Eomes cannot replace its paralog T-bet during expansion and differentiation of CD8 effector T cells

The two T-box transcription factors T-bet and Eomesodermin (Eomes) are important regulators of cytotoxic lymphocytes (CTLs), such as activated CD8 T cells, which are essential in the fight against intracellular pathogens and tumors. Both transcription factors share a great degree of homology based on sequence analysis and as a result exert partial functional redundancy during viral infection. However, the actual degree of redundancy between T-bet and Eomes remains a matter of debate and is further confounded by their distinct spatiotemporal expression pattern in activated CD8 T cells. To directly investigate the functional overlap of these transcription factors, we generated a new mouse model in which Eomes expression is under the transcriptional control of the endogenous Tbx21 (encoding for T-bet) locus. Applying this model, we demonstrate that the induction of Eomes in lieu of T-bet cannot rescue T-bet deficiency in CD8 T cells during acute lymphocytic choriomeningitis virus (LCMV) infection. We found that the expression of Eomes instead of T-bet was not sufficient for early cell expansion or effector cell differentiation. Finally, we show that imposed expression of Eomes after acute viral infection promotes some features of exhaustion but must act in concert with other factors during chronic viral infection to establish all hallmarks of exhaustion. In summary, our results clearly underline the importance of T-bet in guiding canonical CTL development during acute viral infections.


Introduction
The transcription factors (TFs) T-bet and Eomesodermin (Eomes) regulate type 1 immune responses in innate [1] and adaptive lymphocytes [2]. Such responses are broadly characterized by the signature cytokine IFN-γ and cytotoxic effector function against intracellular pathogens [3]. T-bet and Eomes both belong to the large and diverse T-box TF family, that all share the evolutionary conserved DNA-binding domain, the T-box. T-box TFs are present in all metazoans and play important roles in cell fate decisions and organogenesis during development. T-box genes can be grouped into five subfamilies and genes within the same subfamily have arisen from duplication of a single ancestral gene [4]. T-bet and Eomes together with Tbr1 belong to the Tbr1 subfamily and are thus paralogs that share a high degree of sequence homology, especially within their T-box domains. Nevertheless, their regulation and expression within the immune system is quite distinct, which implies unique functions [5][6][7]. T-bet was initially identified as the lineage specifying TF of Th1 CD4 T cells [2], who could also repress alternative cell fates like Th2 or Th17. However, T-bet's role extends beyond CD4 T cells as it is also expressed in activated CD8 T cells [8], innate-like T cells (e.g. NKT cells) [9], CD8αα T cells [10], B cells [11], innate lymphoid cells (ILCs) [1], and even non-lymphoid cells like dendritic cells [12]. Accordingly, T-bet deficiency has broad implications for infectious diseases, immunopathology and anti-tumor immunity. In contrast, Eomes has a much more restricted expression pattern. Although partly expressed in certain CD4 [13,14] and γδ T cells subsets [15], Eomes is largely confined to conventional NK (cNK) cells [5] and activated or memory CD8 T cells [16], both of which also express T-bet. Based on experiments with the respective knockout mice, it has been argued that T-bet and Eomes have partially redundant and overlapping functions in CD8 T cells and can therefore compensate for each other [3]. Initially, this has been investigated during acute viral infections with lymphocytic choriomeningitis virus (LCMV) Armstrong (Arm). This infection leads to the differentiation of naïve CD8 T cells into cytotoxic T cells (CTLs), whose two cardinal features, cytotoxicity and cytokine production, are controlled by T-box TF and are essential for viral control [7]. Interestingly, mice that lack either T-bet or Eomes in their T cell compartment show apparently no defect in the early production of IFN-γ or cytotoxic molecules and, similar to wild type mice, show no viral replication in spleen and kidney after three weeks [3]. Yet, if mice lack both, T-bet and Eomes, they fail to produce sufficient amounts of IFN-γ or to execute sufficient cytotoxicity and are therefore unable to control viral replication. Moreover, activated T cells from these mice show an aberrant type 17 response that leads to progressive multi-organ infiltration with neutrophils and early death [3]. It was therefore concluded that T-bet and Eomes both can regulate early differentiation of CTLs and are therefore redundant during acute infection with intracellular pathogens. However, later it became clear that these two TFs are not simultaneously and uniformly expressed in activated CD8 T cells. Whereas virtually all activated CD8 T cells express T-bet to different degrees, only a subpopulation additionally expresses Eomes [7]. This heterogeneity might be partially explained by increased T cell receptor signaling [17], abundance of inflammatory cytokines, like IL-12 [18], or heightened mTOR kinase activity [19], which favor the induction of T-bet in CTLs while repressing Eomes.
Although numerous models of early T cell differentiation have been developed, T-bet is well known for its role in the formation of short-lived effector cells (SLECs), which are KLRG1 + CD127and do not develop in Tbx21 -/-(Tbx21 encodes for T-bet) mice [20]. This phenotype already indicates that T-bet is essential during acute infection with LCMV Arm and that Eomes only provides partial redundancy here. In contrast, memory precursor effector cells (MPECs), which are KLRG1 -CD127 + , and early effector cells (EECs), which are KLRG1 -CD127are not reduced in Tbx21 -/mice. Intriguingly, Eomes is equally expressed in a subpopulation of SLECs and MPECs and T cell specific Eomes deficient mice show no perturbations in SLEC and MPEC formation during acute infection [7]. However, Eomes becomes vital after viral clearance, after which the percentage of Eomes expressing CD8 T cells increases with time after infection. This process intrinsically requires Eomes, as Eomes deficient memory T cells show defects in long-term maintenance and re-expansion upon secondary challenge when competing with wild type T cells [21].
These conclusions were further corroborated by lineage-tracing experiments in Eomes deficient mice at steady-state. It has been shown that CD44 + CD62L + central memory CD8 T cells (T CM ) show the highest percentage of Eomes expressing cells; followed by CD44 + CD62Leffector memory CD8 T cells (T EM ), whereas CD44 -CD62L + naïve CD8 T cells do not express Eomes. Intriguingly, it was demonstrated that Eomes was not a mere marker of cells when expressed, but that Eomes + memory T cells at steady state, similar to Eomes + memory T cells after an acute infection, depend on Eomes for their survival or development [22].
Further insights into specific functions of T-bet and Eomes came from mice, which were chronically infected with LCMV clone 13 (cl13). The partially inverse expression pattern of Tbet and Eomes can be used to identify two distinct transcriptional states: T-bet high Eomes low and T-bet low Eomes high memory CD8 T cells that cooperatively sustain an effective antiviral immune response [22]. Moreover, using additional markers to discern T-bet low Eomes high CD8 memory T cells it has been demonstrated that Eomes is not only essential for the development of this cellular subset but also for its maintenance, once it is established [23]. In mice chronically infected with LCMV cl13, T-bet low Eomes high cells accumulate with time after infection and show characteristics of exhaustion, such as high expression of PD-1, LAG-3, TIM3, TIGIT, CD39, CD160, CTLA-4, as well as reduced production of pro-inflammatory cytokines and only moderate proliferation potential as compared to T-bet high Eomes low CD8 memory T cells [7,23]. Apparently, this phenotype allows for infection control without causing overt immunopathology during progressive disease.
In summary, previous studies suggested that T-bet preferentially drives a program of terminal effector differentiation while Eomes supports a program of long-term maintenance with adapted effector function. However, it has to be emphasized that activated T-bet high Eomes low and T-bet low Eomes high CD8 T cells engage distinct transcriptional networks that involve various additional transcriptional regulators [24], e.g. Id3, Bcl6, Tcf1, Blimp-1, IRF4, TOX, which contribute to transcriptional and functional differences between these populations. Nonetheless, this plethora of observations clearly indicates that despite their sequence homology and partial redundancy T-bet and Eomes delineate two different effector cell populations. As described above, a key challenge of interpreting previous results with respect to the functional redundancy of T-bet during acute infection is the heterogeneous expression pattern of Eomes in activated CD8 T cells. Bona fide functional redundancy on a transcriptional level would predict that when an activated CD8 T cell cannot express T-bet, e.g. Tbx21 -/-CD8 T cell, it will by default express Eomes, which then coordinates a transcriptional and, eventually, functional program similar to T-bet. Yet, it is not fully clear whether activated CD8 T cells in Tbx21 -/mice uniformly express Eomes. This alternative expression pattern would be an important pre-condition to directly test and compare the transcriptional activities of T-bet and Eomes at the single cell level. However, since no such studies have been performed so far in vitro or in vivo, the degree of functional redundancy of T-box TFs within the immune system remains a matter of debate.
To address the redundancy of T-bet and Eomes, we generated a novel model, in which the Eomes coding sequence was introduced into the first exon of the Tbx21 locus followed by a P2A sequence and mCherry as fluorescent reporter (Tbx21 Eomes-P2A-mCherry/+ , hereafter called Tbx21 E ). This allele results in a null allele for T-bet and the synchronized expression of Eomes in lieu of T-bet as the Tbx21 E allele is entirely under the transcriptional control of the Tbx21 locus. We then went on to stringently test the functional redundancy of T-bet during acute infection with LCMV WE, by comparing Tbx21 +/+ wild type with Tbx21 -/and Tbx21 E/E mice. Finally, to exclude extrinsic effects of our Tbx21 E allele or the state of infection on activated CD8 T cells, we investigated the redundancy of T-bet in an adoptive transfer model.
The results presented here unequivocally show that Eomes cannot properly compensate for the loss of T-bet in activated CD8 T cells during acute infection with LCMV WE but rather promotes an exhaustion-like phenotype. In summary, this study underlines the importance of the diversification of the Tbr1 T-box TF family in the immune system at the onset of vertebrate evolution [25].

Generation of a novel mouse model by putting Eomes expression under control of the endogenous Tbx21 locus
Previously, a Tbx21-Eomes transgenic mouse model had been generated by using a bacterial artificial chromosome (BAC) approach, in which expression of Eomes was put under the transcriptional control of the Tbx21 regulatory elements of the BAC transgene [26]. However, such an approach does not tightly control for copy numbers and insertion, which might produce several off-target effects. Moreover, this strategy does not distinguish between transgenic and endogenous Eomes expression. To study the functional redundancy of T-bet in activated CD8 T cells, we therefore used TALEN-guided knock-in technology [27] to insert the Eomes coding sequence into the Tbx21 locus followed by a P2A peptide-sequence and the mCherry reporter to generate the Tbx21 Eomes-P2A-mCherry/+ allele (hereafter called Tbx21 E ; Fig 1A). Tbx21 E/+ heterozygous mice were interbred to obtain Tbx21 E/E mice, that homozygously express Eomes from the Tbx21 locus. Hence, these mice were also deficient for Tbx21. Mice were born at mendelian ratio and showed no developmental abnormalities or clinical immunopathology during a 3-year observation period. The Tbx21 E allele had no effect on absolute cell numbers in spleen, lymph nodes or thymus (S1A Fig). Furthermore, the thymic distribution of CD4 + single positive (SP), CD8 + SP, CD4 + CD8 + double positive and CD4 -CD8double negative cells was unperturbed (S1B Fig). Similarly, proportions of CD4 and CD8 T cells as well as B cells in spleen (S1C Fig) and lymph nodes (S1D Fig) showed only minor differences. The Tbx21 E allele was associated with more T CM (CD44 + CD62L + ), which was most pronounced in Tbx21 E/E mice and in line with the notion that Eomes promotes maintenance of T CM in wild type mice. Finally, expression of the Tbx21 E allele had no effect on the frequency of CD4 + CD25 + regulatory T cells in spleen and lymph nodes. In summary, Eomes expression from the Tbx21 E allele showed no global effects on the adaptive immune system at steady state. Detailed

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections characterization of innate and innate-like lymphocytes in these mouse strains will be reported elsewhere. To test, whether the Tbx21 E allele was truly under the transcriptional control of the Tbx21 locus, we stimulated splenocytes from Tb21 +/+ , Tbx21 -/and Tbx21 E/E mice with anti-CD3 with or without IL-12. IL-12 signals via STAT4 and it has been shown for CD4 [28] and CD8 T cells [20] that it is a strong inductor of T-bet. Additionally, IL-12 represses Eomes in CD8 T cells [18]. Stimulation with anti-CD3 alone led to uniform expression of T-bet in CD4 T cells from Tbx21 +/+ mice (Fig 1B), which was further increased by IL-12 ( Fig 1B). As control Tbx21 -/and Tbx21 E/E showed no expression of T-bet in any condition. Moreover, stimulation of Tbx21 +/+ cells induced low levels of Eomes in CD4 T cells, which was repressed by additional IL-12 ( Fig 1B). Intriguingly, stimulation of Tbx21 -/cells with anti-CD3 did not lead to an increase of Eomes expression, arguing against a default expression of Eomes in the absence of T-bet ( Fig 1B). As predicted, stimulated CD4 T cells from Tbx21 E/E mice showed uniform and high expression of Eomes, which was further boosted by IL-12 ( Fig 1B). Hence, Eomes expression from the Tbx21 E locus recapitulates the expression pattern of T-bet expression in Tbx21 +/+ CD4 T cells.
Next, we explored the faithfulness of the mCherry reporter in CD8 and cNK cells, as these cells typically express T-bet. At steady state T-bet showed the highest expression in cNK cells, intermediate levels in CD8 memory and was not detectable in naïve wild type CD8 T cells. Correspondingly, mCherry showed the strongest signal in cNK cells and no signal in naïve CD8 T cells (S1E and S1F Fig). Memory CD8 T cells at steady-state showed an intermediate signal-strength. However, this expression could be boosted by polyclonal stimulation with anti-CD3 in combination with IL-12 ( Fig 1C). Similar results were obtained with CD4 T cells ( Fig 1C). In summary we can conclude, that the expression of Eomes and mCherry under the transcriptional control of the Tbx21 locus is reflecting T-bet expression and is dynamically regulated in response to T cell receptor (TCR) stimulation and cytokine signaling.

T-bet deficient Tbx21 E/E mice fail to rescue the expansion and differentiation deficit of Tbx21 -/mice
T-bet plays an important role during the acute phase of an infection by guiding the development of CTLs [8]. We therefore investigated its redundancy in LCMV WE low dose infection, which is usually controlled within 8 days after infection in Tbx21 +/+ wild type mice. Hence, we infected Tbx21 +/+ , Tbx21 -/and Tbx21 E/E mice with 200 plaque forming units (pfu) LCMV WE and analyzed infected mice 8 days later (Fig 2A). Upon gross examination it was apparent, that Tbx21 +/+ wild type mice developed a massive splenomegaly, which was strongly curtailed in either Tbx21 deficient strain (Tbx21 -/and Tbx21 E/E ) (S2A Fig). This was partially the result of a failed CD8 T cell expansion, with Tbx21 +/+ wild type mice having around 5-times more CD8 T cells than Tbx21 -/or Tbx21 E/E mice (Fig 2B), whereas the absolute numbers of CD4 T cells was not different among the three strains (S2A Fig). Correspondingly, we detected a massive reduction of LCMV glycoprotein 33-41 epitope (GP 33 )-specific CD8 T cells (GP 33 -tet + ) in Tbx21 -/and Tbx21 E/E mice ( Fig 2C). Despite the striking difference in absolute numbers the frequencies of GP 33 -tet + CD8 T cells were similar in Tbx21 +/+ and Tbx21 E/E mice ( Fig 2C) and significantly higher than in Tbx21 -/mice. Intranuclear staining for T-bet and Eomes in GP 33tet + CD8 T cells from Tbx21 +/+ wild type mice at day 8 revealed, that they uniformly expressed T-bet, but only a subpopulation was additionally positive for Eomes ( Fig 2D). The latter was also confirmed in Eomes Gfp/+ reporter mice [29], which demonstrated that at day 8 of an acute infection around 15% of GP 33 -tet + CD8 T cells are T-bet + Eomes + (GFP + ), with the remaining cells being T-bet + Eomes -(GFP -) (S2B Fig). GP 33 -tet + CD8 T cells in Tbx21 -/mice showed a significant increase in the percentage of Eomes + cells, however, the majority of cells (~70%) remained Eomes - (Fig 2D). Hence, we can conclude that the absence of T-bet does not result in the mandatory expression of Eomes and therefore Eomes, by definition, cannot compensate for the actions of T-bet in the majority of CD8 T cells during an acute infection with LCMV WE.
As expected, GP 33 -tet + CD8 T cells from Tbx21 E/E mice showed a uniform expression of Eomes and mCherry, but no expression of T-bet ( Fig 2D). A similar expression pattern of Eomes was also seen in activated GP 33 -tet -CD8 T cells from the three genotypes (S2C Fig). However, this expression of Eomes was not sufficient to rescue the development of SLECs, which were also absent in Tbx21 E/E mice ( Fig 2E). Intriguingly, Tbx21 E/E mice had a lower frequency of MPECs than Tbx21 -/mice, but still more than Tbx21 +/+ wild type mice (Fig 2E). Tbet and Eomes guide the differentiation and migration of T cells by regulating their responsiveness towards cytokines and chemokines, respectively. The expression of CD122 (IL-2Rβ chain), which is essential for IL-2 and IL-15 signaling, can be cooperatively regulated by the two TFs [16], however, during acute infection T-bet deficiency resulted in a significant reduction of CD122 expression, which could not be rescued by Eomes in Tbx21 E/E mice (Fig 2F,  left). The reduced expression of CD122 in T-bet deficient strains could have been due to the absence of SLECs in these strains. However, co-staining of CD127 and CD122 in GP 33 -tet + CTLs clearly showed that MPECs (CD127 + ) had the lowest expression of CD122 in T-bet deficient strains, whereas there was no difference in the CD122 expression of CD127 -CD8 T cells,

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections including SLECs and EECs (S2D Fig). Hence, T-bet not only regulates SLEC formation but is also relevant for appropriate CD122 expression in MPECs. In contrast, CD25 (IL-2Rα chain), which is essential for IL-2 signaling, was upregulated in Tbx21 -/mice, which was partially curtailed in Tbx21 E/E mice (Fig 2F, right). The chemokine receptor CXCR4 controls homing of T cells, especially to the bone marrow, where one of its ligands, CXCL12, is highly expressed. Published reports have indicated that CXCR4 expression in immune [30] and non-immune [31] cells can be positively regulated by Eomes. However, this is not an absolute requirement, as even Eomesnaïve T cells or thymocytes express relevant levels of CXCR4 [32]. CXCR4 was downregulated in GP 33 -tet + CD8 T cells from Tbx21 +/+ wild type mice, but remained elevated in CD8 T cells from Tbx21 -/and Tbx21 E/E mice (Fig 2G, left). Interestingly, there was no further increase in Tbx21 E/E mice compared to Tbx21 -/mice. Th1 cells express the chemokine receptor CXCR3, which guides them towards sites of inflammation [33]. High expression was also completely dependent on T-bet and was therefore lost in GP 33 -tet + CD8 T cells in Tbx21 -/and Tbx21 E/E mice (Fig 2G, right). Similar results were also observed in activated polyclonal CD44 + CD4 T cells from the three strains ( Fig 2H). In summary, we show here that despite the strong and uniform induction of Eomes in CD8 T cells from Tbx21 E/E mice, we did not observe a rescue of the Tbx21 -/phenotype in the context of LCMV infection with respect to clonal expansion, effector differentiation and responsiveness towards cytokines or chemokines. This would be in line with a model, in which there is a only limited functional overlap in the transcriptional activity of T-bet and Eomes [34].

T-bet deficient strains show signs of T cells exhaustion and fail to clear acute LCMV WE infection
Recently, it has been shown, that T-bet is able to repress several receptors associated with T cell exhaustion [35], whereas Eomes is inversely expressed in such cells. As a result, retroviral overexpression of T-bet in CD8 T cells results in the downregulation of PD-1, LAG-3 and TIM3. However, it is important to note that these markers are not only associated with exhaustion as they can be expressed transiently during an acute infection [36,37]. Yet, haploinsufficiency for T-bet during chronic infection leads to elevated expression of these markers [35]. In line with that, GP 33 -tet + CD8 T cells from Tbx21 -/mice had high expression of CD39, PD-1, and LAG-3, but not TIM3 at day 8 after acute infection with LCMV WE (Fig 3A). Tbx21 E/E CTLs also showed high expression of CD39, PD-1, LAG-3 and, in addition, a nearly 4-fold increase in TIM3 expression (Fig 3A), which was specific for Tbx21 E/E mice. Functionally, this phenotype in Tbx21 -/and Tbx21 E/E mice corresponded well with a failure of viral clearance in spleen, liver, kidney and lungs as well as significantly reduced cytotoxicity (Fig 3B and 3C). Although there was a stepwise increase in the percentage of IFN-γ producers from Tbx21 +/+ over Tbx21 -/to Tbx21 E/E mice ( Fig 3D and 3E), CTLs from T-bet deficient genotypes produced less IFN-γ on a per cell basis ( Fig 3E, middle bar diagram). Finally, loss of T-bet resulted in an increase of IL-2 production and a de-repression of IL-17 ( Fig 3F and 3G). Even Tbx21 E/E mice showed a significant induction of GP 33 -specific IL-17, implicating that Eomes is unable to repress RORγt induction in CTLs in the absence of T-bet.
We also analyzed CD8 T cell responses against other viral epitopes (NP396 (S3A Fig) and GP276 (S3B Fig), which showed a similar pattern for cytotoxicity and cytokine production. The effects of the Tbx21 E allele were different in LCMV-specific CD4 T cells, which respond to the GP61 epitope. Whereas Tbx21 -/mice showed a dramatic loss of IFN-γ and TNF production, this was partly rescued in Tbx21 E/E mice (S3C Fig). This contrast to CD8 T cells might be best explained by the lower endogenous expression of Eomes from the bona fide Eomes allele in CD4 T cells (Fig 1) as compared to CD8 T cells, which would evidently augment the effects

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections of the Tbx21 E allele in CD4 T cells. Together, the presented data are consistent with a model, in which early induction of T-bet, and not Eomes, during acute infection with LCMV WE governs clonal expansion and functional differentiation of CTLs that leads to viral clearance.

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections

T-bet is intrinsically required for early expansion and differentiation of CTLs after acute LCMV WE infection
Although the analysis of the constitutional knock-out (Tbx21 -/-) and knock-in system (Tbx21 E/ E ) had given us valuable insights on an organism-wide level, two caveats remained. First, it remained unclear whether the observed effects were CD8 T cell-intrinsic, since, beside activated CD8 T cells, various other hematopoietic cells express T-bet, which might affect CTL differentiation. Second, it is known from chronic infection models, that viral persistence has profound effects on CD8 T cell differentiation. This was important, as neither Tbx21 -/nor Tbx21 E/E mice were free of virus on day 8 after infection. Hence, to further elucidate the CD8 T cell-intrinsic redundancy of T-bet we generated three lines of congenically marked (Thy1.1 + ) TCR-transgenic mice: P14 Tbx21 +/+ [38], P14 Tbx21 -/-, and P14 Tbx21 E/E , whose CD8 T cells recognize the immuno-dominant epitope GP33 of LCMV. Next, we adoptively transferred small numbers of P14 CD8 T cells from these mice into Thy1. Hence, persistent viral infection can be excluded as a cause for functional and phenotypical differences between CTLs from P14 Tbx21 +/+ , P14 Tbx21 -/and P14 Tbx21 E/E mice. On day 8 after infection less than 20% of the transferred CTLs from P14 Tbx21 +/+ and Tbx21 -/expressed Eomes, whereas CTLs from P14 Tbx21 E/E mice were uniformly Eomes + (Fig 4A) and mCherry + (Fig 4B). Comparable to non-TCR-transgenic CTLs, P14 Tbx21 -/and P14 Tbx21 E/E CD8 T cells had a profound defect in clonal expansion and we therefore recovered, on average, 6-and 5-times less cells as compared to P14 Tbx21 +/+ transferred mice, respectively ( Fig 4C). Ki-67 expression, a marker for proliferating cells, showed a slightly lower, yet significant reduction in P14 Tbx21 -/-CTLs, whereas its expression was similar in P14 Tbx21 +/+ and Tbx21 E/E CTLs ( Fig 4D). However, these differences were apparently not relevant as CFSE labeled P14 CTLs from the different genotypes showed a comparable division pattern on day 4 after infection ( Fig 4E). Hence, we hypothesized that the reduced accrual of T-bet deficient P14 CTLs on day 8 might be due to accelerated apoptosis. Accordingly, the anti-apoptotic molecule Bcl-2 was strongly upregulated in P14 Tbx21 +/+ CTLs, which presumably contributed to the observed difference in recovered P14 CTLs on day 8 ( Fig 4F). We could not detect an increase in apoptotic CTLs in T-bet deficient CTLs by using Annexin-V/ DAPI staining (Fig 4G), which might be due to their rapid engulfment by macrophages [39]. Collectively, the presented data suggest that Eomes cannot compensate for a loss of T-bet in rapidly dividing CTLs.
Comparable to non-TCR-transgenic T cells, SLEC differentiation was not rescued by the Tbx21 E allele (Fig 4H). Similar to our previous observations, T-bet deficient CTLs showed a bias towards MPECs, whereas P14 Tbx21 E/E CTLs had significantly less MPECs than P14 Tbx21 -/-CTLs ( Fig 4H). Subset-specific analysis of the cytokine receptor chain CD122 revealed that in line with our previous results (S2D Fig) absence of T-bet resulted in reduced levels of CD122 in MPECs, which could not be rescued by Eomes (S4E Fig, upper row). Moreover, Tbet was essential for repression of CD25 in all subsets, a function which Eomes could partly compensate (S4E Fig, lower row). Recently, an additional approach [40] divided CD8 T cells during acute infections into three sub-populations based on the expression levels of the chemokine receptor CX3CR1 and CD27. These two markers combined generated a waterfall-like profile with CX3CR1and CX3CR1 intermediate (int) cells being CD27 + , and CX3CR1 high cells progressively loosing CD27. This distinction was important as CX3CR1 levels correlated with homing and effector properties. Intriguingly, these authors also showed that CX3CR1 int were partially dependent on T-bet for their development, whereas CX3CR1 high cells were absolutely

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections dependent on T-bet, suggesting further heterogeneity of CD27 + CX3CR1 int CTLs. We also detected an absence of CX3CR1 high P14 T cells and significant reduction of CX3CR1 int P14 T cells after transfer of P14 Tbx21 -/-T cells in the spleen (Fig 4I). Transferred P14 Tbx21 E/E cells showed a partial rescue of CX3CR1 int cells, however, they also failed to differentiate into CX3CR1 high cells (Fig 4I). Likewise, transferred P14 Tbx21 E/E cells showed significantly higher expression of CXCR3 as P14 Tbx21 -/cells, but never reached the high levels of P14 Tbx21 +/+ cells (Fig 4J). This partial rescue of CX3CR1 int cells and CXCR3 in P14 Tbx21 E/E CTLs indicates a discrete functional overlap between T-bet and Eomes. Overall, these defects in clonal expansion and differentiation correlated well with immunofluorescence analysis of spleen and lymph nodes that showed reduced infiltration of adoptively transferred CD8 T cells from Tbet deficient backgrounds (S5A Fig).

High-dimensional cytometry data analysis of CTLs reveals heterogeneity in T-box TF dependent clusters
Several models of CTL differentiation based on expression of surface markers like KLRG1 and CD127 have been established. However, recent studies using single-cell RNA sequencing highlight the remarkable heterogeneity of CTLs at day 8 post infection [41,42]. In order to obtain a

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections higher resolution of T-bet dependent differentiation of CTLs, we adapted two recently published workflows [43,44], which have been developed for in-depth analysis of multi-color flow cytometry data. Accordingly, we used t-distributed stochastic neighboring embedding (tSNE) to reduce the dimensionality of our 7-parameter dataset of surface markers from the experiment shown in Fig 4, which all have been implicated in the differentiation of CTLs (S5B Fig). Next, we performed hierarchical clustering on normalized median marker expression of our transferred P14 CTLs using FlowSOM [45] and ConsensusClusterPlus [46] (Fig 5A). To visualize our cluster analysis in a dimensionality-reduced fashion, obtained cluster analysis were depicted as tSNE plots (Fig 5B). At first sight, clusters were very similar distributed between Tbet deficient P14 CTLs but showed a dramatic difference in comparison to wild type P14 CTLs. As revealed by conventional gating in Fig 4H and 4I, SLECs-identified by high expression of KLRG1 and CX3CR1 and the absence of CD127 and CD27 respectively-did not develop in T-bet deficient P14 CTLs (Fig 4H and 4I). Consistently, clusters containing cells with high expression of KLRG1 and CX3CR1 as cluster 17, 19, 21, 22 and 24 were virtually absent from T-bet deficient P14 CTLs, thus validating our approach (Fig 5C).
Intriguingly, whereas conventional gating showed that the frequency of CX3CR1 int CD27 + P14 CTLs could only in part be rescued by the Tbx21 E allele (Fig 4I), in depth cluster analysis revealed that frequencies of clusters containing CX3CR1 int CD27 + P14 CTLs showed considerable heterogeneity. For instance, cluster 10 was around 3-times as frequent in P14 CTLs from

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections Tbx21 E/E mice compared to P14 CTLs from wild type mice. In contrast, cluster 23 was hardly detectable in T-bet deficient mice. Interestingly, cluster 8, which mainly contained cells with high expression of CD27 and CXCR3-a phenotype associated with efficient re-expansion of memory CD8 T cells upon secondary challenge [47,48]-was reduced in P14 CTLs from Tbx21 -/mice, yet, in terms of frequency there was no difference between P14 CTLs from Tbx21 +/+ and Tbx21 E/E mice. However, all of the more or similar abundant clusters in P14 CTLs from Tbx21 E/E mice were less than 6-times as frequent as in P14 wild type CTLs, suggesting a co-dependence on T-bet and Eomes for these clusters to develop or expand. Thus, timely expression of Eomes cannot replace T-bet during clonal expansion of CTLs, but Eomes and Tbet jointly drive the expansion of various CD27 + CTL subsets.

T-bet intrinsically coordinates regulated effector functions in CTLs
Next, we explored cytokine production and cytotoxic effector molecules in the different groups. Contrary to what we observed in the constitutive T-bet deficient strains, CD8 T cells from P14 Tbx21 -/and P14 Tbx21 E/E mice showed no reduction in the cytokine level of IFNγ + TNF + double producers ( Fig 6A). Hence, this was presumably due to extrinsic effects, e.g. insufficient CD4 T cell help, which also requires T-bet for Th1 differentiation. In addition, we discovered three times more cells that produced IL-2 in T-bet deficient P14 CD8 T cells. As expected P14 Tbx21 -/-CD8 T cells showed strong de novo production of IL-17A in around 40% of the cells, which was significantly, albeit slightly, reduced in P14 Tbx21 E/E CD8 T cells (Fig 6A). It also became apparent that T-bet deficient strains had a highly deregulated cytokine

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections profile with more than half of the cells producing more than 3 cytokines and up to 30% of cells that simultaneously produced IFN-γ, TNF, IL-2 and IL-17A (Fig 6B). This wide-spread deregulation is well in line with the initial observation that T-bet is a repressor of IL-2 [49] and the later finding that it represses the development of IL-17 producing Th17 T cells [50]. Hence, these repressive actions cannot be executed by Eomes.
Activated CD8 T cells lyse infected target cells through the cooperative actions of the poreforming protein perforin and a family of proteases termed granzymes. T-bet and Eomes together with Runx3 regulate expression of these proteins [51]. Expression of Granzyme A (GzmA) was strongly dependent on T-bet. P14 Tbx21 -/-CD8 T cells showed the lowest percentage of GzmA expressing cells and those cells that produced GzmA did so at a lower level per cell (Fig 6C and 6D). Similar observations were made in P14 Tbx21 E/E CD8 T cells (Fig 6C  and 6D). In summary, this set of data illustrates that timely expression of T-bet after an acute infection governs important aspects of functional CTL differentiation.

Early expression of Eomes in lieu of T-bet promotes features of exhaustion in CTLs
Besides their roles during acute infection T-bet and Eomes are also important during chronic viral infection. Functionally exhausted Eomes high CD8 T cells become the dominant population as inflammation and viral replication persist during chronic infection [7]. However, it is still a matter of debate, whether high Eomes expression, ongoing viral replication or a combination of both control this dysfunctional state. To discriminate between these options, we took advantage of the fact that in our acute infection model with transferred P14 CD8 T cells mice were virusfree on day 8 but showed high expression of Eomes in P14 Tbx21 E/E CD8 T cells in contrast to P14 Tbx21 +/+ CD8 T cells, which, reciprocally, showed high expression of T-bet. Hence, we further analyzed the phenotype of our transferred P14 CD8 T cells, focusing on inhibitory receptors and transcription factors, which are associated with exhausted CD8 T cells.
As outlined above, CD39, PD-1, LAG-3, TIM3 and CTLA-4 are inhibitory receptors, that can be transiently expressed during acute infections, but whose prolonged and simultaneous expression on virus-specific CD8 T cells during chronic infection identifies cells that are functionally exhausted [23,36,52]. Intriguingly, more than half of P14 CTLs from Tbx21 +/+ mice expressed either CD39 or TIM3 or both and this expression pattern was hardly detectable in T-bet deficient P14 CTLs (Fig 7A and 7B). In contrast, we saw a stepwise increase in the percentage of P14 CTLs that were simultaneously positive for PD-1, CD39, TIM3 and LAG-3 from Tbx21 +/+ over Tbx21 -/to Tbx21 E/E mice (Fig 7A and 7B). Moreover, the immune checkpoint CTLA-4, that inhibits co-stimulation via competition with CD28, was strongly induced in P14 CTLs from T-bet deficient mice (Fig 7C). The fact, that Tbx21 E/E P14 CTLs had a higher percentage of PD-1 + CD39 + TIM3 + LAG-3 + cells than P14 Tbx21 -/-CTLs, clearly indicates that this phenotype cannot solely be attributed to the absence of T-bet, but rather by the additional direct or indirect actions of the Tbx21 E allele. Thus, we gauged the effects of our Tbx21 E allele on other TFs that contribute to the phenotype of exhausted T cells. Previously, it has been shown that Blimp-1 regulates expression of inhibitory receptors [53,56]. In line with our results, we observed a significant increase of Blimp-1 expression [53] in P14 Tbx21 E/E CTLs ( Fig 7D). Moreover, we detected similarly elevated levels of TOX [54] (Fig 7E) and IRF4 [55] (Fig 7F), two recently identified regulators of exhausted T cells, in P14 Tbx21 -/and P14 Tbx21 E/E CTLs as compared to P14 Tbx21 +/+ CTLs. This would argue that TOX and IRF4 are primarily induced upon loss of T-bet and not via expression of Eomes. Collectively, overexpression of Eomes at an early stage of acute viral infection promotes some features of an exhausted phenotype that is independent of viral persistence.

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections

Discussion
Here we have tested whether the synchronized induction of Eomes in T-bet deficient mice can compensate for the loss of T-bet during an acute viral infection. Combining two different model systems, we show that T-bet is intrinsically required for canonical effector cell differentiation and clonal expansion of CTLs during acute infection with LCMV WE. Moreover, we demonstrate that T-bet non-redundantly limits IL-2 production and inhibits aberrant expression of IL-17A in LCMV WE-specific CTLs. In summary, we can conclude that T-bet, has unique and important functions during acute viral infection, which cannot be executed by its paralog Eomes.
The biological complexity of multicellular organism has been facilitated by genome duplication. This allowed for a process of diversification and selection of new genes and functions that provided the organism with many advantages under evolutionary pressure [57]. The T-box TF family is a prime example of such processes. Hence, it is not surprising that the two paralogs T-bet and Eomes, despite their high degree of homology, have only very limited functional overlap in vivo, as we have shown here. Nevertheless, it remains striking that T-bet and Eomes

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections share the highest degree of homology within their DNA-binding sequence; i.e. the T-box (ca. 74%). Here, it is important to point out, that T-box TFs utilize a variety of mechanisms to establish a gene expression program. Among these are context-dependent dimerization, binding of co-factors and sequestration of alternative transcription factors as well as epigenetic modification through recruitment of histone modifying enzymes [58]. Moreover, T-box TFs can compete for promoter binding and, once bound, can exhibit either activation or repression activities [59]. Thus, it is difficult to predict the actions of T-bet and Eomes solely based on their respective consensus DNA-binding motif, which is conspicuously similar. Relating to this, it is of note that T-bet and Eomes show only 29% similarity in their N-and C-terminus [3]. These regions will most likely contribute to the observed functional differences of T-bet and Eomes and studies are under way to test this prediction. In line with this, a recent publication that compared CHIPseq data from T-bet and Eomes in CD8 CTLs demonstrated that Eomes bound only 1292 out of 7534 (17.15%) T-bet target genes [34]. This implicates that more than 80% of the transcriptional signature of T-bet is unique.
The population size of T cells during infection is regulated by a dynamic balance between proliferation and survival, which sequentially results in massive expansion, contraction and memory maintenance of small numbers of antigen-specific T cells [60]. Here, we have extended previous reports by showing that T-bet was essential for early expansion of CTLs and that this defect in Tbx21 -/mice could not be overcome by the timely expression of Eomes in Tbx21 E/E mice. We noted a significant reduction in the expression of the anti-apoptotic molecule Bcl-2 in T-bet deficient strains, which is therefore most likely regulated, either directly or indirectly, by T-bet. In contrast, we detected only mild differences in expression of the proliferation marker Ki-67, which is further supported by similar CFSE division profiles in the three genotypes. Hence, it is most likely that T-bet is important for survival of proliferating CTLs during acute infection. This is further corroborated by data from Prier et al. [61], which showed that Tbx21 -/-CD8 T cells have no early expansion defect following Influenza A virus infection and, in line with our data, this study also did not detect a compensatory increase of Eomes in Tbx21 -/-CD8 T cells after adoptive transfer. Actions of T-bet on survival could be regulated via cytokine responsiveness as T-bet deficient strains showed reduced expression of CD122 in both models. CD122 represents the IL-2Rβ chain, which is important for IL-2 and IL-15 signaling, which in turn can regulate pro-and anti-apoptotic molecules. Although previous experiments have indicated that both, Eomes and T-bet can transactivate CD122 [16] when genetically introduced into T cells, CTLs from P14 Tbx21 E/E mice did not rescue the reduced expression of CD122 in CTLs from P14 Tbx21 -/mice. This difference can be explained by the fact that in contrast to our precisely controlled knock-in model earlier studies used retroviral overexpression systems [16], which often result in supraphysiological doses of TFs. Hence, maximum expression of CD122 during acute infection requires the cooperative actions of T-bet and Eomes. Intriguingly, T-bet deficient strains showed clear evidence of heightened IL-2 signaling. First, CTLs from T-bet deficient mice expressed significantly more CD25, which represent the high-affinity IL-2Rα chain [62]. Second, CD25 can be up-regulated by IL-2 and correspondingly we detected abundant IL-2 production in T-bet deficient CTLs. Prolonged IL-2 signaling, especially in the context of inflammation, drives terminal effector differentiation, whereas curtailed IL-2 signaling during infection favors CD8 memory formation [63]. However, this heightened IL-2 signaling in T-bet deficient strains was not sufficient to rescue SLEC formation in CTLs from P14 Tbx21 E/E mice arguing that IL-2 must act jointly with T-bet to drive SLEC formation. IL-2, in part, acts via the transcriptional regulators Id2 [64] and Blimp-1 [63,65]. Consistent with a previous study, Blimp-1 expression levels did not differ between WT and Tbx21 -/mice [65], whereas Tbx21 E/E mice did upregulate Blimp-1 significantly. This was, however, not sufficient to compensate for the loss of T-bet. In addition to its role during differentiation, early studies have also shown that activation-induced cell death is strongly promoted by IL-2 [66]. Hence, it will be of further interest to determine, whether the heightened IL-2 signaling in T-bet deficient strains contributes to their failure to accumulate at day 8 of infection.
Another important aspect of our findings is the different phenotype of LCMV-specific CTLs in our two model systems. Whereas GP 33 -tet + T cells in Tbx21 -/und Tbx21 E/E failed to produce the high amounts of IFN-γ and TNF per cell as seen in Tbx21 +/+ mice, this difference was not observed in the adoptive transfer system, where P14 Tbx21 +/+ , P14 Tbx21 -/and P14 Tbx21 E/E CTLs all produced high amounts of IFN-γ and TNF. These results are most likely mitigated by extrinsic effects of T-bet deficiency on CTL differentiation in Tbx21 -/und Tbx21 E/E mice. In fact, T-bet is not only expressed by CD8 T cells but also by other lymphocytes and even myeloid cells, that all shape the immune response against LCMV. T-bet was initially discovered as the lineage-defining TF of CD4 Th1 cells [2], as it was able to instruct abundant production of IFN-γ. Although CD4 T cell help is dispensable for short-term acute LCMV infection, LCMV-specific CD4 T cells will get activated during acute LCMV WE infection and provide T cell help. Interestingly, timely expression of Eomes in LCMV-specific CD4 T cells from Tbx21 E/E mice rescued the frequency of non-TCR-transgenic IFN-γ producers as compared to Tbx21 -/mice (S3C Fig), however they produced less IFN-γ per cell as LCMVspecific CD4 T cells from Tbx21 +/+ mice. It is therefore unlikely, that these small differences in the CD4 compartment can explain the differences in CTL responses in our two models. Hence, it will be of future interest to investigate and compare cell types, that usually express Tbet (e.g. B cells, NKT, group 1 innate lymphoid cells, dendritic cells) in Tbx21 E/E mice to understand their contribution to the significant defect in IFN-γ production in CTLs from Tbet deficient strains. Another interesting feature of the adoptive transfer system was the revelation that certain aspects of T-bet could be partially rescued by Eomes; e.g. expression of CD25 (Fig 4E), CXCR3 (Fig 4I) and frequency of CD27 + CX3CR1 int (Fig 4J) in P14 Tbx21 E/E CTLs. However, this never reached the levels seen in P14 Tbx21 +/+ CTLs and in general, P14 Tbx21 E/ E CTLs were more akin to P14 Tbx21 -/-CTLs. We would therefore be very cautious to attribute a biological significance to the partial overlap P14 Tbx21 E/E CTLs share with P14 Tbx21 +/+ CTLs.
Finally, we used our novel model to uncouple the effects of elevated Eomes expression and ongoing viral replication on the phenotype of activated CD8 T cells. Previous reports have shown that Eomes and T-bet are differentially linked to the exhausted phenotype of CD8 T cells in chronic viral infections [7] or tumors [34]. In such disease states T-bet and Eomes are reciprocally expressed and Eomes expression increases with time and ongoing viral replication or tumor burden, respectively. Moreover, Eomes expression correlates with higher expression of inhibitory receptors, e.g. PD-1, LAG-3, CD160, 2B4, TIM3, TIGIT, CTLA-4, and impaired cytokine production, which in concert define the state of exhaustion [24]. Additionally, the HMG-box transcription factor TOX was recently identified as an important regulator of exhausted T cells [54,67], and its conditional deletion resulted in reduced expression of inhibitory receptors and Eomes as well as improved effector functions. We extend this set of data by showing that, even in the absence of viral replication, uniform expression of Eomes in P14 CTLs from Tbx21 E/E mice fosters a partially exhaustion-like phenotype, which was characterized by the highest frequency of CD39 + PD-1 + LAG-3 + TIM3 + P14 CTLs. Moreover, T-bet also represses the inhibitory T cell checkpoint CTLA-4 [68], which is highly expressed in T-bet deficient P14 CTLs (Fig 7C), irrespective of Eomes levels. Of particular interest was the expression pattern of PD-1 and LAG-3, which were simultaneously expressed in a significant proportion of T-bet deficient CTLs in both models (Figs 3A, 7A and 7B). The dual blockade of LAG-3 and PD-1 shows synergistic effects in models of chronic LCMV infection [36] and tumors [69] PLOS PATHOGENS Eomes cannot replace its paralog T-bet during acute infections and has resulted in several ongoing clinical trials [70]. Data from the transfer system in Fig 7A  and 7B clearly illustrates that T-bet represses PD-1 but that Eomes can induce LAG-3 as P14 Tbx21 E/E CTLs show the highest expression levels for LAG-3. Similar results were also observed in our non-TCR-transgenic system (Fig 3A). In summary, we conclude that T-bet and Eomes presumably regulate different aspects of exhaustion in CTLs, which would be in line with recent findings [23,71]. It is important to note that Eomes is not only associated with exhaustion but also an important transcription factor in memory T cells. Eomes expression gradually increases after an acute LCMV infection and as a result, Eomes deficient CTLs have a competitive disadvantage and do not contribute sufficiently to the memory pool. In the future, it will therefore be of interest to determine, whether the enforced expression of Eomes in Tbx21 E/E CTLs accelerates the establishment of a memory program at the cost of an acute effector program. Hence, it is conceivable that Tbx21 E/E CD8 T cells harbor an increased memory population (S1C Fig) after an acute LCMV infection, which potentially still fails to expand upon re-challenge due to the absence of T-bet.
Despite its effect on inhibitory receptors and proliferation T-bet deficiency did not lead to the full spectrum of exhaustion, as transferred P14 Tbx21 -/and Tbx21 E/E T cells showed no impairment in IFN-γ or TNF production. Most importantly, IL-2 production, which is lost early during exhaustion, was much higher in T-bet deficient CTLs, which is well in line with T-bet's ability to repress IL-2. T-bet was also absolutely required to repress IL-17 production in CTLs, which was strikingly deregulated in T-bet deficient strains. Hence, heightened expression of Eomes might explain some features of exhaustion but must act in concert with other factors during chronic infections.
In conclusion, the presented findings underline the non-redundancy of T-bet in acute viral infections and emphasize the necessity for further research to mechanistically dissect the transcriptional activities of T-bet and Eomes. This will help to identify TF-specific targets that could be harnessed therapeutically in the fight against chronic infections and cancer as well as amelioration of immune-mediated diseases.

Ethics statement
All animal experiments were approved and are in accordance with the local animal care committees of the University of Freiburg and the Regierungspräsidium Freiburg (approval number G-19/64).

Mouse strains
Tbx21 -/-, Eomes Gfp/+ , T-cre Eomes flox/flox (Cre transgene is driven by the T (Brachyury) promotor that induces a pan-mesodermal loss of Eomes, including all hematopoietic cells) and P14 TCR tg mice on a C57BL/6 background were bred locally. C57BL/6 mice were purchased from Janvier Laboratories (LeGenest St-Isle, France). Mice of both sex between 8 to 12 weeks of age were used.

Generation of the Tbx21 Eomes-P2A-mCherry/+ (Tbx21 E/+ ) mouse
Targeting of the Tbx21 locus for the generation of the Tbx21 Eomes-P2A-mCherry/+ allele was performed by TALEN-enhanced homologous recombination [27] in ES cells. Two TALENs were directed against sequences close to exon 1 within the first intron of the Tbx21 gene. A targeting vector was generated that contains 500bp of 5' and 3' homology regions surrounding exon 1 of Tbx21, a Eomes-P2A-mCherry cassette introduced at the translational start site (thereby

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections deleting coding sequences of exon 1), and a LoxP-flanked neomycin resistance cassette. Linearized targeting vector together with TALEN-coding vector were electroporated into CCE ES cells using Amaxa 4D Nucleofector protocol for Mouse Embryonic Stem Cells and neomycin resistant ES cell colonies were screened by genomic PCR for correct integration into the gene locus. Generation of chimeras was performed by laser-assisted injections at morula stage (Transgenic Core Facility, Max-Planck-Institute Dresden) and F1 progeny and subsequent generations were genotyped by PCR. The strain was maintained by backcrossing to C57BL/6 mice. For experiments, 8 to 12 weeks old mice, which were backcrossed 6-10 generations to C57BL/6, were used. For steady state analysis, littermates derived from mating Tbx21 E/+ with Tbx21 E/+ mice were used.

Flow cytometry analysis and cell sorting
Single cell suspensions from spleen, lymph nodes and thymus were obtained by meshing spleens, thymocytes and axillary and inguinal lymph nodes through 70 μm cell strainers (BD Biosciences, San Jose, CA, USA). Next, single cell suspensions were depleted of erythrocytes by incubation with red cell lysis buffer. Afterwards, cells were incubated on ice for 30 min with an anti-CD16/CD32 antibody for FC receptor blocking in PBS (Ca2 + and Mg2 + -free) supplemented with 2 mM EDTA and 2% FCS. After a washing step, single cells were stained with murine fluorescence-labelled antibodies for 30 to 60 min on ice in the dark.

Virus and infections
LCMV WE was grown on L929 cells. Mice were inoculated with 200 plaque forming units (pfu) via intravenous application.

Chromium-release assay
Cytolytic activity of CD8 + T cells was determined using a standard 51 Chromium release assay as previously described [72,73]. In brief, GP33, GP276 and NP396 peptide-loaded EL-4 cells were used as target cells and incubated with different concentrations of splenocytes as effector cells for 5 hours at 37˚C. Duplicate wells were assayed for each effector-target ratio and percentages of specific lysis were calculated.

Viral titers
Viral titers were determined using a standard focus-forming assay as previously described [72,73].

Adoptive transfers of CFSE labeled P14 cells
After enrichment of P14 CD8 cells of the different genotypes, cells were washed with PBS and 1x10 7 cells/ml were labeled with 5μM CFSE (Biolegend) in PBS for 10 min at 37˚C. Cells were washed twice in complete RPMI medium and 1x10 6 P14 CD8 T cells were immediately transferred into recipient wild type mice (Thy1.2 + ) one day before infection with 200 pfu LCMV WE. CFSE dilution of P14 CD8 T cells was analyzed by flow cytometry four days after infection.

Immunofluorescence
Spleens and inguinal lymph nodes were removed and cleaned by washing in ice-cold PBS. A piece of the tissue was fixed in 4% PFA on ice for 2h. Afterwards the tissue was washed in icecold PBS before rehydrating the tissue in 30% (w/v) sucrose phosphate buffer overnight. Tissues samples were embedded in O.C.T. Compound (Tissue-Tek) and snap-frozen in liquid

PLOS PATHOGENS
Eomes cannot replace its paralog T-bet during acute infections nitrogen. Cryostat sections of 6 μm thickness were collected on frosted glass slides. Slides were rehydrated in PBS and blocked with 5% BSA (Sigma-Aldrich) for 60 min at room temperature. Samples were incubated with fluorescein isothiocyanate conjugated monoclonal anti-mouse CD90.1/Thy1.1 antibody (clone OX-7, Biolegend, San Diego, CA) and allophycocyanin conjugated monoclonal anti-mouse CD45R/B220 antibody (clone RA3-6B2, Biolegend, San Diego, CA), each diluted 1:500 in 5% BSA for 60 min at room temperature. Slides were washed with PBS and mounted with mounting medium (PermaFlour, Thermo Scientific, Waltham, MA) and each section was visually inspected. Images of representative sections were captured under an Axioplan 2 fluorescence microscope at 20x magnification using an AxioCam camera and Axiovision LE software Rel. 4.9 (all Axio devices were from Zeiss, Oberkochen, Germany).

High-dimensional flow cytometry data analysis
Analysis was adapted from two published workflows [43,44]. In brief, to perform in-depth analysis of polychromatic flow cytometry data, flow cytometry standard (FCS) 3.0 files were imported into FlowJo Version 9. Next, P14 cells (Thy1.1 + CD8 + ) were downsampled to 20,000 cells per sample. After that, data was imported into Cytobank (Santa Clara, CA, US) to determine an appropriate cofactor for arcsinh-transformation of data. Next, data was loaded into R (V3.4.1) and arcsinh-transformed and further analyzed in R as outlined: Data was transformed to values from 0 to 1 and t-distributed stochastic neighbor embedding (t-SNE) was run (perplexitiy = 100, interations = 2000) on surface markers (CX3CR1, KLRG1, CXCR3, CD127, CD44, CD27, CD62L) of P14 cells (20,000 P14 cells per sample, n = 5 samples per experimental group, total of 300,000 P14 cells). Next, P14 cells were hierarchically clustered using FlowSOM [45] and ConsensusClusterPlus [46] based on the mean fluorescence expression (MFI) of aforementioned surface markers transformed to values from 0 to 1. Thus, self-organizing maps (SOM) were generated and after that SOM codes were metaclustered with the Consensu-sClusterPlus package (maxK = 25) and MFI and frequency of obtained clusters were depicted as heatmap using the pheatmap package in R.

Statistical analysis
If not otherwise indicated, p value of data sets was determined by unpaired two-tailed Student's t-test with 95% confidence interval. If indicated, data was Welch-corrected prior unpaired two-tailed Student's t-test. All statistical tests were performed with Graph Pad Prism V4 software (Graph Pad Software, La Jolla, CA). ( � p < 0.05; �� p < 0.01 and ��� p < 0.001; ns, not significant).