The transcription factors Runx3 and ThPOK cross-regulate acquisition of cytotoxic function by human Th1 lymphocytes

Cytotoxic CD4 (CD4CTX) T cells are emerging as an important component of antiviral and antitumor immunity, but the molecular basis of their development remains poorly understood. In the context of human cytomegalovirus infection, a significant proportion of CD4 T cells displays cytotoxic functions. We observed that the transcriptional program of these cells was enriched in CD8 T cell lineage genes despite the absence of ThPOK downregulation. We further show that establishment of CD4CTX-specific transcriptional and epigenetic programs occurred in a stepwise fashion along the Th1-differentiation pathway. In vitro, prolonged activation of naive CD4 T cells in presence of Th1 polarizing cytokines led to the acquisition of perforin-dependent cytotoxic activity. This process was dependent on the Th1 transcription factor Runx3 and was limited by the sustained expression of ThPOK. This work elucidates the molecular program of human CD4CTX T cells and identifies potential targets for immunotherapy against viral infections and cancer.


Introduction
The thymic differentiation of helper CD4 and cytotoxic CD8 T lymphocytes results from the opposite activity of key transcription factors (TF) including ThPOK and Runx3 repressing the expression of CD8 and CD4 T cell lineage genes, respectively. After emigration from the thymus, naive CD8 and CD4 T cells maintain the expression of Runx3 or ThPOK, respectively, suggesting that the lineagedefining role of these TF is also active in the periphery (Vacchio and Bosselut, 2016). However, the repression of a cytotoxic program in peripheral CD4 T cells is not absolute as these cells can acquire perforin-dependent cytotoxic activity (Appay et al., 2002a;van de Berg et al., 2008;Cheroutre and Husain, 2013).
Initially considered as a phenomenon of peripheral importance, the acquisition of cytotoxic function by CD4 T cells is now recognized as a key component of immunity against viruses and tumors  and correlates with positive outcome in multiple human and animal models (Brown et al., 2012;Wilkinson et al., 2012;Johnson et al., 2015;Ma et al., 2015;Weiskopf et al., 2015;Verma et al., 2016;Xie et al., 2010;Quezada et al., 2010;Fu et al., 2013). Beside their role in viral infections and cancer, cytotoxic CD4 (CD4 CTX ) T cells may also have a pathogenic role in chronic inflammatory disorders (Dumitriu, 2015).
The development of CD4 CTX T cells remains incompletely understood. In humans, CD4 CTX T cell function is a hallmark of terminally differentiated antigen-experienced cells producing large amounts of gamma interferon (IFNg) and low levels of interleukin-2 (IL-2) (Appay et al., 2002a;Casazza et al., 2006). This suggests that CD4 CTX T cell differentiation might be induced by the prolonged stimulation of Th1 lymphocytes. On the other hand, recent studies suggest that CD4 CTX T cells form a distinct lineage emerging from precursors expressing class-I restricted T cell-associated molecule (CRTAM) (Takeuchi et al., 2016). Several TF probably contribute to the differentiation of CD4 CTX T cells within or outside the Th1 pathway. In the mouse intestine, the acquisition of cytotoxicity by CD4 T cells is associated with the down regulation of ThPOK and the upregulation of Runx3 Reis et al., 2013;Sujino et al., 2016). In CD8 T cells, the cytotoxic program is activated by the cooperation of TF from the T-box family (T-bet and Eomes) and Runx3, (Cruz-Guilloty et al., 2009;Pearce et al., 2003). In mouse models of cancer or neuroinflammation, Eomes was required for the induction of CD4 CTX T cells (Curran et al., 2013;Raveney et al., 2015). In murine influenza infection, Blimp1 was required for CD4 CTX T cell differentiation (Marshall et al., 2017). The role of these TF in human CD4 CTX T cell differentiation remains to be determined.
The transcriptional program and lineage fate of effector T cells are established and maintained at the epigenetic level through DNA and histone modifications (Wilson et al., 2009;Araki et al., 2008;Sellars et al., 2015). Hypomethylation of the PRF1 promoter, the gene encoding perforin, is associated with increased perforin expression in human CD4 T cells (Kaplan et al., 2004). The epigenetic modifications underlying the differentiation of CD4 CTX T cells have not been determined.
Here, we studied circulating CD4 CTX T cells isolated from the peripheral blood of cytomegalovirus-seropositive (CMV + ) healthy adults. Compared to mouse models of infection (Brown et al., 2012) or cancer (Curran et al., 2013), this situation allows access to a significant number of cells presenting a fully established cytotoxic functional program at steady state. Using transcriptomic and epigenomic approaches, we defined the molecular events that dictate human CD4 CTX differentiation. We further show that the increased expression of Runx3 and T-bet and key epigenetic modifications at the PRF1 promoter, without downregulation of ThPOK, underlie the acquisition of cytotoxic function by human Th1 lymphocytes.

Results
Phenotype and function of in vivo differentiated perforin + human CD4 T cells In healthy humans, chronic CMV infection is associated with the expansion of perforin + /granzyme B + CD4 (Figure 1a-b and Figure 1-source data 1) (van Leeuwen et al., 2004). In order to use this model for transcriptomic and epigenomic analyses of CD4 CTX T lymphocytes, we characterized their phenotype and function in CMV + healthy adults and compared them to cytotoxic CD8 T cells. As previously reported, high perforin expression was observed in terminally differentiated CD4 and CD8 T cells that had downregulated the co-stimulatory molecules CD28 and CD27, respectively ( Figure 1c) (van de Berg et al., 2008;Appay et al., 2002b). Perforin + CD4 T cells were CD8b-negative and a minority expressed low levels of CD8a (Figure 1-figure supplement 1a-b). Further analyses were conducted on sorted naive (CD45RO -CD28 + ) and terminally differentiated (CD28 -) CD4 T cells and on naive (CD45RO -CD27 + ) and terminally differentiated (CD27 -) CD8 T cells (Figure 1d). Increased PRF1 gene expression by CD28 -CD4 and CD27 -CD8 T cells was confirmed by mRNA quantification and was associated with potent cytotoxic activity in a polyclonal cell lysis assay (Figure 1e, Figure 1f and Figure 1-source data 1). This activity was abolished by Concanamycin A, supporting a perforin-dependent mechanism (Kataoka et al., 1996). Bisulphite sequencing indicated an inverse correlation between the expression of the PRF1 gene and the DNA methylation status of its promoter region ( Figure 1g and Figure 1-source data 1). Whereas PRF1 promoter was hypermethylated in a perforinfibroblastic cell line (HEL-299), all CpG sites were hypomethylated in CD28 -CD4 and CD27 -CD8 T cells. Low DNA methylation levels were detected at intermediate (16 to 28; middle grey line on Figure 1g) CpG sites in naive CD8 T cells and at proximal (CpG sites 29 to 34; right grey line) sites in both naive CD4 and CD8 T cells, suggesting that the PRF1 gene is transcriptionally poised in naive T lymphocytes. Together, these results indicate that CD28 -CD4 T cells exert a cytotoxic activity comparable to CD27 -CD8 T cells and that this subset can therefore be used as a relevant model of in vivo differentiated CD4 CTX T cells.

The transcriptional program of CD4 CTX T cells is enriched in CD8 T cell lineage genes without down regulation of ThPOK
In order to elucidate the molecular basis of CD4 CTX T cell differentiation, their transcriptome was first compared to that of naive CD4, naive CD8 and CD8 CTX T cells. Unsupervised analysis of transcriptional programs indicated that naive and cytotoxic T cells formed separate clusters and that CD4 CTX and CD8 CTX T cells were more closely related than their naive counterparts (Figure 2a-b). Gene set enrichment analysis (GSEA) was used to quantify the degree of sharing of the transcriptional program of CD4 CTX T cells with the CD4 and CD8 T cell lineages. Genes expressed at higher levels in CD4 CTX and CD8 CTX T cells as compared to their naive counterparts were identified and their enrichment in naive CD8 and CD4 T cell transcriptomes was assessed (Supplementary file 1 and Figure 2c). As expected, genes that were upregulated in CD8 CTX T cells were significantly enriched in genes of the CD8 T cell lineage ( Figure 2c). Strikingly, genes upregulated in CD4 CTX T cells were also enriched in CD8 rather than CD4 T cell lineage genes. The transcriptional program common to CD4 CTX and CD8 CTX T cells included RUNX3, TBX21 (T-bet) and EOMES, TF known to promote effector and memory functions in CD8 T cells (Supplementary file 1, Figure 2d and Figure 2source data 1) (Cruz-Guilloty et al., 2009). Strikingly, the enrichment in CD8 T cell lineage genes by CD4 CTX T cells was not associated with the downregulation of ZBTB7B (ThPOK) (Figure 2d and Figure 2-source data 1). As expected, ThPOK expression was higher in naive CD4 as compared to naive CD8 T cells. However, in contrast to mouse intestinal CD4 T cells , ThPOK gene expression was not downregulated by human circulating CD4 T cells, CD4 CTX T cells expressing higher levels of ThPOK mRNA than naive CD4 T cells (Figure 2d and We explored the epigenetic basis of the transcriptional program of CD4 CTX T cells by analysing their DNA methylome and comparing it to that of naive CD4 and CD8 CTX T cells. The acquisition of cytotoxic function by CD4 CTX and CD8 CTX T cells was associated with changes in methylation, primarily hypomethylation, of large numbers of genes (Figure 2-figure supplement 1a-d). Unsupervised analysis of DNA methylomes indicated that naive and cytotoxic T cells formed separate The expression of PRF1 mRNA was measured by qPCR in purified T cell subsets of 5 CMV + subjects. Results are median ± interquartile range of log2 fold change as compared to naive CD4 T cells. **:p<0,01 and ***:p<0,01. (f) The cytolytic activity of purified T cell subsets against anti-CD3-loaded target cells was assessed with or without pre-incubation with Concanamycin A (ConA). Data are mean ± SEM of three independent experiments on cells from different donors. (f) The methylation status of the PRF1 promoter was assessed in T cell subsets by bisulphite pyrosequencing. Data are median ± interquartile range of five donors for CD28 -CD4 T cells and HEL-299 and of 9 donors for the other indicated subsets. Grey lines indicate three regions with distinct methylation profiles. See also Figure 1-  . GSEA indicated that genes expressed at higher levels in CD4 CTX or CD8 CTX T cells as compared to their naive counterparts were significantly enriched in hypomethylated CpG located in their promoter regions ( Figure 2f). In contrast, genes that were downregulated in CD4 CTX or CD8 CTX T cells were not enriched in hypermethylated CpG, suggesting distinct epigenetic mechanisms in the up-and downregulation of genes upon differentiation of cytotoxic T cells. Analysis of probes located in the promoter region of TF up regulated in cytotoxic as compared to naive T cells suggested significant hypomethylation for RUNX3 and TBX21 in both CD4 CTX and CD8 CTX (Figure 2g and Figure 2-source data 1), but not for EOMES (data not shown). As observed during thymopoiesis (Vacchio and Bosselut, 2016), ZBTB7B promoter was significantly hypomethylated in naive CD4 as compared to naive CD8 T cells, but no significant difference was observed between CD4 CTX and CD8 CTX cells ( Figure 2g and Figure 2-source data 1). Most analyzed RUNX3 probes (77%) were located in the distal promoter (TSS1500), a region suspected to be involved in the control of RUNX3 mRNA translation (Kim et al., 2015), whereas most TBX21 (77%) and ZBTB7B probes (76%) were located in the proximal promoter (5'UTR, 1st exon and TSS200). Together, these results indicate that the transcriptional program of human CD4 CTX T cells is enriched in CD8 T cell lineage genes. Acquisition of this program involves extensive hypomethylation of the promoter regions of a large number of genes, including TF, and is not accompanied by ThPOK downregulation.

Stepwise differentiation of CD4 CTX T cells within the Th1 lymphocyte lineage
In order to decipher the molecular pathways involved in the differentiation of CD4 CTX T cells, we measured the expression of cytotoxicity-related genes in subsets of memory CD4 T cells. The production of perforin was associated with the expression of the Th1 chemokine receptors CCR5 and, to a lower extend, CXCR3 but not with Th17 or Th2 receptors CCR6, CCR4 or CRTh2 ( Figure 3a). (Sallusto et al., 1998;Cosmi et al., 2000;Rivino et al., 2004;Cohen et al., 2011;Couturier et al., 2014). To determine at which stage of their differentiation Th1 cells initiate the production of perforin, central memory (CM) and CD28 + effector memory (EM) T cells expressing Th1 chemokine receptors were compared to naive and CD4 CTX T cells (Figure 3b). No perforin + cells were detected by flow cytometry among CM CD4 T cells. A small proportion of perforin + cells were detected among CD28 + EM CD4 T cells in some donors and these cells were CCR5 + . Mean fluorescence intensity analysis indicated high perforin expression by CD4 CTX cells and low and comparable expression in naive, CCR5 + CM (CM Th1 ) and CCR5 + CD28 + EM (EM28 + Th1 ) CD4 T cells ( Figure 3c and Figure 3-source data 1). In contrast, PRF1 mRNA expression analysis of sorted cells (sorting strategy shown in Figure 3-figure supplement 1a) indicated that the PRF1 gene was already expressed at higher levels in CM Th1 as compared to naive cells and was further upregulated in EM28 + Th1 cells and in CD4 CTX cells (Figure 3d and Figure 3-source data 1). Notably, this pattern cell subsets purified from four to six donors, as indicated, was assessed by qPCR (upper panels). Results are expressed as median ±interquartile range of the log2 fold change as compared to naive CD4 or naive CD8 T cells. #:p<0.05 and ##:p<0.01 as compared to CD4 T cell counterparts; *:p<0.05 and **:p<0.01 as compared to naive counterparts. (e) TF protein expression was analyzed in T cell subsets by flow cytometry Upper panels: co-expression with perforin from one representative subject. Gated populations include naive CD4 (red), CD4 CTX (green), naive CD8 (purple) and CD8 CTX (blue) T cells. Lower panels: individual median intensity of fluorescence (MFI) of 5 CMV + subjects. *:p<0.05. NS: not significant. Naive and cytotoxic T cell subsets were gated using the markers and strategy illustrated in Figure 1d. As observed in naive and CD4 CTX T cells, proximal CpGs (sites 29 to 34) were hypomethylated in CM Th1 and CD28 + EM Th1 T cells. H3K4me3 (active promoter) and H3K27ac (active promoter and enhancer [Shlyueva et al., 2014]) enrichment was analyzed in previously identified putative regulatory regions ( Figure 3f, left panel) (Pipkin et al., 2010;Adams et al., 2012). In agreement with PRF1 gene expression and promoter methylation, high enrichment of H3K4me3 and of H3K27ac was detected at the PRF1 proximal promoter region in CD4 CTX T cells (Figure 3f, middle and right panels and Figure 3-source data 1). Interestingly, an H3K27ac enrichment was observed in a region located 5,5 to 7,7 kb upstream of the TSS in Th1 cell subsets, suggesting the presence of an active enhancer.
Together, these results indicate the progressive acquisition of PRF1 gene expression from CM Th1 to EM Th1 lymphocytes. Single-cell PCR analysis revealed that this process is related to a progressive increase in the proportion of PRF1 mRNA + Th1 cells ( Figure 4a and Figure 4-source data 1). Gene co-expression analysis at the single-cell level indicated that PRF1 mRNA was co-expressed with distinct sets of TF in Th1 cell subsets ( Figure 4b and Figure 4-source data 1). In CM Th1 cells, PRF1 expression was co-expressed with a relatively restricted set of TF, including PRDM1, RUNX3 and EOMES. In CD4 CTX T cells, a larger set of co-expressed TF was identified, including TBX21, HOPX, ZNF683 (Hobit), PRDM1 and RUNX3 and EOMES. Notably, a lower proportion of perforin + CD4 CTX T cells co-expressed EOMES as compared to the other co-expressed TF. In conclusion, the analysis of PRF1 gene expression in vivo suggests a model in which Th1 lymphocytes acquire permissive modifications of the local chromatin environment and a network of TF factors that could underlie the acquisition of cytotoxic functions.

Transcriptional program underlying the expression of perforin in Th1 cells
In order to identify the key steps that drive the acquisition of cytotoxic functions along the Th1 pathway, we compared the gene expression profile of CD4 CTX T cells to that of CM Th1 cells. Unsupervised gene expression analysis indicated that CM Th1 cells formed a cluster separated from naive and CD4 CTX T cells and were more closely related to CD4 CTX T cells than to naive cells ( Figure 5a). Including CD8 T cell subsets in this analysis indicated that the transcriptome of CM Th1 cells was more distant to CD8 CTX T cells than CD4 CTX T cells (  gene numbers were downregulated in the two subsets. As expected, genes upregulated in CD4 CTX T cells as compared to CM Th1 cells included cytotoxicity-related molecules, among which GNLY (granulysin), granzymes, CD107a (LAMP1), CX3CR1 and CD8A as well as TF RUNX3 and EOMES (Figure 5c-d). T-bet was not up regulated in the transcriptome dataset, in line with reported lack of sensitivity of the Illumina array for this gene (Dimova et al., 2015). Expression of IFNG, TNF and Most selected TF were induced by cell activation but their pattern of expression was differently associated with polarizing conditions (Figure 6e and Figure 6-source data 1). TBX21 and HOPX expression was specifically induced under Th1 conditions. RUNX3 and ZBTB7B mRNA were upregulated in all conditions but reached higher levels in Th1 cells. EOMES was induced in both Th1 and Based on these results, we further evaluated the role of Runx3, T-bet, Eomes, Hopx and ThPOK in this model using shRNA silencing (Figure 7a). Significant knockdown was achieved for each target TF at mRNA and protein levels (Figure 7-figure supplement 1a), whereas the non-silencing (N-S) shRNA and the empty vector (EV) had no significant effect on any of the studied genes ( Together, these results indicate that the acquisition of a cytotoxic program by naive CD4 T cells is dependent on Runx3 and, to a lesser extend, T-bet and is limited by the sustained expression of the CD4 lineage TF ThPOK. Because ThPOK was upregulated in CD8 CTX T cells in vivo (Figure 2d-e and Figure 2-source data 1), we studied its expression and role in the in in vitro differentiation of CD8 T cells. In vitro activation of naive CD8 T cells induced the differentiation of perforin + cells that co-expressed ThPOK (Figure 7-figure supplement 2a). Knockdown of ThPOK in differentiated CD8 T cells did not significantly influence the expression of perforin but significantly increased the expression of granzyme B as compared to N-S shRNA (Figure 7-figure supplement 2b), suggesting that ThPOK may limit the cytotoxic function of human CD8 T cells.

Discussion
This study demonstrates that the acquisition of cytotoxic function by human CD4 T cells is an integral part of the Th1 linear differentiation pathway. Several concordant observations support this conclusion. First, PRF1 gene expression was detected in all subsets of memory Th1 cells. The proportion of PRF1 expressing cells increased from CM to terminally differentiated EM Th1 cells and this process was associated with the diversification of co-expressed TF networks. The epigenetic modifications detected at the PRF1 gene promoter reflect this increase in perforin expression and suggest that the local chromatin environment becomes progressively more favorable from naive T cells to CM Th1 cells to terminally differentiated EM Th1 cells. Progressive acquisition of specific transcriptional and epigenetic marks is a hallmark of the linear differentiation model of CD4 T cell memory development (DEEP Consortium et al., 2016). Our work suggests that acquisition of cytotoxic functions by CD4 T cells follows a similar stepwise program.
The accessibility and expression of PRF1 gene in all subsets of Th1 cells also provide a basis for the classical observation that memory Th1 cells of diverse antigen-specificities that do not express the perforin protein ex vivo become perforin positive and cytotoxic upon in vitro expansion and cloning (Parronchi et al., 1992;Riaz et al., 2016). This in vitro acquisition of perforin expression likely reflects pre-established PRF1 chromatin modifications in precursors of CD4 CTX T cells.
The cytotoxic potential of Th1 cells is also supported by the in vitro model of CD4 CTX T cell differentiation. In this model, human naive CD4 T cells stimulated in the presence of Th1, and not Th2 or Th17, polarizing cytokines differentiated in perforin + granzyme B + cells with potent cytotoxic activity. This could be due to the promotion of the cytotoxic phenotype by Th1 cytokines or to its repression by the Th2 or Th17 transcriptional programs, or both (Parronchi et al., 1992;Xiong et al., 2013;Ciucci et al., 2017). In contrast to the rapid acquisition of IFNG expression, naive CD4 T cells expressed high levels of PRF1 1 to 2 weeks after stimulation. This observation also contrasts with the more rapid acquisition of cytotoxic function by CD8 T lymphocytes (Araki et al., 2008) and suggests that the initial steps of Th1 cell differentiation provide the required epigenetic and transcriptional signals promoting the expression of PRF1. The relatively delayed acquisition of cytotoxicity also suggests that CD4 CTX T cells are induced by prolonged antigen stimulation in vivo and may intervene when Th1 and CD8 T cells do not adequately control pathogens.
The acquisition of cytotoxic function within the Th1 lineage was promoted by Runx3 and T-bet. Runx3 knockdown reduced the expression of perforin, granzyme B and granulysin by CD4 CTX and decreased their cytotoxic activity. In contrast to mouse Th1 cells, Runx3 did not influence IFN-g mRNA expression by human Th1 cells (Djuretic et al., 2007;Wang et al., 2014). This discrepancy may be related to inter-species differences or to incompleteness of the knockdown in our experimental conditions. T-bet knockdown reduced the expression of IFN-g and cytotoxic molecules but had a more moderate impact than Runx3 knockdown on the acquisition of cytotoxic activity. This result is in line with a recent report indicating a role of T-bet in the induction of cytotoxic molecules by TCR-engineered tumor-specific CD4 T cell lines (Jha et al., 2015). Together these results indicate that human Th1 cells acquire cytotoxic functions under the control of the master regulator Runx3 cooperating with T-bet (Cruz-Guilloty et al., 2009;Jha et al., 2015). Runx3 and T-bet also controlled the expression of other TF that were not directly involved in the acquisition of cytotoxic function in these experimental conditions, including Eomes and Hopx.
Eomes was upregulated in CD4 CTX T cells as compared to naive cells in vivo but its expression was less correlated with perforin than Runx3 and T-bet. In vitro, Eomes was neither sufficient nor necessary to induce cytotoxicity as it was upregulated in Th2 cells that did not express perforin and its knockdown did not impact the expression of perforin in Th1 cells. Together, these results suggest that the role of Eomes in the acquisition of cytotoxic function by human CD4 T cells may be limited. This contrasts with a report indicating that overexpression of Eomes induces cytotoxic function in CD4 T cell lines, suggesting that this TF may promote cytotoxicity when expressed at high levels in CD4 T cells, as observed in CD8 T cells (Pearce et al., 2003;Eshima et al., 2012;Intlekofer et al., 2008). The role of Eomes could also be restricted to specific conditions of co-stimulation as recently reported (Mittal et al., 2018). The expression of Hopx by CD4 CTX T cells showed a similar pattern as T-bet and Runx3 both in vivo and in vitro. Its knockdown reduced the expression of granzyme B but did not significantly impact perforin. Therefore, the primary role of Hopx in CD4 CTX T cells may not be the induction of cytotoxicity but may include other functions, including survival (Albrecht et al., 2010).
Two other TF, Hobit (ZNF683) and ZEB2, were specifically expressed by CD4 CTX T cells in vivo but not in vitro. Hobit is upregulated in human effector CD8 T cells and murine NKT cells and therefore appears to be part of a signature common to cytotoxic lymphoid cells (van Gisbergen et al., 2012;Vieira Braga et al., 2015a2015a;Vieira Braga et al., 2015b). A recent study revealed that Hobit induces a transcriptional program promoting tissue residency of memory T cells and suggests that it could regulate cytotoxic function in murine NKT1 cells (Mackay et al., 2016). Its role in the promotion of cytotoxic function by CD4 T cells in vivo therefore remains to be established. Similarly, further studies should establish the role of ZEB2 in terminally differentiated CD4 + T lymphocytes (Dominguez et al., 2015;Omilusik et al., 2015).
In parallel with the positive regulation operated by T-bet and Runx3, the cytotoxic function of human CD4 T cells was negatively regulated by ThPOK. In CD4 + CD8 lo thymocytes, ThPOK decreases the expression of Runx3 and perforin and promotes the development of the CD4 T cell lineage (Liu et al., 2005). In the periphery, ThPOK also inhibits the expression of Runx3, T-bet and Eomes by mouse CD4 T cells and restricts their cytotoxic differentiation (Wang et al., 2008). We observed that the expression of Runx3 and ThPOK is not mutually exclusive in human CD4 CTX and CD8 CTX T cells. Co-expression of Runx3 and ThPOK has been observed in murine Th1 cells (Djuretic et al., 2007) and in simian MHCII-restricted CD8 aa T cells after CD4 downregulation (Vinton et al., 2017). Also, ThPOK is up regulated by mouse effector CD8 T cells during acute viral infection and promotes their expansion and effector function upon rechallenge (Setoguchi et al., 2009). On the other hand, ThPOK is required for the development of other murine lymphoid subsets with cytotoxic potential including CD4 + NKT cells and gd T cells (Wang et al., 2010;Park et al., 2010). Together, these observations indicate that the expression of high levels of ThPOK is compatible with the expression of Runx3 and with cytotoxic function. Yet, the in vitro model of T cell differentiation revealed that ThPOK is an important regulator of the cytotoxic activity of human CD4 and possibly CD8 T lymphocytes. This observation suggests that ThPOK may contribute to the regulation of antiviral, anticancer and immunopathological properties of Th1 cells in vivo.
In conclusion, this study shows that Runx3 and ThPOK cross-regulate the acquisition of cytotoxic function by Th1 lymphocytes and therefore represent targets for interventions against viral infections, cancer and autoimmune disorders.  The number of samples analyzed in each experiment was defined on the basis of previous experience of the investigators or on published literature. No sample size was computed. Peripheral blood mononuclear cells (PBMC) were purified by density gradient centrifugation and stained with titrated conjugated antibodies. Cells were sorted on a BD FACS Aria III or acquired with a BD LSR Fortessa cytometer and data were analyzed using the FlowJo software (v9.2). The RAW 264.7 murine macrophage cell line (RRID:CVCL_0493; obtained from ATCC) and the HEK-293 human kidney cell line (RRID:CVCL_0045, obtained from ATCC) were cultured in DMEM (Lonza) supplemented with 10% fetal calf serum, 1% AAG, 1% Na Pyruvate and 1% Pen/strep. The HEL-299 fibroblastic cell line (RRID:CVCL_2480; obtained from ATCC) was cultured in EMEM (Lonza) supplemented with 10% fetal calf serum, 1% NEAA, 1% Hepes, 1% Glutamine, 1% Na Pyruvate and 1% Pen/strep. All cell lines were tested negative for mycoplasma infection (MycoAlert, Lonza). Because they were used only as tools to produce lentivirus particles, as targets of cytotoxic cells, and as a negative control in one methylation analysis, they were not re-authenticated after purchase.

FACS-staining
For membrane staining, cells were washed with PBS containing 0.1% bovine serum albumin (BSA). Antibodies were incubated in PBS + 0.1% BSA for 10 min at 37˚C or 15 min at room temperature. Cells were then washed with PBS + 0.1% BSA before addition of Cellfix (BD) or intracellular staining. Cells were permeabilized for intranuclear and intravesicular staining using the FoxP3 staining kit (eBiosciences). Active caspase3 staining was performed using the cytofix-cytoperm and Permwash buffers (BD). References of used antibodies are presented as Supplementary file 2b.

FACS-sorting
Naive CD4 T cells were isolated by negative selection for in vitro stimulation. Before cell sorting, fresh PBMC were enriched in CD4 T cells with the Miltenyi human CD4 +T Cell Isolation Kit. Membrane staining was then performed as mentioned above with a dump channel in PE including CD14, CD19, CD16, CD56, TCRgd and CD8 mAbs. CD25-, CD45RO-and CXCR3-negative cells were further selected in order to exclude regulatory T cells, memory and stem cell memory CD4 T cells, respectively. A small fraction of these untouched naive CD4 T cells (CD3 + CD4 + CD45RO -CCR7 + -CD28 + ) cells were then stained to verify their naive phenotype. Cell purity was 96 [94-97]% (Median [IQ]). CD4 and CD8 T cell subsets were isolated by positive selection. Before cell sorting, fresh PBMC were depleted of glycophorin A-, CD19-and CD14-positive cells as well as CD8-or CD4-positive cells using an Automacs instrument (Miltenyi). Membrane staining was then performed as mentioned above. Cells were resuspended in complete antibiotics-containing medium, sorted and collected in the same complete medium, centrifuged and lysed in RLT Plus buffer +10 ml betamercaptoethanol for later nucleic acid extraction or immediately tested for cytotoxic activity. Cell purity was 98 [95-99]% (Median [IQ]). Single cells were sorted on a FACS Aria III cell sorter (BD) following staining and suspension in an EDTA-containing sorting buffer. Quality of sorting was assessed using the staining index from the DIVA software version 8.0.

Cytotoxicity assay
Effector cells were pre-incubated for 1 hr with or without 100 nM Concanamycin A (Sigma-Aldrich-Merck, Germany). RAW target cells were labelled with PKH-26 (Sigma-Aldrich-Merck, Germany) as previously described (Sheehy et al., 2001). Effector cells were added to 5000 RAW cells at appropriate effector/target ratios and incubated for 4 hr in the presence of 2 mg/ml mouse anti-human CD3 antibody (clone OKT3). RAW cells incubated with the anti-CD3 antibody but without effector cells were used as controls. Percentage of lysis was calculated as the percentage of caspase3-positive RAW cells after subtraction of the % of active caspase3 in the control wells (He et al., 2005).

Quantitative PCR
qPCR was performed using the Taqman RNA Amplification kit or the LightCycler Multiplex RNA Virus Master and a LightCycler 480 instrument (Roche). Raw data were analyzed using the fit points method and fold change was calculated with the Delta-Delta Cp method using the housekeeping gene EEF1A1 (EF1) as a reference. Primers and fluorescent probes were designed using Primer3 and purchased from Eurogentec. A Taqman Gene Expression assay was used for RUNX3 (Hs00231709_m1, Thermo Fisher) and PLZF (ZBTB16, Hs00232313_m1, Thermo Fisher) analyses. Oligonucleotide sequences are presented as Supplementary file 2c.

Bisulphite pyrosequencing
The PRF1 promoter was divided into 11 amplicons covering 34 CpG sites as previously described (Narasimhan et al., 2009). Genomic DNA was bisulphite-converted using the Qiagen FAST Epitect bisulfite kit and sequenced using a Pyromark Q96 device after isolation of single strand biotinylated DNA from the PCR product using streptavidin and a pyromark Q96 vacuum prep station. Quality assessment and methylation level calculation were performed using the software Pyro Q-CpG and the CpG assay 1.0.9 (Biotage).

Nucleic acid material
DNA and RNA were extracted using the Qiagen AllPrep DNA/RNA kit. Concentration and purity were assessed by spectrophotometry (nanodrop -Thermoscientific). Median [interquartile range] of A260/A280 ratios were 1.93 [1.77-2.10] and 1.82 (1.71-1.91) for RNA and DNA samples, respectively. For microarray analyses, RNA integrity number (RIN) was measured using the Eukaryote Total RNA Nano assay and a Bioanalyzer (Agilent). One sample out of 25 had a RIN below seven and was excluded from the analyses.

Gene expression and methylation microarrays
Total RNA was amplified with the Illumina TotalPrep RNA Amplification Kit (Ambion) and hybridized with the HumanHT-12 v4 array containing 47,323 probes for 44,053 annotated genes, according to the instructions of the Whole-Genome Gene Expression Direct Hybridization Assay (Illumina). Chips were scanned with the HiScan Reader (Illumina). For methylation analyses, genomic DNA was bisulphite-converted using an EZ DNA methylation Kit (ZYMO). DNA methylation level was measured using the Illumina Infinium HD Methylation Assay. Bisulphite converted DNA was hybridized with the Illumina HumanMethylation450 BeadChip 450K array (12 samples/chip), as described previously (Dedeurwaerder et al., 2011). Data from both arrays are available on GEO (https://www.ncbi.nlm. nih.gov/geo/) under the accession number GSE75406.
Chromatin immunoprecipitation (ChIP) and ChIP-qPCR MACS-purified CD4 T cells were stained and fixed with 1% formaldehyde. Glycine was added at a final concentration of 0.125 M to quench the crosslinking reaction. Cells were washed twice with icecold PBS and resuspended in complete medium for FACS sorting. Dry pellets of sorted cells were frozen at À80˚C. Thawed pellets were lysed in 1% SDS-containing buffer and sonicated to obtain DNA fragments of 300-800 bp using a Bioruptor device (Diagenode). Chromatin of 200,000 cells was immunoprecipitated with an anti-histone antibody and protein G magnetic-activated beads. Chromatin was incubated overnight at 4˚C with the following antibodies: 1 mg anti-H3K4me3 (Millipore 17-614 rabbit monoclonal), 0.5 mg anti-H3K27ac (abcam ab4729 rabbit polyclonal) or anti-H3 (diagenode C15310135 rabbit polyclonal). 1% of the IP reaction was collected before adding the antibody and the beads and was used as total chromatin input. Beads were washed five times: once with low-salt buffer, once with high-salt buffer, once with lithium chloride containing buffer and twice with TE buffer. After washing and reverse crosslinking (incubation with NaCl 200 mM for 4 hr at 65˚C), chromatin was eluted with a buffer containing 1% SDS and 100 mM NaHCO 3 and treated with RNAse A and Proteinase K for 1 hr at 45˚C. IP-DNA was purified using the MinElute PCR purification kit (Qiagen) and then analyzed by qPCR using the Probe Master 480 kit with primers encompassing different regulatory regions of the perforin locus (sequences are presented as Supplementary file 2d). The DeltaCp method was used to calculate the % of input for each IP. Results were normalized for the DeltaCp of H3-IP DNA.

Singe-cell qPCR assay
Single cells were collected in 5 mL lysis buffer (CelluLyser micro lysis buffer from Tataa biocenter), immediately frozen on dry ice and stored at À80˚C until used (Svec et al., 2013). Reverse transcription (RT) was performed using the CelluLyser Micro Lysis and cDNA Synthesis Kit following manufacturer's instructions (Tataa biocenter). RT step was validated using the Universal RNA Spike (TATAA Universal RNA Spike I from Tataa biocenter) in each well to ensure the absence of inhibitions. Wells most likely to contain single cells were selected on the basis of housekeeping gene expression and exclusion of outliers. cDNA was then pre-amplified for 20 cycles using the TATAA PreAmp Grand-Master Mix kit from the same company. Single-cell qPCR was performed on 43 cells per subset in 96.96 Dynamic Array IFC plates for Gene Expression (BMK-M-96.96) using the fluidigm technology (Biomark HD). Primer sequences are presented in Supplementary file 2e.

Transcription factor knockdown
Lentiviral particles were produced by transient transfection of packaging HEK 293 T cells. pMD2.G and psPAX2 were used as envelop and core packaging plasmids, respectively, together with the gene transfer plasmid (Supplementary file 2f). Before transduction, 50,000 freshly isolated naive CD4 T cells were stimulated during 46 hr in the presence of feeders and PHA (5 mg/ml) in Th1-cytokines containing X-Vivo15 medium (Lonza). For transduction, viral particles were added at a MOI of 10, together with IL-2 and IL-15 in 50 ml of fresh medium. Cells were amplified during 10 days before sorting of GFP + transduced cells for down-stream analysis.

TCR CDR3 sequencing
Purified cDNA (AMPure XP Beads (Agencourt)) was obtained from total RNA and used in templateswitch anchored RT-PCR experiment with specific alpha and beta chain primers. PCR products were then submitted to high-throughput sequencing as previously described (Van Caeneghem et al., 2017). Briefly, V2 300 kit with 200 bp at the 3' end (read 2) and 100 bp at the 5' end (read 1) were used on the Illumina MiSeq platform.

Statistical analyses
Data were analyzed with the GraphPad Prism software unless otherwise specified. After one-way ANOVA analysis of variance, a Mann-Whitney test was performed for selected two-by-two comparisons and a Dunnet's test for multiple comparisons when appropriate. For grouped analysis, we used two-way ANOVA with multiple Tukey's tests. Differences were considered statistically significant at p-values<0.05.

Illumina Expression HT12 Arrays
Raw data were quantile normalized using the normalization method from the lumi package (Du et al., 2008). Unsupervised clustering (Uc) analysis of gene expression datasets was performed using the pvclust package of the R software (R) (Suzuki and Shimodaira, 2006). The robustness of the Uc tree was tested by multiscale bootstrap resampling using Pearson's correlation as distance and Ward.D2 as clustering method, with 1000 iterations. An AU (approximately unbiased) p-value (percentage) was calculated and placed on the nodes of the cluster dendrogram. Principal component analysis (PCA) on the expression dataset was performed using the MultiExperiment Viewer (MeV) tool and the scatter plot function in R. The GeneSign module of the BubbleGUM software (Spinelli et al., 2015) was used with the Min/max method to identify lists of genes specifically expressed in cell subsets. A probe was considered specific of a given cell subset if its minimal expression value across the replicates of the cell subset of interest (test population) was higher than its maximal expression value across the replicates of the cell population used as reference (reference population). To obtain a limited number of genes, probes for which the ratio between the maximal and minimal expression values across all samples was below 1.2 were considered not regulated in any cell subsets and thus excluded from the analysis. Finally, probe lists were transformed into gene lists (or GeneSets). Heatmaps were generated using the heatmap.2 function of gplots package of R.

Illumina HumanMethylation450 BeadChip arrays
Raw data were filtered using a detection p-value<0.05. Cross-reactive probes were filtered out while probes containing SNPs, which do not introduce an important confounder in intra-individual studies, were kept in the analysis as previously detailed. (Dedeurwaerder et al., 2014) b-values were computed using the following formula: b-value = M/[U + M] where M and U are the raw 'methylated' and 'unmethylated' signals, respectively. The b-values were corrected for type I and type II bias using the peak-based correction (Dedeurwaerder et al., 2011;Dedeurwaerder et al., 2014). The differential analyses were applied according to published recommendations (Dedeurwaerder et al., 2014): first the methylation values were converted to M-values using the following formula: M-value = log2 (b-value/(1-b-value)). The statistical significance of the differential methylation was then assessed using a paired t-test applied on these M-values. In parallel, a Delta-b was computed as the absolute difference between the median b-values. Cytosine showing a p-value<1e-4 together with an absolute delta-b>0.1 were reported as differentially methylated. Heatmaps were generated based on the scaled beta values of all the probes located in the promoter region of each represented gene, using the heatmap.2 function of gplots package of R. Promoter regions included 5'prime, TSS1500, TSS200 and 1 st exon.

Gene Set Enrichment Analysis (GSEA)
GSEA was used to analyze the enrichment of GeneSets obtained by GeneSign on the pairwise comparisons of our gene expression microarray data (Subramanian et al., 2005). Catalog c3 from MsigDB was added to our GeneSets for robust statistical analysis. GSEA was used with 1000 Gene-Set-based permutations and 'difference of classes' as a metric for ranking the genes since the expression values were in Log2 scale. The software quantifies enrichments by computing the Normalized Enrichment Score (NES) and the False Discovery Rate (FDR). FDR below 0.25 was considered significant (Subramanian et al., 2005). Genes showing no differential expression between CD8 CTX and CD4 CTX were used as negative control. GSEA was also used to quantify the correlation between gene expression and methylation. The identifiers of the genes included in the GeneSets were transformed to match the DNA methylation probe identifiers (Perl scripts included in Supplementary file 3). GSEA Pre-ranked analysis was then used to assess the enrichment of our expression-based Gene-Sets on the pairwise comparisons of the pre-ranked methylation delta b values. Genesets of catalog c3 from MsigDB were similarly converted to probe identifiers and added to our GeneSet file for robust statistical evaluation.

Single-cell qPCR
Data mining was performed using the Fluidigm Real-time PCR analysis (V1.4.3), and data analysis was performed using the Genex6 MultiD software as previously described (Ståhlberg et al., 2013). Heat Map was created using the standard function in R.

TCR repertoire
Raw sequencing reads from fastq files (both reads) were aligned to reference V, D and J genes specifically for 'TRA' or 'TRB' to build CDR3 sequences using the MiXCR software version 1.7. . CDR3 sequences were then assembled and clonotypes were exported and analyzed using the tcR package (Nazarov et al., 2015). The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Ethics
Human subjects: Volunteers were recruited by the research centre ImmuneHealth, CHU Tivoli, La Louviè re. Clinical staff informed the volunteers about the objectives of the study and obtained their written consent to use the human material for research purposes. The study and the informed consent form were approved, following approval by the Ethics committee of the CHU Tivoli, La Louviè re, Belgium (Reference B09620097253). The study followed the Good Clinical Practice (ICH/GCP) guidelines, the Belgian Law and the declaration of Helsinki ("World Medical Association Declaration of Helsinki; Ethical Principles for Medical Research Involving Human Subjects"). Additional files

Supplementary files
. Supplementary file 1. Lists of genes included in the 12 GeneSets obtained from CD4 versus CD8 T cell comparison ( Figure 2) and naive CD4 T cell versus CMTh1 cell versus CD4CTX T cell comparison ( Figure 4). Genes expressed at higher level by the first listed subset as compared to the subset indicated between brackets were identified using the min/max method. Genes are ranked according to mean log2 fold change calculated using the Limma package in R.