL differentiation, Related to Figure 5.

4

Table S4. Sequences of shRNAmirs, Related to Figure 2 and STAR Methods.

Highlights.

TCR signals and Runx3 initiate accessibility to the CTLmem cis-regulatory landscape Runx3 promotes accessibility of cis-regions enriched with IRF, bZIP, and Prmd1 motifs Runx3-deficiency impairs MP CTL transcription, differentiation and CTLmem formation Increased Runx3 dosage enhances differentiation of MP CTL and CTLmem