key: cord-0330297-r3vi3u0c
authors: Popović, Branka; Guislain, Aurélie; Engels, Sander; Nicolet, Benoît P.; Jurgens, Anouk P.; Paravinja, Natali; Freen-van Heeren, Julian J.; van Alphen, Floris P.J.; van den Biggelaar, Maartje; Salerno, Fiamma; Wolkers, Monika C.
title: Time-dependent regulation of cytokine production by RNA binding proteins defines T cell effector function
date: 2021-11-04
journal: bioRxiv
DOI: 10.1101/2021.11.03.467112
sha: b6f5e4c59589e46b9ca81a80a2174d88f36cae63
doc_id: 330297
cord_uid: r3vi3u0c

Effective T cell responses against target cells require controlled production of the key pro-inflammatory cytokines IFN-γ, TNF and IL-2. Post-transcriptional events determine the magnitude and duration of cytokine production in T cells, a process that is largely regulated by RNA binding proteins (RBPs). Here we studied the identity and mode of action of RBPs interacting with cytokine mRNAs. With an RNA aptamer-based capture assay from human T cell lysates, we identified >130 RBPs interacting with the full length 3’untranslated regions of IFNG, TNF and IL2. The RBP landscape altered upon T cell activation. Furthermore, RBPs display temporal activity profiles to control cytokine production. Whereas HuR promotes early cytokine production, the peak production levels and response duration is controlled by ZFP36L1, ATXN2L and ZC3HAV1. Importantly, ZFP36L1 deletion boosts T cell responses against tumors in vivo, revealing the potential of the RBP map to identify critical modulators of T cell responses.

T cells are critical players to fight infections and to clear malignant cells. To do so, they produce effector molecules such as granzymes and proinflammatory cytokines.

Interferon gamma (IFN-γ) and Tumor necrosis factor (TNF) are major contributors to the anti-microbial and anti-tumoral effects of T cells (Zhang et al., 2008; Patel et al., 2017) . Furthermore, T cells that co-produce IFN-γ and TNF together with the proliferation and survival-inducing cytokine IL-2 are considered the most potent antiviral and anti-tumoral CD8 + and CD4 + T cells in humans (Almeida et al., 2007; Ciuffreda et al., 2008; Quezada et al., 2010; Wimmers et al., 2016; Oh et al., 2020) .

less is known about the molecular switches that drive their production. Recent studies revealed that post-transcriptional regulation (PTR) is essential in determining the magnitude and duration of cytokine production (Yoshinaga et al., 2019; Salerno et al., 2020; Jurgens et al., 2021) . PTR defines RNA splicing and RNA modifications, the nuclear export and further subcellular localization, mRNA stability and translation efficiency. These post-transcriptional events are controlled by signals that T cells receive. We previously showed that the TCR signal strength together with costimulatory signals define the mRNA synthesis and degradation rate and regulate translation and translation efficiency of cytokines in T cells (Salerno et al., 2016 (Salerno et al., , 2017 .

The importance of PTR in immune cell function was revealed when AU-rich sequences (AREs) were deleted from the germ-line Tnf and Ifng 3'Untranslated region (3'UTR) in mice. This resulted in hyperinflammation and immunopathology due to aberrant cytokine production (Kontoyiannis et al., 1999; Hodge et al., 2014) .

Conversely, tumor-infiltrating T cells (TILs) fail to produce IFN-γ despite their continuous expression of Ifng mRNA, which points to post-transcriptional dysregulation of cytokine production within tumors . Indeed, removing AREs from the Ifng 3'UTR restored the IFN-γ protein production in murine TILs, and it augmented and prolonged protein production in human T cells Freen-van Heeren et al., 2020) . These studies thus underpin the critical role of 3'UTRs as key regulators for cytokine production in T cells.

RNA binding proteins (RBPs) are critical mediators of PTR (Akira et al., 2021) . For instance, the RBPs ZC3H12A (Regnase-1) and Roquin-1 prevent naïve CD4 + T cells from exiting their quiescent state. By interacting with the 3'UTR, they destabilize mRNAs that encode key regulators of CD4 + T cell differentiation and function (Hoefig et al., 2018) . In contrast, Arid5a modulates T cell differentiation by stabilizing Stat3 and Ox40 mRNA upon T cell activation (Masuda et al., 2016; Hanieh et al., 2018) .

We recently showed that the ARE-binding protein ZFP36L2 blocks translation of preformed cytokine mRNAs in memory T cells, which prevents aberrant protein production from these ready-to-deploy mRNAs in the absence of TCR stimuli (Salerno et al., 2018) . Importantly, re-activating memory T cells released pre-formed Tnf and Ifng mRNA from ZFP36L2, allowing for immediate cytokine production (Salerno et al., 2018) . Thus, RBP binding to target mRNAs can rapidly change, which in turn can have a major impact on the acquisition and execution of T cell effector function. A comprehensive study on RBP interactions to target mRNAs in T cells is however lacking.

To decipher RNA targets of specific RBPs, Cross-linked immunoprecipitation (CLIP) methods have been developed (König et al., 2010) , and our understanding of RBP-mRNA interactions substantially advanced with the recently generated 150 RBP interaction maps (Van Nostrand et al., 2020) . However, these studies performed in epithelial cell lines are not directly translatable to T cells. Not only different target mRNAs are expressed in T cells, but also RBP expression is cell-type specific and alters during T cell differentiation and activation (Salerno et al., 2020; Jurgens et al., 2021; Zandhuis et al., 2021) . More than 2000 RBPs have been identified Castello et al., 2016; Perez-Perri et al., 2018) . Which of these RBPs interact with cytokine mRNA, and how RNA binding alters upon T cell activation is to date unexplored. To answer this question, an RNA-centric approach is required.

Here, we provide the first comprehensive analysis of RBP-mediated regulation of T cell effector function. With an unbiased RNA aptamer pull-down from primary human T cell lysates with the full-length 3'UTRs of IFN-γ, TNF and IL-2, we identified >130 RBPs that interact with cytokine 3'UTRs. RNA-RBP interactions depended on the T cell activation status, and RBPs employed RBP-specific post-transcriptional 4 mechanisms to control cytokine production. Importantly, genetic deletion of the RBP ZFP36L1, which dampened cytokine production in activated T cells, resulted in improved anti-tumoral T cell responses in vivo, revealing the critical role of RBPs in defining T cell effector function.

To determine how cytokine 3'UTRs contribute to protein production, we retrovirally transduced OT-I T cell receptor (TCR) transgenic CD8 + T cells with reporter constructs containing the full length 3'UTR of the murine Gzmb, Ifng, Tnf or Il2 mRNA fused to the coding sequence of GFP ( Figure 1A ). The empty GFP construct was used as control (GFP control ). After 4 days of culture, we measured the GFP geometric mean fluorescence intensity (gMFI). In resting cells, the Ifng, Tnf and Il2 3'UTRs substantially reduced the GFP protein expression levels compared to GFP control , or to GFP-Gzmb-3'UTR ( Figure 1A , Figure S1A ). Gzmb 3'UTR also lowered GFP expression levels compared to GFP control , but much less so than cytokine 3'UTRs.

Activating T cells with OVA 257-264 peptide showed limited to no alterations in protein expression for GFP control and GFP-Gzmb-3'UTR. In sharp contrast, the GFP protein expression significantly increased for GFP-Ifng-3′UTR, GFP-Tnf-3′UTR and GFP-Il2-3'UTR at 6 h, and even more so at 16 h of activation ( Figure 1A ). Similar results were obtained with the full length 3'UTRs of human IFNG, TNF and IL2: the GFP protein expression was blocked in resting CD8 + and CD4 + peripheral blood-derived human T cells, and T cell activation with PMA/Ionomycin or α-CD3/α-CD28 boosted the protein expression ( Figures 1B, Figure S1 ). This profound regulation of protein expression by the 3'UTRs of IFNG, TNF, and IL2 is thus conserved between mouse and human.

RNA binding proteins (RBPs) are critical modulators of mRNA fate and metabolism, yet their binding to cytokine 3'UTRs in T cells remains ill defined. To identify RBPs that specifically interact with cytokine 3'UTRs, we generated in vitro transcribed streptavidin-binding 4xS1m RNA aptamers (Leppek et al., 2014) fused to the full length 3'UTRs of IFNG, TNF, and IL2. An empty 4xS1m RNA aptamer was used as control ( Figure 2A ). Because RBPs are expressed in a cell-type specific manner (Zandhuis et al., 2021) , we used human T cell lysates (Figure 2A ). Upon capture of RNA-RBP complexes with streptavidin beads, RNA-interacting proteins were identified by mass spectrometry (MS). Pull-down methods enrich for specific interactors, which results in different protein concentrations and data distribution between pull-down samples and controls ( Figure S2A , B). Therefore, we analysed protein raw intensities to identify putative binders. We detected in total 1808 proteins in all replicates of at least one condition in one independent experiment ( Figure S2A , Table S1 ). Due to the non-normal data distribution in one of the triplicates of IFNG 3'UTR, we excluded this sample from the analysis ( Figure S2B ). Of the 1808 detected proteins, 598 proteins detected in at least 2 of the 3 replicates were enriched with log 2 fold change (LFC)>4 (>16-fold) in cytokine 3'UTR aptamer pulldowns compared to empty aptamer control. Of these, 307 proteins (51.3%) were experimentally confirmed or computationally predicted RBPs (n=222 and n=85, respectively; see Methods). In the downstream analysis, we only included RBPs that were enriched in at least two independently performed experiments. This resulted in 93 RBPs that were identified with the IFNG 3′ UTR, 69 RBPs with the TNF 3′ UTR, and 82 RBPs with the IL2 3′ UTR ( Figure 2B and 2D, and Table S1 ). When we performed Gene ontology (GO) analysis on all 138 detected RBPs, we found that the terms RNA processing and (m)RNA metabolic processes were enriched ( Figure 2C ). Translation was in particular enriched in IFNG-and TNF-associated RBPs, and RNA splicing in IFNG and IL2associated RBPs ( Figure 2C , Table S2 ).

Interestingly, some RBPs were identified with more than one cytokine 3'UTR. For instance, NOLC1 and CHERP interacted with the IFNG and TNF 3'UTRs ( Figure 2D , Table S1 ). The splicing factor SYNCRIP was detected with the IFNG and IL2 3'UTRs, and nuclear mRNA export protein ZC3H11A and HNRNPAB with TNF and IL2 3'UTRs ( Figure 2D , Table S1 ). RBPs interacting with all three cytokine 3'UTRs included components of the cleavage and polyadenylation specificity factor complex CPSF1, CPSF2, CPSF4 and FIP1L1 and the ARE-binding proteins ZFP36L1 and HuR (ELAVL1) ( Figure 2D , Table S1 ). ZFP36L2, which we previously identified to bind to preformed Ifng and Tnf mRNA in memory T cells (Salerno et al., 2018) , was also detected, but with LFCs of 2.5 (IFNG), 1.9 (TNF) and 3.3 (IL2), and did not meet our stringent cut off of LFC>4 (Table S1 ). Other RBPs were enriched for only one cytokine 3'UTR, such as FMR1 autosomal homolog 2, FXR2, zinc-finger protein ZC3HAV1 and RBM27 for the IFNG 3'UTR, stress granule-components ATXN2 and ATXN2L and poly(A)-tail-binding proteins PABPC1 for the IL2 3'UTR, and Pumilio homolog 1, PUM1, and RBM15 for the TNF 3'UTR ( Figure 2D , Table S1 ). In conclusion, we here identified 138 RBPs in primary human T cells interacting with one or more full length cytokine 3'UTRs.

T cell activation drives cytokine production, and post-transcriptional events are critical to tightly control the production of these perilous molecules. External triggers can alter the expression and/or post-translational modifications of RBPs, which in turn can alter their RNA binding capacity (Coelho et al., 2017; Grammatikakis et al., 2017; Salerno et al., 2018) . To determine how T cell activation modulated the RBP binding landscape to cytokine 3'UTRs, we not only performed the aptamer pull-down with cytosolic lysates from resting T cells, but also from paired samples of T cells activated with PMA/Ionomycin for 2 h ( Figure 3A ). We detected in total 1596 proteins in all replicates of at least one pulldown condition ( Figure S2C , Table S3 ). Of these 1596 proteins, 443 proteins were enriched with a LFC>4 in cytokine 3'UTR samples compared to empty 4xS1m aptamer control. 244 proteins were annotated as RBPs, of which 77 were detected in two independently performed experiments ( Figure 3A , Table S3 ). 58 of these 77 RBPs displayed enriched binding with an LFC>4 to IFNG 3′UTR, 63 RBPs to TNF 3′UTR, and 22 RBPs to the IL2 3′UTR ( Figure 3A ). We then studied GO terms of the intified RBPs. IL2 was left with only 3 RBPs that were reproducibly identified in both experiments with an LFC>4, and was therefore excluded from this analysis. RBPs interacting with IFNG and TNF 3'UTRs 3'UTRs in activated T cells were enriched for GO terms related to RNA processing and (m)RNA metabolic processes ( Figure 3B , Table S4 ).

We next determined whether RBPs alter their RNA binding profile upon T cell activation. Interestingly, the interaction of the ARE-binding protein ZFP36L1 to all three cytokine 3'UTRs considerably increased after T cell activation ( Figure 3A and 3C). Other RBPs with enriched binding included m6A-methylation reader YTH domain-containing family protein 1 (YTHDF1) to IFNG 3'UTR, and ribosomal RNA processing protein 1 homolog B (RRP1B) to TNF 3'UTR. RBPs enriched for the IL2 3′UTR included RRM-containing protein (RBM42), PC4 and SFRS1-interacting protein (PSIP1) and WARS ( Figure 3C ). In contrast, T cell activation resulted in undetectable RNA binding by ZC3H12D (Regnase-4) and uridylyltransferase ZCCHC11 to all three cytokine 3'UTRs, PABPN1 to IFNG 3'UTR, ATP-dependent RNA helicase DDX39A and ZNHIT6 to TNF 3'UTR, and DDX46 and USO1 to IL2 3'UTR ( Figure 3C ).

To define whether the alterations of RBP binding to cytokine 3'UTRs resulted from altered RBP expression levels, we compared the RBP protein expression of the total cytosolic cell lysates from activated and non-activated T cells by MS analysis ( Figure   3D , Table S5 ). Of the RBPs that showed increased interaction with cytokine 3'UTRs upon T cell activation, only ZFP36L1 also showed increased protein expression ( Figure 3D , Figure S2D ). WARS and YTHDF1 increased their binding to cytokine 3'UTRs upon T cell activation, even though their protein expression was unaltered ( Figure 3D , Figure S2D ). Conversely, the decreased interaction of ZC3H12D with cytokine 3'UTRs upon T cell activation coincided with decreased protein expression, yet DDX39A and DDX46 lost their interaction to cytokine 3'UTRs without altering their expression levels ( Figure 3D , Figure S2D ). In conclusion, T cell activation results in dynamic changes of the RBP binding landscape to cytokine 3'UTRs, which can only partially be explained by altered RBP protein expression.

We next questioned which of the RBPs interacting with cytokine 3'UTRs modulated the cytokine production. We turned our attention to ZFP36L1 (TIS11B), one of the top candidates that displayed increased interaction with cytokine 3'UTRs upon T cell activation ( Figure 2D , 3C). ZFP36L1 is an ARE-binding zinc-finger RBP that promotes mRNA degradation and represses translation (Lykke-Andersen et al., 2005; Bell et al., 2006; Moore et al., 2018) . Together with ZFP36L2, ZFP36L1 limits cell cycle progression in thymocytes and enables T cell maturation (Hodson et al., 2010; Vogel et al., 2016) .

We first confirmed ZFP36L1 binding to endogenous IFNG, TNF and IL2 mRNA using native RNA immunoprecipitation (RIP) with cytosolic lysates from human PMAionomycin activated T cells ( Figure 4A ). We then generated ZFP36L1-deficient primary human T cells by CRISPR/Cas9 gene-editing to determine whether ZFP36L1 was functionally relevant (Table S6, Figure 4B ). Non-targeting crRNA was used as control. ZFP36L1-knockout (KO) in resting T cells did not result in aberrant cytokine production ( Figure S3A ). However, ZFP36L1 KO CD8 + T cells reactivated for 4 h with α-CD3/α-CD28 produced more cytokines, as defined by higher percentages of IFN-γ, 9 TNF, and IL-2-producing T cells, and by higher cytokine production (gMFI levels) per cell when compared to control T cells ( Figure 4B , Figure S3B ).

We and others previously showed that IFN-γ, TNF, and IL-2 follow individual production kinetics (Han et al., 2012; Nicolet et al., 2017; Salerno et al., 2017) . To investigate whether ZFP36L1 modulates the cytokine production kinetics, we reactivated T cells with α-CD3/α-CD28 for 1 h to up to 8 h, and measured temporal snapshots of protein secretion by adding brefeldin A for the last 2 h of activation (Salerno et al., 2017) . This assay revealed that ZFP36L1 deletion did not alter the early production kinetics of IFN-γ, TNF, and IL-2 in CD8 + T cells ( Figure 4C -D, Figure   S3C ). Consistent with its protein expression profile in activated T cells ( Figure S3D ), ZFP36L1 deletion showed the greatest effects on cytokine production between 4-8 h of activation, a feature that was partially conserved in CD4 + T cells ( Figure 4C -D, Figure S3E ). Zooming in on 8h of activation when the effects of ZFP36L1 deletion were most prominent, compiled data from 6 donors revealed significant increases in percentages and in magnitude of the protein production for all 3 cytokines (gMFI levels) ( Figure 4E ). Interestingly, whereas IFN-γ and IL-2 displayed the largest differences in percentages of cytokine-producing T cells in ZFP36L1 KO T cells, the TNF production was primarily altered in magnitude, as determined by the gMFI ( Figure 4D , E). Thus, ZFP36L1 modulates the cytokine production of human T cells, and does so in a time-and cytokine-dependent manner.

To investigate whether ZFP36L1 affects the stability of cytokine mRNAs, we activated human T cells with for 3h α-CD3/α-CD28 and then treated them with Actinomycin D (ActD) to block de novo transcription. qRT-PCR analysis revealed that the half-life of IFNG and IL2 mRNA was between 50-60 min in control T cells, and that of TNF mRNA between 20-30 min ( Figure 4F ). ZFP36L1 deletion substantially increased the half-life of all three cytokine mRNAs to >120 min for IFNG and IL2, and to 40-50 min for TNF mRNA ( Figure 4F ). The increased mRNA stability also resulted in higher overall cytokine mRNA expression levels in activated ZFP36L1 KO T cells compared to control T cells ( Figure 4G ). Intriguingly, a co-immunoprecipitation assay with α-ZFP36L1 from activated human T cell lysates identified several components of the mRNA degradation machinery, including members of the CCR4-NOT complex, i.e. CNOT3, CNOT4, CNOT6L, CNOT7, CNOT10, CNOT11, and of the RNA 3'->5' exosome complex, i.e. EXOSC1, EXOSC2, EXOSC5, EXOSC6, EXOSC7, EXOSC8, EXOSC9 ( Figure 4H , Figure S3F ). These proteins were only detected upon immunoprecipitation with α-ZFP36L1, but not with the IgG control. Thus, ZFP36L1 dampens cytokine production in T cells by destabilizing their mRNA.

We next investigated other RBPs that were identified with the pull-down assay. We followed up on FXR2, ZC3HAV1, ATXN2L and HuR, four RBPs that showed strong binding to the cytokine 3'UTRs in resting, and in activated T cells ( Figure 2D ). CRISPR/Cas9 gene-editing in human T cells substantially reduced the protein expression of all four RBPs ( Figure 5A ). Deleting FXR2 in α-CD3/α-CD28 activated T cells had no effect on cytokine expression when compared to the donor-matched control ( Figure 5A , Figure S4A ). ATXN2L and ZC3HAV1 deletion, however, resulted in higher percentages of IFN-γ, TNF and IL-2 producing CD8 + T cells at 4 h after activation ( Figure 5A , Figure S4A ). In contast, reduced protein expression of HuR diminished the cytokine production in CD8 + T cells ( Figure 5A , Figure S4A ). In line with their measured effect on cytokine production, ATXN2L, ZC3HAV1 and HuR directly interacted with endogenous cytokine mRNA, as determined by native RIP ( Figure 5B ). Interestingly, the preference for specific cytokine RNAs observed in the aptamer pull-down assay ( Figure 2D ) was also reflected by the native RIP: the ATXN2L RIP showed highest enrichment for IL2 mRNA, and the HuR RIP for TNF mRNA ( Figure 5B ).

To determine at which time points the RBPs ZC3HAV1, ATXN2L and HuR exert their effects on cytokine production, we again performed the snapshot analysis ( Figure   5C ). Intriguingly, this analysis revealed different kinetics in RBP activity. Similar to the ZFP36L1 KO T cells ( Figure 4C ), ATXN2L KO increased the cytokine production in activated CD8 + and CD4 + T cells from 4 h onwards, both in terms of percentage and production per cell ( Figure 5C and 5D, Figure S4B ). ZC3HAV1 deletion also increased the cytokine production from 4 h of activation onwards, yet increases of TNF and IL-2 production per cell were already observed at 1-2 h post T cell activation 11 ( Figure 5C and 5D, Figure S4B ). Conversely, HuR deletion decreased the cytokine expression in particular between 1-4 h after T cell activation, and did so most effectively in CD8 + T cells ( Figure 5C and 5D, Figure S4B , C).

Intriguingly, whereas ZFP36L1 deletion resulted in stabilized cytokine mRNA, this effect on RNA stability was not observed when ATXN2L, ZC3HAV1, or HuR were deleted ( Figure S5A ). Also the overall expression of cytokine mRNAs in ATXN2L KO, ZC3HAV1 KO and HuR KO T cells was comparable with the control cells ( Figure   S5B ). In conclusion, CRISPR/Cas9 gene-editing of primary human T cells confirmed the regulatory activity on cytokine production for 4 out of 5 tested RBPs identified in the RBP pull-down assay. These 4 RBPs differentially modulate cytokine production, and at different time points: whereas ATXN2L, ZC3HAV1, and ZFP36L1 dampen the production of IFN-γ, TNF and IL-2 at later stages of T cell activation, HuR boosts the early cytokine production.

We next determined how RBP depletion modulated T cell responses to target cells.

To test this, we retrovirally transduced T cells with the MART1-specific TCR recognizing the HLA-A*0201-restricted MART1 26-35 epitope (Gomez-Eerland et al., 2014) . Subsequent CRISPR/Cas9 gene editing resulted in efficient RBP deletion ( Figure S6A ). To study the cytokine production in response to antigen-expressing target cells, we exposed RBP-deficient, and control MART1 TCR expressing T cells to a MART1 hi HLA-A*0201 + melanoma cell line (MART1 + ), or to a MART1 lo HLA-A*0201melanoma control cell line (MART1 -) ( Figure S6B ) (Nicolet et al., 2020) . ZFP36L1 KO CD8 + T cells were superior cytokine producers compared to control T cells, as revealed by substantially higher IFN-γ, TNF and IL-2 production after 6 h of co-culture ( Figure 6A, B) . Of the other two tested RBPs, only ZC3HAV1 KO T cells displayed slight, but significant increases in IFN-γ producing cells ( Figure 6A , B). In line with their reduced production kinetics at later time points (Han et al., 2012; Nicolet et al., 2017) , all T cells showed reduced cytokine production after 24 h of coculture, in particular of IL-2 and TNF. However, ZFP36L1 KO T cells were able to maintain their IFN-γ and TNF production at this late time point ( Figure 6B ). This was also evidenced by the polyfunctional cytokine production profile. Whereas only slight increases of T cells producing 2 or 3 cytokines were found at 6 h of co-culture in ZFP36L1 KO and ZC3HAV1 KO T cells, at 24 h of co-culture, ZFP36L1 KO T cells, and to a lesser extent ATXN2L KO T cells clearly maintained their capacity to (co)produce IFN-γ and TNF compared to control T cells ( Figure 6C ).

To determine if different antigen levels are relevant for cytokine production by RBP deficient T cells, we loaded the HLA-A*0201 + MART1-negative uveal melanoma cell line 2A14 with increasing amounts of MART1 26-35 peptide, and we measured the cytokine production after 6h of co-culture with MART1 TCR-expressing T cells.

cytokine production compared to control T cells at high antigen load (100nM), ZFP36L1 KO T cells were superior in producing all three cytokines even at low antigen levels (0.1nM), and this effect was maintained at intermediate and high

antigen levels (1 and 100nM, respectively; Figure 6D , 6E, S6C). Thus, the activity of individual RBPs requires different signal strength to exert their modulatory function.

Our data reveal that RBP depletion can boost cytokine production in response to target cells. Gradual loss of effector function, and in particular of IFN-γ and TNF is a major hurdle of effective T cell responses (van der Leun et al., 2020; Philip et al., 2021) . RBPs are well conserved in mammals . Therefore, to study whether RBP deletion rescues cytokine production of tumor infiltrating lymphocytes (TILs) in vivo, we turned to the murine B16F10 melanoma model. We focused our attention on ZFP36L1, which showed the most potent effects in human T cells. We generated Zfp36l1-deficient OT-I TCR transgenic murine T cells by CRISPR/Cas9 gene-editing ( Figure S7A , Table S6 ), and injected 1x10 6 Zfp36l1 KO, or control OT-I T cells into tumor-bearing mice that had received 7 days earlier B16-OVA tumor cells expressing Ovalbumin (B16-OVA; Figure 7A ) (de Witte et al., 2006) . At 14 days post T cell transfer, we analysed the capacity of splenic and tumorinfiltrating OT-I T cells to produce cytokines. In line with our previous findings , we only detected cytokine production in splenic OT-I T cells when reactivated with OVA 257-264 peptide ( Figure S7C ). We found a slight, but not significant increase in the percentage of IFN-γ and IL-2 producing Zfp36l1 KO splenic T cells compared to control T cells ( Figure S7C ).

When we analysed the tumor-infiltrating T cells, we found that the absolute numbers of Zfp36l1 KO and control OT-I TILs did not substantially differ ( Figure 7B ). To measure the capacity to produce cytokines within the tumor digest, we first incubated the tumor digest ex vivo for 2 h with brefeldin A and monensin, without the addition of exogenous peptide. The production of TNF and IL-2 in control OT-I TILs was undetectable, and of IFN-γ very limited, as previously reported ( Figure 7C and S7B; ). Treatment with exogenous OVA 257-264 peptide for 4 h only marginally increased the cytokine expression in control OT-I TILs, in particular when compared to reactivated splenic OT-I T cells ( Figure 7C , S7B and S7C). Zfp36l1 KO OT-I TILs, however, produced all three cytokines, a feature that was apparent in the presence and absence of exogenous antigen, and for the percentages of cytokine producing TILs as well as for the cytokine production per cell ( Figure 7C , S7B and S7C). Thus, the inhibitory role on cytokine production by ZFP36L1 in T cells is conserved between mouse and man. Intriguingly, also other effector molecules such as the degranulation marker CD107a and the cytotoxic molecule Granzyme B were increased in Zfp36l1 KO TILs when compared to control TILs ( Figure 7D ). This feature appeared to be unique to the tumor environment, because no differences in CD107a and Granzyme B expression were observed in spleen-derived T cells ( Figure S7D ).

KO TILs, we questioned if Zfp36l1 deficiency also influenced the tumor outgrowth.

We therefore injected a low number (0.65x10 6 cells) of Zfp36l1 KO or control OT-I T cells into B16-OVA tumor-bearing mice and followed the tumor growth. Remarkably, even with these low numbers of transferred T cells, the tumor outgrowth of this aggressive tumor was delayed in mice that had received Zfp36l1 KO T cells ( Figure   7E ). In fact, at 14 days post T cell transfer, i.e. the time point when we measured superior cytokine production and degranulation in Zfp36l1 KO TILs ex vivo ( Figure 7C and 7D), 3/8 of the control mice and none of the Zfp36l1 KO mice reached the human endpoint of 1000mm 3 of tumor size ( Figure 7F ). In conclusion, Zfp36l1 geneediting in T cells enhances the therapeutic potency of T cell therapy.

The fine-tuning of cytokine production by post-transcriptional events is key for appropriate T cell effector function. However, little is known about the mediators responsible for modulating cytokine production. Here, we provide the first RBP catalogue of >130 RBPs that interact with the full length 3'UTRs of IFN-γ, TNF and IL-2 in human T cells.

Notably, RBP interactions with the cytokine 3'UTRs can alter upon T cell activation, as observed for increased binding by ZFP36L1, YTHDF1, WARS and decreased RNA binding by ZC3H12D and DDX39A. In some cases, alterations in RNA binding coincides with altered RBP protein expression (i.e. ZFP36L1, ZC3H12D). However, this is not observed for all RBPs. For example, YTHDF1, WARS and DDX39A change their binding potential upon T cell activation while maintaining their overall protein expression levels. This finding points to other regulatory mechanisms. For instance, RBPs can undergo post-translational modifications, as exemplified by stress-induced HuR phosphorylation that resulted in reduced target mRNA binding in HeLa cells (Kim et al., 2008; Yoon et al., 2014) . RBPs can also compete with each other for target binding. For instance, Arid5a and ZC3H12A interact with the same stem loop in the 3'UTR of STAT3 mRNA in HEK293T cells (Masuda et al., 2016) .

Similarly, HuR and ZFP36 (Tristetraprolin, TTP) compete for binding to TNF and GM-CSF mRNA in macrophages and HeLa cells, respectively (Raghavan et al., 2001; Tiedje et al., 2012) . Lastly, competition of RBPs with other RNA regulatory molecules such as miRNAs is conceivable (Srikantan et al., 2012) . Understanding the mechanisms that define RBP expression and function will be thus critical to further decipher their role in modulating T cell responses.

In this study, we tested 5 identified RBPs, of which 4 modulate the cytokine production in human T cells. Intriguingly, these 4 RBPs display different effects on cytokine production and with different activity kinetics. HuR promotes the cytokine production in human T cells, and does so most prominently at the onset (1-2 h) of T cell activation. This early boost in cytokine production may in particular be relevant for antigen-experienced T cells. These cells contain pre-formed mRNA, and immediate cytokine production is licensed by releasing the ready-to deploy mRNA from ZFP36L2 (Salerno et al., 2017 (Salerno et al., , 2018 . Our data here suggest that this early cytokine production is further facilitated by HuR. The mode of action of HuR is yet to be defined. Because we found no effects on mRNA stability, HuR may rather influence translation efficiency directly, or indirectly by rendering the mRNA accessible to the translation machinery through altering its subcellular localization, or through competing with other RBPs.

In contrast to HuR, the RBPs ZFP36L1, ATXN2L and ZC3HAV1 block cytokine production, and they do so most effectively at the peak (4-8 h), and ZFP36L1 (and ATXN2L) also at prolonged (24 h) T cell activation. ATXN2L and ZC3HAV1 do not affect RNA stability, and thus employ other post-transcriptional events to dampen cytokine production. In epithelial cell lines, ATXN2L regulates stress granules and processing bodies (Kaehler et al., 2012) . ZC3HAV1 was shown to destabilize IFNB and IFNL3 mRNAs in hepatocytes, and human cytomegalovirus RNA in fibroblasts, and it inhibits programmed ribosomal frameshifting of the SARS-CoV-2 virus (Schwerk et al., 2019; Gonzalez-Perez et al., 2021; Zimmer et al., 2021) . The mechanisms that ATXN2L and ZC3HAV1 employ in T cells to supress cytokine production, however, await further elucidation. ZFP36L1 destabilizes all 3 cytokines, and it interacts with regulators of mRNA degradation in activated T cells. One component of the CCR4-NOT complex, CNOT7, was previously shown to interact with ZFP36L1 in 293T cells (Adachi et al., 2014) , and is critical for degradation of ZFP36 targets in HeLa cells (Marchese et al., 2010; Sandler et al., 2011) . We here identified CNOT7 and several other CNOT proteins together with several exonucleases that possibly contribute to the degradation of target mRNAs of ZFP36L1 in primary T cells. Irrespective of the mode of action, the time point of activity of ZFP36L1, ATXN2L, and ZC3HAV1 clearly indicates their critical contribution in preventing excessive production of cytokines in activated T cells. Having multiple RBP regulators with different modes of action also highlights the need for fine-tuning the cytokine production for appropriate T cell function.

T cells lose their capacity to produce cytokines in the tumor environment. We previously showed that post-transcriptional events substantially contribute to the loss of IFN-γ production Freen-van Heeren et al., 2020) . Here, we report that deleting cytokine-impeding RBPs increases the cytokine production of human T cells when exposed to tumor cells. Importantly, the deletion of ZFP36L1 does not only delay the shutdown of cytokine production, but also boosts the effector function of TILs in vivo, which leads to better anti-tumoral responses. Deletion of ZFP36L1 does not only improve the production of IFN-γ, but also that of TNF and IL-2, and of the cytotoxic mediators Granzyme B and CD107a, indicating that ZFP36L1 instructs several T cell effector programs. Therefore, ZFP36L1 represents an attractive target to boost anti-tumoral T cell effector responses.

In conclusion, we here provide the first catalogue of dynamic RBP interactions to cytokine mRNAs in human T cells. Our results reveal that this map can serve as a valuable resource for studying the role of RBPs in regulating T cell responses, which can be tested in vivo against tumors to identify RBPs as potential targets for therapeutic purposes. Considering the association of RBPs with human genetic disorders (Gebauer et al., 2021) , our work should pave the way to support further studies on the role of RBPs in other immune-related disease settings.

To reduce false positive calls of RBP interaction, we have used a very stringent cutoff of LFC>4. This cut-off does by no means mean that a cut off <4 is not a valid target (as exemplified by ZFP36L2), or that >4 is per se an RBP interacting with the cytokine mRNAs. All identified RBPs require further validation. Furthermore, the noncrosslinking approach applied here for the RNA aptamer pulldown and the RIP assays may select for high-affinity RNA-RBP interactions, and weak interactions with RNA may therefore remain undetected. Irrespective of this restriction, we identify >100 RBPs interacting with cytokine 3'UTRs. This high number of identified RBPs could stem from heterogeneity of RBP-RNA interactions between different T cells, or from heterogeneous binding to specific mRNAs within the same T cell. It may thus reflect a broad spectrum of mRNA-RBP interactions that are possible in the T cells. Figure 4C ).

(C-E) IFN-γ, TNF, and IL-2 protein production kinetics of ZFP36L1 KO (red) and control (black) CD8 + T cells stimulated with α-CD3/α-CD28 for indicated time points.

Brefeldin A was added for the last 2h of activation (for the 1h time point: from the beginning). Figure 5C ). Ratio paired t test (*P <0.05; **P<0.01, ****P < 0.0001).

(D) Fold increase of % of cytokine production in human RBP KO T cells compared to control. Compiled data of 6 donors from 2 independently performed experiments.

Unpaired t test (*P <0.05; ***P<0.001 ****P < 0.0001). with Dunnett's multiple comparison to the control; *P < 0.05; **P < 0.01; ***P < 0.001).

CD4 + T cells (E) left untreated (resting, gray histograms), or that were reactivated for

and CD4 + T cells (E). Data are presented as mean ± s.d. of 3 donors and representative of at least 2 independently performed experiments (one-way ANOVA with Dunnett's multiple comparison to the control; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001). 

Gene-specific full length 3′UTRs were amplified from genomic DNA of a C57BL/6J mouse or from human genomic DNA. Human and mouse 3′ UTR variants were cloned into the BamHI and NotI sites of pRETRO-SUPER GFP (Table S6 ) (Brummelkamp et al., 2002) downstream of GFP, all under the control of the murine Pgk1 promoter. 4xS1m RNA aptamers containing the full length 3'UTR of human IFNG, TNF and IL2 were cloned into the pSP73-4xS1m vector (Leppek et al., 2014) as previously described (Salerno et al., 2018) . All sequences were confirmed by All animals were housed in individually ventilated cage systems under specificpathogen-free conditions. Mice were used at 6-12 weeks of age.

Studies with human T cells from anonymized healthy donors were performed in accordance with the Declaration of Helsinki (Seventh Revision, 2013) after written informed consent (Sanquin). Peripheral blood mononuclear cells (PBMCs) were isolated through Lymphoprep density gradient separation (Stemcell Technologies).

Cells were used after cryopreservation. Human T cells were activated as previously described (Nicolet et al., 2020) . Briefly, 24-well plates were pre-coated overnight with 0.5 μ g/mL rat α-mouse IgG2a (clone MW1483, Sanquin) at 4 ºC. Plates were washed once with PBS, coated with 1 μ g/mL α-CD3 (clone Hit3a, Biolegend) for a minimum of 4 h at 37 ºC. 1x10 6 PBMCs/well were seeded with 1 μ g/mL α-CD28 

For virus production, FLYRD18 (ECACC 95091902) cells for human and Plat-E cells (Morita et al., 2000) for mouse T cells were used. Both packaging cell lines were cultured at 37 °C and 5 % CO 2 in IMDM supplemented with 10 % FBS, 100 U/mL Penicillin, 100 µg/mL streptomycin and 2 mM L-glutamine. Virus packaging cells were plated into 6-well plates at 5-6 x 10 5 cells per well. Next day, cells were transfected with 1 μ g of retroviral vector using GeneJammer (Agilent), according to the manufacturer's protocol. After 48 h, virus supernatant was harvested, filtered through a 0.45 μ m filter and stored at -80 °C. For retroviral transduction of human T cells, PBMCs from individual donors were activated for 48 h with α-CD3/α-CD28 as described above. Transduction was performed with Retronectin (Takara) as described (Nicolet et al., 2020) . Briefly, non-tissue culture treated 24-well plates were coated with 50 μ g/mL Retronectin overnight at 4 °C and washed once with PBS.

Subsequently, 300-500 μ L viral supernatant was added per well and was centrifuged for 30 min at 4 °C at 4500 rpm (2820 g). 1x10 6 freshly activated T cells were added per well, spun for 5 min at 1000 rpm (180 g), and cultured at 37 °C and 5 % CO 2 in a humidified incubator. Cells were refreshed after 24 h and cultured in standing T25/75 flasks at a concentration of 0.8x10 6 cells/mL for 5 to 14 days in presence of rhIL-2 and rhIL-15. Medium was refreshed every 3 days.

For retroviral transduction of mouse T cells, preactivated CD8 + OT-I T cells were harvested and retrovirally transduced as described above. T cells were maintained with 10 ng/ml rm IL-7 and reactivated with 100 nM OVA 257-264 peptide (Genscript).

α-CD3/α-CD28 activated human T cells were cultured for 5 days. T cells that were activated with PMA/Ionomycin for 2 h or left untreated were washed twice with icecold PBS, and the cell pellet was snap frozen in liquid nitrogen. Cells were homogenized using 5-mm steel beads and a tissue lyser (Qiagen TissueLyser II) 6x

at 25 Hz for 15 s. The homogenate was then solubilized and precleared as previously described (Salerno et al., 2018) . Cell lysates were incubated with RNA-aptamer- Peptides were desalted and concentrated using Empore-C18 StageTips and eluted with 0.5 % (v/v) acetic acid, 80 % (v/v) acetonitrile. Sample volume was reduced by SpeedVac and supplemented with 2 % acetonitrile and 0.1 % TFA.

Cytoplasmic lysates of human T cells activated for 2 h with PMA/Ionomycin (100x10 6 cells per condition) were prepared using lysis buffer (140 mM NaCl, 5 mM MgCl 2 , 20 mM Tris/HCl pH7.6, 1 % Digitonin) that was freshly supplemented with 1% of protease inhibitor cocktail (Sigma). Protein A Dynabeads (Invitrogen) were prepared according to the manufacturer's protocol. The lysate was immunoprecipitated for 4 h at 4 °C with 10 μg polyclonal rabbit α-ZFP36L1 (ABN192, Sigma-Aldrich) or with a polyclonal rabbit IgG isotype control (12-370, Sigma-Aldrich). Beads were washed twice with wash buffer (150 mM NaCl, 10 mM Tris/HCl pH7.6, 2 mM EDTA, protease/phosphatase inhibitor cocktail), and then twice with 10 mM Tris/HCl pH7.6.

Immunoprecipitated proteins were reduced and alkylated on-beads. Proteins were detached from the beads by incubation with 250 ng trypsin for 2 h at 20 °C. Beads were removed and proteins were further digested into peptides with 350 ng trypsin for 16 h. Peptides were prepared for MS analysis, as described above.

Mass spectrometry data acquisition. The instrument was run in top speed mode with 3 s cycles. All data were acquired with Xcalibur software.

Mass spectrometry data analysis.

The raw mass spectrometry files were processed with the MaxQuant computational platform, 1.6.2.10 (Cox et al., 2008) . Proteins and peptides were identified using the Andromeda search engine by querying the human Uniprot database (downloaded February 2017 and February 2019, 89,796 entries) for the RNA pull-down. Standard settings with the additional options match between runs, and unique peptides for quantification were selected. The generated 'proteingroups.txt' data were imported in R and processed with the Differential Enrichment analysis of Proteomics data (DEP) R package (Zhang et al., 2018) . Identified peptides were filtered for potential contaminants, only identified by site and reverse hits. The raw intensity values were transformed in log2 scale and averaged, and log2 fold change (LFC) was calculated.

To identify enriched proteins, we used a cut-off of LFC> 4, compared to empty aptamer control. We only included proteins that were idenitifed in at least 2 out of 3

replicates in this analysis. To select for RBPs, we compiled the 1153 RBPs identified by RNA-interactome capture on HeLa and Jurkat cells (Castello et al., 2016; Perez-Perri et al., 2018) with 1542 computationally predicted RBPs based on the presence of a defined list of RNA-binding domains (RBDs) , resulting in a list of 2026 unique RBPs (Table S7) . Filtered data are shown as log2 mediancentered intensities. The MS data of the aptamer pulldown and the co-IP have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD028171.

Gene ontology analysis was performed with the Panther database (version 16.0) (Mi et al., 2019) . A statistical overrepresentation test (Fisher's exact with FDR multiple test correction) was performed with a reference list composed of all Homo Sapiens genes in the database. Overrepresented GO terms (FDR<0.001) were filtered for RNA-related functions and manually curated. Due to space constrains, only selected terms were included in the main figures. Full lists of overrepresented GO terms are provided in Table S2 and Table S4 . The R package ggplot2 was used to generate the graphical representations.

crRNAs were designed using the CRISPR design tools in Benchling (https://benchling.com, Table S6 ). Sequences were verified to be specific for the target of interest via BlastN (NCBI).

Cas9 RNP production and T cell nucleofection was performed as previously described (Freen-van Heeren et al., 2020 after electroporation by Western blot as described below.

Total RNA was extracted using Trizol (Invitrogen) and cDNA was synthesized with SuperScript III (Invitrogen). RT-PCR was performed with duplicate reactions using SYBR Green on a StepOne Plus (both Applied Biosystems). Ct values were normalized to 18S levels (Nicolet et al., 2017) .

To determine the half-life of cytokine mRNA, T cells were activated for indicated timepoint (1 or 3 h) with α-CD3 and α-CD28, and then treated with 5 μ g/ml actinomycin D (ActD) (Sigma-Aldrich).

Cytoplasmic lysates of human T cells activated with PMA/Ionomycin (300x10 6 cells per condition) were prepared using lysis buffer ( 

Human T cells were stimulated with 10ng/ml PMA and 1μM Ionomycin (both Sigma-Aldrich) or 1 μ g/mL pre-coated α-CD3 and 1 μ g/mL soluble α-CD28 according to manufacturer's protocol. Expression levels were acquired using FACSymphony (BD Biosciences) and data were analyzed using FlowJo (BD Biosciences, version 10).

Results are shown as mean ± s.d. Statistical analysis between groups was performed with 

ZFP36L1 and ZFP36L2 control LDLR mRNA stability via the ERK-RSK pathway

Control of RNA Stability in Immunity

Superior control of HIV-1 replication by CD8+ T cells is reflected by their avidity, polyfunctionality, and clonal turnover

The RNA binding protein Zfp36l1 is required for normal vascularisation and posttranscriptionally regulates VEGF expression. Developmental dynamics : an official publication of the

Stable suppression of tumorigenicity by virus-mediated RNA interference

Comprehensive Identification of RNA-Binding Domains in Human Cells

Polyfunctional HCV-specific T-cell responses are associated with effective control of HCV replication

Promotes Tumor Immunoresistance by Stabilizing PD-L1 mRNA

MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification

Human T cells employ conserved AU-rich elements to fine-tune IFN-γ production

RNA-binding proteins in human genetic disease

Evolutionary conservation and expression of human RNA-binding proteins and their role in human genetic disease. Advances in experimental medicine and biology

A census of human RNA-binding proteins

Manufacture of gene-modified human T-cells with a memory stem/central memory phenotype

The Zinc Finger Antiviral Protein ZAP Restricts Human Cytomegalovirus and Selectively Binds and Destabilizes Viral UL4/UL5 Transcripts

Posttranslational control of HuR function

Polyfunctional responses by human T cells result from sequential release of cytokines

Arid5a stabilizes OX40 mRNA in murine CD4+ T cells by recognizing a stem-loop structure in its 3'UTR

IFN-gamma AU-rich element removal promotes chronic IFN-gamma expression and autoimmunity in mice

Deletion of the RNA-binding proteins ZFP36L1 and ZFP36L2 leads to perturbed thymic development and T lymphoblastic leukemia

Posttranscriptional regulation of T helper cell fate decisions

T cells at work: How post-transcriptional mechanisms control T cell homeostasis and activation

Ataxin-2-like is a regulator of stress granules and processing bodies

Nuclear HuR accumulation through phosphorylation by Cdk1

iCLIP reveals the function of hnRNP particles in splicing at individual nucleotide resolution

Impaired on/off regulation of TNF biosynthesis in mice lacking TNF AU-rich elements: implications for joint and gut-associated immunopathologies

An optimized streptavidin-binding RNA aptamer for purification of ribonucleoprotein complexes identifies novel ARE-binding proteins

Recruitment and activation of mRNA decay enzymes by two ARE-mediated decay activation domains in the proteins TTP and BRF-1

MAPKAP kinase 2 blocks tristetraprolin-directed mRNA decay by inhibiting CAF1 deadenylase recruitment

Arid5a regulates naive CD4+ T cell fate through selective stabilization of Stat3 mRNA

PANTHER version 14: more genomes, a new PANTHER GO-slim and improvements in enrichment analysis tools

ZFP36 RNAbinding proteins restrain T cell activation and anti-viral immunity

Plat-E: an efficient and stable system for transient packaging of retroviruses

CD29 identifies IFN-γproducing human CD8+ T cells with an increased cytotoxic potential

Combined Single-Cell Measurement of Cytokine mRNA and Protein Identifies T Cells with Persistent Effector Function

A large-scale binding and functional map of human RNA-binding proteins

Intratumoral CD4+ T Cells Mediate Anti-tumor Cytotoxicity in Human Bladder Cancer

Identification of essential genes for cancer immunotherapy

Discovery of RNA-binding proteins and characterization of their dynamic responses by enhanced RNA interactome capture

CD8+ T cell differentiation and dysfunction in cancer

Tumor-reactive CD4+ T cells develop cytotoxic activity and eradicate large established melanoma after transfer into lymphopenic hosts

HuA and tristetraprolin are induced following T cell activation and display distinct but overlapping RNA binding specificities

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

Costimulation through TLR2 Drives Polyfunctional CD8 + T Cell Responses

TLR-Mediated Innate Production of IFN-γ by CD8 + T Cells Is Independent of Glycolysis

Critical role of post-transcriptional regulation for IFN-γ in tumor-infiltrating T cells

Distinct PKC-mediated posttranscriptional events set cytokine production kinetics in CD8(+) T cells

Dynamic Post-Transcriptional Events Governing CD8+ T Cell Homeostasis and Effector Function

Not1 mediates recruitment of the deadenylase Caf1 to mRNAs targeted for degradation by tristetraprolin

RNA-binding protein isoforms ZAP-S and ZAP-L have distinct antiviral and immune resolution functions

Functional interplay between RNA-binding protein HuR and microRNAs. Current protein & peptide science

Dynamic programming of CD8+ T lymphocyte responses

The p38/MK2-driven exchange between tristetraprolin and HuR regulates AU-rich element-dependent translation

We thank the animal caretakers of the NKI and the Sanquin FACS facility. We thank G. Stoecklin (University of Heidelberg) for the 4xS1m aptamer construct; T. 

The authors declare no competing interests.