Translational repression of pre-formed cytokine-encoding mRNA prevents chronic activation of memory T cells

The 3’UTR of Ifng determines protein expression levels in T_M cells

We first examined if the Ifng 3’UTR regulated protein production in T_M cells. We fused the murine Ifng 3’UTR to a GFP reporter gene (hereafter GFP-Ifng3’UTR ) under the control of the murine PGK-1 promoter. We retrovirally transduced OTI TCR transgenic CD8⁺ T cells expressing the congenic marker CD45.1 with GFP-Ifng3’UTR, or with GFP that lacked a 3’UTR (hereafter GFP_control). The next day, 1000 sorted GFP-Ifng3’UTR OTI or GFP_control OTI cells were transferred into naive C57BL/6J/CD45.2 recipient mice followed by infection with the intracellular bacterium Listeria monocytogenes genetically engineered to express ovalbumin (LM-OVA)²⁹ the next day. We found identical percentages of GFP-Ifng3’UTR and of GFP_control OTI cells in the blood of recipient mice at all time points measured (Fig. 1a). The GFP_control OTI cells expressed constant levels of GFP throughout the infection as determined by the GFP mean fluorescence intensity (GFP-MFI; Supplementary Fig. 1a). In contrast, the GFP-MFI in GFP-Ifng3’UTR OT-I cells increased at day 9 post infection, and dropped at day 13, when the infection was resolved (Supplementary Fig. 1a). At day 35 post infection, the GFP-MFI was about six-fold lower for GFP-Ifng3’UTR OTI cells than for GFP_control OTI cells isolated from blood, liver and spleen (Fig. 1b,c and Supplementary Fig.1b). Reactivation of OTI cells isolated at day 35 post infection with the OVA_257–264 peptide did not alter the MFI of GFP_control OTI cells (p=0.4), but increased the expression of GFP-Ifng3’UTR compared to non-activated cells (p=0.01, Fig. 1b,c).

Spleen-derived GFP_control CD4⁺ and CD8⁺ T cells from C57BL/6J mice showed high GFP-MFI when cultured in IL-7 for several days in the absence of antigen (hereafter ‘resting’), and reactivation for 4h with PMA+ionomycin did not alter the GFP-MFI. In contrast, GFP-Ifng3’UTR T cells had low GFP-MFI when resting, and reactivation led to GFP induction (Supplementary Fig. 1c–f). The MFI of endogenous IFN-γ protein was equal in GFP⁺ and GFP^- T cells (Supplementary Fig. 1g), indicating that the factors that control protein production were not limiting. Similar results were obtained using the minimal murine Ifng promoter³⁰ (Supplementary Fig. 1h), suggesting that the regulatory capacity of Ifng 3’UTR was independent of a specific promoter. Combined, these data implied that the Ifng 3’UTR controls protein production in T_M cells.

Conserved AREs in the Ifng 3’UTR determine protein expression in T cells

We next defined the regulatory region within the Ifng 3’UTR using deletion mutants. Only the reporter constructs containing the first 241 nucleotides of Ifng 3’UTR reduced GFP-MFI similar to the GFP-Ifng3’UTR reporter in resting OTI cells, and maintained the induction upon 6h reactivation with OVA_257–264 peptide (Fig. 2a), indicating that this domain was essential to mediate protein expression. The murine Ifng 3’UTR contains six class I AREs, of which five are located within the first 241nt (Fig. 2b). Mutating the ARE located outside of the 241nt region of the Ifng 3’UTR (GFP-Ifng_MUT6) did not alter the regulatory capacity of the Ifng 3’UTR (Fig. 2c), while the combined site-directed mutation of all five AREs located within the proximal 241nt region (GFP-Ifng_MUT1–5) completely abrogated its regulatory activity (Fig. 2c). Resting OTI cells transfected with GFP-Ifng_MUT1–5 had similar GFP-MFI as GFP_control, which did not increase upon activation with OVA_257–264 peptide (Fig. 2c).

Single or double mutants of the five ARE within the proximal 241nt region only marginally altered the GFP-MFI in resting T cells compared to the GFP-Ifng 3’UTR, and activation with OVA_257–264 peptide resulted in an incomplete, but significant increase of GFP-MFI (Supplementary Fig. 2a,b). Mutating the first three AREs (GFP-Ifng_MUT1–3) clustered at position 74–98nt fully abrogated the induction of GFP-MFI in OVA_257–264-activated OTI cells, and this deletion only partially blocked protein expression in resting T cells compared to GFP_control, and GFP-Ifng_MUT1–5 (Fig. 2d,e). Mutation of the AREs at position 137–141nt and 192–196nt (GFP-Ifng_MUT4 and GFP-Ifng_MUT5, respectively) were required to completely block protein production in resting T cells (Fig. 2d).

The ARE-containing region of the Ifng 3’UTR is highly conserved in man, mouse and many other mammals (Supplementary Fig. 2c). Human CD4⁺ and CD8⁺ T cells transduced with the human GFP-IFNG 3’UTR had lower GFP-MFI than the T cells transduced with GFP_control (Fig. 2f and Supplementary Fig. 2d). As for mouse Ifng_MUT1–3 alike, human GFP-IFNG_MUT1–3 T cells showed higher GFP-MFI than wild-type GFP-IFNG 3’UTR T cells, but 4h reactivation with PMA+Ionomycin did not increase GFP-MFI (Fig. 2f and Supplementary Fig. 2d). In contrast, human GFP-IFNG_MUT1–5 T cells expressed GFP-MFI comparable to GFP_control T cells (Fig. 2f and Supplementary Fig. 2c). These findings indicate that the highly conserved proximal region of the IFNG 3’UTR in humans and Ifng 3’UTR in mice blocks protein production in resting T cells through an ARE-dependent mechanism.

Deletion of Ifng 3’UTR AREs causes chronic IFN-γ production in T_M cells

To investigate the effect of deleting AREs from the endogenous Ifng encoding mRNA in T_M cells, we crossed OTI mice with mice carring a germline replacement of the ARE-containing region with a random sequence (hereafter IFN-γ-ARE-Del OTI mice)¹⁶. To compare the expansion kinetics and cytokine production of IFN-γ-ARE-Del OTI cells with wild-type OTI cells in response to LM-OVA infection, we co-transferred equal numbers (500 cells) of sorted naive CD62L^hiCD44^loIFN-γ-ARE-Del CD45.2 OTI cells and wild-type CD45.1 OTI cells into C57BL/6J CD45.1xCD45.2 recipient mice. This ensured an equivalent bacterial load and a similar cytokine milieu during the course of infection. One day after T cell transfer, recipient mice were infected with LM-OVA. We followed the OTI responses to LM-OVA infection by measuring the percentage of IFN-γ-ARE-Del and wild-type OTI cells of the CD8⁺ T cell population in the blood. At day 6 and 9 post infection, the percentage of IFN-γ-ARE-Del and wild-type OTI cells was similar. At day 13 when the infection was cleared, OTI cells equally contracted (Fig. 3a). At day 35 post LM-OVA infection, similar percentages of OTI cells were found in blood, spleen, liver and bone marrow of C57BL/6J CD45.1xCD45.2 recipient mice (Fig. 3a and data not shown). In addition, the percentage of CD44^hiCD62L^loCD127^lo effector T cells (T_eff), CD44^hiCD62L^loCD127^hi effector memory (T_EM) and CD44^hiCD62L^hiCD127^hi central memory (T_CM) T cells, and the percentage of KLRG1^hiCD127^lo short-lived effector cells (SLEC) and KLRG1^loCD127^hi memory precurser cells (MPEC) was indistinguishable between IFN-γ-ARE-Del and wild-type OTI cells (Fig. 3b, Supplementary Fig. 3a).

Despite the normal differentiation into T_EFF and T_M cells, blood-derived IFN-γ-ARE-Del OTI cells isolated from CD45.1xCD45.2 recipient mice on day 35 post infection produced IFN-γ even in the absence of peptide restimulation, as defined by intracellular cytokine staining after 3h of incubation with brefeldin A (BFA) (Fig. 3c). The IFN-γ production was also detected in spleen, liver and bone marrow-derived IFN-γ-ARE-Del OTI cells 35 days post infection (Fig. 3c), and in spleen and liver-derived IFN-γ-ARE-Del OTI cells 48 days post infection (Supplementary Fig. 3b). In contrast, wild-type OTI cells that were co-transferred into the same CD45.1xCD45.2 recipient mice did not produce IFN-γ (Fig. 3c, Supplementary Fig. 3b). Wild-type and IFN-γ-ARE-Del T cells also did not produce TNF (Fig. 3c). At day 35 post infection all liver and spleen-derived IFN-γ-ARE-Del T cells -T_EFF, T_EM, T_CM cells, SLECs and MPECs- produced IFN-γ, as determined by the percentage and MFI of IFN-γ protein (Fig. 3d; Supplementary Fig. 3c, d).

The expression of Ifng mRNA was similar in sorted IFN-γ-ARE-Del and wild-type OTI cells at day 35 post infection (Fig. 3e), indicating that IFN-γ production of IFN-γ-ARE-Del OTI cells did not correlate with increased mRNA expression. To determine whether this leaky IFN-γ production also occurs in other settings, we studied naturally occurring CD44^hi memory-like T cells. Similar to infection-induced T_M cells, spleen-derived CD44^hi sorted OTI cells mediate protection to LM-OVA infection by rapidly producing IFN-γ upon antigen exposure³¹. CD44^hi OTI cells expressed more Ifng mRNA than naive CD44^lo OTI cells (Supplementary Fig. 3e). Only CD44^hi IFN-γ-ARE-Del cells produced IFN-γ ex vivo when cultured for 4h in the absence of stimulation (Fig. 3f,g; Supplementary Fig. 3f,g). Yet, the Ifng mRNA levels of IFN-γ-ARE-Del OTI cells were comparable to wild-type CD44^hi OTI cells (Fig. 3f,g). Again, the IFN-γ production was observed in spleen-derived CD44^hiCD62L^lo T_Eff/T_EM cells and in CD44^hiCD62L^hi T_CM cells from IFN-γ-ARE-Del mice (Supplementary Fig. 3f,g), and not from wild-type mice, whereas T cells from both genetic backgrounds expressed identical Ifng mRNA levels (Supplementary Fig. 3h). As such, the ARE region within the Ifng mRNA controls the production of IFN-γ in Tm cells by post-transcriptional mechanisms.

The ARE-binding protein ZFP36L2 interacts with Ifng mRNA

To investigate how AREs prevent the production of IFN-γ in T_M cells, we pulled down the ARE-binding proteins from cytoplasmic cell lysates of in vitro expanded, resting primary human T cells using the streptavidin-binding RNA aptamer 4xS1m³², which expresses the 189nt long ARE region of the human IFNG-3’UTR, and analyzed them by mass spectrometry. To reduce false positive results, we included the empty 4xS1m RNA aptamer lackig the ARE region as control (4xS1m-empty), and performed the purifications in triplicate. ARE binding was scored positive when a protein was identified in all three samples. This analysis identified 83 proteins that bound IFNG-4xS1m but not 4xS1m-empty. 57 proteins were significantly enriched (Student t-test, false discovery rate [FDR]=0.05, S0=0.4) upon purification with the IFNG-4xS1m construct (Fig. 4a). To validate ARE-specific binding, we performed a second purification that included the human IFNG_MUT1–5 variant, IFNG_MUT1–5-4xS1m. 80 out of the 83 proteins found in the first purification were also detected in the second purification. Of these, we identified two proteins, ZFP36L1 and ZFP36L2, that interacted with IFNG-4xS1m, but not with IFNG_MUT1–5-4xS1m or with 4xS1m-empty (Fig. 4a and Supplementary Fig. 4a–d).

ZFP36L1 (TIS11B) and ZFP36L2 (TIS11D) are ZFP36 (tristetraprolin or TTP) family proteins, which bind AREs^33,34. ZFP36 induces cytokine-encoding mRNA degradation in T cells, and ZFP36-deficient mice develop severe autoimmunity due to excessive cytokine production^35,36. We detected more zfp36l2 than zfp36l1 mRNA in spleen-derived CD44^hiCD8⁺ T cells, along with more ZFP36L2 than ZFP36L1 protein (Fig. 4b,c). Furthermore, ZFP36L2 protein expression was comparable in naïve T cells, in CD44^hi T_M cells and in PMA+Ionomycin reactivated T_M cells, while ZFP36L1 was only detectable upon T cell activation (Supplementary Fig. 4e). Native RNA immunoprecipitation (i.e. without cross-linking) showed that Ifng mRNA was enriched in ZFP36L2 pull-downs from resting T cells, but not from resting IFN-γ-ARE-Del T cells (Fig. 4d). As such, ZFP36L2 interacts specifically with the ARE region of the Ifng 3’UTR.

ZFP36L2 represses IFN-γ production independently of mRNA stability

Next, we used mice with a conditional deletion of ZFP36L2 in the T cell lineage³⁷ to assess the role of ZFP36L2 deficiency in CD44^hi memory-like T cells. CD4-Cre Zfp36l2^fl/fl mice (hereafter Zfp36l2^cKO) had a similar percentage of CD8⁺ and CD4⁺ T_EFF, T_EM and T_CM cell subsets in the spleen as the wild-type littermates (Supplementary Fig. 5a). However, spleen-derived CD44^hi CD8⁺ and CD4⁺ T cells from Zfp36l2^cKO mice, but not from wild-type littermates, produced IFN-γ in the absence of stimulation (Fig. 5a,b), whereas Ifng mRNA amounts were comparable to wild-type CD44^hi T cells (Fig. 5a,b).

We next tested if loss of ZFP36L2 also affected the GFP reporter constructs. We transduced wild-type and Zfp36l2^cKO T cells with GFP_control, GFP-Ifng3’UTR or GFP-Ifng_MUT1–5. The GFP-MFI of GFP-Ifng_MUT1–5 T cells and GFP_control T cells was comparable in resting wild-type and Zfp36l2^cKO T cells (Fig. 5c,d). In contrast, GFP-MFI was substantially increased in GFP-Ifng 3’UTR CD8⁺ and in CD4⁺ Zfp36l2^cKO T cells compared to wild-type T cells (Fig. 5c,d), indicating that ZFP36L2 repressed IFN-γ production in T cells.

AREs are primarily associated with the regulation of mRNA stability^16,17,35,38. The overall amount of mRNA depends on the transcription rate and the rate of mRNA decay. We measured the rate of mRNA decay by treating resting OTI cells with the polymerase II transcription inhibitor actinomycin D (ActD). Ifng mRNA had a longer half-life in IFN-γ-ARE-Del OTI cells (t_1/2 >120min) than in wild-type OTI cells (t_1/2 ~30min; Fig. 5e). The half-life of Ifng dropped to t_1/2 ~60min in sorted CD44^hi IFN-γ-ARE-Del OTI cells, compared to t_1/2 ~30min in wild-type OTI cells (Fig. 5f). Following 120 min of ActD treatment the decay of Ifng mRNA in IFN-γ-ARE-Del OTI cells was similar to that of wild-type OTI cells (Fig. 5f), indicating that the Ifng mRNA stability ex vivo is substantially lower than in vitro. This was not the case for other transcripts, because Tnf remained unstable, and 18S remained stable (Supplementary Fig. 5b–e). Similar data were obtained using polyclonal CD4⁺ and CD8⁺CD44^hi T cells from IFN-γ-ARE-DEL mice (Fig. 5g; Supplementary Fig. 5f). The rate of Ifng mRNA decay in Zfp36l2^cKO CD4⁺ and CD8⁺ CD44^hi T cells was comparable to that of IFN-γ-ARE-DEL T cells (Fig. 5h; Supplementary Fig. 5g), indicating the ARE-independent degradation of Ifng mRNA did not require ZFP36L2. In addition, the amount of miR-29a/b, which can affect Ifng mRNA expression^39,40, and the expression of Tbx21 and Eomesodermin mRNA, which encode the transcription factors T-bet and Eomes, were similar in IFN-γ-ARE-Del and wild-type CD44^hi T_M cells (Supplementary Fig. 5h), suggesting that the expression of these factors was not influenced by IFN-γ-ARE-Del-mediated alterations in T_M cells. As such, Ifng-AREs determine the Ifng mRNA stability depending on the activation status of T cells, and they do so in a ZFP36L2-independent manner.

ZFP36L2 blocks the association of Ifng mRNA to ribosomes in T_M cells

We next examined if AREs modulated transcription and translation of Ifng mRNA in T_M cells. We treated spleen-derived IFN-γ-ARE-Del CD44^hi memory-like T cells with ActD, or the translation inhibitor cycloheximide (CHX) and measured the IFN-γ production in the absence of stimulation. Blocking transcription with ActD did not interfere with IFN-γ production in IFN-γ-ARE-Del CD44^hi T cells (Fig. 6a,b). We did not detect TNF production in these cells, and cell viability or the surface markers CD8 and CD44 were unaltered (Supplementary Fig. 6a). In contrast, treatment with CHX, which inhibits protein translation, blocked the production of IFN-γ in IFN-γ-ARE-Del CD44^hi T cells compared to ActD-treated cells (Fig. 6a,b), indicating that the IFN-γ production depended on translation and that the pre-formed Ifng mRNA serves as template for protein production in IFN-γ-ARE-Del CD44^hi T_M cells.

To test whether AREs impaired the translation of pre-formed Ifng mRNA, we measured the association of Ifng mRNA with the ribosomes in spleen-derived IFN-γ-ARE-Del or wild-type CD44^hi T cells. The heavy fraction of cytosolic extracts was pelleted by sucrose cushion centrifugation to enrich for ribosomes⁴¹. As control, half of the each cytosolic extract was treated with 20mM EDTA, which dissociates ribosomal and ribosome-associated proteins from mRNA⁴¹ (Supplementary Fig. 6b). mRNA levels measured from this control sample included residual binding of Ifng mRNA to ribosomes and unbound Ifng mRNA, and was substracted from the pellets of EDTA-untreated sucrose cushions. Only the Ifng mRNA from IFN-γ-ARE-Del CD44^hi T cells, but not from wild-type CD44^hi T cells associated with ribosomes (Fig. 6c). The association of control mRNAs Tnf and 18S was comparable between wild-type and IFN-γ-ARE-Del CD44^hi T cells (Supplementary Fig. 6c,d). Notably, also the Ifng mRNA from CD44^hiCD4⁺ and CD44^hiCD8⁺ Zfp36l2^cKO T cells associated with ribosomes (Fig. 6d,e), indicating that the recruitment of Ifng mRNA to the ribosome depended at least in part on ZFP36L2. Thus, AREs within the Ifng mRNA and ZFP36L2 suppress the translation of IFN-γ protein from pre-existing mRNA in T_M cells.

ZFP36L2 binds to pre-formed mRNA in memory T cells

To determine if ZFP36L2 blocked the translation of other ARE-containing mRNAs that are rapidly translated in T_M cells upon activation, we sorted spleen-derived CD44^hi memory-like OTI cells and performed mass spectrometry on untreated cells, and on T cells activated for 2h with OVA_257–264 peptide. Of the 5490 identified proteins, 54 significantly altered their expression profile upon T cell activation (Student t-test, FDR=0.05, S0=0.4; Fig. 7a, Supplementary Fig. 7a). 17 proteins were significantly less abundant in stimulated T cells, including CD62L (Sell, Fig. 7a), which is rapidly downregulated upon T cell activation. Conversely, 37 proteins were significantly induced in stimulated CD44^hi OTI cells, including the early activation marker CD69 and the pro-inflammatory cytokines IFN-γ, TNF and IL-2 (Fig. 7a). Of note, 32 of these 37 proteins were generated from ARE-containing mRNAs (Fig. 7a; Supplementary Fig. 7a).

We next determined the mRNA levels of these rapidly induced proteins in T_M cells. 52 out of the 54 proteins that altered their expression profile upon activation of CD44^hi OTI cells were annotated in T_EM and T_CM cells specific for the lymphochoriomeningitis virus (LCMV)⁴². The mRNAs for the 17 proteins that were enriched in the CD44^hi OTI cells were also expressed in LCMV-specific T_EM and T_CM cells (Fig. 7b, Supplementary Fig. 7b). The mRNAs of the 35 proteins that were rapidly induced in CD44^hi OTI cells greatly varied in expression in T_EM and T_CM cells (Fig. 7b, Supplementary Fig. 7b). LCMV-specific T_EM and T_CM cells expressed the mRNA for 29 of the 35 (82.9%) rapidly generated proteins (Fig. 7b, Supplementary Fig. 7b). For 22 of these 35 proteins (62.9%), the peptide abundance was below detection limit in non-activated CD44^hi T cells (Supplementary Fig. 7a), indicating that these 22 mRNAs are putatively blocked from translation in T_M cells.

ZFP36L2 binds to ARE-containing mRNA with a tandem zinc finger⁴³ that requires two AREs for interaction. Of the 35 rapidly-induced proteins, 26 ARE-containing transcripts encompassed at least 2 AREs (defined as AUUUA) within the 3’UTR (Supplementary Table 1). We focused on 11 putative target genes that fulfilled the following criteria: they contained 2 AREs in most of their transcript variants (Supplementary Table 1), and they expressed pre-formed mRNA in T_M cells (Supplementary Fig. 7b). We measured ZFP36L2 binding to these endogenously expressed mRNAs using native RNA-immunoprecipitation in resting OTI cells. Of the 11 tested mRNAs, 6 mRNAs were significantly enriched in resting T cells compared to the IgG control pull down: Ifng, Tnf, Irf4, Junb, Zfp36l1 and Pim1 (Fig. 7c), indicating that ZFP36L2 targets several ARE-containing pre-formed mRNAs T_M cells .

ZFP36L2 rapidly releases pre-formed mRNA upon activation

Because pre-formed mRNA drives rapid cytokine production upon T cell activation¹⁵, we tested if T cell activation supports the release of pre-formed mRNA from ZFP36L2. As determined by RNA-immunoprecipitation, ZFP36L2 binding to Ifng mRNA was significantly reduced in OTI cells reactivated with OVA_257–264 for 2h compared to non-activated resting OTI cells (Fig. 7d). Of note, loss of mRNA binding upon reactivation coincided with rapid downregulation of Zfp36l2 mRNA, but ZFP36L2 protein remained unaltered (Supplementary Fig. 7d). ZFP36L2 also rapidly dissociated from Tnf and Pim1 mRNA upon T cell activation, while its binding to Irf4, Junb and Zfp36l1 mRNA did not change compared to non-activated T cells (Fig. 7d). Peptides for IFN-γ, TNF and Pim1 were below detection limit in non-activated CD44^hi T cells in the proteomics analysis (Supplementary Fig. 7a), further pointing to a block of mRNA translation. Tnf mRNA was significantly enriched in the ribosome-enriched fraction of Zfp36l2^cKO CD8⁺ and CD4⁺ CD44^hi T cells compared to wild-type T cells (Fig. 7e), and TNF production was found in CD8⁺ and CD4⁺ CD44^hi T cells in the absence of stimulation (Fig. 7f). These observations thus show that ZFP36L2 suppresses the translation of several target mRNAs in memory T cells.

Mice and cell culture

Mice were obtained from Charles River, or housed and bred in-house at the Netherlands Cancer Institute (NCI). IFN-γ ARE-Del mice were generated as previously described¹⁶ and crossed to C57BL/6J.OTI T cell receptor (TCR) transgenic mice (OTI). Mice carrying a conditional allele of Zfp36l2 have been previously described³⁷ and were crossed to mice in which the Cre recombinase is expressed under the control of mouse CD4 regulatory elements (TG(CD4-cre)1Cwi) to obtain deletion specifically in T cells⁵⁵. Experiments were performed in accordance with institutional and national guidelines and approved by the Experimental Animal Committee at the NCI.

Cells were cultured at 37°C and 5% CO₂ in IMDM (GIBCO-BRL) containing 8% FCS, 2mM L-glutamine and Pen/Strep, and for mouse cells supplemented with 15μM 2-mercaptoethanol.

Cloning

Gene-specific 3’UTRs were amplified from genomic DNA of a C57BL/6J mouse, or from human genomic DNA (sequences upon request). The 3’UTR variants were cloned into the ClaI site of pRETRO-SUPER GFP⁵⁶ downstream of GFP. 189nt of the human IFNG 3’UTR were cloned into the pSP73–4xS1m³². The region containing the streptavidin binding sites was reversed using the two BglII sites, and the IFNG 3’UTR was inserted to the 5’ end of the S1m motifs. Prior to in vitro transcription, the pSP73–4xS1m construct was linearized with NdeI. ARE-mutants with a substitution of AUUUA sequence with AAAAA were obtained from GeneScript, and subcloned into pRS-GFP or pSP73–4xS1m. Sequences were confirmed by sequence analysis.

T cell activation and retroviral transduction

1.5×10⁶ human PBMCs were activated in a 24-well plate for 40h with plate-bound anti-CD3/CD28 (1μg/ml each, eBioscience). Retroviral transduction was performed with Retronectin (Takara) as described⁵⁷, and cells were maintained in culture in the presence of 20ng/ml recombinant human Interleukin 7 (PeproTech).

1×10⁶ CD8-purified OTI T cells (Miltenyi CD8 isolation kit; 90–99% purity) were co-cultured for 20h with 0.1×10⁶ pre-seeded MEC.B7.SigOVA cells as described⁵⁸. 3×10⁶ C57BL/6J splenocytes were activated for 48h with Concanavalin A (2μg/ml, Sigma) and recombinant murine Interleukin 7 (rmIL-7, 1ng/ml, PeproTech), or with plate-bound anti-CD3 (2μg/ml, eBioscience) and soluble anti-CD28 (1μg/ml, Bioceros) antibodies. Cells were harvested and retrovirally transduced as described⁵⁹. T cells were maintained with 10ng/ml rmIL-7 and reactivated with 100nM or 1μM OVA_257–264, or with 10ng/ml PMA/1μM ionomycin (Sigma).

FACS-sorted CD8⁺ and CD4⁺ CD44^hi T cells were isolated from spleen or liver of 6–9 week-old mice. T cells were incubated for 4h with brefeldin A (BFA), with or without 1μg/ml Actinomycin D (ActD) or 10μg/ml cycloheximide (CHX) (Sigma).

Adoptive T cell transfer and Listeria monocytogenes-OVA infection

GFP-expressing OTI T cells were purified with Lympholyte M (Cedarlane) and FACS-sorted for GFP expression. Naive CD8⁺CD44^loCD62L^hi WT and IFN-γ-ARE-Del OTI T cells were FACS-sorted from splenocytes. 1000 GFP⁺ OTI T cells, or 500 naive WT OTI/ CD45.1⁺ together with 500 naive IFN-γ-ARE-Del OTI/ CD45.2⁺ T cells were adoptively transferred into recipient mice. The next day, mice were infected i.v. with 2.5×10³ or 2.5×10⁴ of L. monocytogenes strain expressing Ovalbumin (LM-OVA). Peripheral blood was drawn to monitor T cell responses. 35–48 days after infection, spleen, liver and bone marrow cells were incubated with brefeldin A alone, or reactivated for 6h with 1μg/ml OVA_257–264, and intracellular staining for IFN-γ was performed.

Flow cytometry

Mouse T cells were labeled for 20min at 4°C with anti-CD3 (17A2), anti-CD4 (GK1.5), anti-CD8 (53–6.7), anti-CD45.1 (A20), anti-CD45.2 (104), anti-Vβ_5.1–5.2 (MR9–4), anti-CD44 (IM7), anti-CD62L (MEL-14), anti-CD127 (A7R34), KLRG1 (MAFA), anti-IFN-γ (XMG1.2) and anti-TNF-α (MP6-XT22) (all eBioscience). Human T cells were labeled with anti-CD3 (UCHT1, Beckman Coulter), anti-CD4 (SK3, BD Biosciences), anti-CD8 (SK1, BD Biosciences). Dead cells were excluded with 1ng/ml TO-PRO-3 iodide (Sigma-Aldrich), or Near-IR (L10119, Life Technology). Intracellular cytokine staining was performed with the cytofix/cytoperm kit (BD Biosciences). Flow cytometry analysis was performed on LSR-II and LSR Fortessa (BD Biosciences). Data were analyzed with FlowJo software (Tree Star, version 7.6.5 and version 10). Representative flow cytometry gating strategies are found in Fig S9.

Quantitative PCR analysis

Total RNA was extracted using Trizol (Invitrogen) and cDNA was synthesized with SuperScript III (Invitrogen), unless differently specified. MiR were extracted from FACS-sorted CD8⁺CD44^hi WT and IFN-γ-ARE-Del OTI T cells using Total RNA Purification Plus Kit (Norgen Biotek), and cDNA was synthesized with miScript II RT kit (Qiagen). RT-PCR was performed with triplicate reactions using SYBR green on a StepOne Plus (both Applied Biosystems) (primer sequences see⁵⁸). C_t values were normalized to L32 levels and confirmed with 18S.

RNA immunoprecipitation and Western Blot

Cytoplasmic lysates of resting WT or IFN-γ-ARE-Del OTI T cells (250×10⁶ cells per condition) were prepared using lysis buffer (10mM HEPES pH 7.0, 100mM KCl, 5mM MgCl₂, 0.5% NP40) freshly supplemented with 1mM DTT, 40U/ml RNase OUT (both Invitrogen), 0.4mM vanadylribonucleoside complex RNase inhibitor (NEB) and 1% EDTA-free protease/phosphatase inhibitor cocktail (Thermo Scientific). Protein G dynabeads (Thermo Scientific) were prepared as previously described⁶⁰. The lysate was immunoprecipitated for 2h at 4°C with a rabbit anti-ZFP36L2 polyclonal antibody (ab70775) or a rabbit polyclonal IgG isotype control (ab27478, both Abcam). RNA was extracted directly from beads by using Trizol, and mRNA expression was measured by RT-PCR as described above. Specificity of the RNA-IP assay was tested by western blot using a rat pan-ZFP36 antibody generated in the laboratory of M. Turner (unpublished data). This antibody is directed against the C terminus of ZFP36L1 and recognizes all three ZFP36 family members.

1×10⁶ FACS-sorted CD8⁺CD44^low and CD8CD44^hi T cells were lysed according to standard procedures. Proteins were separated on a 10% SDS-PAGE gel and transferred onto a nitrocellulose membrane by iBlot (Thermo). Rabbit anti-ZFP36L1 polyclonal antibody (ab42473), rabbit anti-ZFP36L2 polyclonal antibody (ab70775, both Abcam), mouse anti-RhoGDI monoclonal antibody (MAB9959, Abnova), mouse anti-GAPDH monoclonal antibody (MAB374, Millipore), and a rat pan-ZFP36 antibody were used as primary antibodies, followed by goat anti-rabbit-HRP (4050–05), goat anti-mouse-HRP (1031–05, both Southern Biotech), or goat anti-rat-HRP (31470, Invitrogen) secondary antibodies.

Sucrose cushion

1×10⁶ FACS-sorted CD8⁺ CD44^hi WT or IFN-γ-ARE-Del OTI T cells were lysed in modified polysome buffer (20mM Tris-HCl pH 7.5, 300 mM KCl, 5mM MgCl₂, 0.50% NP40) freshly supplemented with 2mM DTT, 40U/ml RNase OUT, and 10μg/ml CHX. Total cell lysate was treated or not with 20mM EDTA pH 8.0 (Invitrogen), layered over a 20% sucrose cushion and then centrifuged at 37,500 rpm at 4°C for 2 hours in a SW60Ti swing out rotor ultracentrifuge (Beckman coulter). RNA was isolated from the pellet using Total RNA Purification Plus Kit (Norgen Biotek), and mRNA expression levels were analyzed by RT-PCR. The distribution of transcripts within the ribosome-enriched fraction was calculated as fold increase relative to the amount of transcript precipitated in the corresponding EDTA treated sample.

4xS1m RNA aptamer-protein pull down

Human resting T cells were washed with PBS and the cell pellet was snap frozen in liquid nitrogen. The cell pellet was homogenized using 5mm steel beads and a tissue lyser (Qiagen TissueLyser II) at 25 Hz for 15 sec per 6 times. The homogenate was then solubilized and pre-cleared as previously described³². The 4xS1m RNA aptamers containing or not the 189 nt of the human IFNG 3’UTR were in vitro transcribed using ampliscribe T7-flash transcription kit (Epicentre) and coupled to Streptavidin Sepharose High Performance beads (GE Healthcare) as described³². RNA-coupled beads and cell lysate were incubated for 3.5h at 4°C under rotation in the presence of 60U RNasin (Ambion). For each pull-down, about 30μg in vitro transcribed RNA and 10mg protein were used. RNA-bound proteins were eluted by adding 1μg RNaseA (Thermo Scientific) and 100μl 100mM Tris-Hcl pH 7.5 (Gibco-Invitrogen). Proteins were reduced, alkylated and digested into peptides using trypsin. Peptides were desalted and concentrated using Empore-C18 StageTips⁶¹ and eluted with 0.5% (v/v) acetic acid, 80 % (v/v) acetonitrile. Sample volume was reduced by SpeedVac and supplemented with 2% acetonitrile, 0.1% TFA.

Proteomic analysis of OTI memory T cells

Triplicates of 2×10⁶ FACS-sorted CD8⁺ CD44^hi OTI T cells were incubated for 2h in IMDM containing 5% FCS and 1μg/ml BrfA with or without the presence of 100nM OVA_257–264 peptide. Cells were washed twice with ice-cold PBS and cell pellets were snap frozen in liquid nitrogen. Cells were lysed in 40μl 1% Sodium Deoxy Cholate, 40mM Chloro Acetamide (both Sigma Aldrich), 10mM TCEP (Thermo Scientific), and 100mM Tris-Hcl pH 8 (Life Technologies), boiled at 95° for 5 minutes and sonicated for 10 minutes in a Sonifier bath (Branson). An equal volume of 50mM ammonium bicarbonate (Sigma Aldrich) was added, containing 600ng Trypsin Gold (Promega). Samples were digested overnight at room temperature, acidified by addition of 1μl trifluoroacetic acid (Thermo Scientific) and loaded on in-house prepared SDB-RPS StageTips (Empore). Peptides were desalted and eluted in three fractions by increasing concentrations of ammonium formate (100mM and 150mM) or 5% (v/v) ammonium hydroxide and acetonitrile (40%, 60% and 80% v/v)⁶². Sample volume was reduced by SpeedVac and supplemented with 2% acetonitrile, 0.1% TFA.

Mass spectrometry data acquisition

Tryptic peptides were separated by nanoscale C18 reverse chromatography coupled on line to an Orbitrap Fusion Tribrid mass spectrometer via a NanoElectroSpray Ion Source (both Thermo Scientific). Peptides were loaded on a 20 cm 75–360μm inner-outer diameter fused silica emitter (New Objective) packed in-house with ReproSil-Pur C18-AQ, 1.9μm resin (Dr Maisch GmbH). The column was installed on a Dionex Ultimate3000 RSLC nanoSystem (Thermo Scientific) using a MicroTee union formatted for 360μm outer diameter columns (IDEX) and a liquid junction. The spray voltage was set to 2.15 kV. Buffer A was composed of 0.5% acetic acid and buffer B of 0.5% acetic acid, 80% acetonitrile. Peptides were loaded for 17 min at 300nl/min at 5% buffer B, equilibrated for 5 minutes at 5% buffer B (17–22min) and eluted by increasing buffer B from 5–15% (22–87min) and 15–38% (87–147min), followed by a 10 minute wash to 90% and a 5min regeneration to 5%. Survey scans of peptide precursors from 400 to 1500 m/z were performed at 120K resolution (at 200 m/z) with a 1.5×10⁵ ion count target. Tandem mass spectrometry was performed by isolation with the quadrupole with isolation window 1.6, HCD fragmentation with normalized collision energy of 30, and rapid scan mass spectrometry analysis in the ion trap. The MS2 ion count target was set to 104 and the max injection time was 35ms. Only those precursors with charge state 2–7 were sampled for MS2. The dynamic exclusion duration was set to 60s with a 10ppm tolerance around the selected precursor and its isotopes. Monoisotopic precursor selection was turned on. The instrument was run in top speed mode with 3s cycles. All data were acquired with Xcalibur software.

Mass spectrometry data analysis

The RAW mass spectrometry files were processed with the MaxQuant computational platform, 1.5.0.25 (RNA pull down) or 1.6.0.13 (memory OTI T cells)⁶³. Proteins and peptides were identified using the Andromeda search engine by querying the human Uniprot database (downloaded February 2015, 89796 entries) for the RNA pull down, or the mouse Uniprot database (downloaded August 18, 2017, 51434 entries) for analysis of memory OTI T cells. Standard settings with the additional options match between runs, Label Free Quantification (LFQ), and unique peptides for quantification were selected. The generated ‘proteingroups.txt’ table was filtered for potential contaminants and reverse hits using Perseus 1.5.0.31 (RNA pull down) or 1.5.1.6 (memory OTI T cells). The LFQ values were transformed in log2 scale, the triplicates per experimental condition grouped, and proteins were filtered for at least three valid values in one of the experimental groups. Missing values were imputed by normal distribution (width=0.3, shift=1.8), assuming these proteins were close to the detection limit. To identify the proteins with the most prominent differences, we performed a two-sided t-test using an FDR of 5% and S0 of 0.4 (Volcano plot). Mass spectrometry data of resting and activated memory T cells (Fig. 7 and Supplementary Fig. 7) are deposited at PRIDE: PXD008051.

mRNA expression analysis and ARE determination

mRNA expression of LCMV-specific spleen-derived central memory (CM) and effector memory (EM) T cells was extracted from Mackay et al.⁴² (GEO accession number: GSE70813). Reads per million mapped reads (RPM) were transformed as log2-normalized counts using DESeq2⁶⁴.

To determine the presence of ARE sequences in mRNAs, the 3’UTR sequences of all murine transcripts were downloaded from Ensembl BioMart (release May 2015) and compared to transcripts present in the RNAseq data set of memory T cells⁴². AREs (sequence motif ATTTA) were counted in all 3’UTR variants. When multiple transcripts were present for a gene, we used the minimum, maximum, and average count of AREs. Resulting data were combined with gene expession data by gene symbols.

Statistical analysis

Statistical analysis between groups was performed with GraphPad Prism 6, using the 2-tailed Student t test when comparing 2 groups, or one-way ANOVA test with Dunnett correction when comparing > 2 groups. P values <0.05 were considered statistically significant.