A central role for Notch in effector CD8(+) T cell differentiation.

Activated CD8(+) T cells choose between terminal effector cell (TEC) or memory precursor cell (MPC) fates. We found that the signaling receptor Notch controls this 'choice'. Notch promoted the differentiation of immediately protective TECs and was correspondingly required for the clearance of acute infection with influenza virus. Notch activated a major portion of the TEC-specific gene-expression program and suppressed the MPC-specific program. Expression of Notch was induced on naive CD8(+) T cells by inflammatory mediators and interleukin 2 (IL-2) via pathways dependent on the metabolic checkpoint kinase mTOR and the transcription factor T-bet. These pathways were subsequently amplified downstream of Notch, creating a positive feedback loop. Notch thus functions as a central hub where information from different sources converges to match effector T cell differentiation to the demands of an infection.

The adaptive immune system must simultaneously curb the acute threat of microbial infections and generate immunological memory. Adaptive immunity to intracellular pathogens largely depends on CD8+ T cells. Protective CD8+ T cell responses require expansion of naive antigen-specific cells and their differentiation into cytolytic and cytokine-producing effector cells[1]. Different types of effector cells are generated, responsible for acute protection or generation of memory[1, 2]. At one extreme, a population of terminally differentiated KLRG1+CD127− effector cells exists (terminal effector cells, TECs), which are mostly short-lived[1, 2]. The capacity to differentiate into long-lived memory cells is found predominantly in a KLRG1−CD127+ population, collectively referred to as memory precursor cells (MPCs)[1]. Apart from ability to survive, marked differences exist between these populations. TECs express higher amounts of effector molecules than MPCs[3] and favor migration to non-lymphoid tissues and splenic red pulp. In contrast, MPCs preferentially home to lymph nodes and splenic white pulp[4]. Both lineages express distinct transcriptional programs. The transcription factors Blimp1 and T-bet are critical regulators of the TEC gene expression program[2, 3, 5], whereas Eomesodermin, Tcf-1, FoxO1, Stat3 and Id3 control various aspects of MPC biology[6-12].

TECs and MPCs can derive from a single naive CD8+ T cell or from different precursors[13]. Cells committed to either lineage can be identified among KLRG1−CD127− early effector cells (EECs) within three days after infection[9, 14, 15]. An important question is how differentiating CD8+ T cells choose between effector cell fates. This choice may be affected by asymmetric segregation of fate determining factors[16]. It stands to reason that generation of TECs is proportional to the severity of the infection[17]: greater infectious load requires generation of more fully armed effector cells, without increasing the demand for memory cells. Such proportionality would require instructive signals that relay information regarding the severity of infection. Candidates include inflammatory cytokines, such as interleukin 12 (IL-12) and type I interferons (IFN), help by CD4+ T cells, and strong IL-2 receptor signaling, all of which promote generation of TECs[2, 14, 15, 18, 19]. Given the potentially harmful consequences from erroneous generation of highly cytotoxic TECs, it makes intuitive sense that their generation requires licensing by multiple signals.

The cell surface receptor Notch is a conserved regulator of binary cell fate decisions[20]. Notch responds to membrane bound ligands of the Delta-like (DLL) and Jagged families. Its intracellular domain (NICD) acts as a transcriptional activator after ligand induced cleavage and nuclear translocation[20]. All four mammalian Notch receptors activate a pathway involving recruitment to DNA via RBPJ molecules[21]. Notch reportedly regulates the gene encoding Eomesodermin, a factor required for the formation of stable CD8+ T cell memory[10, 22]. Furthermore, during asymmetric division of naive CD8+ T cells, an inhibitor of Notch signaling called Numb is targeted towards the proximal cell, destined to become a TEC[16]. These findings suggest that Notch may promote MPC and inhibit TEC differentiation. On the other hand, in vitro experiments have implicated Notch in control of the genes encoding IFN-γ, Perforin, Granzyme B and T-bet[22-26], more consistent with a positive role in differentiation of TECs. To establish whether Notch governs the cell fate decision between TEC and MPC, we here used the well-characterized influenza virus infection model and mice with T cell specific genetic deficiencies in the Notch pathway. We find that Notch is an essential hub in a feed forward network to integrate multiple signal inputs and translate these into differentiation of fully protective TECs.

Results

TEC promoting signals induce Notch expression

To study how the degree of viral infection affects effector CD8+ T cell differentiation, we infected mice intranasally with different concentrations of the A/HK×31 (HK×31) influenza strain.influenza strain. Influenza specific CD8+ T cells were identified at the peak of the response (day 10-results not shown) using Db tetramers loaded with the immunodominant 366–374 peptide of the influenza nucleoprotein (H-2 Db–NP)[27]. Increasing viral loads across a 100-fold range resulted in great elevation of TEC numbers, whereas the number of MPCs remained constant (Fig. 1a). This result suggests that signals exist, which couple the severity of the infectious threat to generation of TECs.

Figure 1

TEC promoting signals induce Notch on CD8+ T cells. (a) KLRG1+CD127− (top) and KLRG1−CD127+ (bottom) H-2 Db–NP+CD8+ T cell numbers in spleens of mice infected with different doses of A/HK×31 10 days earlier (3 mice/group). (b) Expression (MFI) of Notch1 (top) and Notch2 (bottom) on naïve CD8+ T cells (white bars) or after 16h co-culture with BM-DC with (black bars, black histograms) or without R-848 (grey bars, grey histograms)(triplicates) (c) Rbpj mRNA expression (relative to Actb) in CD8+ T cells stimulated o/n with anti-CD3 and control (white bar) or R-848 DC sup (black bar) (cumulative of 2 experiments, each in duplicate). (d) Notch1 (top) and 2 (bottom) expression on OT-I CD8+ T cells activated o/n by BMDC and Ovalbumin without (white circles) or with (black circles) R-848. Isotype control staining was subtracted (ΔMFI) (duplicates). (e–i) Notch1 expression on (e) CD8+ T cells stimulated o/n with anti-CD3 and control (white circles) or R-848 DC sup (black circles) (duplicates), (f) wild-type and type I IFN receptor knock-out (IFNAR1ko) CD8+ T cells, unstimulated (US-white bars) or stimulated with control (grey bars) or R-848 DC sup (black bars) (duplicates), (g) CD8+ T cells, unstimulated (white bar) or stimulated as in (f) with (grey bars) or without Rapamycin (black bars) (3 to 6 replicates/group), (h) wild-type (black bars) or Tbx21−/− CD8+ T cells stimulated as in (c) (triplicates), (i) CD8+ T cells stimulated with anti-CD3 and IL-2 (duplicates and quadruplicates). (a,f,h,i) representative of two ,(d,e,g) of three and (b) of more than five experiments. Mean + s.e.m. *P < 0.05; **P < 0.01; ***P < 0.001, two-tailed t-test or one-way ANOVA with Bonferroni corrections.

To examine whether Notch could be involved in this process, we determined whether expression of Notch receptors on CD8+ T cells is regulated by signals known to promote TEC differentiation. Important among these are inflammatory mediators produced by antigen presenting cells (APCs). We incubated naive CD8+ T cells with dendritic cells (DCs) and added the RNA analog R-848, a mimic of RNA viruses. These conditions indeed induced surface expression of Notch1 on the naïve CD8+ T cells within 24 h (Fig. 1b). Expression of Notch2 was only marginally induced. Addition of R-848 to naive CD8+ T cells without DCs did not elevate expression of Notch1, but supernatant from DCs treated with R-848 (R-848 DC sup) did (Supplementary Fig. 1a and b). This induction required the presence of the TLR adapter Myd88 in DC, but not in T cells (Supplementary Fig. 1b and c). Lipopolysaccharide (LPS) similarly induced expression of Notch1 on naïve CD8+ T cells via a Myd88-dependent pathway in DCs (Supplementary Fig. 1b). Thus, TLR activation stimulates DCs to produce soluble factors that in turn induce surface expression of Notch receptors on naïve CD8+ T cells. Such soluble factors also elevated the expression of RBPJ mRNA (Fig. 1c), suggesting general enablement of the Notch pathway.

T cell receptor (TCR)-mediated activation of naive OT-I CD8+ T cells by DCs presenting the Ovalbumin peptide (amino acids 257–264) resulted in modest induction of Notch1 and Notch2 (Fig. 1d). Induction of both these receptors was markedly enhanced by addition of R-848 (Fig. 1d). This induction was not due to improved antigen presentation, as R-848 DC supernatant also enhanced Notch1 expression on CD8+ T cells activated with antibodies to CD3 (Fig. 1e).

To identify the soluble mediators responsible for induction of Notch, we focused on type I IFN. These cytokines are produced by APCs upon recognition of viral nucleic acids and promote differentiation of TECs[15, 19, 28]. CD8+ T cells lacking expression of IFNAR1, the receptor for type I IFN, failed to elevate Notch1 in response to R-848 DC supernatant (Fig. 1f). Thus, CD8+ T cells respond to type I IFN by elevating surface expression of Notch1.

We further tested whether signaling pathways known to regulate differentiation of TECs might control expression of Notch. The rapamycin-sensitive TORC1 complex is required for differentiation of TECs[29-31] and is activated by type I IFN receptor signaling[32]. Rapamycin almost abrogated induction of Notch1 expression on CD8+ T cells by R-848 DC supernatant (Fig. 1g). Differentiation of TECs also depends on T-bet[2]. Expression of this factor was induced in naive CD8+ T cells by R-848 DC supernatant (results not shown) and is partially responsible for induction of Notch1 expression, as Notch1 surface expression was reduced in Tbx21−/− T cells (Fig. 1h). Finally, strong IL-2 receptor signaling enhances TEC differentiation[14, 18] and IL-2 also synergized with TCR stimulation to elevate expression of Notch1 (Fig. 1i).

Consistent with a role in driving TEC differentiation, Notch abundance was also transiently elevated in vivo on early KLRG1+ effector cells compared to KLRG1− effector cells (Supplementary Fig. 2). Thus, promoters of TEC differentiation also induce receptivity of the Notch pathway in CD8+ T cells. Convergence of all these signals on expression of Notch suggests that perhaps Notch plays a role in guiding the decision between TEC and MPC differentiation.

Notch ligands are induced on APC by viral infection

Two major populations of APCs in lungs are CD11chiMHCIIint alveolar macrophages and CD11chighMHCIIhigh migratory DC (characterized by intermediate and large size, respectively) (Fig. 2a, middle and Supplementary Fig. 3a)[33, 34]. These migratory DCs (mDCs) were shown to carry influenza antigen to the mediastinal lymph nodes and are responsible for priming naive CD8+ T cells through direct presentation[34, 35]. Both mDCs and alveolar macrophages expressed undetectable amounts of Dll4 and Jagged2 and low amounts of Dll1 and Jagged1 in lungs from uninfected mice (Fig. 2a, bottom). Expression of Dll1 and Jagged1 was increased on both types of APCs upon infection with influenza virus (Fig. 2a, bottom). This elevated expression persisted on lung-derived migratory DCs (but not lymph node resident macrophages) isolated from the lung-draining mediastinal lymph nodes from infected mice (Fig. 2b and Supplementary Fig. 3b). Yet, Notch ligands were virtually undetectable on other populations of MHCII+ cells present in lungs, even upon infection with influenza (Supplementary Fig. 3c). Therefore, expression of Notch ligands is induced by influenza infection on the APC that prime naive CD8+ T cells, positioning these molecules for a potential role in differentiation of virus specific CD8+ T cells.

Figure 2

Notch ligands are expressed on APC during influenza infection. C57BL/6 mice were infected intranasally with A/HK×31 influenza. (a) Five days post infection (p.i.) expression of DLL1, DLL4, Jagged1 and Jagged2 expression on lung APC subsets was measured by flow cytometry. Two main APC subsets in the lung were defined as mDCs (migratory DC, MHCIIhiCD11chi, red gate) or macrophages (Alveolar Mf, SSChiMHCIIintCD11chi, green gate). Filled histograms represent staining with isotype control mAb (Isotype co.), grey lines show Notch ligand expression on APC isolated from non-infected lungs (Uninfected co.), and black lines indicate expression on APC isolated from infected lungs (influenza infected). (b) Mean Fluorescence Intensity of DLL1, DLL4, Jagged1 and Jagged2 on mDC isolated from the lung (left) or isolated from the draining mLN (right). Shown is Δ-MFI (corrected for background staining with isotype control mAb) for non-infected (white bars) and infected mice (black bars). Results represent 2 (uninfected) or 4 (infected) separately processed mice from a representative of 5 experiments. Mean + s.e.m. *P < 0.05, two-tailed t-test.

Notch controls acquisition of effector function

The above results show that the Notch signaling module is assembled on both sides of the T cell-APC interface under conditions favoring differentiation of TECs. To investigate whether Notch controls differentiation of these cells, we used mice lacking expression of Notch genes in the T cell lineage. Notch1 and Notch2 are both expressed in CD8+ T cells (Fig. 1), whereas expression of Notch3 and Notch4 is undetectable[26]. We therefore generated mice lacking expression of both Notch1 and Notch2 by crossing a Cd4-Cre transgene with floxed alleles of the Notch1 and Notch2 genes. Notch1flox/floxNotch2flox/floxCd4-Cre mice (Notch1-2-KO mice) lack expression of both Notch receptors in mature CD4+ and CD8+ T cells (results not shown). Thymic T cell development was not overtly affected in these mice[36].

Upon infection with HK×31, similar numbers of H-2 Db–NP-binding CD8+ T cells were found in blood, spleens or lungs from wild-type and Notch1-2-KO mice at the peak of the response (Fig. 3a–c). However, the proportion of influenza specific CD8+ T cells was consistently higher in mediastinal lymph nodes from Notch1-2-KO than from wild-type mice (Fig. 3d). Also, Notch1-2-KO CD8+ T cells preferentially localized to sites in spleens not accessible to antibodies injected into the blood stream, suggesting localization in the white pulp (Fig. 3e)[37]. Although the proportion of IL-2 producing cells in response to NP366–374 peptide was similar between wild-type and Notch1-2-KO CD8+ effector T cells, production of IFN-γ or tumor necrosis factor (TNF) were reduced in the latter (Fig. 3f). Production of the cytolytic effector molecule granzyme B was almost undetectable in Notch1-2-KO CD8+ T cells and mRNA abundance for both granzyme B and perforin were reduced in these cells (Fig. 3g). Low expression of these molecules translated into reduced cytolytic effector function in an in vivo cytolysis assay. Whereas nearly all CFSElo NP peptide–loaded splenocytes were killed by pre-infected wild-type mice, a proportion of peptide carrying cells could still be detected in infected Notch1-2-KO mice (Fig. 3h). As expected from these defects in effector function, viral clearance and weight recovery were compromised in Notch1-2-KO mice infected with the aggressive A/PR/8/34 influenza strain (Fig. 3i,j). Although Notch1-2-KO mice also lack expression of Notch in CD4+ T cells, defective viral clearance in these mice was not caused by ineffective help to B cells, as titers of neutralizing antibodies to A/PR/8/34 were similar in both wild-type and Notch1-2-KO mice (Fig. 3k). Thus, Notch is required for generation of fully functional CD8+ effector T cells.

Figure 3

Notch is required for effector function. Mice were infected with A/HK×31 and analyzed after 10 days. (a) Flow cytometry profiles and (b) frequencies of H-2 Db–NP+ cells of CD8+ T cells in blood from (15) WT (grey bars) and (10) Notch1-2-KO (white bars) mice (3 pooled experiments). (c) Frequencies (left) and numbers (right) of H-2 Db–NP+ cells of CD8+ T cells in spleens and lungs (2 mice per genotype, representative of over 10 experiments), or (d) mediastinal LN (10 wild-type and 11 Notch1-2-KO mice/group from 3 experiments). (e) Percentage H-2 Db–NP+ of CD8+ T cells in spleens unstained by intravenous anti-CD8 (11 mice/group from 3 experiments). (f) Percentages NP366–374 stimulated lung CD8+ T cells producing IL-2, IFN-γ and TNF (14 or 15 mice/genotype from 4 experiments) or (g) Granzyme-B (WT-black line; Notch1-2-KO-grey line) (representative of 4 experiments, minimally 3 mice/group). (right) Granzyme B (Gzmb) and perforin (Prf1) mRNA in splenic H-2 Db–NP+CD8+ T cells (RNAseq-average expression of 3 experiments, each 3 pooled mice/group). (h) In vivo cytolysis assay. Uninfected (UI-left) or A/HK×31 infected (10 days earlier) WT (middle) or Notch1-2-KO mice (right) were injected with syngeneic splenocytes labeled with NP366–374 peptide and low CFSE or without peptide and high CFSE (1:1 ratio). Numbers represent percentage of cells per CFSE peak in spleens 4h after injection (representative of 4 mice/group). (i–k) 10 days after 200xTCID50 A/PR/8/34 infection. Symbols represent individual mice (closed-WT; open-Notch1-2-KO). (i) Lung viral loads. (j) Relative weights (normalized to day 0) day 0–7 n=12; day 8 n=8; day 9–10 n=4. (k) Neutralizing antibodies to A/PR/8/34 in sera. Mean + s.e.m. *P < 0.05; **P < 0.01, (d,e,f) two-tailed t-test; 2-tailed Mann Whitney test (i,k); two-way ANOVA, Sidak's multiple comparisons test (j)).

Notch controls differentiation of TECs

High expression of effector molecules and residence outside lymph nodes and splenic white pulp are characteristics of TECs. In contrast, MPCs preferentially localize inside these locations and tend to exclusively produce IL-2 (refs.[3, 4, 37]). Our results are therefore consistent with a defect in the TEC compartment and an increase in MPCs in Notch1-2-KO mice. Indeed, hardly any KLRG1+CD127− CD8+ TEC T cells were detectable in Notch1-2-KO mice at the peak of the response, whereas numbers of KLRG1−CD127+ CD8+ MPC T cells were increased (Fig. 4a). TECs were also absent in Notch1-2-KO mice at earlier and later time points, demonstrating that their absence did not reflect altered response kinetics (Fig. 4b). Partial loss of TECs was observed in mice lacking expression of only Notch1 or Notch2 (Fig. 4c) or mice that were heterozygous for both these genes (Supplementary Fig. 4). These results show that Notch1 and Notch2 act redundantly and that the presence of TECs is sensitive to Notch gene dosage. Furthermore, TECs were virtually non-existent when the canonical Notch effector RBPJ was deleted in T cells (Fig. 4d and Supplementary Fig. 5). Dependence on Notch was not unique to the response to HK×31, as no TECs were detectable after infection with the A/PR/8/34 influenza strain (Supplementary Fig. 6).

Figure 4

Notch is required for TEC differentiation. (a–f) Mice were infected with A/HK×31. (a) Flow cytometry plots for KLRG1 and CD127 on splenic H-2 Db–NP+CD8+ T cells after 10 days. (bottom) average frequencies KLRG1+CD127− TECs (left) and KLRG1−CD127+ MPCs (right) of CD8+ T cells (WT-grey bars; Notch1-2-KO -white bars) (3 mice/group). (b) Time course percentages CD127− KLRG1+ TECs among H-2 Db–NP+CD8+ T cells in blood (closed symbols-WT; open symbols- Notch1-2-KO) (3 to 20 mice per data point). (c) Percentages KLRG1+CD127− TECs in blood of wild-type (black bar), Notch1-2-KO (dark grey bar), Notch1 (light grey bar) or Notch2 (white bar) single knockout mice (minimally 6 mice/group from 3 experiments) or (d) RBPJ knockout mice, 10 days after infection (3 mice/group). (e,f) Mixed chimeras of CD45.1+ WT + CD45.2+ WT (left) or CD45.1+ WT + CD45.2+ Notch1-2-KO bone marrow were made and analyzed 10 days after infection. (e) CD45.1 versus KLRG1 on H-2 Db–NP+CD8+ T cells. (f) Percentages KLRG1+ H-2 Db–NP+CD8+ T cells (black bars-CD45.1+ WT with CD45.2+ WT; white bars-CD45.1+ WT with CD45.2+ Notch1-2-KO; grey bar-CD45.2 Notch1-2-KO) (8 or more mice/group from 3 experiments). (g) Percentages KLRG1+ OT-I T cells from Notch1flox/floxNotch2flox/flox mice, expressing control (white bars) or hCre (black bars) retrovirus, 5 days after transfer into OVA-influenza infected mice. Untransduced (UT) and transduced (TD) T cells from the same cultures are shown (3 mice/group). (h) Percentages KLRG1+ Notch1-2-KO OT-I T cells, expressing control (white bars-EV) or NICD1 (black bars) retrovirus, 5 days after transfer into OVA-influenza infected mice (2 mice per group). Mean + s.e.m. *P < 0.05; **P < 0.01; ***P < 0.001, two-tailed t-test or one-way ANOVA with Bonferroni corrections. (d) representative of two, (e,g,h) of three and (a) of more than ten experiments.

To test whether defective development of KLRG1+CD127− CD8+ T cells in Notch1-2-KO mice depends on a direct role for Notch in CD8+ T cells, we generated mixed bone marrow chimeras and infected these with HK×31 influenza. Notch deficiency did not affect the efficacy of reconstitution (data not shown). However, CD45.2+ Notch1-2-KO CD8+ T cells failed to produce KLRG1+CD127− CD8+ T cells even in the presence of CD45.1+ wild-type cells (Fig. 4e,f), showing that TEC differentiation depends on a CD8+ T cell intrinsic role for Notch. Likewise, physiological numbers of Notch1-2-KO OT-I TCR transgenic CD8+ T cells failed to generate TECs after transfer into mice infected with the WSN-OVAI influenza strain, which expresses the H2Kb-restricted OVA257–264 peptide (Supplementary Fig. 7). These data show that Notch controls TEC development in a CD8+ T cell–intrinsic manner and independent of influenza strain and TCR specificity.

As developmental defects could potentially indirectly perturb TEC differentiation in Notch1-2-KO mice, we inactivated Notch1 and Notch2 expression in mature CD8+ T cells. To this end, we retrovirally introduced Cre recombinase in Notch1flox/floxNotch2flox/flox OT-I CD8+ T cells and transferred these into mice infected with the WSN-OVAI influenza strain (for full expansion in vivo-results not shown). Expression of Cre abrogated development of KLRG1+ cells (Fig. 4g). Vice versa, introduction of NICD1, the constitutively active intracellular domain of Notch1, restored the ability of Notch1-2-KO OT-I T cells to generate KLRG1+ cells (Fig. 4h). We conclude that Notch1 and Notch2 act directly in mature CD8+ T cells to drive the development of cells phenotypically resembling TECs.

Notch controls the gene expression signature of TECs

To examine whether Notch is required for adoption of full TEC identity, we determined the global gene expression profiles of Notch1-2-KO and wild-type CD8+ effector cells by whole transcriptome RNA sequencing. To this end, H-2 Db–NP366–374 binding CD8+ T cells were sorted by flow cytometry from mice infected with HK×31 influenza 10 days earlier. This population consists for well over 99% pure effector cells, as evidenced by their CD44+ phenotype (data not shown).

Gene set enrichment analysis for immunological signatures was performed, using the c7 gene set from the Molecular Signature Database (http://www.broadinstitute.org/gsea/msigdb/), to obtain an unbiased view of the types of processes affected by Notch deficiency. Consistent with a role for Notch in control of TEC differentiation, nine of the top ten most significantly enriched gene sets pertained to CD8+ effector T cell differentiation in general (labeled in yellow in Supplementary data set S1), and differentiation of TECs and MPCs in both viral and bacterial models in particular (labeled in red in Supplementary data set 1).

To more specifically determine whether the TEC program is controlled by Notch, we generated molecular definitions of TECs and MPCs by whole transcriptome sequencing of sorted KLRG1−CD127+ and KLRG1+CD127− H-2 Db–NP specific CD8+ T cell populations from wild-type mice. We found 132 TEC specific genes, whereas another 250 genes were preferentially expressed in MPCs (Supplementary data set 2). Assignment of these identities to the genes with differential expression between wild-type and Notch1-2-KO H-2 Db–NP specific CD8+ T cells showed that a large proportion of Notch dependent genes (higher in wild-type) was TEC specific (labeled in green in Fig. 5a). Indeed, more than 40% of the TEC specific transcriptome was reduced in Notch1-2-KO CD8+ effector T cells (green bar on left - Fig. 5b). Conversely, less than 2% of MPC genes were reduced in Notch1-2-KO CD8+ effector T cells (magenta bars - Fig. 5b). Instead, expression of a large proportion (36%) of MPC specific genes was elevated (magenta in Fig. 5a,b). These results strongly suggest that Notch promotes TEC fate and may inhibit formation of MPCs.

Figure 5

Notch regulates the TEC transcriptome. RNAseq was performed on splenic H-2 Db–NP+CD8+ T cells from WT and Notch1-2-KO mice 10 days post-infection with A/HK×31. (a) Mean expression versus log2FoldChange (Notch1-2-KO/wild-type) plot of differentially expressed (DE) genes. TEC-(green) and MPC-specific genes (magenta) were determined by RNAseq of CD8+ H-2 Db–NP+ KLRG1+CD127− TECs and KLRG1−CD127+ MPCs from WT mice (Supplementary dataset 2). Grey, magenta and green dots represent significantly (adjusted P-value < 0.05) DE genes comparing WT versus Notch1-2-KO. Also shown are TEC- or MPC-specific genes not significantly DE between wild-type and Notch1-2-KO (light colored dots). (b) Percentages of TEC-specific (green bars) or MPC-specific genes (magenta bars), significantly (padj<0.05) reduced (left) or induced (right) in Notch1-2-KO H-2 Db–NP+CD8+ T cells. (c) GSEA Enrichment Map (using GO terms for Biological Pathways-see Supplementary dataset 3 for enriched GO-terms). Node core and periphery represent gene set enrichment in the comparison of Notch1-2-KO versus WT (core) or MPC versus TEC (periphery). Node color represents significance of enrichment (magenta-enriched, white-not enriched). Connecting line length represents similarity between gene sets and thickness the number of genes shared. DE genes with an FDR of <0.5 were selected for GSEA. Gene categories shown are enriched at an FDR<0.1. (d–i) Log2-fold ratio of expression between MPC/TEC (dark blue bars) and Notch1-2-KO/WT (light blue bars) for (d) transcription factors, (e) chemokine receptors, (f) adhesion molecules, (g) S1PR family members, (h) cytolytic effector molecules and (i) killer cell lectin-like Receptor genes. (e,f) restricted to genes with minimally 2-fold difference between MPC vs TEC or Notch1-2-KO vs WT (note: CD44 and Sell included for completion). RNAseq data available at Array Express (E-MTAB-2255). Cumulative from 3 experiments, each with 3 pooled mice/genotype.

As Notch controls expression of many, but not all, TEC signature genes (Fig. 5b), we examined whether Notch selectively controls specific functional categories. To this end, a side by side analysis was made of the Gene Ontology (GO) terms (Biological Pathways) enriched in the comparison of the wild-type versus Notch1-2-KO transcriptome and those enriched in the comparison of the MPC versus TEC transcriptome. We focused on the major functional categories, which define the difference between MPCs and TECs, including those associated with effector function, migration, viability, activation and differentiation (Fig. 5c). The near complete overlap of enriched Gene Ontology terms (shown by the red color of both the core and periphery of each node) revealed a representation of Notch-dependent genes in all these categories (Fig 5c). Thus, Notch does not control individual aspects of TEC differentiation, but exerts broad control throughout the TEC gene expression program. This conclusion was further supported by a comparison of the relative expression patterns of specific genes encoding transcription factors (involved in effector cell differentiation), chemokine receptors, adhesion molecules, cytotoxic effector molecules or Killer Lectin-like Receptors. Expression of TEC-specific genes across all these categories was reduced in Notch1-2-KO effector cells (Figs. 5d–i and Supplementary data set 3, compare expression ratios of MPC/TEC in dark blue and knock out/wild-type in light blue). Although some genes were either unaffected or only weakly affected, no TEC gene was significantly elevated in Notch1-2-KO effector cells. Reciprocally, expression of MPC specific genes was generally elevated in Notch1-2-KO effector cells or unchanged, but never reduced (Figs. 5d–i). Together, these results clearly show that Notch is a major regulator of the decision between TEC and MPC differentiation.

Notch feeds back onto TEC promoting pathways

Early TECs are characterized by high expression of CD25 (the IL-2 receptor α chain) and IL-2 receptor signaling promotes differentiation of TECs[14, 18]. CD25 is induced on CD8+ T cells by constitutively active Notch[38], suggesting that this IL-2 receptor component is a downstream target of Notch. Indeed, expression of Il2ra mRNA was reduced in Notch1-2-KO effector CD8+ T cells (Fig. 6a) as was CD25 expression on Notch1-2-KO OT-I CD8+ T cells at early stages of the response (Fig. 6b), when differentiation takes place.

Figure 6

Notch regulates pathways controlling TEC differentiation. (a) Il2rα mRNA in WT (grey bar) and Notch1-2-KO (white bar) cells determined as in Figure 5. (b) CD25 surface profile (top) and average MFI (7 mice/group, from 2 experiments) on transferred wild-type (black line, grey bar) or Notch1-2-KO (grey line, white bar) OT-I CD8+ T cells, 5 days post-infection with OVA-influenza. (c) Flow cytometry profiles (left) and average ΔMFI (right-isotype control background subtracted) for phosphorylated AKT (T308) and mTOR (S2448) in transferred wild-type (black line, grey bars) and Notch1-2-KO (grey line, white bars) OT-I T cells or endogenous CD44−CD8+ T cells (black bars), 5 days post-infection with OVA-influenza (pooled from 6 mice/group, 2 experiments, representative of 3). (d) KLRG1 expression on Notch1-2-KO OT-I T cells expressing control (white bars) or myr-AKT (black bars) IRES-GFP retrovirus, 5 days after transfer into OVA-influenza infected mice. Minimally 5 mice/group, 2 pooled experiments. (e) Percentages KLRG1+CD127− TECs in blood of wild-type (black bar) and Tbx21−/− (white bar) mice, 10 days after A/HK×31 infection (minimally 7 mice/genotype). (f) Tbx21 mRNA expression (relative to Actb) in wild-type (grey bar) or Notch1-2-KO (white bar) splenic H-2 Db–NP+CD8+ T cells, 10 days after A/HK×31 infection (4 mice/group, 2 experiments). (g) KLRG1 expression on wild-type (left) and Tbx21−/− (right) P14 T cells, expressing control (white bars) or NICD1 (black bars) IRES-GFP retrovirus, 5 days after transfer into gp33–41-influenza infected mice (minimally 5 mice per group, 2 experiments). (h) KLRG1 expression on Notch1-2-KO OT-I T cells, expressing control (white bars) or T-bet (black bars) encoding retrovirus, analyzed as in (g) (2 mice/genotype, representative experiment of 3). Shown is the mean + s.e.m. *P < 0.05; **P < 0.01; ***P < 0.001; unpaired, two-tailed t-test or One-way ANOVA with Bonferroni corrections.

The Akt-mTOR pathway, an important regulator of TEC differentiation[29–31, 39, 40], is activated by IL-2 receptor signaling and functions downstream of Notch in thymocytes and T cell leukemia[41, 42]. We therefore examined whether Notch controls activity of this pathway during CD8+ T cell responses. Activation of Akt requires phosphorylation of the 3-phosphoinositide-dependent kinase 1 (PDK1) substrate site T308 and this modification positively correlates with high CD25 expression in early CD8+ effector T cells in vivo[39, 40]. Phosphorylation of Akt T308 was clearly detectable in OT-I effector cells isolated from mice five days after infection with WSN-OVAI influenza (Fig. 6c, top). In contrast, in Notch1-2-KO OT-I T cells, phosphorylation of this residue was reduced to background found in unactivated CD44−CD8+ T cells (Fig. 6c, top bar graph). Similar results were obtained for Akt-mediated phosphorylation of mTOR (S2448) and Foxo1/3 (Thr24/Thr32), which were both reduced to background in Notch1-2-KO OT-I cells (Fig. 6c, bottom bar graph; Supplementary Fig. 8a). Of note, expression of constitutively active Akt[41] in Notch1-2-KO OT-1 T cells was sufficient to restore differentiation of a KLRG1+ population (Fig. 6d). These results support the conclusion that the Akt-mTOR pathway functions downstream of Notch in differentiation of TECs.

T-bet is an important transcriptional regulator of TEC development[2] (Fig. 6e). The Tbx21 gene (encoding T-bet) is a direct target of Notch in CD4+ T cells[24, 25] and correspondingly, the abundance of Tbx21 mRNA was also reduced in Notch1-2-KO effector CD8+ T cells (Fig. 6f). Genetic deficiency for T-bet largely abrogated the ability of retrovirally expressed NICD1 to induce development of KLRG1+ effector CD8+ T cells (Fig. 6g), whereas ectopic expression of T-bet restored differentiation of such cells in Notch1-2-KO effector CD8+ T cells (Fig. 6h). These results strongly suggest that T-bet is a downstream effector of Notch in driving differentiation of TECs.

In conclusion, expression of the IL-2 receptor α chain and T-bet as well as activity of Akt and mTOR all depend on Notch. As these factors also act upstream of Notch by inducing its expression on naïve CD8+ T cells (Fig. 1), these results show that Notch functions as a central hub in a network, which integrates signal input from different sources (inflammation, IL-2, Notch ligands) to induce generation of TECs (Supplementary Fig. 8b).

Discussion

How activated CD8+ T cells decide between committing to the memory or the terminal effector cell lineage represents an intriguing conceptual question with implications for the vaccine design. Important is the requirement to balance effective rejection of invading microorganisms and the risk of immunopathology. Given their destructive potential, generation of TECs must be regulated by signals that shape the response to the threat posed by the infection. Indeed, inflammatory signals promote the development of TECs[2, 15, 17, 19]. We now show that Notch bridges signaling by innate pattern recognition receptors and inflammation to TEC development. Type I IFN induce expression of Notch receptors and RBPJ on CD8+ T cells, while TLR and inflammatory cytokines induce expression of Notch ligands on APC[43, 44]. Thus, microbial and inflammatory stimuli activate both sides of the T cell-APC interface to assemble the Notch signaling module. Notch receptors and ligands never reach very high surface levels. However, it is important to note that these levels are often actively kept low, for instance by internalization[45, 46], a property that may promote dosage sensitivity of the pathway[47]. TEC differentiation is indeed sensitive to Notch dose, as shown in mice lacking Notch1 or Notch2 genes individually or heterozygous mice for both. Importantly, combined induction of receptors, ligands and the downstream mediator RBPJ likely results in cumulative amplification of signal amplitude.

Our results show that Notch, IL-2 receptor, mTOR and T-bet together form a positive feedback module, which integrates signals from various sources. These signals may signify the detection of living microorganisms (type I IFN), direct interaction of the APC with those micro-organisms (Notch ligands) and perhaps help by CD4+ T cells (IL-2). Such a system requiring integration of multiple signals is presumably less prone to spuriously develop the potentially destructive TECs than one depending on an individual signal. Together, these factors (and possibly yet others) could serve as a licensing code, which allows full differentiation of TECs only when justified by the severity of the infection. The fact that delivery of all these signals together sets up a positive feedback loop, presumably ensures maximal separation between the TEC and MPC fates.

Notch and Akt promote survival in multiple cell types[48,49], but this does not appear to explain their roles in effector CD8+ T cells. Activation of Akt in fact reduces survival of CD8+ effector T cells in vivo, presumably by promoting differentiation of the short lived TEC fate[39, 50]. Likewise, our results are not consistent with a major role for Notch in survival of TECs. First, Notch deficiency does not affect the magnitude of the H-2 Db–NP366–374 specific response, inconsistent with loss of a major population. Second, if Notch deficiency compromised TEC survival, one would expect loss of the entire TEC signature in Notch1-2-KO mice. Instead, we find that approximately half of the TEC gene expression signature is unaffected by N1/2 deficiency. Furthermore, there are marked differences in the degree to which different TEC genes are affected. These data strongly suggest that Notch controls TEC differentiation by regulating a large proportion of the TEC specific gene expression program. In this program, some genes rely more heavily on Notch than others. Yet, in the absence of Notch, the program as a whole collapses, resulting in a severe differentiation defect. Direct Notch target genes have been identified with dedicated roles in effector CD8+ T cells. These include the genes encoding granzyme B, perforin, IFN-γ and T-bet[22, 24–26]. Furthermore, Notch controls expression of a critical component of the IL-2 receptor as well as the amplitude of Akt and mTOR signaling during effector CD8 T cell differentiation. All of these promote differentiation of effector CD8+ T cells and regulate additional transcription factors involved in this process, including STAT5, Foxo and HIF1 proteins[7, 39, 40, 51–53]. Thus, besides directly regulating key effector genes, Notch mobilizes additional signaling pathways and thereby coordinates an extensive differentiation program. Still, the fact that approximately half the TEC specific transcriptome is unaffected by the absence of Notch shows that the TEC differentiation program is modular. It seems likely that the other module(s) are controlled by additional signals activated by inflammation.

Loss of part of the TEC signature in Notch1-2-KO mice is mirrored by elevation of roughly a third of the MPC specific transcriptome. This unlikely reflects a relative increase in MPC numbers due to the absence of TECs, as that would result in elevation of the entire MPC transcriptome. The simplest interpretation of these data is therefore that, apart from inducing a TEC gene expression program, Notch represses part of the MPC program. This program may further be suppressed by additional TEC promoting signals and/or may require provision of MPC promoting signals, such as Wnts and activators of STAT3, including IL-10 and IL-21 (refs.[6, 11, 12]). Nonetheless, our results do suggest that avoiding (strong) Notch signals is a condition to allow differentiation into the MPC lineage. How and where this condition exists during an infection will be an important question for future research.

Methods

Mice

All mice were on a C57BL/6 background.

Notch1flox/floxNotch2flox/floxCd4-Cre mice or Rbpjflox/floxCd4-Cre mice were used[36, 43]. Cre-negative littermates were used in all experiments. Transgenes for the OT-I TCR (003831), P14 TCR (004694) as well as knock out mice for Myd88 (009088), IFNAR1 (032045) and Rag1 (002216) are all available from Jackson Laboratories. Mice were bred and housed in specific pathogen-free conditions at the Animal Centers of the Academic Medical Center (AMC, Amsterdam, The Netherlands), Mount Sinai School of Medicine (New York, NY, USA) and Yale University, School of Medicine (New Haven, CT, USA). Mice (both male and female) were between 8–16 weeks of age at the start of the experiment. During infection experiments, wild-type and Notch1-2-KO mice were housed together to avoid cage bias. No intentional method for randomization was used. No formal method for blinding was used, except for determination of viral loads and hemagglutination assay, where the operator did not know mouse genotypes. Mixed-bone marrow (BM) chimeras containing wild-type and Notch1-2-KO BM at a 1:1 ratio were generated via intravenous injection of 5–10 × 106 donor BM cells into lethally irradiated RAG1-deficient mice. Wild-type and Notch1-2-KO cells of donor origin were identified with the congenic CD45.1/2 markers. BM chimeras were used at 12 weeks after engraftment. All mice were used in accordance of institutional and national animal experimentation guidelines. All procedures were approved by the local Animal Ethics Committees.

Media, reagents and mAbs

Culture medium was Iscove’s modified Dulbecco’s medium (IMDM; Lonza) supplemented with 10% heat-inactivated FCS (Lonza), 200 U/ml penicillin, 200 µg/ml streptomycin (Gibco), GlutaMAX (Gibco) and 50 µM β-mercaptoethanol (Invitrogen) (IMDMc). All directly conjugated monoclonal antibodies used for flow cytometry were purchased from eBioscience, San Diego, CA, unless stated otherwise: anti-CD3ε (clone 145-2C11), anti-CD4 (clone GK1.5), anti-CD8α (Ly-2, clone 53-6.7), anti-CD8b (Ly-3, clone eBio341), anti-CD25 (anti-IL2Rα, clone 7D4), anti-CD28 (clone 37.51), anti-CD44 (clone IM7), anti-CD45.1 (clone A20, BD Biosciences), anti-CD45.2 (clone 104), anti-CD62L (clone MEL-14), anti-CD69 (clone H1.2F3), anti-CD127 (anti-IL7Rα, clone A7R34), anti-DLL1 (clone HMD1-5), anti-DLL4 (clone HMD4-1), anti-Granzyme B (clone GB-11, Sanquin PeliCluster), anti-IL-2 (clone JES6-5H4), anti-IFN-γ (clone XMG1.2), anti-Jagged1 (clone HMJ1-29), anti-Jagged2 (clone HMJ2-1), anti-KLRG-1 (clone 2F1), anti-Notch1 (clone HMN1-12, Biolegend), anti-Notch2 (clone HMN2-35, biolegend) and anti-TNFα (clone MP6-XT22), anti-p-AKT [T308] (cat. # 13038) and p-mTOR [S2448] (cat. # 5536), anti-FoxO1/3 [T24/T32] (cat.# 2599S); isotype control (cat. #3900S) (Cell Signaling Technology).

Influenza infection

Mice were intranasally infected with 100–200 × 50% tissue culture effective dose (TCID50) of the H3N2 influenza A virus HK×31[27], influenza A/WSN/33, A/WSN/33–OVA(I)[54], A/PR/8/34 (H1N1) or the recombinant A/PR/8/34 expressing the LCMV gp33–41 epitope[55]. Stocks and viral titers were obtained by infecting MDCK or LLC-MK2 cells as described previously[56].

At indicated time intervals, blood samples were drawn from the tail vein or mice were sacrificed and organs were collected to determine numbers of influenza-specific CD8+ T cells. Influenza-specific CD8+ T cells were enumerated using anti-CD8 (53-6.7) and PE- or APC-conjugated tetramers of H-2Db containing the influenza-A-derived nucleocapsid protein (NP) peptide NP366–374 ASNENMETM (produced at the Sanquin Laboratory for Blood Research). A/PR/8/34 viral loads in lungs of infected mice were determined by isolating lung mRNA and detection of viral mRNA by quantitative PCR using the following primers and probe specific for the A/PR/8/34 M gene. Sense primer: 5’-CAAAGCGTCTACGCTGCAGTCC-3’; antisense primer: 5’-TTTGTGTTCACGCTCACCGTGCC-3’; Probe: 5’-AAGACCAATCCTGTCACCTCTGA-3’.

Sera were tested for the presence of neutralizing antibodies to this virus by hemagglutination inhibition (HI) assay as described previously using four hemagglutinating units of virus and turkey erythrocytes[57]. Values represent the maximum serum dilution at which agglutination was completely inhibited.

In vitro CD8+ T cell activation

Single cell suspensions were prepared by grinding organs over 70 µm nylon sieves (BD Biosciences). Contaminating red blood cells were removed by hypotonic lysis (155 mM NH4Cl, 10 mM KHCO3, and 1 mM EDTA), and cell counts were determined by an automated cell counter (CasyCounter, Innovatis). CD8+ T cell purification from spleen and peripheral lymph nodes was performed by MACS in accordance with manufacturer’s directions (Miltenyi Biotec) and naïve CD44−CD62L+ CD8+ T cells were sorted by flow cytometry. To activate CD8+ T cells in vitro, cells were cultured for 16–18 h in the presence of indicated concentrations of OVA-protein and CD11c-MACS purified splenic DCs (T:DC ratio was 1:1) or plate bound anti-CD3 (145-2C11) and soluble anti-CD28 (37.51) (50,000 T cells/well in 96-well plate). BMDC-conditioned supernatant was generated by stimulating GM-CSF–derived BMDCs in the absence or presence of 10 µg/ml R-848 or 100 ng/ml LPS. After overnight culture, BMDC-conditioned supernatant was isolated, centrifuged and passed through a 0.45 µM filter to remove remaining BMDCs and added to CD8+ T cell cultures for 16–18 h.

Dendritic cell flow cytometry

Single cell suspensions were prepared by cutting lungs into small fragments followed by enzymatic digestion. For this, 1 ml IMDMc with 100U/ml collagenase type IV (Sigma-Aldrich) was added per lung lobe for 60 min incubation at 37 °C. Cold IMDMc containing 10 mM EDTA and 20 mM HEPES pH 7.5 was then added, and cells were filtered though 70 µm nylon sieves (BD Biosciences). Red blood cells were lysed by hypotonic lysis. Fc-receptors were blocked with anti-CD16/CD32 for 30 min, followed by staining DC- and macrophage-subsets with a cocktail of mAbs specific to CD11c, CD11b and MHC class II (I-A/I-E) and the different Notch ligands (for clone identifiers see at list of mAbs above).

Flow cytometry and cell sorting

For intracellular cytokine and granzyme B staining, splenocytes and total lung samples were stimulated with 1 µg/ml of the MHC class I restricted influenza-derived peptide NP366–374 ASNENMETM for 4 h in the presence of 10 µg/ml brefeldin A (Sigma) to prevent cytokine release. Cells were stained with the relevant fluorochrome-conjugated mAbs for 30 min at 4 °C in PBS containing 0.5% BSA and 0.02% NaN3. For intracellular staining, cells were fixed and permeabilized using the Cytofix/Cytoperm (BD Biosciences). For Phospho-specific flow cytometry, cells were fixed and permeabilized using Phosflow lysis and Phosflow PermWash I reagents (BD Biosciences) according to the manufacturer’s recommendations and subsequently stained with the phosphor-specific antibodies for 1 h at 20°C. When required, cells were washed and additionally incubated with secondary antibody (Goat anti-Rabbit ALEXA488) for 40 min. Data acquisition and analysis was done on a FACSCanto (Becton Dickinson) and FlowJo software.

To isolate H-2 Db–NP tetramer-positive CD8+ T cells from influenza infected mice, single cell suspensions of spleens were stained with influenza-specific tetramers and various markers. Cells were sorted using FACSAria cell sorters (BD Biosciences).

To discriminate between circulating T cells and T cells in splenic white pulp intravascular CD8+ T cells were stained by injecting 1 µg anti-CD8β-PE (eBio341) i.v. 8 min before the animals were sacrificed. Organs were processed as described above, and CD8+ T cells were stained in vitro with a different, noncompeting anti-CD8α antibody (53-6.7) along with antibodies to other surface markers.

Retroviral transductions and adoptive transfers of CD8+ T cells

Virus was produced in PlatE cells as described[43]. Total splenocytes from CD45.2+ OT-I wild-type or OT-I Notch1-2-KO mice were incubated with 1 nM OVA257–264 peptide, and next day cells were spin-infected (700 × g for 90 min at 37°C) with viral supernatant (with 8 µg/ml polybrene), followed by 5 h at 37°C. Medium was replaced and next day, live T cells were isolated by density centrifugation (Lymphoprep, Axis-shield PoC) and between 7.5 × 102 and 5 × 104 cells (as indicated in Figure legends) were transferred into timed influenza-OVA infected CD45.1+ mice. Donor OT-1 T cells were detected 5–10 days after transfer as CD45.2+CD8+ and Thy1.1 or GFP triple positive cells.

In vivo cytotoxicity assay

Target spleen cells from C57BL/6 mice were pulsed with 1 µg/ml influenza-derived peptide NP366–374 for 35 min and subsequently labeled with 0.2 µM CFSE (Invitrogen) for 15 min at 37 °C (CFSElow, specific target cells) or were not pulsed with peptide and labeled with 2 µM CFSE (CFSEhigh, non-specific target cells). The two target populations were mixed in equal numbers and 5 × 106 cells were transferred i.v. into mice that had been infected with influenza A virus HK×31 10 d before or into untreated control mice. Mice were killed 4 h later and the ratio between peptide-loaded versus empty target cells was quantified by flow cytometry.

Gene expression profiling

H-2 Db–NP366–374+CD8+ T cells were isolated from spleens of influenza infected mice by flow cytometry. Total RNA was extracted with TRIzol reagent (Invitrogen) according to the manufacturer’s protocol. For Deep sequencing analysis, total RNA was further purified by nucleospin RNAII columns (Macherey-Nagel) and RNA was amplified using the Superscript RNA amplification system (Invitrogen) and labeled with the ULS system (Kreatech), using either Cy3 or Cy5 dyes (Amersham). Sequences were obtained by pooling 10 samples in one lane on a HiSeq2000 machine. Between 17 and 27 million reads were obtained per sample.

Read mapping (TopHat) and determining differentially expressed genes (DESeq) was done as described in[58]. Reads were mapped against the mouse reference genome (build mm9) using TopHat (version 1.4.0), which allows to span exon-exon junctions. TopHat was supplied with a known set of gene models (NCBI build 37, version 64). In order to obtain per sample genecounts HTSeq-count was used. This tool generates genecounts for each gene that is present in the provided Gene Transfer Format (GTF) file. Genes that have zero counts across all samples were removed from the dataset. Statistical analysis was performed using the R package DESeq. Differentially expressed genes were determined between the SLEC and MPEC samples, and between the wild type and knock-out samples. DESeq assumes that gene counts can be modelled by a negative binomial distribution. For sample normalisation the ‘size factors’ were determined from the count data. The empirical dispersion was determined with the ‘pooled’ method, which used the samples from all conditions with replicates to estimate a single pooled dispersion value. Subsequently, a parametric fit determines the dispersion-mean relationship for the expression values resulting in two dispersion estimates for each gene (the empirical estimated, and the fitted value). Using the ‘maximum sharingMode’ we selected the maximum of these two values to be more conservative. Finally, p-values and FDR corrected p-values were calculated.

To highlight biological processes that are over-represented in the set of differentially expressed genes we used Bioconductor package GOseq[59], which was developed for the analysis of RNA-seq data. First we selected all genes with an FDR<0.5 from the SLEC-MPEC and WT-KO comparisons. Subsequently, the GO ‘Biological Processes’ gene sets were used to determine over-represented processes. In addition we used the ‘C7’ gene set from the Molecular Signatures Database (MSigDB; http://www.broadinstitute.org/gsea), which is a collection of annotated gene sets. Gene set C7 comprises immunologic signatures composed of gene sets that represent cell types, states, and perturbations within the immune system. The signatures were generated by manual curation of published microarray studies in human and mouse immunology. This gene set was generated as part of the Human Immunology Project Consortium (HIPC; http://www.immuneprofiling.org). An in-house R script was developed to convert the C7 gene set into a format that could be used by GOseq.

To visualize the results for the GOseq analysis based on the GO biological processes, we used the Cytoscape Enrichment Map plugin[60]. We first constructed a gmt file containing the GO terms and corresponding ensembl gene identifiers. GO terms associated with more than 250 genes were removed because these represent too general terms that are difficult to interpret. In addition, very small gene sets (<15 genes) were removed because these are more likely to become significant by chance alone. Subsequently, from the results of the GOseq analysis we selected biological processes related to chemokines, cytokines, cytotoxicity, differentiation, adhesion, migration, apoptosis, activation and proliferation. Selected GO terms with FDR<0.5 were visualized as an enrichment map. RNAseq data can be accessed at Array Express (E-MTAB-2255)

Quantitative RT-PCR

cDNA was made with Oligo (dT) and random hexamers using the First Strand cDNA synthesis kit (Fermentas). Quantitative PCR using SYBRgreen (Bio-Rad) was performed using the C1000 Thermal Cycler (Bio-Rad). Relative concentrations were determined by normalizing to β-actin using the Bio-Rad CFX Manager software. Melt curves ensured amplification of a single product. Sequences of primers used: β-actin (5’-GAAGTCCCTCACCCTCCCAA-3’ and 5’-GGCATGGACGCGACCA-3’), RBPJ (5’-CGGTCTGCTTATCAACTTTCC-3’ and 5’-ATTCACAGTCCGAGATGGCTA-3’), Tbx21 (5’-CAACAACCCCTTTGCCAAAG-3’ and 5’- TCCCCCAAGCAGTTGACAGT-3’)

Statistical analysis

Figures represent means and error bars denote standard error of the mean (s.e.m.). Standard Student's t-tests (unpaired, two-tailed) was applied with GraphPadPrism software. If 3 or more groups were compared One-way ANOVA with Bonferroni correction was used. P < 0.05 was considered statistically significant.