Rapid in vitro generation of bona fide exhausted CD8+ T cells is accompanied by Tcf7 promotor methylation

Exhaustion is a dysfunctional state of cytotoxic CD8+ T cells (CTL) observed in chronic infection and cancer. Current in vivo models of CTL exhaustion using chronic viral infections or cancer yield very few exhausted CTL, limiting the analysis that can be done on these cells. Establishing an in vitro system that rapidly induces CTL exhaustion would therefore greatly facilitate the study of this phenotype, identify the truly exhaustion-associated changes and allow the testing of novel approaches to reverse or prevent exhaustion. Here we show that repeat stimulation of purified TCR transgenic OT-I CTL with their specific peptide induces all the functional (reduced cytokine production and polyfunctionality, decreased in vivo expansion capacity) and phenotypic (increased inhibitory receptors expression and transcription factor changes) characteristics of exhaustion. Importantly, in vitro exhausted cells shared the transcriptomic characteristics of the gold standard of exhaustion, CTL from LCMV cl13 infections. Gene expression of both in vitro and in vivo exhausted CTL was distinct from T cell anergy. Using this system, we show that Tcf7 promoter DNA methylation contributes to TCF1 downregulation in exhausted CTL. Thus this novel in vitro system can be used to identify genes and signaling pathways involved in exhaustion and will facilitate the screening of reagents that prevent/reverse CTL exhaustion.

Introduction Cytotoxic CD8+ T cells (CTL) play a critical role in eliminating viral infection and controlling cancer development. During chronic viral infection and cancer, CTL acquire a state of dysfunction that is often referred as CTL exhaustion which was originally described in chronic Lymphocytic Choriomeningitis Virus (LCMV) infection of mice [1]. CTL exhaustion has been documented in humans in chronic viral infections such as human immunodeficiency virus (HIV), hepatitis B virus (HBV) and hepatitis C virus (HCV) infections and in most human cancers [2][3][4][5][6][7][8][9] and is thought to be a central mechanism behind the failure of CTL to eliminate chronically infected and cancerous cells. Preventing and/or reverting exhaustion therefore constitutes a promising approach to restore the function of these CD8+ T cells. This requires however an in depth understanding of the mechanisms that lead to exhaustion and the stimuli that affect exhausted CD8+ T cells.
To study T cell exhaustion, in vivo mouse models are commonly used where T cell exhaustion is either induced through chronic viral infection or cancer. LCMV infection is a well-characterized mouse model for CD8+ T cell exhaustion [10,33]. The exhausted phenotype is also detected among tumor infiltrating lymphocytes (TIL) in tumor mouse models [34,35]. The persistence of the antigen has been shown to be the crucial element [36] and chronic antigen stimulation alone is sufficient to induce CTL exhaustion in vivo [37,38]. Previous in vitro exhaustion protocols have failed to validate T cell exhaustion, assuming inhibitory receptor expression and reduction of IL-2 production as surrogates for dysfunction [39]. Some of these changes, however, such as expression of PD-1 and CD160 also accompanies T cell activation while loss of IL-2 production is a feature of effector differentiation. Thus a fully validated in vitro exhaustion system has yet to be established.
The importance of CTL exhaustion in chronic infections and cancer demands rapid in vitro methods to screen new approaches that can prevent or revert CTL exhaustion. Inducing CTL exhaustion in vivo is time-consuming, requiring more than 30 days and yields limited numbers of exhausted cells as experimental material. Most importantly the in vivo milieu in these models is characterized by inflammation, high viral loads, suppressive cytokines or cells; all of which can obscure the phenotype of exhausted CTL and therefore make it hard to dissect true exhaustion-associated changes. Our newly developed in vitro method circumvents all of the above mentioned issues and allows gene manipulation and screening of small molecules or antibodies that could restore the function of exhausted CTL and/or prevent the induction of exhaustion. Using this system, we can show that the Tcf7 promotor is methylated in exhausted CTL and using DNA transmethylase inhibitors one can prevent the downregulation of TCF1. Therefore, this 5-day in vitro exhaustion system enables the rapid testing of new approaches to prevent and/or revert exhaustion in T cells and therefore may facilitate the development of new therapies for chronic infections and cancer.

Repeat stimulation with cognate peptide to induce CTL exhaustion
Because of the critical role sustained antigen stimulation plays in driving CD8+ T cells to exhaustion [36], we utilized repeat peptide stimulations of OVA (257-264) −specific TCR transgenic OT-I cells to induce CTL exhaustion in vitro. To induce exhaustion, CD8+ T cells purified from OT-I mice were stimulated daily for five days with 10ng/ml OVA (257)(258)(259)(260)(261)(262)(263)(264) peptide in the presence of IL-15 and IL-7 (Repeat peptide stimulation, Fig 1A). As controls, cells were either left unstimulated or stimulated only once with OVA (257-264) peptide for 48 hours and then washed and cultured without peptide for an additional 3 days. All cells were cultured with IL-7 and IL-15. On day 5, cells were either harvested for analysis or in some instances cells were cultured an additional 3 days in the presence of IL-7 and IL-15 without any additional peptide stimulation ( Fig 1A).
An important feature of this in vitro exhaustion system is that it rapidly yields large numbers of exhausted cells within 5 days (Fig 1B). Single peptide stimulations yielded comparable numbers while unstimulated culture numbers remained largely unchanged. The expansion in the numbers of exhausted T cells represents a~9-fold increase from the T cell numbers seeded. Thus this in vitro exhaustion system can yield large numbers of exhausted CTL that can be used to further study and manipulate the pathways that drive exhaustion.

Repeat peptide stimulation reduces cytokine production and leads to loss of polyfunctionality
Since hierarchical loss in the capacity to produce cytokines is one of the most critical characteristics of exhausted CD8+ T cells [10], we first examined the capacity for cytokine release upon peptide re-stimulation. After five days of culture, the cells were harvested and stimulated with OVA (257-264) peptide in the presence of Golgi plug for 6 hours. More than 40 percent of the unstimulated cells could produce IFN-γ upon re-stimulation (46±4%, mean ± standard error, Fig 1A and 1B). Single peptide stimulated cells were also able to produce IFN-γ (58±4 %). In contrast, repeat peptide stimulated cells had impaired IFN-γ production with only 27± 3% of cells able to produce IFN-γ after re-stimulation (Fig 2A and 2B). In addition, very few repeat peptide stimulated cells (4.8±0.4%) could produce TNF-α compared to single peptide stimulated cells (44±4%) and unstimulated cells (71±4%) (Fig 2A and 2B). IL-2-producing cells could barely be detected in the repeat peptide stimulated cells after re-stimulation, while IL-2 was readily produced by single peptide stimulated cells (37±3%) and unstimulated cells (60±4%) (Fig 2A and 2B). Ultimately, these results demonstrate that repeat peptide stimulation of CD8+ T cells in vitro leads to impaired cytokine production in a manner that has been previously reported in exhausted CD8+ T cells, first IL-2 production is lost followed by TNF-α and finally IFN-γ.
Exhausted CD8+ T cells, unlike memory cells, cannot produce multiple cytokines at the same time, and we therefore assessed the polyfunctionality of our in vitro generated cells. For this, the percentage of single, double and triple cytokine producing cells were determined and visualized using the SPICE software. In unstimulated cultured cells, 33% produce all three cytokines simultaneously upon peptide stimulation while significant percentages of these cells were double producer ( Fig 2C). Only 15% of the cells did not produce any of these three cytokines. Similarly, in the single peptide stimulated cells, 18% were triple positive for all three cytokines, large numbers were double cytokine producers, while 26% were incapable of releasing any of these cytokines ( Fig 2C). In contrast, 70% of the repeat peptide stimulated cells could not produce any cytokines after re-stimulation. IFN-γ single producers were 25% of the cells, while very few cells produced other cytokines (Fig 2C). To demonstrate the persistence of the cytokine profile of repeat peptide stimulated cells, we rested cells without peptide for another 3 days before examining their cytokine production. We found that the reduced cytokine production of repeat peptide stimulated cells was maintained after resting (S1A Fig).
Degranulation is an important step for the CD8+ T cell cytotoxicity. We therefore analyzed the degranulation marker CD107a and Granzyme B (GzmB) expression in our cells. When cells were analyzed without peptide re-stimulation, cells either unstimulated or single peptide stimulated were negative for CD107a and GzmB (Fig 3A and 3B), while repeat peptide stimulated cells showed a significantly higher level of both markers (Fig 3A and 3B). When cells were re-stimulated with OVA (257-264) peptide, cells cultivated in the presence of no peptide and single peptide stimulated cells upregulated CD107a (Fig 3A). In contrast, repeat peptide stimulated cells failed to increase CD107a expression after re-stimulation, indicating that these cells are not able to degranulate further upon re-stimulation. Granzyme B was increased in repeat peptide stimulated cells in agreement with the terminal differentiation of exhausted cells ( Fig 3B). No increase of GzmB was detected after 6 hours of re-stimulation for any of the conditions analyzed ( Fig 3B). Repeat peptide stimulated cells exhibited reduced cytotoxic capacity against OVA (257-264) peptide-loaded tumor cells (Fig 3C). These findings clearly show that after repeat stimulation in vitro, CTL lose their ability to make cytokines, have reduced polyfunctionality, cannot degranulate further, have increased GzmB and reduced cytotoxicity. All of these features are characteristics of the dysfunctional state of in vivo exhausted CTL.

Multiple inhibitory receptors were upregulated following repeat peptide stimulation
An increase in inhibitory receptor expression corresponds to a more exhaustive state [17]. After harvesting on day five, the cells were stained immediately for the surface expression of the inhibitory receptors PD-1 (CD279), CD244, CD160, Lag3 (CD223), Tim-3 (CD366) and TIGIT. As expected, inhibitory receptors were barely expressed on the cells without peptide stimulation and single peptide stimulated cells (Fig 4A and 4B), except for PD-1, where 20±4% of the single
We next examined the simultaneous co-expression of multiple inhibitory receptors (PD-1, Lag3, CD160 and CD244) using SPICE ( Fig 4C). Repeat peptide stimulated cells were 39% double positive for the inhibitory receptors and another 39% of the cells co-expressed three of these inhibitory receptors simultaneously. Furthermore, 11% of the cells expressed all 4 inhibitory receptors ( Fig 4C). In contrast, very few of the no peptide and single peptide stimulated cells expressed two or more inhibitory receptors ( Fig 4C). The differences in inhibitory receptor expression remained even after resting repeat peptide stimulated cells for 3 days (S1B Fig).
To exclude that differences in inhibitory receptor expression were due to a different activation status, all the cells were re-stimulated for 6 hours after harvesting them on day 5. Although reactivation induces a slight upregulation of some inhibitory receptors on the no peptide and single peptide cultures, they still remain much lower than the repeat peptide stimulated cultures (S2A Fig). The exception was CD160 which was upregulated to similar levels in all cells. Overall, these findings confirm that multiple inhibitory receptors are expressed on the repeat peptide stimulated cells.

Expression of transcription factors is altered in repeat peptide stimulated cells
Previous studies have reported that exhausted CTL express and utilize transcription factors (TF) differently compared to effector and memory cell [11,12,40]. To characterize the expression pattern of TF in the in vitro exhausted cells, four of the core TFs (TCF1, TOX, T-bet, and

PLOS PATHOGENS
Generate exhausted CD8+ T cells in vitro EOMES) were measured. TCF1 has been reported to play a critical role in identifying subsets of exhausted T cells [22,41]. Early exhausted or the progenitor exhausted cells maintained TCF1 expression, while terminally exhausted T cells downregulate its expression. In comparison to single peptide stimulation, repeat peptide stimulation induced down-regulation of TCF1 expression and upregulation of GzmB (Fig 5A and 5B). TOX expression was reported to be increased on exhausted cells [2,[23][24][25]40], and indeed we also found TOX to be significantly upregulated in the repeat peptide stimulated cells in comparison to unstimulated and single peptide stimulated cells (Fig 5A and 5B). These differences in TOX and TCF1 expression persisted even after resting 5 day cultures for another 3 days (S1C and S1D Fig In vitro peptide stimulation downregulated EOMES, with no detectable difference of EOMES expression comparing repeat with the single peptide stimulated cells (Fig 5A and 5C). The increased TOX and T-bet accompanied by decreased TCF-1 expression in repeat peptide stimulated CTL is in agreement with the TF profile of in vivo exhausted CTL [25,41,42].

Repeat peptide stimulated cells have decreased in vivo expansion capacity
To evaluate the in vivo expansion capacity of repeat peptide stimulated cells, we sorted live CD8+ T cells from in vitro cultures (no peptide, single peptide and repeat peptide stimulated cells) and adoptively transferred 10 4 live CD45.1+ CD8+ T cells into wild type CD45.2+ mice that were infected 3h later with WSN-OVA influenza virus. Freshly isolated naïve CD8+ T cells from an OT-I spleen were also transferred as controls. At day 10 post infection, mice were harvested and the number of donor CD45.1+ cells in the lung were measured. We found a significant reduction in the frequency (23% versus 68% and 66%, for repeat peptide, single peptide and no peptide, respectively, Fig 6A) as well as absolute cell numbers (3.2x10 6 , 11.8x10 6 and 12.6x10 6 , for repeat peptide, single peptide and no peptide respectively, Fig 6B) for the repeat peptide stimulated cells compared to all controls. Freshly isolated naïve cells did not differ from cells of single and no peptide cultures. The numbers of repeat peptide stimulated donor cells were also reduced in mediastinal lymph nodes (MLN) and spleens (S3 Fig). This indicated that repeat peptide stimulated cells are less capable of expanding after being exposed to their specific antigen. Meanwhile, the mice that received the repeat peptide stimulated cells possessed a larger frequency and number of endogenous CD45.2+ OVA (257-264) -specific CTL in their lungs than mice that received unstimulated or single peptide stimulated cells (frequency: 4.5% versus 0.8% and 1.1%, Fig 6C; cell numbers: 0.6x10 6 versus 0.1x10 6 and 0.2x10 6 , for repeat peptide, single peptide and no peptide respectively, Fig 6D). This indicates that the adoptively transferred repeat peptide stimulated cells were less efficient than the other donor cells to compete with the endogenous antigen-specific CTL response. These results indicated Pooled data of frequency of CTL expressing individual inhibitory receptors within the different culture conditions. (C) SPICE figures depicting the frequency of CD8 + T cells expressing one, two, three or four inhibitor receptors (PD-1, CD244, CD160, Lag3) simultaneously. Each symbol representative one animal (n = 7-11), with 7-9 independent experiments performed. Line depicts mean ± SE. Between the groups, Wilcoxon signed rank test was performed with exception of % Tim-3+ and % CD160+ for which a Student's t test with Welch's correction was used. � <0.05, �� P<0.01, ��� P<0.001, ���� P<0.0001. https://doi.org/10.1371/journal.ppat.1008555.g004

PLOS PATHOGENS
Generate exhausted CD8+ T cells in vitro that in vitro repeat peptide stimulation resulted in a reduced capacity of CTL to expand when re-exposed to their cognate antigen. This further supports our notion that in vitro repeated antigen stimulation results in exhaustion of these cells.

Repeat peptide stimulated cells have the transcriptional profile of exhausted CTLs
To determine whether repeat peptide stimulated cells also have a distinct transcriptional profile, we performed RNAseq on sorted live CD8+ T cells from the different culture conditions.

Generate exhausted CD8+ T cells in vitro
As shown in the principle component analysis (PCA) (Fig 7A), the samples clustered together and where distinctly separated depending upon their in vitro treatment. We identified 1196 genes with more than 2-fold increased expression and 1218 genes with more than 2-fold downregulation in the repeat peptide stimulated cells relative to that in single peptide stimulated cells. The transcriptional prolife of the repeat peptide stimulated cells is clearly more distinct than those of the single and non-stimulated cells (Fig 7B and S1 Table). Among the upregulated genes, multiple inhibitory receptors encoding genes, including PD-1 (Pdcd1), Lag3, Tim3 (Havcr2), CD160, Tigit and CTLA4 were presented on the top of this list. As expected, the genes for the markers of the terminal differentiated effector cells, like GzmB, GzmC and Prf1 were also significantly upregulated on the repeat peptide stimulated cells ( Fig  7C). The transcription factors, Eomes and Tcf7 were downregulated by 4.0 and 6.6 fold, respectively, while Tox was found to be 2.9 fold increased in the repeat peptide stimulated cells compared to single peptide stimulated cells (Fig 7C), confirming our flow cytometry findings of these TFs (Fig 5B and 5C). Furthermore, TFs, which are associated with CTL exhaustion such as IRF4 [28], NR4a [29] and Batf [30] were also upregulated in the repeat peptide stimulated cells in comparison to single peptide and no peptide stimulated cells (Fig 7C).
To assess whether differentially expressed genes in repeat peptide stimulated cells were enriched for genes that characterize in vivo exhausted CTL from LCMV cl13 infections, we

PLOS PATHOGENS
Generate exhausted CD8+ T cells in vitro performed gene set enrichment analysis (GSEA). Differentially expressed genes between repeat peptide stimulated cell and single peptide stimulated cells were compared to public gene sets of upregulated or downregulated in in vivo exhausted CTLs (gene set GSE87646). Gene sets upregulated in exhaustion were found to be significantly more enriched in the repeat peptide stimulated upregulated genes compared to single peptide stimulated cells ( Fig 7D). Conversely, the gene sets that were reported to be downregulated in exhausted cells were more enriched in the genes that were downregulated in repeat peptide stimulated cells versus the single peptide stimulated cells (Fig 7D). Thus, GSEA demonstrated significant transcriptional similarity between in vitro repeat peptide stimulated cells and in vivo exhausted cells. To further investigate signaling pathways that distinguish repeat peptide stimulated cells from single peptide stimulated cells, we performed IPA analysis. The most significant pathway with upregulated activity in repeat peptide stimulated cells was the T cell exhaustion signaling pathway (Fig 7E). The above findings clearly show that in vitro repeat peptide stimulation results in the transcriptional changes of in vivo exhausted CD8+ T cells from LCMV cl 13 infections which serve as the benchmark of CTL exhaustion.

Repeat peptide stimulation results in hyper-methylation of the Tcf7 transcriptional regulatory region
Changes in methylation of transcription regulatory region have been described in exhausted CTLs [18,43]. In order to identify whether in vitro repeat peptide stimulated cells also possess similar epigenetic characteristics of exhausted CTLs, whole genome methylated DNA sequencing (MeD-seq) [44] was performed on the sorted live CD8+ T cells. Stimulated T cells undergo distinct genome wide DNA methylation changes depending on the type of treatment (Fig 8A  and S2 Table). In comparison to unstimulated and single peptide stimulated cells, the transcriptional regulatory region of Pdcd1 had significantly less DNA methylation in the repeat peptide stimulated cells (Fig 8B, left). When comparing the promotor methylation status (2kb region surrounding the TSS) of Tcf7, there was more DNA methylation detected in repeat peptide stimulated than in the single peptide stimulated cells or unstimulated cells (Fig 8B, center). Meanwhile, in the promotor region of the GzmB, significantly less DNA methylation was found in the repeat peptide stimulated cells than in the other cells (Fig 8B, right), which was in line with the higher expression of the protein upregulated in the repeat peptide stimulated cells (Fig 3B). Besides Pdcd1, none of the other inhibitory receptor genes' regulatory or promotor regions were found to possess significant differences in methylation status, although some of them showed expected trends in DNA methylation changes at their TSS. Interestingly, the DNA methylation states of the cytokine genes IL-2, IFN-γ and TNF-α were not significantly different. This indicated that other gene expression control mechanisms, like histone modifications or transcription factor abundance, might regulate the differential expression of these

PLOS PATHOGENS
Generate exhausted CD8+ T cells in vitro cytokines. Overall, these findings indicate that the repeat peptide stimulated cells have distinct DNA methylation patterns and reveal that the downregulation of TCF1 expression is accompanied by increased promotor methylation of Tcf7.

Generate exhausted CD8+ T cells in vitro
To further confirm that DNA promotor methylation contributes to the downregulation of TCF1 expression in exhausted cells we treated cells with a DNMT inhibitor for the last 3 days of culture and examined whether TCF1 expression was modified in the repeat peptide stimulation cultures. Indeed, treatment of these cells with DNMT inhibitor resulted in an increase in TCF1 levels in exhausted CTL (Fig 8C). This further supports that DNA methylation plays a role in silencing TCF1 as exhausted cells progress from "progenitor exhausted" to the "terminally exhausted" subpopulation.
DNMT inhibitor also reduced the expression of PD-1 and Tim3 in repeat peptide stimulated cells (S4 Fig). Because PD-1 and Tim3 expression are controlled by promoter region DNA methylation [43,45], our findings suggest that DNMT inhibitor prevented exhaustion rather than reverted the exhaustion of T cells. These experiments suggest that the in vitro exhaustion system we describe can be used to test reagents that modulate T cell exhaustion.

Anergy is not a feature of in vitro and in vivo exhausted CTL
The high purity of T cells in our culture system raises the question whether the absence of CD28 costimulation induces anergy [46][47][48] and this explains some of the phenotypes we observe. However, this is unlikely as our cells are cultured in the presence of IL-7 that has been suggested to prevent anergy [49] and furthermore we still see exhaustion when cells are cultured with IL-2 (S5A and S5B Fig) another anergy preventing cytokine [50,51]. Some CD28 costimulation, however, may be present in our culture system, as repeat peptide stimulated cells express CD80 mRNA levels 74-fold and 13-fold higher compared to unstimulated and single peptide stimulated cultures, respectively. Flow cytometry confirmed the high surface expression of CD80 on repeat peptide stimulated cells (S5C Fig).
To further exclude that our repeat peptide stimulated cells have features of anergy, we performed GSEA for sets of genes identified by in vitro anergy induction (gene set GSE5960) [46]. We find that both genes that are upregulated and downregulated in anergic T cells, are enriched in our repeat peptide stimulated cells (S6A and S6B Fig). Thus there is no evidence for the presence of anergy. We found a similar enrichment for genes both upregulated and downregulated in anergic T cells in in vivo exhausted gp33-specific CTL from LCMV clone 13 infected mice (gene set GSE41867) [12] (S6C and S6D Fig).
Finally, we performed pathway enrichment analysis by IPA on the genes differentially expressed between repeat peptide stimulated cells and the unstimulated cells. We found that anergy related signaling pathway was neither up-or down-regulated in the repeat peptide stimulated cells (S6E Fig). In this comparison, the T cell exhaustion signaling pathway was again the most significant upregulated pathway. Taken together, these findings argue that anergy is not a major feature of our in vitro exhausted CTL.

Discussion
Persistent antigen stimulation is a key driver of T cell exhaustion [37,38]. In this study, we show that repeat stimulation with peptide is sufficient to induce T cell exhaustion in vitro. Repeat peptide stimulation rapidly, within 5 days, induced all characteristics of CTL exhaustion including loss of cytokine production and polyfunctionality, the expression of multiple inhibitory receptors, reduced expansion capacity and competitive fitness, transcription factor and gene expression patterns all compatible with in vivo generated exhausted CTL [10,11]. This phenotype was largely preserved when cells were rested for several days. Our in vitro repeat peptide stimulation culture conditions include cytokines IL-7 and IL-15 which are known to have anti-apoptotic potential and ensure T cells do not undergo activation induced cell death (AICD) [52]. In particular, IL-15 has been shown to promote the survival of exhausted HIV-specific CD8+ T cells by upregulating Bcl-2 and Bcl-xL [7,53]. Its inclusion in the repeat peptide stimulations, we argue, promotes the survival of these cells as they become gradually more exhausted while avoiding the increased apoptosis mediated by IL-2, a known promotor of T cell AICD [54].
The in vitro exhausted CTL we generate have many of the molecular features of exhaustion. In particular transcription factors that associate and orchestrate T cell exhaustion are clearly modulated in the same manner as seen with in vivo induced CTL exhaustion of chronic infection models like LCMV clone 13 infections and cancer [11,31]. The transcription factor TOX is known to be increased in in vivo exhausted CTL [23,24,26], while TCF-1 is downregulated in exhausted cells and its presence is associated with precursor exhausted CTL that retain some proliferative capacity [11,40]. All in vitro exhausted CTL we generate have high protein expression for TOX and Tox mRNA is upregulated by 2.6-fold. Furthermore, in vitro exhausted CTL have downregulated TCF-1 and at the mRNA level its gene Tcf7 is decreased by 6.6-fold. A recent report indicated that Prdm1 and c-maf transcription factors control a coinhibitory module in T cells [55]. Although we found that Prdm1 expression was upregulated by 4.5-fold in our in vitro exhausted CTL, we did not find any changes in c-maf transcription factor. Activated protein C receptor (PROCR) and podoplanin (PDPN), recently reported to be co-inhibitory receptors [55], were not upregulated in our in vitro exhausted cells. This may not be surprising as these markers were only found in tumor infiltrating T cells and IL-27 may be mediating this profile [55]. Exhausted T cells from chronic infection however are not affected by IL-27 signaling [56] and thus PROCR and PDPN expression may depend on the tumor microenvironment and not characterize all exhausted CTL. Our in vitro exhaustion method generates exhausted T cells which may more closely reflect the core features of exhaustion. Our method is independent of environmental factors that may occur in the in vivo setting of chronic infection or tumor microenvironment. Such factors can obscure direct exhaustion related transcriptional changes from those induced by bystander effects such as high viral loads.
Our in vitro exhaustion method utilizes peptide-stimulated purified CD8+ T cells and therefore one could question whether TCR stimulation in the absence of costimulation induces anergy in our T cells. Although, T cell anergy and exhaustion may share some traits and potential signaling pathways of unresponsiveness to restimulation, the mechanisms of their induction is very different. Analyzing gene expression and signaling pathways we found no evidence for anergy in our in vitro exhausted CTL. Similar data were obtained with in vivo exhausted CTL. The reason we do not induce anergy in our cultures is because they are carried out in the presence of IL-7 and IL-15. It is well established that γc chain cytokines such as IL-2 and possibly IL-7 and IL-15 can override the induction of anergy when CD8+ T cells are stimulated in the absence of CD28 costimulation [47,[49][50][51]. Cells were exhausted even when cultured with IL-2, albeit IL-2 in these cultures increases cell death significantly. Interestingly, our in vitro exhausted CTL express high levels of mRNA and surface CD80 and this may also contribute to preventing anergy. This is not surprising as activated mouse T cells can express CD80 both in vitro but also in vivo in autoimmune mice [57,58]. Thus in vitro exhausted CTL do not present key anergy traits although one cannot exclude that some downstream signaling or inhibitory pathways are shared between exhaustion and anergy.
Previous studies have reported that there are subsets of exhausted CTLs, one known as "progenitor exhausted" cells which express TCF1 and TOX, corresponding to cells that retain the capacity to expand with checkpoint blockade therapy and the other subpopulation known as the "terminally exhausted" cells which are TCF1 low and cannot respond to checkpoint blockade [26,40,41,59]. In our in vitro exhausted cell cultures, we observed both subsets; a subset of cells that express low levels of TCF1 and high levels of TOX and GzmB, which is in line with the characteristics of "terminally exhausted" T cells and a subset that is TCF1+TOX + corresponding to the "progenitor exhausted" cells. All cells are T-bet hi PD1+, thus our TOX + cells may retain some "progenitor exhausted" characteristics as T-bet hi cells are reported to retain proliferative capacity in the exhaustion setting [42]. Thus, these two subsets exist in our exhausted cultures and this allows us to perform future in depth characterization of these cells and establish their relationship and the factors that control them. CXCR5+ T cells were described as "less exhausted" and responders to check-point blockade in chronic viral infection [21,59]. Although "progenitor exhausted" TCF1+TOX+ are present in our repeat peptide stimulated T cells, we could not detect any CXCR5 expression. CXCR5+ exhausted CD8+ T cells are found in lymphoid organs but not in peripheral blood or other infected organs of LCMV clone13 infected animals [21], and therefore our CXCR5in vitro exhausted cells may resemble more the later than those found in lymphoid organs.
TCF1 downregulation in exhausted cells was reported to be accompanied by reduced chromatin accessibility [24,25]. Our findings suggest that DNA methylation also contributes to reduced TCF1 expression as both the Tcf7 promotor was found to be hyper-methylated and the inhibition of DNA methyltransferases during the last 3 days of culture results in the retention of TCF1 expression in our exhausted CTL. Whether these TCF1 high cells generated by DNMT inhibitor treatment retain all the characteristics of the "progenitor exhausted" cell subset remains to be determined. If true it may indicate that such inhibitors can be combined with checkpoint inhibition therapy as such TCF1+ cells are the major responders to therapy.
By studying these in vitro exhausted CTLs, the accumulation of inhibitory receptor expression on exhausted cells can be better understood. PD-1, Tim-3 and TIGIT are rapidly expressed on the early exhausted cells, while the other receptors, like CD160 and CD244, were induced more slowly by repeat antigen stimulation. Because of the promising effects of the checkpoint blockade therapy, such as anti-PD1/PD-L1, on the treatment of human cancer [60], there is considerable interest to revive the function of exhausted T cells by co-blockade of multiple inhibitory receptors simultaneously. Our in vitro method, which induces multiple inhibitory receptors in just a few days, can serve as a feasible tool to test blocking approaches before performing in vivo animal experiments. For example, TIGIT blockade can be tested, because these cells not only upregulated TIGIT, but also downregulated its competitor of the same ligand, CD266, on their surface. Moreover, the underlying mechanisms of checkpoint blockade on reversal of T cell exhaustion need to be further investigated as do the pathways that restrain these cells from developing into terminally exhausted cells. Clearly the balance of these exhausted cells can be modulated with large effects on the numbers and potentially the function of these cells. As we recently have demonstrated, by overexpressing a single miRNA, namely miR-155, we can ameliorate the attrition of virus-specific CTL during chronic infection and in vivo increase their numbers by 2 logs [61]. Our in vitro system of T cell exhaustion can be used to screen the overexpression of genes with retroviruses or their inhibition or deletion using CRISPR-Cas9 to discover new targets which modulate T cell exhaustion.
The method we describe allows for the rapid, within 5 days, generation of large numbers of fully exhausted CTL and can therefore be used for medium to high throughput screening of compounds, reagents or gene modifications that can prevent, ameliorate, reverse or accelerate T cell exhaustion. In particular, the extreme loss of IL-2 and TNFα production in our repeat peptide stimulated cells allows for the first time a truly medium to high throughput approach for drug screening since changes in IL-2 or TNFα production can be easily determined in the supernatants.
In conclusion, we have established a rapid in vitro system of CD8+ T cell exhaustion. These exhausted CTL exhibit all the known molecular and functional characteristics of exhaustion yet can be induced in large numbers within 5 days as opposed to the small numbers generated after 30 days of chronic infection of in vivo mouse models of exhaustion. This in vitro method can not only be used as a screening system to prevent/revert CD8+ T cell exhaustion, thereby identifying new therapies, but also for research aiming at revealing mechanisms of CD8+ T cell exhaustion. Using this in vitro method we show that TCF1 silencing during exhaustion is in part controlled by DNA methylation. Overall, this in vitro method makes the future study of CTL exhaustion more feasible and reduces the need for in vivo studies.

Repeated antigen stimulation in vitro
For single peptide stimulation, cells were cultured in the presence of OVA (257-264) peptide for 48 hours. The peptide was then removed by washing the cells two times with complete media. For the remaining 3 days, the cells were cultured in the complete media with cytokines. For repeat peptide stimulation, 10ng/ml OVA (257-264) peptide was added daily for five days. The cells were washed also on day 2 to allow for comparable culture conditions. Unstimulated control cells were cultured in media with cytokines but without adding peptide. Cells from all three conditions were checked daily, and when the cells were confluent, they were split and cultured with fresh complete media containing cytokines. After day 5, some of the cell were washed two times with complete media and maintained in the media only with cytokines for another three days. In some experiments cell cultures were treated on day 2 with 20μM DNA methyltransferase (DNMT) inhibitor SGI-1027 (Tocris, Bio-techne) that targets DNA methyltransferases DNMT3B, DNMT3A and DNMT1.
On day five, cells were harvested and counted using an automated counting system (Countess, Life Technologies). Cells were stained with DAPI Viability dye (Beckman Coulter, Cat B30437) and Acridine Orange (Biotium, Cat 40039) to distinguish live and dead cells.

Flow cytometry
To investigate phenotypic and functional changes, cells were stained with monoclonal antibodies and analyzed using flow cytometry. On day 5, cells were harvested and immediately stained for surface and intracellular antigens. For surface staining, cells were washed with FACS wash (HBSS containing 3% FBS and 0.02% sodium azide) and incubated with 20μL mix of the pre-determined optimal concentrations of the fluorochrome-conjugated monoclonal antibodies at 4˚C in the dark for 20 minutes. The cells were then washed once with FACS wash and fixed with 1% PFA. For the transcription factor staining, cells were first stained for surface antigens as described above. Following the washing step, cells were fixed with FoxP3 Fixation Buffer (005523, eBioscience) for 60 minutes in the dark at 4˚C and then washed with Perm/Wash buffer (008333, eBioscience) and stained with a mix of antibodies against transcription factors for 1 hour at 4˚C in the dark. Cells were then washed twice with Perm/Wash buffer and fixed with 1% PFA. Appropriate isotype controls were included for staining of transcription factors.
To analyze cytokine production, cells were re-stimulated with the 10μg/ml OVA (257-264) SIINFEKL peptide for 6 hours at 37˚C, 5% CO 2 in the presence of GolgiPlug (BD Biosciences) and anti-CD107a-APC-Cy7 antibodies (ID4B, Biolegend). Cells were then stained with surface marker antibodies as described above. After washing with FACS wash, cells were fixed with IC Fixation Buffer (88-8824, eBioscience) at 4˚C overnight, washed with Perm/Wash buffer and stained for intracellular cytokines for 45 min in the dark at 4˚C. After staining, cells were washed twice with Perm/Wash buffer and fixed with 1% PFA.
Cells were measured on a LSRFortessa (BD Biosciences) using application settings and at least 200,000 events were collected. Data was then analyzed with FlowJo software (Version 9.9.4, Treestar, Ashland, OR, USA).

In vivo influenza model
For influenza virus infection studies, live CD45.1+ OT-I cells were sorted from in vitro cultures on a FACSAria III (BD Biosciences) using fluorochrome-conjugated Annexin V and anti-CD8 antibodies and 10,000 cells were intravenously transferred into 8-12 weeks old CD45.2 + C57BL/6J wild-type recipient mice. Naïve CTL were also freshly isolated from spleen of an OT-I mouse and adoptively transferred. Three hours later, mice were anesthetized with 2.5% isoflurane gas and infected intranasally with influenza virus strain A/WSN/33-expressing OVA (WSN-OVA, a gift from Dr. D. Topham, University of Rochester Medical Center). Body weight was measured daily to track the influenza infection.
Ten days post infection, the lung, spleen and mediastinal lymph nodes were harvested and single cell suspensions were obtained after processing the tissues. As described previously [62], lungs were digested for 2 h at 37˚C with 3.0 mg/ml collagenase A and 0.15 μg/ml DNase I (Roche) in RPMI 1640 containing 5% heat-inactivated FBS, 2 mM L-glutamine, 100 IU/ml penicillin, 100 μg/ml streptomycin. Digested lungs were then filtered through a 40-μm cell strainer (Falcon) and washed in the same media as above. PE-conjugated tetramers of H-2K b major histocompatibility complex class I loaded with OVA (257-264) were used to identified the antigen-specific CTLs in the lungs [63].

RNA sequencing
To compare the gene expression between the different culture conditions, RNA sequencing was performed. On day 5, 1x10 6 live CD8+ T cells were sorted from the three different culture conditions and immediately lysed with TRIzol LS reagent (Life Technologies) and stored at -80˚C. RNA was extracted according to manufacturer's instructions and a bioanalyzer (Agilent) was used to determine the integrity of the extracted RNA. Barcoded sequencing libraries were generated using the KAPA RNA HyperPrep kit (Roche Diagnostics) with RiboErase (HMR) rRNA depletion. Library quality was assessed with the bioanalyzer and KAPA qPCR was performed for quantification before cluster generation and 100-bp paired-end sequencing on a Hiseq2500 machine (Illumina). Data sets are deposited in the Gene Expression Omnibus (GEO) database repository under accession numbers GSE150120.

Analysis of differential transcript abundance, normalization of read counts by gene size, and downstream analyses
Four independent biological replicates were analyzed for each condition. The quality of the sequencing was verified using the FastQC software (http://www.bioinformatics.babraham.ac. uk/projects/fastqc/). The demultiplexed fastq files were aligned using STAR software (v.2.5.3e) [64] with default settings and mus musculus GRCm38 as alignment reference. Then bam files, generated during alignment, were annotated using FeatureCount software (v1.6.1) [65] to obtain the annotated files (count files). The annotation reference was genecode.vM15.gtf.
Count data was preprocessed to remove low expressed genes. Then, rlog transformation was applied to the clean counts for visualization and comparison purposes which include correlation and clustering analysis generating the heatmaps and PCA plots during the process. For differential expression analysis, DESeq2 (DESeq2 R package, v1.22.2) [66] was used directly in the clean count data.
The list of differentially expressed genes was used to perform Ingenuity Pathway Analysis (IPA, Qiagen, USA version 01-12) to further discern which pathways are involved in the CD8 + T cell exhaustion process. Pathway enrichment P-values (Fisher's exact test) and activation Z-scores were calculated by IPA and used to rank the significant pathways.
The same list of significantly differentiated genes was used for enrichment analysis (GSEA Desktop Application, v2.2.1). The CD8+ T cell exhaustion gene-sets were downloaded from GMO datasets (gene set GSE87646) based on the publication of Bengsch B, et al [2]). Enrichment analysis in CTL exhaustion genes was determined for the upregulated and downregulated genes separately. Normalized Enrichment Scores (NES) values were used to determine whether an expression gene-set was enriched or not in CTL exhaustion genes. For anergy GSEA analysis, gene set GSE5960 [46] and gene set GSE41867 [12] were used.

DNA methylation profiling detection
DNA methylation profiling was done as previously described by the MeD-seq method [44]. For MeD-seq sample preparation LpnPI (New England Biolabs) digestions were carried out on DNA samples according to manufacturer's protocol. Reactions contained 50 ng and digestion took place overnight in the absence of enzyme activators. Digests of genomic DNA with LpnPI resulted in snippets of 32 bp around the fully-methylated recognition site that contains CpG. The DNA concentration was determined by the Quant-iT High-Sensitivity assay (Life Technologies) and 50 ng dsDNA was prepared using the ThruPlex DNA-seq 96D kit (Takara). Twenty microliters of amplified end product were purified on a Pippin HT system with 3% agarose gel cassettes (Sage Science). Stem-loop adapters were blunt end ligated to repaired input DNA and amplified (4 +10 cycles) to include dual indexed barcodes using a high fidelity polymerase to yield an indexed Illumina NGS library. Multiplexed samples were sequenced on Illumina HiSeq2500 systems for single read of 50 base pairs according to manufacturer's instructions. Dual indexed samples were demultiplexed using bcl2fastq software (Illumina).
MeD-seq data processing was carried out using specifically created scripts in Python version 2.7.5. Raw fastq files were subjected to Illumina adaptor trimming and reads were filtered based on LpnPI restriction site occurrence between 13-17 bp from either 5' or 3' end of the read. Reads that passed the filter were mapped to mm10 using bowtie2.1.0. Multiple and unique mapped reads were used to assign read count scores to each individual LpnPI site in the mm10 genome. BAM files were generated using SAMtools for visualization. Gene and CpG island annotations were downloaded from UCSC (MM10). Genome wide individual LpnPI site scores were used to generate read count scores for the following annotated regions: transcription start site (TSS) (1 kb before and 1 kb after), CpG islands and genebody (1 kb after TSS till TES).
MeD-seq data analysis was carried out in Python 2.7.5. DMR detection was performed between two datasets containing the regions of interest (TSS, genebody or CpG islands) using the Chi-Squared test on read counts. Significance was called by either Bonferroni or FDR using the Benjamini-Hochberg procedure. Differently methylated regions were used for unsupervised hierarchical clustering, the Z-score of the read counts was used for normalization and is also shown in the heatmaps. In addition, a genome wide sliding window was used to detect sequentially differentially methylated LpnPI sites. Statistical significance was called between LpnPI sites in predetermined groups using the Chi-squared test. Neighbouring significantly called LpnPI sites were binned and reported, DMR threshold was set at a minimum of ten LpnPI sites, a minimum size of 100 bp and either a twofold or fivefold change in read counts. Overlap of genome wide detected DMRs was reported for TSS, CpG island and gene body regions.

Statistical analysis
Statistical analyses were performed using Prism software (GraphPad Prism5 for Windows, Version 5.04). The normality of data distribution was assessed using the Shapiro-Wilk

PLOS PATHOGENS
Generate exhausted CD8+ T cells in vitro normality test. Homogeneity of variance was tested with Bartlett's test. When data were normally distributed and group variances were equal, an ANOVA with Tukey's Multiple Comparison Test was performed. When data were normally distributed but group variances were unequal, a Student's t test with Welch's correction was performed. If data were not normally distributed a Wilcoxon signed rank test or a Mann-Whitney U test was performed. P values equal or lower than 0.05 were considered statistically significant with the numbers of stars in the figures indicating the p value: � = P � 0.05, �� = P � 0.01, and ��� = P � 0.001. In repeat peptide stimulated cells the anergy related pathway in IPA analysis was neither upregulated not downregulated. IPA analysis was performed on the differentially expressed genes in repeat stimulated cells (compared to no peptide). The "Regulation of IL-2 expression in activated and anergic T lymphocytes" pathway is showed enlarged and the overall positioning in the global IPA pathway analysis is indicated by the arrow and red circle. Significantly upregulated pathways (orange) and downregulated pathways (blue), no activity pattern available (grey) are depicted. Bar graphs depict P values (hypergeometric test) presented as-Log10. (TIFF) S1 Table. Differentially expressed genes in repeat peptide stimulated T cells determined by RNAseq. Log2 fold change is show for repeat peptide OT-I versus single peptide stimulated OT-I. After 5 day cultures, cells were sorted and RNAseq was performed. Genes with >2 fold change and padj value < 0.05 shown. (PDF) S2 Table. Differentially methylation of TSS regions was determined by MeD-seq. Fold change is show for repeat peptide OT-I versus single peptide stimulated OT-I. After 5 day cultures, cells were sorted and MeD-seq was performed. Greater than 2-fold changes shown. (PDF)