The transcription factor T-bet is critical for cytotoxic T lymphocyte (CTL) differentiation, but it is unclear how it operates in a graded manner in the formation of both terminal effector and memory precursor cells during viral infection. We find that, at high concentrations, T-bet induced expression of Zeb2 mRNA, which then triggered CTLs to adopt terminally differentiated states. ZEB2 and T-bet cooperate to switch on a terminal CTL differentiation program, while simultaneously repressing genes necessary for central memory CTL development. Chromatin immunoprecipitation sequencing showed that a large proportion of these genes were bound by T-bet, and this binding was altered by ZEB2 deficiency. Furthermore, T-bet overexpression could not fully bypass ZEB2 function. Thus, the coordinated actions of T-bet and ZEB2 outline a novel genetic pathway that forces commitment of CTLs to terminal differentiation, thereby restricting their memory cell potential.
CD8+ T cells are a critical component of cell-mediated immunity against intracellular pathogens, such as viruses, and can provide long-term protection from reinfection for decades after the initial infection is cleared (Ahmed and Gray, 1996; Jameson and Masopust, 2009). Despite the importance of cytotoxic T lymphocyte (CTL) immunity in controlling viral infections, a successful T cell–based vaccine has yet to be developed. Many intracellular pathogens for which we still lack effective vaccines, such as HIV, involve pathogens that can escape neutralizing antibody; a T cell–based vaccine strategy may improve protection from such pathogens. Harnessing this potential requires greater immunological insight into how T cell memory forms after infection and vaccination.
Our understanding of effector and memory T cell development has advanced considerably over the past decade. In response to acute infections, CD8+ T cells expand into a heterogeneous population of effector cells that can be phenotypically, functionally, and anatomically distinguished. Importantly, the long-term fates of the effector cells also differ after infection in that the majority of cells (∼90–95%) die and a minority persist to give rise to longer-lived memory T cells (Ahmed and Gray, 1996; Jameson and Masopust, 2009; Kaech and Cui, 2012). Often, increased IL-7 receptor α (IL-7R) expression on effector cells identifies those with a higher potential to persist and seed diverse populations of central memory (TCM), effector memory (TEM), and resident memory (TRM) T cells (Sallusto et al., 1999; Schluns et al., 2000; Kaech et al., 2003; Huster et al., 2004; Joshi et al., 2007; Jameson and Masopust, 2009; Kaech and Cui, 2012; Mackay et al., 2013). These antigen-specific IL-7R+ CD8+ T cells, commonly referred to as memory precursor (MP) cells, are endowed with longevity and the ability to self-renew and regenerate new clonal bursts of effector cells (i.e., they are multipotent). Conversely, terminally differentiated effector (TE) cells, often identified by killer-cell lectin-like receptor G1 (KLRG1) expression, are potent killers and IFN-γ secretors that have decreased longevity, proliferative potential, and restricted plasticity (Voehringer et al., 2001; Thimme et al., 2005; Joshi et al., 2007, 2011; Olson et al., 2013). This divergence in long-term fates raises the questions: How is the process of terminal differentiation programmed and how is plasticity maintained in CTLs as they differentiate during infection?
Gene expression profiling experiments have identified unique transcriptional signatures for MP cells (KLRG1lo IL7-Rhi) and TE cells (KLRG1hi IL7Rlo; Joshi et al., 2007; Rutishauser et al., 2009; Best et al., 2013; Arsenio et al., 2014). Further, T-bet (encoded by Tbx21), B lymphocyte-induced maturation protein-1 (Blimp-1, encoded by Prdm1), inhibitor of DNA binding 2 (Id2), and signal transducer and activator of transcription 4 (Stat4) have been identified as critical drivers of TE cell differentiation (Joshi et al., 2007; Kallies et al., 2009; Rutishauser et al., 2009; Yang et al., 2011; Mollo et al., 2013; Shin et al., 2013). Conversely, expression of eomesodermin (Eomes), B cell lymphoma-6 (Bcl6), T cell factor-7 (Tcf7), Id3, forkhead box O 1 (Foxo1), and Stat3 promote development of memory CD8+ T cells and their progenitors (Ichii et al., 2002, 2004; Jeannet et al., 2010; Zhou et al., 2010; Cui et al., 2011; Yang et al., 2011; Hess Michelini et al., 2013; Kim et al., 2013; Tejera et al., 2013). However, little is known about how these transcription factors interact or affect each other’s expression or function to develop distinct subsets of CTLs with diverse cell fates. Small differences in the amounts of some of these transcriptional regulators can have profound effects on CTL fate. For example, T-bet operates in a graded manner in effector CTLs, with moderate levels permitting memory cell fates but relatively higher levels promoting terminal differentiation (Joshi et al., 2007). Mechanistically, how modest differences in T-bet expression translate into distinct changes in gene expression, function, and specification of long-term fates in CTLs is not clear.
This study identifies a novel role for the transcription factor ZEB2 as one such translator of high T-bet expression. We find Zeb2 mRNA is highly expressed in terminally differentiated CTLs, in agreement with results from studies profiling gene expression in CTLs (Rutishauser and Kaech, 2010; Wirth et al., 2010; Best et al., 2013; Arsenio et al., 2014), and that this occurs in a T-bet–dependent manner. Deletion of ZEB2 reveals that it is necessary for normal TE cell expansion and transcriptional programming. Whole-transcriptome RNA sequencing (RNA-seq) analysis of WT, Tbx21−/−, and Zeb2−/− CTLs identified a set of genes whose expression was dependent on both T-bet and ZEB2, and suggested that they cooperate to promote TE genes while repressing MP genes. ZEB2 deficiency also affected T-bet binding to TE and MP cell signature loci, resulting in enhanced T-bet binding at several MP cell signature gene loci, suggesting T-bet alone cannot repress these genes. Motif enrichment analysis revealed that many of these MP loci contained putative ZEB2-binding sites, suggesting coordinated regulation of T-bet binding by ZEB2. Furthermore, overexpression of T-bet could not fully restore TE differentiation in ZEB2-deficient CTLs, particularly the repression of several MP cell signature genes, suggesting ZEB2 is necessary for the repression of this genetic program along with T-bet. Moreover, Zeb2 was necessary in TE cells to promote maximal expression of TE cell signature genes in conjunction with T-bet. Thus, we have identified ZEB2 as a developmental trigger, induced by higher amounts of T-bet, that works in conjunction with T-bet to switch off MP- and switch on TE-signature genes; ultimately, this process drives terminal CTL differentiation by restricting effector cell plasticity and memory cell potential.
Zeb2 is expressed in KLRG1hi CD8+ T cells in a T-bet–dependent manner
To examine the Zeb2 mRNA expression pattern in virus-specific CTLs during lymphocytic choriomeningitis virus (LCMV) infection, B6 mice were infected with the Armstrong strain of LCMV (LCMV-Arm), which causes an acute viral infection. Naive (CD44lo) CD8+ T cells were sorted from uninfected mice and MHC class I tetramer+ LCMV-specific CTLs were purified from infected mice at 6, 8, and >60 d postinfection (d.p.i.). The virus-specific CTLs were further subdivided to compare Zeb2 mRNA levels between more differentiated (KLRG1hi) and less differentiated (KLRG1lo) CD8+ T cells (Fig. 1 a). Zeb2 mRNA was not detectable in naive CD8+ T cells, but was induced in effector and memory CD8+ T cells, most dramatically in the KLRG1hi cells (Fig. 1 a). Tbx21 mRNA transcript expression showed an expression pattern similar to Zeb2 mRNA across both subsets and over time (Fig. 1 b), demonstrating the direct correlation of Zeb2 and Tbx21 expression in CTL subsets. In our hands, commercially available antibodies could not detect murine ZEB2 protein in CTLs, and lentiviral overexpression of a flag-tagged ZEB2 proved toxic in activated CD8+ T cells (unpublished data); thus, we were unable to measure or isolate ZEB2 protein by conventional methods in this study.
As T-bet controls the development of TE cells in a graded manner (Joshi et al., 2007, 2011), we next evaluated whether the induction of Zeb2 mRNA in effector CD8+ T cells was dependent on the amount of T-bet expressed by the CTLs. To do this, we generated P14 chimeric mice by transferring small numbers (∼5,000) of Tbx21+/+ WT), Tbx21+/−, or Tbx21−/− P14 CD8+ T cells, specific for the DbGP33-41 epitope of LCMV, into B6 recipient mice that were subsequently infected with LCMV-Arm. This transfer system ensured the analysis of only CD8+ T cell–intrinsic effects of T-bet deficiency on Zeb2 mRNA expression. Similar to the polyclonal CTLs, Zeb2 mRNA was most abundant in the KLRG1hi P14 CTLs, particularly the KLRG1hi IL-7Rlo subset (Fig. 1 c). Although fewer KLRG1hi cells formed in Tbx21+/− and Tbx21−/− CTLs, a sufficient number of KLRG1hi IL-7Rlo effector cells could be obtained for qPCR analysis, and this revealed a direct correlation between T-bet copy number and Zeb2 mRNA levels. That is, KLRG1hi IL-7Rlo cells lacking one copy of T-bet (Tbx21+/−) expressed 30–60% less Zeb2 mRNA than their WT counterparts, and those lacking both copies of T-bet (Tbx21−/−) expressed 90% less (Fig. 1 d). These data suggested that increased amounts of T-bet are needed for maximal Zeb2 mRNA expression in CTLs.
To determine whether T-bet directly bound to the Zeb2 locus, chromatin immunoprecipitation-sequencing (ChIP-seq) was performed on virus-specific CD8+ T cells with anti–T-bet antibodies. This identified predominant regions of T-bet binding at the Zeb2 transcriptional start site (TSS) and 3 kb upstream of the TSS (Fig. 1 e). Both of these sites contained histone H3K27 acetylation, suggesting the locus was transcriptionally active (unpublished data). Additionally, these binding sites were validated by ChIP-qPCR in WT and Tbx21−/− virus-specific CD8+ T cells (the Ifng promoter and a third region in exon 8 of Zeb2 served as positive and negative controls, respectively; Fig. 1 f). These data show that T-bet binds directly to the Zeb2 locus, providing stronger evidence that T-bet acts as a direct transcriptional activator of ZEB2 in virus-specific CTLs.
Effector and memory cell development in ZEB2-deficient CD8+ T cells
To investigate the function of ZEB2 in CD8+ effector T cell differentiation, Zeb2 was deleted from activated CD8+ T cells by crossing mice with floxed Zeb2 alleles (Zeb2flox/flox; Higashi et al., 2002) to mice expressing Cre recombinase under control of the Granzyme B promoter (GzmB-cre; Jacob and Baltimore, 1999). In this conditional knock-out system, the Zeb2 locus remains intact until a naive CD8+ T cell is activated and expresses GzmB and therefore also GzmB-cre. For simplicity, the Zeb2flox/flox; GzmB-cre+ mice will be referred to as Zeb2−/− mice and Zeb2flox/flox; GzmB-cre− or Zeb+/+; GzmB-cre+ littermate controls will be referred to as WT mice. It was important to use a conditional knock-out strategy because Zeb2−/− mice show embryonic lethality and Zeb2 is expressed in many cell types, including hematopoietic stem cells, NK cells, monocytes, and mast cells (Tylzanowski et al., 2003; van Grunsven et al., 2006; Goossens et al., 2011; Barbu et al., 2012; Best et al., 2013). Zeb2 was efficiently deleted in ∼85–90% of the polyclonal CD44hi CD8+ T cells using this system (unpublished data).
ZEB2-deficient CD8+ T cells underwent clonal expansion and CTL differentiation during LCMV infection, but the numbers of Zeb2−/− tetramer-specific CTLs was reduced by ∼50% at 8 d.p.i. Furthermore, the Zeb2−/− effector CTLs displayed accelerated rates of contraction, although by 60 d.p.i. similar numbers of Zeb2−/− and WT memory cells persisted (Fig. 2 a). The ZEB2-deficient CTLs appeared to have no gross defects in cytotoxicity based on in vivo CTL-killing assays (Fig. 2 b). Similarly, the ability of Zeb2−/− and WT CTLs to produce IFN-γ and TNF was comparable (Fig. 2 c, left). However, CTLs lacking Zeb2 produced more IL-2 than WT cells (Fig. 2 c, right). These results indicated that ZEB2 promoted effector cell clonal expansion and survival during acute viral infection, but was dispensable for cytotoxicity as well as IFN-γ and TNF production. However, ZEB2 appeared to restrict polyfunctionality by suppressing IL-2 production in CTLs.
Because increased IL-2 production is a property of TCM cells (Sallusto et al., 1999; Wherry et al., 2003b), we postulated that ZEB2 might repress the development of TCM cells. Indeed, analysis of memory P14 Zeb2−/− CTLs showed an accumulation of memory T cells with TCM phenotypes, such as decreased KLRG1 and increased l-selectin (CD62L), B cell lymphoma 2 (BCL-2), and interleukin receptor 2 β (IL-2Rβ) expression (Fig. 2 e). The phenotype of Zeb2-deficient memory CD8+ T cells suggested the rate of TCM formation after LCMV infection was accelerated in these mice. Functionally, there were no gross defects in the ability of ZEB2-deficient memory cells to produce cytokines or homeostatically turn over (Fig. 2, d and e). Transferred P14 Zeb2−/− memory CTLs retained the ability to protect recipient animals upon challenge with the chronic strain of LCMV clone13, expanding and controlling serum viremia similar to WT memory cells (Fig. 2 f). These data indicated that memory CD8+ T cell homeostasis and recall responses were largely unperturbed by Zeb2 deficiency.
ZEB2 represses MP gene expression in TE cells
Examination of virus-specific Zeb2−/− CTL subsets by flow cytometry revealed a significant reduction in the frequency of KLRG1hi IL-7Rlo TE cells and, consequently, an increase in the frequency of all other populations (Fig. 3 a). However, when cell numbers were calculated, there was only a significant reduction in the number of TE cells, without a corresponding increase in any other population (Fig. 3 a). Furthermore, the TE-like cells that were present in the Zeb2−/− effector population actually expressed less surface KLRG1 (based on mean fluorescence intensity [MFI]) and Gzma mRNA (based on RT-PCR) and had ectopic expression of several MP cell–associated genes, such as CD27 (encoded by Tnfrsf7) and the chemokine receptor C-X-C receptor 3 (CXCR3; Fig. 3 b). Therefore, we next analyzed the transcription factors involved in MP and TE cell fate determination, which revealed that the Zeb2−/− TE-like cells contained substantially lower amounts of T-bet protein and Prdm1 (Blimp-1) mRNA, two transcription factors necessary for TE cell development (Joshi et al., 2007; Kallies et al., 2009; Rutishauser et al., 2009; Shin et al., 2013), and conversely, increased amounts of TCF7, Eomes, and BCL6 protein, as well as Klf4 and Id3 mRNA, five transcription factors important for memory CD8 T cell development (Ichii et al., 2002, 2004; Kallies et al., 2009; Jeannet et al., 2010; Zhou et al., 2010; Cui et al., 2011; Shin et al., 2013; Fig. 3 c). In contrast, the expression of such pro–memory genes was not affected by Zeb2 deficiency in the MP-like cells (Fig. 3, b and c).
ZEB2 and T-bet regulate an overlapping set of genes to promote TE cell transcriptional programming
Comparison of Tbx21−/− and Zeb2−/− CTLs revealed a high degree of phenotypic similarity (Fig. 4, a and b). Although substantially fewer KLRG1+ cells formed in the absence of T-bet, de-repression of CD27, CXCR3, and Eomes was observed in these cells, similar to the Zeb2−/− TE cells (Fig. 4 b). Overall, this suggested that ZEB2 and T-bet may perform similar roles in TE cell differentiation. To investigate this in greater detail and to identify potential gene targets, we evaluated the genome-wide transcriptional profiles of Tbx21−/− and Zeb2−/− CTLs using RNA-seq (Fig. 4, c–f).
First, we defined MP and TE gene expression signatures by identifying genes differentially expressed between MP and TE cells at 8 d.p.i. (≥1.5 fold change with a false discovery rate [FDR] ≤ 0.1). We identified 1,458 genes that met these criteria, of which 777 were up-regulated in MP relative to TE cells, whereas conversely 687 genes were up-regulated in TE relative to MP cells. Next, we examined how the expression of the MP and TE cell signature genes was affected by T-bet or ZEB2 deficiency. We designated genes that were differentially expressed (≥1.5 fold change with a FDR ≤ 0.1) between the Tbx21−/− and WT CTLs or Zeb2−/− and WT CTLs as being T-bet or ZEB2 dependent, respectively (Fig. 4, e and f, red and blue gene sets). We referred to the set of genes differentially expressed in common between both knock-outs versus WT CTLs as the codependent gene set (Fig. 4, e and f, purple gene sets). These analyses showed that within both knock-out CTL populations there was a marked increase in the expression of many MP-signature genes and decrease in the expression of several TE-signature genes, with T-bet affecting a larger portion of these genes (Fig. 4 e). Importantly, supporting the notion that T-bet and ZEB2 cooperate to regulate CTL differentiation, a significant portion of TE and MP genes were codependent on both ZEB2 and T-bet function (Fig. 4, e and f, see purple codependent gene set).
To further analyze how T-bet and ZEB2 regulated the MP and TE cell signatures, we compared the differentially expressed genes in the knock-out CTLs to the MP- and TE-signature genes using volcano plots (Fig. 4 f). This clearly showed that genes up-regulated in Tbx21−/− or Zeb2−/− CTLs relative to their WT counterparts were preferentially enriched with MP-signature genes and conversely, those genes down-regulated were preferentially enriched with TE-signature genes. This pattern became even more apparent within the T-bet and ZEB2 codependent set of genes whose expression is altered in both Zeb2−/− and Tbx21−/− CTLs. Because the Tbx21−/− or Zeb2−/− CTL populations are comprised of different percentages of TE- and MP-like CTLs, we isolated the subsets of KLRG1hi IL-7Rlo and KLRG1lo IL-7Rhi effector cells from WT and Zeb2−/− CTLs to directly compare the MP- and TE-gene signatures within each cell subset by microarray analysis (Fig. 3 d). This analysis verified that Zeb2 was required specifically in the KLRG1hi IL-7Rlo cells for significant repression (* = FDR < 0.01) of several additional MP-signature genes, such as Socs3, Il7r, Pou6f1, Ctla4, Tcf7, and Ltb, and promotion of several TE-signature genes, such as Klra3, Klre1, and Itgam, in addition to the others shown in Fig. 3 (b and c). In contrast, none of the MP- or TE-signature genes were significantly differentially expressed in Zeb2−/− MP cells compared with WT MP cells, in line with the low amounts of Zeb2 mRNA in this cell subset (Fig. 3 d).
The genes affected by T-bet and ZEB2 deficiency can be assigned to various functional categories involved in CTL differentiation, including transcriptional regulation, T cell migration, and adhesion (Fig. 4 f). The codependent gene set affected in both Zeb2−/− and Tbx21−/− CTLs included known regulators of CTL differentiation, such as Prdm1 (Blimp-1), Runx2, and Tcf7, underscoring the importance of this codependent gene set (Fig. 4 f). Notably, when the codependent gene signature was compared with previously described transcriptional networks of coordinately regulated genes in CD8+ T cells, we found that they supported previous predictions of T-bet– and ZEB2-controlled gene networks (Best et al., 2013). Overall, these data suggest T-bet and ZEB2 similar roles in driving terminal CTL differentiation, through the regulation of several pathways, including previously described transcriptional networks.
T-bet binding is enriched in genes dependent on both T-bet and ZEB2 and modulated in the absence of ZEB2
We performed T-bet ChIP-seq to profile genome-wide T-bet binding patterns in WT CTLs 8 d.p.i. In total, there were 9,374 significant T-bet–binding sites (TBS; FDR < 1 × 10-5) across the genome. These binding sites were annotated to the nearest gene, and comparison to our gene expression data showed that T-bet bound to 48% of TE and 37% of MP cell signature genes (Fig. 5 a). Overall, T-bet bound to 47% of T-bet–dependent genes, which increased to nearly 60% in the ZEB2-dependent and codependent gene sets (Fig. 5 a). This result suggested a bias in T-bet binding and activity at genes co-regulated by ZEB2 and T-bet.
We then examined whether ZEB2 affected T-bet binding across the genome by performing T-bet ChIP-seq in Zeb2−/− CD8+ T cells isolated at 8 d.p.i. This comparative analysis identified 9787 T-bet binding sites (TBSs) across both sets of samples, of which the majority (94%) were not affected by Zeb2 deficiency (Fig. 5 b, regions within dotted lines,). However, 136 TBS showed a significant decrease, whereas 413 showed a significant increase in T-bet binding in Zeb2−/− compared with WT CTLs. Interestingly, we observed that several TE-signature loci (gold dots) displayed a loss in both T-bet binding and mRNA expression (Fig. 5, b and c, bottom left quadrant) and, conversely, many MP-signature genes (green dots) displayed an increase in both T-bet binding and gene expression in Zeb2−/− CD8+ T cells (Fig. 5, b and c, top right quadrant). Closer inspection of genes whose expression was dependent on T-bet (red dots), ZEB2 (blue dots), or codependent on both (purple dots) as previously defined (Fig. 4, e and f) showed strong correlations between T-bet binding and mRNA expression in the knockout CTLs (Fig. 5 b, middle and bottom plots).
To confirm the specificity of T-bet binding, we performed motif enrichment analysis at significant TBS common to WT and Zeb2−/− and found, as predicted, a significant central enrichment of known T-box motifs, T and Eomes (Fig. 5 d, top; Teo et al., 2011; Mathelier et al., 2014). Moreover, this was confirmed by performing a de novo motif search, which revealed a significant enrichment of a GSTGTGR motif (E-value = 4.6 × 10-83) that most closely resembles the T motif (E-value = 4.5 × 10-2; Fig. 5 d, bottom). Interestingly, Runx1- and Runx2-binding sites also appear to be enriched at TBS, which is in agreement with previous studies showing functional interactions between these transcription factors (Lazarevic et al., 2011), but also resemble T-box motifs in their consensus binding to G(G)TGTG sequences.
Next, we sought to examine computationally if ZEB2 or other transcription factors may bind coordinately with T-bet at certain loci associated with CTL differentiation by performing motif enrichment analysis of the differentially bound T-bet sites between WT and Zeb2-deficient CD8+ T cells (see diagram in Fig. 5 e). The most significantly enriched motif discovered within DNA regions associated with decreased T-bet binding (WT > Zeb2−/−) was a KTGTGA motif (E-value = 3 × 10−3), which most resembles the T motif (E-value = 0.026; Fig. 5 e). In contrast, the most significantly enriched motif discovered within the DNA regions associated with increased T-bet binding (Zeb2−/− > WT) was a CAGGTRW motif (E-value = 6.5 × 10−29), which most highly resembles the ZEB1 and ZEB2 E-box DNA binding domain motif (Comijn et al., 2001; Fig. 5 e). Lastly, sites with increased T-bet binding containing the CAGGTRW motif were identified and showed significant overlap with MP genes up-regulated in Zeb2−/−(Fig. 5 f). One could infer from these computational correlations that ZEB2 and T-bet coordinately bind to several MP-signature gene loci, and that ZEB2 is required to both restrict T-bet binding and to repress transcriptional activation of these genes. Direct testing of this hypothesis requires the ability to ChIP ZEB2, which is not feasible at this time.
T-bet regulation of several MP and TE genes is ZEB2 dependent
The gene expression profiling experiments showed that ZEB2 and T-bet regulate the expression of a common set of genes in CTLs (Fig. 4, c–f), and the T-bet ChIP-seq data suggested ZEB2 influences T-bet binding to its targets (Fig. 5, b–f). However, considering that ZEB2 is necessary for maximal T-bet expression, and that T-bet binding to the Tbx21 locus is somewhat lower in Zeb2−/− CTLs (Figs. 3 c and 5 c), it was still unclear whether ZEB2 mainly acted to promote T-bet expression or whether ZEB2 itself was directly necessary for the regulation of CTL gene expression.
To further address the cooperative functions between ZEB2 and T-bet, we overexpressed (OE) T-bet in WT and Zeb2−/− P14 CTLs using a T-bet expressing retrovirus (RV) and determined whether Zeb2 was necessary for the ability of T-bet to induce or repress TE- and MP-signature genes in CTLs during LCMV infection (Fig. 6, a and b). Intracellular staining for T-bet protein showed that it was expressed to a similar extent in both WT and Zeb2−/− P14 CTLs transduced with T-bet RV (Fig. 6 c). T-bet OE sufficiently induced expression of KLRG1 in both WT and Zeb2-deficient CTLs, and thus KLRG1hi TE-like cells were sorted to compare gene expression genome-wide using RNA-seq. This showed that when T-bet was OE in WT CTLs, it further enhanced expression of S1pr5, Zeb2, Notch3, and Cx3cr1 above and beyond that of the empty vector control cells (Fig. 6 a, compare columns 1 and 2). T-bet OE could also intensify repression of certain MP genes such as Cxcr3, Tcf7, Eomes, Ltb, IL-7r, and ccr7 in WT KLRG1hi CTLs compared with empty vector control cells (Fig. 6 a, compare columns 1 and 2). However, in the absence of Zeb2, T-bet OE was largely ineffective at altering gene expression at the majority of MP- and TE-signature genes (Fig. 6 a, compare columns 2 and 4), as well as genome-wide (Fig. 6 b). We refer to this category of genes (clusters I and II) as Zeb2 dependent. Of course, there were some Zeb2-independent genes (clusters III and IV) whose expression could be altered by T-bet OE in the absence of Zeb2, but this was a minority of genes (Fig. 6, a and b). The effects of T-bet OE on protein expression in WT and Zeb2-deficient CTLs were confirmed for a subset of genes by flow cytometry (Fig. 6 c). These data showed quite strongly that T-bet and ZEB2 act in a cooperative, nonredundant manner because T-bet was heavily dependent on Zeb2 for its ability to repress a large number of MP-signature genes and, more surprisingly, to induce the many TE-signature genes in KLRG1hi CTLs. Thus, in addition to augmenting T-bet protein levels in TE cells, ZEB2 is also necessary downstream of or in conjunction with T-bet to drive the transcriptional programming of terminal differentiation in CD8+ T cells.
During infection, the adaptive immune response has two primary goals; a short-term goal of eradicating the present infection and a long-term goal of establishing immunological memory, all of which needs to occur without excessive collateral damage to host tissues. Elucidating the factors that control effector T cell differentiation to fulfill these two goals is essential for understanding how long-lived immunity to pathogens forms and potentially allowing for therapeutic manipulation of effector and memory CD8+ T cell function. Previous work identified T-bet as a central regulator of terminal differentiation in CD8+ T cells, which operates in a graded manner with higher amounts driving TE and lower amounts supporting MP cell differentiation (Takemoto et al., 2006; Joshi et al., 2007). In this study, we found that high amounts of T-bet induced Zeb2 mRNA expression and that ZEB2 was critical for the transcriptional programming of TE cell differentiation. To our knowledge, this study, along with that of Omilusik et al. (2015), is the first to describe the function of ZEB2 in T cells. Overall, these data suggest that ZEB2 and T-bet enforce terminal effector differentiation through their cooperation by simultaneously inducing TE and repressing MP cell gene expression. Thus, this study not only identifies a novel regulator of TE fates, but also provides new insights into the mechanisms by which graded T-bet drives terminal effector differentiation through its cooperative functions with ZEB2.
The differentiation of stem cells into specialized cell types with distinct functions and phenotypes involves the activation of differentiation genes and the suppression of stemness genes (Bernstein et al., 2006; Sánchez Alvarado and Yamanaka, 2014). A similar process occurs in CD8+ T cells as terminal effector cells develop apart from memory precursor CD8+ T cells, and this study has identified key roles for T-bet and ZEB2 in controlling these differentiation programs. Given that ZEB2 has primarily been described as a transcriptional repressor, known to recruit histone deacetylases (Verstappen et al., 2008; Wang et al., 2009; Kim et al., 2012), we predicted that ZEB2 would repress MP cell genes and this was confirmed by analysis of gene expression in Zeb2−/− KLRG1hi CTLs. Based on the data presented herein, we postulate that once T-bet levels surpass a given threshold, T-bet switches on ZEB2, and then T-bet and ZEB2 bind in a coordinated fashion at certain MP regulatory loci to mediate their repression. This model is supported by the proximity and overlap of ZEB1/2 and T-box motifs in TBS at several MP loci that are derepressed in Zeb2−/− TE-like cells. Several of these loci displayed both increased T-bet binding and expression in Zeb2-deficient CTLs, and their mRNA expression was further augmented by T-bet OE. These data suggest that ZEB2 may serve dual roles at distinct MP loci by both restricting T-bet binding and recruiting gene silencing machinery. Possibly T-bet directly recruits ZEB2 to such loci similar to the model recently proposed for transcriptional repressor Bcl6 (Oestreich et al., 2011). It is worth noting that some of these MP genes that show both increased T-bet binding and mRNA expression in Zeb2−/− CTLs are also increased in expression in the absence of T-bet, suggesting increased T-bet binding is not necessary for their expression. One possibility is that, the T-box TF homologue of T-bet, Eomesodermin, acts redundantly to promote MP signature gene expression in Tbx21−/− CD8 T cells.
A potential caveat with the experiments comparing T-bet binding between WT and Zeb2−/− CTLs is that the two populations contained different frequencies of TE and MP cells, and this may have biased the results. We argue that this confounding issue is likely insignificant because T-bet binds to ∼35% of MP-signature genes in CTLs, yet only 6% of these genes demonstrated significant gains in T-bet binding in the absence of Zeb2 (if this was simply an artifact of overrepresentation of MP cells in the Zeb2−/− cell population, one would expect that all of the MP cell loci with TBSs would have shown increased T-bet binding). Moreover, many of the TBSs that showed increased binding in Zeb2−/− cells also contained predicted ZEB2-binding sites, yet this was not observed in the TE genes that displayed decreased T-bet binding in Zeb2−/− cells. Thus, there was a selectivity of increased T-bet binding at MP loci that also had neighboring predicted ZEB2-binding sites. Rigorous molecular testing of this model in murine T cells would require better reagents to study ZEB2 protein, or possibly future studies in human T cells to determine the conservation of ZEB2 function.
The Zeb2−/− CTL phenotype does not simply reflect a reduction in T-bet protein, as T-bet overexpression could not fully restore the TE cell genetic program. Rather, we find that Zeb2 is a necessary downstream mediator of T-bet activity in driving terminal effector cell differentiation. As T-bet overexpression failed to induce several TE-cell signature genes in the absence of Zeb2, this suggests that ZEB2 is necessary to either (a) repress a pro–memory transcriptional repressor that restricts TE-signature gene expression (e.g., Bcl6) or (b) it has an alternative function as a transcriptional activator in TE cells. Indeed, there are reports that ZEB2 can act as a transcriptional activator when complexed with SP1, driving the expression of vimentin and integrin α-5 in certain cancer cell lines without directly binding DNA (Bindels et al., 2006; Nam et al., 2012).
An interesting observation made during the course of this study was that Zeb2 mRNA remained elevated in the subset of KLRG1+ IL-7R+ TEM cells after effector cell contraction. Although cells of this phenotype display a limited lifespan and proliferative capacity compared with KLRG1lo IL-7Rhi memory CD8+ T cells in circulation, they can offer enhanced protection to some types of pathogens due to elevated cytotoxic activity (Jabbari and Harty, 2006; Olson et al., 2013) and, interestingly, are more prominently observed after serial infections (Masopust et al., 2006; Wirth et al., 2010; Joshi et al., 2011; Fraser et al., 2013; Olson et al., 2013). Thus, it will be important to determine if sustained ZEB2 expression is necessary for maintenance of this population of TEM cells. Furthermore, Zeb2 mRNA levels are elevated in CD8+ T cells during chronic LCMV infection (Doering et al., 2012) and reduced in TRM cells relative to circulating TEM cells (Wakim et al., 2012; Mackay et al., 2013; Tse et al., 2013). Perhaps Zeb2 mRNA is repressed in TRM cells because it enhances CTL expression of Cx3cr1 and S1pr5, which may affect the tissue-restricted localization of TRM cells. In the future, it will be of great interest to uncover the role of ZEB2 in regulating CTL differentiation in different tissues and settings of infection. Moreover, ZEB2 is also expressed in CD4+ T cells, NK cells, monocytes, and other immune cells, and it might function in a similar manner, possibly in cooperation with T-bet, to promote terminal differentiation of other lymphocytes. Elucidating the genetic pathways that switch on terminal differentiation in T cells will enhance our understanding of how specialized types of T cells are established and stabilized during immune responses, which could inform therapies aimed at enhancing effective CTL memory development.
MATERIALS AND METHODS
C57BL/6 (B6) mice were obtained from the National Cancer Institute (Frederick, MD). Zeb2flox/flox mice were originally generated by D. Hoylebroeck (University of Leuven, Leuven, Belgium; Higashi et al., 2002) and obtained from R. Aslopp (John A. Burns School of Medicine, University of Hawaii, Honolulu, HI). Granzyme B-Cre (GzB-Cre+) mice were provided by J. Jacobs (Emory University, Atlanta, GA) via R. Flavell’s laboratory (Yale University School of Medicine, New Haven, CT) and were crossed to Zeb2flox/flox mice for generation of GzB-cre+; Zeb2flox/flox (Zeb2−/−) mice and GzB-cre+; Zeb2+/+ or GzB-cre−; Zeb2flox/flox (Zeb2+/+) mice. Zeb2−/− and Zeb2+/+ mice were further crossed to P14 TCR transgenic mice so that P14 Zeb2−/− and P14 Zeb2+/+ mice could be obtained. To generate “P14 chimeric mice” 10–50,000 P14 CD8+ T cells were transferred into B6 mice by i.v. injection. All animal experiments were done with approved institutional animal care and use committee protocols.
Infections and treatments
For infections of mice, 2 × 105 PFU of the LCMV Armstrong strain were administered i.p. For recall experiments, mice were infected with 2 × 106 PFU of the LCMV clone 13 strain i.v. by retroorbital injection. Viral titers were measured by plaque assay (Wherry et al., 2003a).
For BrdU labeling to determine homeostatic proliferation, BrdU (1 mg/ml) was administered in the drinking water daily for 10 d.
Antibodies for surface and intracellular staining
Lymphocyte isolation, along with surface and intracellular staining, was performed as described previously (Joshi et al., 2007). For in vitro stimulation, splenocytes were stimulated with GP33-41 and GPNP396-404, peptides (100 ng/ml) for 5 h in the presence of Brefeldin A. Antibodies were purchased from eBioscience, BD, or BioLegend and Cell Signaling Technology. Class I MHC tetramers were generated as described previously (Kaech et al., 2003). Anti-BRDU antibody was purchased from BD and used according to instructions. Flow cytometry data were acquired on BD LSRII with Diva software and analyzed with Flow Jo software (Tree Star). Sorting was performed on a FACS Aria (BD).
Naive splenocytes were isolated from B6 mice and either labeled with 1 µM CFSE and pulsed with GP33-41 (100 ng/ml) or labeled with 0.1 µM CFSE alone. In total, 5 million of each target cell population were adoptively transferred i.v. to day 8 LCMV-infected WT or Zeb2−/− mice. 1 h later, the mice were sacrificed and flow cytometry was used to determine the presence of transferred cells in the spleen.
ChIP and ChIP-sequencing (ChIP-seq)
ChIP experiments were performed on Thy1.1 P14 CTLs enriched by selection of the congenic marker Thy1.1 from 8 d.p.i. splenocytes using Easy Sep biotin selection kits (STEMCELL Technologies Inc.) in conjunction with anti-Thy1.1 mAbs (eBioscience). Sample purity was >90%.
One additional sample (Fig. 1 e) was stimulated with IL-12 (10 ng/ml) to enhance T-bet expression. Chromatin was recovered from batches of 10 million cells, after cross-linking with 1% formaldehyde for 10 min and sonication to obtain DNA fragment ∼200–500 bp. ChIP was performed with anti–T-bet (Santa Cruz Biotechnology, Inc.) from 20 million cells chromatin was isolated, amplified, and processed into a library. Illumina HiSeq 2000 was used for sequencing. Sequence reads from each cDNA library (paired-end, 75 bp) were mapped onto the mouse genome build mm10 using bowtie2.
Conventional ChIP was performed as described but anti–mouse IgG was used as a negative control. Three independent experiments were performed. Immunoprecipitated DNA was analyzed by qPCR. Sybr-based qPCR was performed with the following primers: Ifng promoter forward, 5′-GCTTTCAGAGAATCCCACAAGAAT-3′, Ifng promoter reverse, 5′-GCTATGGTTTTGTGGCATGTTAGA-3′; Zeb2TSS promoter forward: 5′-CGGCTGCTTCATTGATAAGA-3′, Zeb2TSS promoter reverse: 5′-CGCTGTGTTTGGTTGCTAGA-3′; Zeb2 3-kb upstream forward: 5′-GATGCAGGGGGCTGATTAT-3′, Zeb2 3kb upstream reverse: 5′-CCCCCTTTTGTGAGACTGA-3′; Il-7r promoter forward: 5′-TTGCTGCTACCAATCAGTAAGAAT-3′, Il-7r promoter reverse: 5′-TGGGGCTCTTTTACGAGTGA-3′.
Differential binding analysis and quantification of ChIP-Seq
Using the MACS v2.1.0 peak caller (Feng et al., 2012), peaks were first called relative to input using the callpeak module with the –B,–nomodel, and –extsize 200 options set. Peaks from replicates within groups were intersected to identify consensus peaks, and then the union of peaks between groups was taken to serve as features used for further analysis. To identify differential binding events, these features were quantified for signal using the bedgraph output and bedtools map utility to sum the reads normalized for sequencing depth over the enriched regions. HOMER2 was then used to annotate the peaks to the nearest gene using the GRCM38 (mm10) reference genome. All processing thereafter was performed in R, importing the annotated peaks and tag quantification per peak. For the purpose of visualization, the two replicates per group (WT and Zeb2−/−) were pooled and visualized using the UCSC Genome Browser.
Motif enrichment analysis
Motif enrichment analysis was performed using the MEME 4.10 suite of tools. Genomic sequences at the summit of T-bet binding peaks ±100 bp were analyzed for central enrichment using Centrimo. 200 bp genomic sequences of regions identified as differentially bound were compared against a background of sites with no differential binding (common) to search for the enrichment of de novo motifs. Using the DREME algorithm with default settings on each group, discovered motifs were then submitted to Tomtom to find the most similar motif for each de novo motif. For the CAGGTRW motif enriched in the Zeb2−/− > WT group, sites containing this motif were identified using the FIMO algorithm, and visualized using R.
Viral supernatants for the retroviral constructs described in these studies were obtained by transfection of 293T cells with the respective retroviral construct and Eco-helper. Transfections were performed using Fugene6 (Promega) or Xtremegene9 (Roche); no difference was noted in transfections performed with these reagents. P14 donor mice were directly infected with 2 × 106 PFU LCMV-Armstrong i.v. and, 1 d later, mice were sacrificed and splenocytes were isolated. Splenocytes were spin-transduced for 90 min at 34°C with viral supernatant in the presence of 8 µg/ml polybrene. After transduction, 0.5–1 × 105 P14 CTLs were transferred i.v. to recipient mice B6 mice that were subsequently infected with 2 × 105 PFU LCMV i.p.
Gene expression by qRT-PCR
For qRT-PCR, RNA was isolated from 200,000–1,000,000 sorted cells by QIAshredder and RNeasy kits (QIAGEN); the use of QIAzol and RWT buffers with the RNeasy kits; or TRIzol extraction (Life Technologies) followed by ethanol precipitation. CDNA was synthesized using SSRTII (Life Technologies) and qRT-PCR was performed on a Stratagene Mx3000P with iTaq Universal SYBR Green super mix (Bio-Rad Laboratories). Relative fold changes were calculated using Rpl9 (L9) expression.
The following primers were used in these studies: Zeb2 forward: 5′-GAGCAGGTAACCGCAAGTTC-3′ Zeb2 reverse: 5′-TGTTTCTCATTCGG-3′; Rpl9 forward: 5′-TGAAGAAATCTGTGGGTCG-3′ Rpl9 reverse: 5′-GCACTACGGACATAGGAACTC-3′; Prdm1 forward: 5′-GACGGGGGTACTTCTGTTCA-3′ Prdm1 reverse: 5′-GGCATTCTTGGGAACTGTGT-3′; Eomes forward: 5′-ATGTACGTTCACCCAGAATC-3′ Eomes reverse: 5′-GTGCAGAGACTGCAACACTA-3′S; Klf4 forward: 5′-CCACACTTGTGACTATGCAG-3′ Klf4 reverse: 5′-CCAGTCACAGTGGTAAGGTT-3′ Cxcr3; forward: 5′-CAGCCAAGCCATGTACCTTGAG-3′ Cxcr3 reverse: 5′-TCAGGCTGAAATCCTGTGGGCA-3′; Id3 forward: 5′-ACTTACCCTGAACTCAACGC-3′ Id3 reverse: 5′-CTCCAAGGAAACCAGAAGAA-3′; Tbx21 forward: 5′-CAACAACCCCTTTGCCAAAG-3′ Tbx21 reverse: 5′-TCCCCCAAGCAGTTGACAGT-3′; Gzma forward: 5′-TCAGCTCCCTCTGAAACTCT-3′ Gzma reverse: 5′-TCTCCACCAAAAGAGGTGAT-3′.
RNA-seq library preparation and data analysis
Total RNA was purified with the use of QIAzol and RNeasy Mini kit (QIAGEN), in which an on-column DNase treatment was included. Purified RNA was submitted to the Yale Center for Genomic Analysis where it was subjected to mRNA isolation and library preparation. Libraries were pooled, six samples per lane, and sequenced on an Illumina HiSeq 2500 (75-bp paired end reads), and aligned using STAR to the GRCm38 (mm10) reference genome. A count-based differential expression protocol was adapted for this analysis (Anders et al., 2013); mappable data were counted using HTSeq, and imported into R for differential expression analysis using the DESeq2. To find differentially regulated sets of genes for signature generation, an absolute 1.5-fold-change difference between samples and FDR (Benjamini-Hochberg) ≤ 0.1 was used.
Prism 6 (GraphPad Software) was used to calculate statistics for all bar graphs shown. For comparisons of two groups, two-tailed Student’s t test was performed. For multiple group comparisons, one-way ANOVA was used with Tukey's multiple comparisons test. For grouped multiple comparisons two-way ANOVA with Sidak's multiple comparison test was used. *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001.
We thank the members of the Kaech, Craft, and Kleinstein laboratories for helpful comments and suggestions, in particular Drs. W. Cui and I. Parish. We'd like to thank Drs. J. Noonan and J. Cotney for advice on ChIP-seq analysis and visualization. We also thank Dr. R. Allsopp for providing us with Zeb2flox/wt mice and Dr. L. Glimcher for the T-bet RV construct.
This work was supported by grants from the National Institutes of Health (3R01AI07469905 and 3R01AI07469905S; S.M. Kaech) and National Research Service Award (National Institute of Allergy and Infectious Diseases, Ruth Kirschstein Predoctoral Fellowship F31 AI084500-01 to C.X. Dominguez), the Howard Hughes Medical Institute Gilliam Fellowship (R.A. Amezquita), the Howard Hughes Medical Institute International Student Research Fellowship (T. Guan) and the Howard Hughes Medical Institute (S.M. Kleinstein).
The authors declare no competing financial interests.
cytotoxic T lymphocyte
false discovery rate
killer-cell lectin-like receptor G1
lymphocytic choriomeningitis virus
T-bet binding site
central memory T cell
effector memory T cell
resident memory T cell
transcriptional start site
C.X. Dominguez, R.A. Amezquita, and T. Guan contributed equally to this paper.