Developmentally regulated PERK activity renders dendritic cells insensitive to subtilase cytotoxin-induced integrated stress response

In stressed cells, phosphorylation of eukaryotic initiation factor 2α (eIF2α) controls transcriptome-wide changes in mRNA translation and gene expression known as the integrated stress response (ISR). We show here that dendritic cells (DCs) display unusually high eIF2α phosphorylation, which is mostly caused by a developmentally regulated activation of the ER kinase PERK (EIF2AK3). Despite high p-eIF2α levels, differentiated DCs display active protein synthesis and no signs of a chronic ISR. eIF2α phosphorylation does not majorly impact DC differentiation nor cytokines production. It is however important to adapt protein homeostasis to the variations imposed on DCs by the immune or physiological contexts. This biochemical specificity prevents translation arrest and expression of the transcription factor ATF4 during ER-stress induction by subtilase cytotoxin or upon DC stimulation with bacterial lipopolysaccharides. This is also exemplified by the influence of the actin cytoskeleton dynamics on eIF2α phosphorylation and the migratory deficit observed in PERK-deficient DCs.


Introduction
Dendritic cells (DC) are key regulators of both protective immune responses and tolerance to selfantigens (Dalod et al., 2014). DCs are professional antigen presenting cells (APC), equipped with pattern recognition receptors (PRRs), capable of recognizing microbe-associated molecular patterns (MAMPs) (Akira et al., 2006) and enhance their immuno-stimulatory activity (Steinman, 2007). MAMPs detection by DCs triggers the process of maturation/activation, that culminates in the unique capacity of priming naïve T cells in secondary lymphoid organs. Lipopolysaccharide (LPS) detection by Toll-Like-Receptor 4 (TLR4) promotes DCs maturation by triggering a series of signaling cascades resulting in secretion of polarizing and inflammatory cytokines, upregulation of co-stimulatory molecules, as well as enhanced antigen processing and presentation (Mellman, 2013). All these functions are accompanied by major remodeling of membrane trafficking and actin organization to favor both antigen phagocytosis and DC migration to the lymph nodes (West et al., 2004, Bretou et al., 2017, Chabaud et al., 2015, Arguello et al., 2016. TLR4 is particularly expressed in the group 2 of conventional DCs (cDC2), characterized by the expression of surface CD11c, CD172a, as well as CD11b and endowed with a strong MHC II restricted antigen presentation capacity (Dalod et al., 2014).
One the major features of DC activation is a large increase in protein synthesis (Lelouard et al., 2007, Reverendo et al., 2019. This is required for the up-regulation of co-stimulatory molecules at the cell surface and acquire T cell immune-stimulatory function. Among the many pathways controlling mRNA translation, phosphorylation of eukaryotic initiation factor 2 (eIF2) is a central hub for regulating protein synthesis during stress. In homeostatic conditions, eIF2 mediates the assembly of the translation initiation complex and regulates start codon recognition in the mRNA by promoting the recruitment of other components necessary for active protein synthesis. During stress, phosphorylation of the alpha subunit of eIF2 (eIF2α) on serine 51 is mediated by a group of four eIF2α kinases (EIF2AK1-4), that specifically sense physiological imbalance (Claudio et al., 2013, Arguello et al., 2016. Phosphorylation of eIF2α converts eIF2 into a competitive inhibitor of the GDP-GTP guanidine exchange factor (GEF) eIF2B, impairing the GDP-GTP recycling required to form new translation initiation complexes and proceed with active mRNA translation (Yamasaki and Anderson, 2008). High levels of eIF2α phosphorylation impacts cells in two main ways: i) By reducing the rate of translation initiation and thus global protein synthesis levels; ii) By favoring the translation of the activating transcription factor 4 (ATF4) (Fusakio et al., 2016, Han et al., 2013, that in turn activates the transcription of many genes involved in the integrated stress response (ISR). The ISR aims at protecting cells from amino acid deprivation, oxidative, or mitochondrial stress and is incorporated as a branch of the ER unfolded protein response (UPR) upon PERK-activation. The ISR comprises a negative feedback loop that causes eIF2α dephosphorylation, through the induction of GADD34 (also known as PPP1R15a), a phosphatase 1 (PP1c) co-factor (Novoa et al., 2001. Consequently, dephosphorylation of p-eIF2α by GADD34/PP1c complexes, and associated protein synthesis restoration, signal ISR termination and a return to cellular homeostasis (Novoa et al., 2001). If stress persists, despite GADD34 activity, long term ATF4 expression promotes programmed cell death, through the induction of the pro-apoptotic transcription factor CHOP (Marciniak et al., 2004).
Interestingly, ATF4 also regulates the expression of Rho GTPases and can control cell motility (Pasini et al., 2016), while globular actin has been shown to be part of the PP1c/GADD34 complex and provide additional targeting specificity for dephosphorylating p-eIF2α (Chambers et al., 2015, Chen et al., 2015. The process of MAMPs sensing severely affects cellular homeostasis by boosting protein synthesis and results in a panoply of biochemical insults, that trigger various responses including autophagy and the ISR (Valecka et al., 2018, Arguello et al., 2016. the ISR can enter in a crosstalk with specialized microbe sensing pathways, which turns on or amplifies inflammatory cytokines production (Claudio et al., 2013. Toll-like receptor (TLR) activation in macrophages undergoing an ISR, suppress CHOP induction and protein synthesis inhibition, preventing apoptosis in activated cells (Woo et al., 2009). Moreover, ATF4 binds Interferon Regulatory factor-7 (IRF7) to block its capacity to induce transcription of type-I Interferon (IFN) induction (Liang et al., 2011). Several key innate immunity signaling cascades are believed to be dependent for their signalosome assembly on the chaperone HSPB8 and the kinase heme-regulated inhibitor (HRI/EIF2AK1) (Pierre, 2019). Microbe-activated HRI was shown to mediate phosphorylation of eIF2α and increase ATF4-dependent expression of HSPB8, thus promoting the signal transduction pathways leading to inflammatory cytokines transcription in macrophages (Abdel-Nour et al., 2019).
We show here that DCs obtained from spleen or derived from Fms-related tyrosine kinase 3 ligand  treated-bone marrow cultures, display unusually high levels of phosphorylated eIF2α. Using Cre/lox recombination to generate mice specifically lacking GADD34 (PPP1R15a) or PERK (EIF2AK3) activity in DCs, we demonstrate that PERK-dependent eIF2α phosphorylation is acquired during DC development. PERK drives high eIF2α phosphorylation in steady state DCs with a relatively low impact on protein synthesis. We found that mRNA translation in DCs, differently to what has been shown during chronic ISR , is mediated despite high p-eIF2α levels by an eIF4F-dependent mechanism. We also found that primary DCs, conversely to macrophages, do not rely on eIF2α phosphorylation to activate cytokines transcription nor IL-1β secretion in response to MAMPs (Abdel-Nour et al., 2019), (Chiritoiu et al., 2019). Importantly, GADD34 antagonizes PERK activity to maintain functional protein synthesis levels in non-activated DCs. GADD34 upregulation in response to LPS, occurs in an ATF4-independent and TBK1-dependent manner, and contributes therefore directly to DC function by modulating the production of pro-inflammatory cytokines. Surprisingly, PERK activity impacts positively DC migration speed, correlating with the regulation of p-eIF2α levels by the synergistic action of GADD34 and actin cytoskeleton reorganization. Thus, DCs require PERK and GADD34 activity to coordinate protein synthesis, activation and migration capacity in response to MAMPs. These developmentally regulated features also endow DC with increase resistance to acute ER stress, preventing ATF4 induction in response to stressors like the bacterial subtilase cytotoxin (SubAB).

Cell culture
Bone-marrow (BM) was collected from 6-9 weeks old female mice, and differentiated in DCs or macrophages during 7 days. The culture was kept at 37°C, with 5% CO2 in RPMI (GIBCO), 10% FCS (Sigma Aldrich), 100 units/ml penicillin, 100 units/ml streptomycin (GIBCO) and 50µM βmercaptoethanol (VWR) supplemented with Flt3-L, produced using B16-Flt3-L hybridoma cells (Naik et al., 2005) for DC differentiation or M-CSF for macrophages, as described previously . For the migration assays, granulocyte-macrophage colony stimulating factor (GM-CSF) was used instead of Flt3-L and cells were cultured during 10-12 days with changes in the medium each 3 days. GM-CSF was obtained from transfected J558 cells (Pierre et al., 1997).To obtain splenocytes, spleens were collected, and injected with Liberase TL (Roche) and incubated 25 min at 37°C to disrupt the tissues. DC purification was performed using a CD11c+ positive selection kit (Milteny), according to manufacturer's instructions and CD8α+ T cells isolation was performed with a Dynabeads TM untouched TM mouse CD8 T cells kit from Thermo Fischer Scientific. CD8α+ T where incubate overnight with anti-CD3 (10µg/ml) and anti-CD28 (5µg/ml) antibodies, to mimic activation by APCs. Mouse embryonic fibroblasts used in this work (MEFs), ATF4-/-and matched WT (129 SvEv) were a kind gift from Prof. David Ron (Cambridge Institute for Medical Research, UK). PERK KO-/-and matched WT were a kind gift from Prof. Douglas Cavener (Penn State University). MEFs were cultured in DMEM medium (GIBCO) with 5% FBS (Sigma) and 50 µM 2-mercaptoethanol. For the experimental assays, cells were plated from 16-24h before stimulation in 6 well plates, at 150 000 cells/ml in 2 ml of the same medium. After stimuli, cells were treated with trypsin-EDTA for 2 minutes at 37°C before washing in order to detach cells from the wells.

Reagents
Lipopolysaccharide (Escherichia coli O55:B5), Cycloheximide, Puromycin, MRT67307 and Rocaglamide and Thapsigargin were purchased from Sigma Aldrich. Harringtonine is from ABCAM, Latrunculin A and Jasplakinolide are from Merck-Millipore. Low molecular weight Polyinosinicpolycytidylic acid (LMW poly(I:C)) was from InvivoGen. SAR1 was kindly provided by Sanofi (France) and Integrated Stress Response Inhibitor (ISRIB) was a gift from Carmela Sidrausky and Peter Walter, UCSF, San Francisco). Subtilase cytotoxin (Shiga toxigenic Escherichia coli strains) was purified from recombinant E. coli, as previously described (Paton et al., 2004). 4EGI-1 was purchased by Bertin bioreagent. from BDBiosciences. These antibodies were used in combination with the LIVE/DEAD ® Fixable Aqua Dead Cell Stain (ThermoFisher). For intracellular staining, cells were next fixed with BD phosflow fix buffer I (BD Biosciences) during 10 min at room temperature and washed with 10% Perm/wash Buffer I 1X (BD Biosciences). Permeabilized cells were blocked during 10' with 10% Perm/wash buffer 1X, 10% FCS, before staining with primary antibodies. When the primary antibody was not coupled, cells were washed after and blocked during 10 min with Perm/wash buffer 1X, 10% FCS, 10% of serum from the species where the secondary antibody was produced. Then, the incubation with the secondary antibody was performed at 4°C during 30 min. p-eIF2α(S51) was purchased from ABCAM and p-eEF2(Thr56) from Cell signaling and Deoxyribonuclease (DNAse I) was purchased from Invitrogen.

Flow cytometry analysis
Data were acquired on an LSR-II/UV instrument using FACS Diva software. The acquired data were analyzed with FlowJo software (BD Biociences).

Translation intensity and speed measurement
SUnSET technique to measure the intensity of protein synthesis was used as previously described (Schmidt et al., 2009). Puromycin was added in the culture medium at 12.5 μg/ml and the cells were incubated for 10 min at 37°C and 5% CO2 before harvesting. Cells were washed with PBS prior to cell lysis and immunoblotting with the anti-puromycin 12D10 antibody (Merck Millipore). For flow cytometry (flow) cells were processed, as described below for the intracellular staining, using the α-puromycin 12D10 antibody directly conjugated with Alexa 488 or A647 from Merck Millipore. The SUnRISE technique was performed as described . Samples were treated with 2µg/ml of harringtonine at different time-points (90,60,45,30,15, 0 sec), and then treated with 12.5µg/ml of puromycin all at the same time. For the measurement of Cap-dependent translation cells were treated with 4EGI-I (100µM) or Rocaglamine (RocA-1) (100nM) for 0.5, 1, 2, or 4h. Cells were then incubated for 15 min at 37°C and stained with the 12D10 antibody (Merk-Millipore).

Gene expression analysis
Total RNA was extracted from the DCs using the RNeasy mini kit (QIAGEN), including a DNA digestion step with RNAse-free DNAse (QIAGEN) and cDNA was synthesized using the Superscript II reverse transcriptase (Invitrogen). Quantitative PCR amplification was performed using SYBR Green PCR master mix (Takara) using 10ng of cDNA and 200nM of each specific primer on a 7500 Fast Real-PCR system (Applied Biosystems). cDNA concentration in each sample was normalized to GAPDH

Gene set enrichment analysis
The gene set enrichment analysis has been performed using published murine microarray datasets accessible through the Gene Expression Omnibus (GEO) repository under the references GSE9810 (Robbins et al., 2008) and GSE2389 (Fontenot et al., 2005). Raw microarray data describing CD8∝ + cDCs (DC1), CD11b+ cDC (DC2), pDC (plasmacytoid dendritic cells), B cells, NK cells, CD8+ T cells (Robbins et al., 2008) and CD4+ T cells (Fontenot et al., 2005) in mice spleen have been download For each of these cell types, the hybridation has been performed using Affymetrix mouse 4302.0 gene chips. Microarray data have been normalized by Robust Multi-array Average (RMA) algorithm (Irizarry et al., 2003) using the oligo BioconductorR package (Carvalho and Irizarry, 2010). Normalization consists of a background correction of raw intensities, a log2 transformation followed by quantile normalization in order to allow the comparison of each probe for each array. Prior data usage, the absence of batch effect was assessed by Principal Component Analysis (PCA) using ade4R package (Bougeard and Dray, 2018 were from Jackson ImmunoResearch Laboratories.

Cytokines measurement
The IL-6, the IFN-β, and IL1β quantifications from the cell culture supernatant were performed using the mouse Interleukin-6 ELISA kit (eBioscience), the mouse IFN-β ELISA kits (PBL-Interferon Source or Thermos Fisher) according to manufacturer's instructions.

Immunohistochemistry
Spleens were snap frozen in Tissue Tek (Sakura Finetek). Frozen sections (8 μm) were fixed with acetone permeabilized with 0,05% saponin. The following antibodies were used for the staining: CD11c Invitrogen (Ref 14-0452-81). Images were collected using a Zeiss LSM 510 confocal microscope. Image processing was performed with Zeiss LSM software.
PERK loxp/loxp mice were the kind gift of Dr Doug Cavener  and purchased from Jackson Laboratories. GADD34ΔC loxp/loxp mice were developed at the Centre d'Immunophénomique (CIPHE, Marseille, France). PERK loxp/loxp and GADD34 loxp/loxp were crossed with Itgax-Cre+ mice (Caton et al., 2007) and backcrossed, to obtain stable homozygotic lines for the loxp sites expressing Cre. For all studies, age-matched WT and transgenic 6 to 9 weeks females were used. All animals were maintained in the animal facility of CIML or CIPHE under specific pathogen-free conditions accredited by the French Ministry of Agriculture to perform experiments on live mice. These studies were carried out in strict accordance with Guide for the Care and Use of Laboratory Animals of the European Union.
All experiments were approved by the Comité d'Ethique PACA and MESRI (approval number APAFIS#10010-201902071610358). All efforts were made to minimize animal suffering.

Preparation of microchannels and speed of migration measurement
Microchannels were prepared as previously described (Vargas et al., 2016). For velocity measurements (carried out in 4µm-by-5µm microchannels), phase contrast images of migrating cells were acquired during 16 hours (frame rate of 2 min) on an epifluorescence Nikon Ti-E video microscope equipped with a cooled charge-coupled device (CCD) camera (HQ2, Photometrics) and a 10× objective. Kymographs of migrating cells were generated and analyzed using a custom program.

Statistical analysis
Statistical analysis was performed using GraphPad Prism Software. The most appropriate statistical teste was chosen according to the set of data. Mainly, we used Student's t test and Kruskal-Wallis test were used. * p < 0.5, ** p <0.01, *** p < 0.001, **** p < 0.0001.

Steady-state dendritic cells display high levels of eIF2a phosphorylation
Given the importance of eIF2a in cell homeostasis, we monitored the physiological levels of phosphorylated eIF2a (p-eIF2a) in mouse spleen sections by immunohistochemistry. All DC subsets, identified by virtue of CD11c expression together with CD8a (cDC1), CD11b (cDC2) or B220 ( proportions (circa 30%, 60% and 10% respectively of the CD11c+ population) and with phenotypes equivalent to those of spleen DC subsets (Brasel et al., 2000). This was further confirmed upon sorting and analysis of the different populations, which were submitted to p-eIF2a immunoblot in comparison to isolated primary CD8+ T cells or mouse embryonic fibroblasts (MEFs) stimulated or not with the ERstress inducing drug thapsigargin for 2h (Fig. 1D). eIF2a phosphorylation was evident in all DC subsets, even when compared to thapsigargin-treated control T cells and MEFs. Quantification of p-eIF2a/eIF2a ratios indicated that steady state DCs display 2 to 4 times more p-eIF2a, than stressed MEFs, with the cDC1 population displaying the highest ratio of phosphorylation (Fig. 1D). We next evaluated when eIF2a phosphorylation was acquired during DC differentiation. The daily immunoblot analysis of differentiating Flt3-L BMDCs, allowed us to establish that high p-eIF2a levels were displayed by DCs from 4 days of culture ( Fig. 1E). This suggests that this event is an integral part of the DC development program induced by Flt3-L (Fig. 1E). Given the described dominant negative effect of p-eIF2a on translation initiation, we next sought to monitor the levels of protein synthesis in the different DC subsets. CD8+ T cells were used as a reference, since in these cells, p-eIF2a is barely detectable. We used puromycilation and detection by flow cytometry (flow) (Schmidt et al., 2009 to establish that, despite high eIF2a phosphorylation levels, all resting DC subsets displayed higher level of translation compared with resting and CD3/CD28 stimulated CD8+ T cells (Fig. 1F). We next monitored protein synthesis every two days of culture to establish precisely the influence of eIF2a phosphorylation over protein synthesis activity during DC differentiation in vitro. After puromycilation and flow analysis using various DC differentiation markers, we applied dimensionality reduction using t-distributed stochastic neighbor embedding (tSNE) to visualize DC differentiation and the evolution of protein synthesis activity within the different subpopulations over time ( Fig. 2A). We confirmed that protein synthesis levels steadily increased with differentiation in all three DC subsets, this despite high eIF2a phosphorylation. These observations suggest that steady state DCs have adapted their translation machinery to overcome the dominant negative effect on translation initiation of eIF2a phosphorylation on Ser51, which is clearly associated with their differentiation from bone marrow progenitors.
eIF2B expression is strongly up-regulated upon DC differentiation P-eIF2a inhibits translation initiation by forming a stable inhibitory complex with eIF2B, that reduces the GEF activity of eIF2B. eIF2B levels are generally lower than those of eIF2 and partial eIF2a phosphorylation is sufficient to attenuate protein synthesis initiation in most cells (Adomavicius et al., 2019). We therefore monitored eIF2Be expression levels during DC differentiation and observed that eIF2Be levels were strongly increased both transcriptionally and translationally in differentiated cells ( Fig. 2B & 2C). The progression of the ratio eIF2Be over eIF2a proteins during differentiation (Fig. 2C), suggests that the quantity of p-eIF2a required to inhibit translation initiation in DCs is superior to what is needed in other cells like MEFs. This developmentally regulated expression of eIF2B and eIF2S1, could therefore explain the adaptation of DCs to relatively high p-eIF2a levels and the progressive acquisition of high translation levels upon differentiation.

Flt3-L BMDCs activation by LPS promotes eIF2a-dephosphorylation.
Phosphorylation of eIF2a and of elongation factor 2 (eEF2) were gradually lost during E. coli lipopolysaccharide (LPS)-stimulation of Flt3-L BMDCs, indicating a tight regulation of this pathway during DC activation (Fig. 3A). Complementary to p-eIF2a, phosphorylated eEF2 is a major repressor of translation in adverse growth conditions, such as starvation, or accumulation of misfolded proteins in the endoplasmic reticulum (ER) (Ryazanov, 2002), (Lazarus et al., 2017). Stress activates elongation factor 2 kinase (eEF2K), which phosphorylates eEF2 and inhibits translation elongation through an arrest of codon translocation from ribosomal A to P sites (De Gassart andMartinon, 2017, Richter andColler, 2015). Similarly to eIF2a, eEF2 phosphorylation levels were found high in steady-state Flt3-L BMDCs and down-regulated after LPS stimulation (Fig. 3B). Given their sensitivity to LPS-activation, we monitored more specifically TLR4 expressing DC2 by flow, and observed a profound loss of p-eEF2 after 30 min of stimulation, thus demonstrating the strong impact of LPS signaling on this major translation factor and on protein synthesis intensity, which was doubled over 6h of stimulation (Fig. 3C).
Given the rapidity and intensity of eIF2α and eEF2 dephosphorylation upon activation, we applied to cDC2, the SunRISE technique, a method for monitoring the translation level and elongation rate of multiple cell types in parallel . SunRiSE was applied using detection by immunoblot (Fig. 3D) and by flow (Fig. 3E), confirming a striking augmentation of translation intensity in LPS-activated DCs compared to the steady-state situation (T=0s). Polysomes elongation speed, indicated by the rate of puromycin staining decay after harringtonine treatment, remained however identical among steady-state and LPS-activated cDC2 (Fig. 3E). eIF2α and eEF2 dephosphorylation in activated DCs is therefore correlated with a strong increase in translation initiation, rather than elongation, to allow mRNA translation to reach its maximum level concomitantly with the acquisition of DC immune-stimulatory capacities (Lelouard et al., 2007).

PPP1R15a (GADD34) controls eIF2α dephosphorylation in activated DCs.
The inducible PP1c co-factor PPP1R15a, known as GADD34, is key in mediating p-eIF2a dephosphorylation in the resolution phase of the ISR (Novoa et al., 2001, Novoa et al., 2003. GADD34 is also absolutely necessary to re-establish protein synthesis in MEFs undergoing PKR activation upon dsRNA exposure or viral infection (Clavarino et al., 2012a). We have also shown that while induction of GADD34 transcription during ER stress is ATF4-dependent (Walter and Ron, 2011), expression of GADD34 upon viral sensors activation is Interferon Regulatory Factor 3 (IRF3 dependent (Dalet et al., 2017). We therefore investigated if the p-eIF2a dephosphorylation observed in Flt3-L BMDCs was GADD34 dependent, and whether TANK-binding kinase 1 (TBK1)/IRF3 signaling cascade, which is triggered upon LPS stimulation, was responsible for its transcriptional induction during DC activation.
The IKKϵ/TBK1 inhibitor (MRT67307, TBKin) (Clark et al., 2011) was used to follow the impact of IKKϵ/TBK1/IRF3 activation on LPS-induced p-eIF2a dephosphorylation (Fig. 4A). Upon efficient treatment with TBKin, indicated by the loss of p-IRF3 detection, p-eIF2a levels were clearly augmented even in absence of LPS stimulation, suggesting that baseline IKKϵ/TBK1/IRF3 signaling decreases p-eIF2a levels by driving steady state GADD34 expression. LPS-dependent induction of GADD34 mRNA was also severely decreased upon IKKϵ/TBK1 inhibition (Fig. 4B), suggesting that in DCs, Ppp1r15a/GADD34 transcription is partially dependent on IRF3 activation and not only on the ATF4dependent transcriptional axis. This hypothesis was reinforced by the observation that ATF4 and CHOP expression, the two main transcription factors controlling Ppp1r15a/GADD34 transcription during ERstress, remained stable after 6h of LPS stimulation, and in the case of CHOP even decreased in the first hours of activation (Fig. 4C). Surprisingly, IKKϵ/TBK1 inhibition also augmented considerably MHC II surface expression (Fig. 4D), presumably by changing the levels of secreted cytokines influencing DC maturation in the culture. This was further exemplified by the absence of IRF3-dependent IFN-β expression in TBKin-treated cells (Fig. 4E), while NF-κB-dependent transcription of IL-6 remained unaffected (Fig. 4E), as well as CD86 surface appearance (Fig. 4D). Importantly, despite augmenting p-eIF2a levels, IKKϵ/TBK1 inhibition also strongly boosted protein synthesis suggesting that DCs conserve the capacity to augment their protein synthesis irrespective of TBK activation and despite a low rate eIF2a dephosphorylation (Fig. 4F). Artificial inhibition of IKKϵ/TBK1, prevents therefore GADD34 and IFN-β expression, while augmenting eIF2a phosphorylation. However it reveals also the existence of a signaling pathway downstream of TLR4, capable of overcoming augmented levels of p-eIF2a and of increasing protein synthesis, independently of GADD34 expression. Given the amount of eIF2B expressed by differentiated DCs (Fig. 2B and 2C), this pathway could be equivalent to the TLRdependent activation of the eIF2B GEF through PP2A-mediated serine dephosphorylation of the eIF2Bϵ-subunit, that was observed by Woo and colleagues in macrophages (Woo et al., 2012).
We further investigated the impact of GADD34 on the eIF2a pathway in DCs by generating, a novel transgenic mouse model with floxed alleles for PPP1R15a, allowing the deletion of the third exon coding for the C-terminal PP1 interacting domain, which upon Cre recombinase expression create a null phenotype for GADD34-dependent eIF2a dephosphorylation ) (Sup Fig. S1A).
Ppp1r15a loxp/loxp C57/BL6 mouse were crossed with a Itgax-cre deleter strain (Caton et al., 2007) to specifically inactivate ppp1r15a activity in CD11c expressing cells, including all DC subsets. Despite inducing a light splenomegaly, ppp1r15a inactivation had no obvious consequences for splenocytes development in vitro and in vivo (Sup. Fig. S2). Flt3-L BMDCs derived from WT and Itgaxcre/Ppp1r15a loxp/loxp (GADD34ΔC) mice were LPS activated prior detection of translation factors by immunoblot (Fig. 5A). GADD34 inactivation prevented LPS-dependent eIF2a dephosphorylation, however no impact was observed on the phosphorylation levels of the activator beta subunit of eIF2 (eIF2β), eEF2, nor ribosomal S6 protein, demonstrating that among these translation factors subunits, GADD34 only targets eIF2a with high specificity (Fig. 5A). Consequently, p-eIF2a levels do not influence directly the fate of the eIF2β subunit and eEF2 in DC, nor their respective impact on translation. eIF2β phosphorylation has been previously shown to counteract p-eIF2a dominant negative effect and promote mRNA translation (Gandin et al., 2016). However, it was neither impacted by LPS activation nor GADD34 activity in DCs, and thus unlikely to interfere with eIF2a regulation. Functional deletion of GADD34 inhibited translation initiation in both steady-state and LPS-activated cDC2 (Fig.   5B), and also reduced moderately translating polysomes speed in non-stimulated cells. GADD34 expression seems therefore to prevent exaggerated protein synthesis inhibition linked to abundant eIF2a phosphorylation in steady-state DCs. The amount of eIF2B present in the DC seems, however, sufficient to maintain a lower but still overall active protein synthesis despite GADD34 inactivation and increased P-eIF2a, as already suggested by our results obtained with TBKin treatment (Fig. 4F). These results confirm that GADD34 is produced and active in steady-state DC, suggesting the existence of a basal IRF3-dependent Ppp1r15a mRNA transcription in DCs. Interestingly these results also imply that GADD34 mRNA translation occurs independently of any acute stress in DCs, despite its upstream uORF-dependent translational regulation, as phenomenon recently observed in MEFs (Reid et al., 2016).

PERK mediates eIF2a phosphorylation in steady state DCs.
We next investigated the consequences of inactivating known eIF2A kinases on the p-eIF2a levels observed in steady-state Flt3-L BMDCs (Krishna and Kumar, 2018). We tested pharmacological and genetic inactivation of PKR (EIF2AK2) and GCN2 (EIF2AK4) (Sup Fig. S3), without observing any major disturbances in eIF2a phosphorylation levels. We next turned towards the ER-stress kinase PERK (EIF2AK3) by crossing PERK loxp/loxp mice with the Itgax-cre strain (Caton et al., 2007) allowing for the deletion of the exons 7-9, coding for the kinase domain (PERKΔK), and consequently PERK inactivation in CD11c-expressing cells (Sup Fig. S1B). PERK protein levels were enriched in WT CD11c+ splenic DC compared to other splenocytes (Fig. 6A), further suggesting the importance of this molecule for DC function. PERK expression was efficiently abrogated in CD11c+ splenocytes and Flt3-L BMDCs derived from animals bearing the floxed-PERK alleles (Fig. 6A), with no obvious consequences for splenocytes development in vitro and in vivo (Sup. Fig. S4). PERK inactivation decreased p-eIF2a levels by 60% in steady-state DCs (Fig. 6B), while p-eEF2 levels remained unchanged (Fig. 6C). Interestingly, LPS stimulation increased PERK levels in WT Flt3-L BMDCs, suggesting that this kinase could be further activated during DC maturation. However, despite increased PERK levels, p-eIF2a levels were nevertheless decreased with activation, reaching a minimum at 4h of LPS stimulation (Fig. 6B). This low phosphorylation intensity of eIF2a was equivalent to the basal levels observed in PERK-deficient DCs, and likely resulting from other eIF2AKs activities (Krishna and Kumar, 2018) and PP1c mediated dephosphorylation. Importantly, and conversely to GADD34-deficient DCs, PERK deletion caused an increase in translation initiation and protein synthesis intensity as measured by SunRISE in both steadystate and LPS stimulated cDC2 (Fig. 6D). PERK is therefore the EIF2AK responsible for the majority of eIF2a phosphorylation in Flt3-L BMDCs DCs, and mirrors GADD34 activity to regulate active protein synthesis at steady state and during DC activation. These features presumably contribute to the acquisition of their immuno-stimulatory functions by DCs. which is one of the major transcription factors induced during the UPR (Han et al., 2013). Transcription factors ATF4 and ATF6 mRNA expression increased moderately during DC differentiation, their activation occurring mostly at the post-transcriptional level during stress (Walter and Ron, 2011). We next wondered if the relatively high PERK activity observed in developing DCs, could induce a chronic ISR in these cells. Recently, translational and transcriptional programs that allows adaptation to chronic ER stress has been described by Guan et al. (Guan et al., 2017). This chronic ISR (cISR) operates via PERK-dependent mechanisms, which allow simultaneous activation of stress-sensing and adaptive responses while allowing recovery of protein synthesis. Given the biochemical similarities with artificial cISR displayed by steady state DCs, we took advantage of available transcriptomic data (GSE9810, GSE2389) and of ATF4/CHOP-dependent gene (Han et al., 2013) to perform a Gene Set Enrichment Analysis (GSEA) and define the level of common gene expression found in DCs and potentially shared with a cISR. GSEA was followed by multiple testing correction (Subramanian et al., 2005) using the BubbleGUM software, which allows statistical assessment and visualization of changes in the expression of a pre-defined set of genes in different conditions (Spinelli et al., 2015). GSEA revealed no significant enrichment of ATF4-and CHOP-dependent genes expression in the DC transcriptome (FDR >0.25) (Sup Fig. S6A and S6B). Next, we compared cISR dependent transcription  with steady state splenocytes transcriptome, and found again no significant enrichment of cISR genes in the DC transcriptome compared to reference samples exposed to 1h or 16h of thapsigargin (Sup Fig. S6C and S6D) (FDR >0.25). DCs have therefore adapted to the consequences of high eIF2α phosphorylation to allow for translation of the DC transcriptome, without induction of chronic ISR.
PERK-and eIF2α-P-mediated translational reprogramming during cISR appears to bypass cap-mediated translation and therefore to be largely eIF4F independent . We therefore used 4EGI-1, an inhibitor that interferes with eIF4F assembly (Moerke et al., 2007), and ROCA, an eIF4A inhibitor (Iwasaki et al., 2019), to treat WT and PERK-deficient DCs and confirm the dependency of their protein synthesis on eIF4F activity. Both small compounds had a profound inhibitory effect on DCs translational activity (80% of reduction), irrespective of their subsets or activation state (Sup Fig.   S5B and S5C). This important level of inhibition indicates that DC mostly depend on eIF4F-dependent cap-mediated translation, thus again contrasting from cells undergoing chronic ISR .
Despite high p-eIF2α levels, steady state DCs do not display a cISR-like response, ATF4-or CHOPdependent gene transcription programs being unlikely part of DC development and function.
This lack of ATF4-dependent gene signatures in DCs when compared with other CD45+ cell types, lead us to wonder whether the high eIF2-P levels observed at steady state could interfere with DC capacity to respond to ER-stress. We thus exposed Flt3-L BMDCs to subtilase cytotoxin (SubAB), a bacterial AB5 toxin that promotes the release of the ER-chaperone BiP (HSPA5, Heat Shock Protein Family A (Hsp70) Member 5) and a PERK-dependent ISR (Paton et al., 2006). When WT and PERKΔK DCs were submitted to SubAB treatment a modest PERK-dependent induction of p-eIF2a was observed in WT cells (Fig. 7A), with limited consequences on translation (Fig. 7B). Conversely, thapsigargin treatment, arrested translation efficiently and triggered eIF2a phosphorylation potentially through a different EIF2AK than PERK, since high p-eIF2a levels were induced in PERKΔK cells ( Fig. 7A and B).
Little ATF4, could be detected in the cytosolic or nuclear fractions of control or toxin-treated DC, reflecting the modest induction of eIF2α phosphorylation ( Fig. 7C and 7D), and confirming the limited impact of SubAB on DC translation and ISR induction. The efficacy of SubAB treatment was tested in MEFs, in which ATF4 was strongly induced by the toxin, and of course absent from control ATF4 -/-cell ( Fig. 7C). DCs are therefore resistant to SubAB treatment and are unable to induce the ISR, despite the activation of other UPR branches, as demonstrated by augmented IRE1a-dependent splicing of the XBP1 mRNA (Fig. 7E). BiP mRNA levels were moderately augmented during DC differentiation.
However, the similar expression ranges observed for DCs and MEFs (Fig. 7F) suggest that BiP transcriptional regulation is not involved in the DC resistance to SubAB. The relatively high PERK and GADD34 activity observed in steady-state DCs, together with high eIF2B expression, seems therefore to prevent the consequences of further activation of the ER kinase, leading to a minor increase in eIF2a phosphorylation and avoiding full ISR induction, translation arrest and associated ATF4 and CHOP synthesis.

Importance of the ISR for innate receptors signaling in DCs.
In macrophages, p-eIF2a-and ATF4-dependent expression of HSPB8 is required for the assembly of signaling adaptors such as mitochondrial antiviral-signaling protein (MAVS), or TIR domain-containing adapter protein inducing interferon-a (TRIF), but not of Myeloid Differentiation Primary Response Gene 88 (MyD88). HSPB8 seems necessary for TRIF and MAVS to be incorporated into protein aggregates that constitute signalosomes for different innate immunity signaling pathways triggered by MAMPs (Abdel-Nour et al., 2019). Given the lack of ATF4 synthesis and the resistance of active DCs to mount an acute ISR, we investigated their capacity to produce pro-inflammatory cytokines and type-I IFN in a perturbed eIF2a-phosphorylation context. Importantly, we have previously shown that GADD34deficient GM-CSF BM-derived and spleen DCs, have a reduced capacity to produce different cytokines, like IFN-β (Clavarino et al., 2012b, Perego et al., 2018, further suggesting that GADD34 activity and potentially eIF2a dephosphorylation is an important process for TLR signaling and cytokine production in DCs, although contrasting with the data obtained in macrophages (Abdel-Nour et al., 2019). LPS and the dsRNA synthetic mimic polyriboinosinic:polyribocytidylic acid (poly(I:C)), that is recognized by TRIFdependent TLR3 and the MAVS-dependent cytosolic RNA helicases RIG-I and MDA5 (RLRs) Kagan, 2020, Ablasser andHur, 2020) were used to stimulate DCs (Fig. 8). Treatment with ISRIB, a pharmacological ISR inhibitor (Sidrauski et al., 2015), that prevents inhibition of eIF2B and promotes translation initiation (Fig. 8A), was included to prevent the induction of the ISR in activated Interleukin-1 beta (IL-1β) secretion in BMM. PERK was proposed to control the caspase-1-dependent proteolysis of pro-to mature-IL-1β, while IRE1a to keep mature IL-1β soluble to allow its secretion (Chiritoiu et al., 2019). LPS-activated WT and PERKΔK-DCs were co-stimulated with ATP to promote conversion of cytosolic pro-to mature IL-1β and its secretion (Fig. 8D). Surprisingly, we did not observe any impairment of IL-1β expression in PERKΔK-DCs, but rather an increase by 25% of both IL-1β mRNA transcription and mature IL-1β secretion compared to WT cells (Fig. 8D). This increased expression implies that PERK inactivation in DCs is not detrimental to IL-1β processing, but on the contrary, likely reflects the general protein synthesis augmentation displayed by the PERKΔK-DCs (Fig.   6D). The positive feedback loop created by augmented IL-1β secretion, could in turn increase IL1B mRNA transcription . Thus, conversely to macrophages, DCs do not rely on the ISR nor on PERK activity to produce cytokines in response to MAMPs, despite high p-eIF2a levels, that seems to prevent these cells to mount an acute stress response.

PERK and actin polymerization coordinates p-eIF2α levels and migration in DCs.
Recently, the importance of globular actin in the formation of a tripartite holophosphatase complex assembled with GADD34 and PP1c to dephosphorylate eIF2a was revealed (Crespillo-Casado et al., 2017, Chambers et al., 2015, Crespillo-Casado et al., 2018. Given the unusual regulation of eIF2α phosphorylation in DCs, we wondered whether actin organization could also impact this pathway in a different setting than artificial ER-stress induction. Actin depolymerizing and polymerizing drugs, respectively Latrunculin A (Lat A) and Jasplakinoloide (Jaspk) had opposite effects on eIF2a phosphorylation in Flt3-L BMDCs (Fig. 9A). Globular actin accumulation induced by Lat A was strongly correlated with eIF2a dephosphorylation (Fig. 9A and 9B), while actin polymerization induced by Jaspk resulted in a massive increase of eIF2a phosphorylation, together with a reduction in protein synthesis ( Fig. 9A and 9C), similarly to what was observed in GADD34-deficient DCs. We next tested the impact of the two drugs on WT and PERKΔK Flt3-L BMDCs activated or not by LPS (Fig. 9D). LPS activation or Lat A treatment resulted in the same levels of eIF2a dephosphorylation and had a synergistic effect when used together (Fig. 9D). Jaspk dominated the LPS effect and strongly increased p-eIF2a levels in all conditions tested. PERK inactivation decreased the levels of p-eIF2a, but had no obvious consequences on the efficacy of the drugs, since both induced similar responses in WT DCs. We next tested the impact of actin remodeling and translation regulation on the acquisition by DC of their immune-stimulatory phenotypes. Surface MHC II and co-stimulatory molecule CD86 expression were up-regulated by LPS stimulation but remained unaffected by Jaspk treatment (Fig. 9E). Similarly, transcription levels of key cytokines like IL-6 and IFN-β were insensitive to this actin polymerizing drug ( Fig. 9F & 9G). However, while IL-6 secretion remained identical, IFN-β levels were found reduced by Jaspk treatment, suggesting that IL-6 mRNA translation occurs normally despite high eIF2aphosphorylation conditions. Interestingly, the IL-6 gene was described to bear an upstream uORFdependent translational regulation, that could allow IL-6 mRNA translation during Jaspk treatment (Sanchez et al., 2019). Taken together, these observations suggest that extensive actin polymerization in DCs increases strongly eIF2a phosphorylation, affecting protein synthesis and ultimately controlling translationally specific cytokines expression, similarly to what has been observed with Cdc42 or Wiskott-Aldrich syndrome protein (WASp) mutants (Pulecio et al., 2010, Prete et al., 2013. These results further suggest that actin dynamics and its effect on eIF2a phosphorylation could be a key regulating factors for specific mRNAs homeostasis and translation, like IFNs, that are generally secreted in a polarized fashion or needs to be coordinated with cell migration (Pulecio et al., 2010, Prete et al., 2013. Finally, given the interplay between eIF2a phosphorylation and actin polymerization, we wondered whether PERK-deficient cells could display some migratory deficits. We used microfabricated channels, that mimic the confined geometry of the interstitial space in tissues (Heuze et al., 2013, Bretou et al., 2017, to find that PERK-deficient cells were not able to increase their migration speed in response to LPS (Fig. 9G), confirming the link between eIF2a phosphorylation and actin dynamics. PERK activity is therefore necessary for DCs to acquire normal immune-stimulatory and migratory activities, presumably by coordinating protein synthesis and translation specific mRNAs with actin polymerization.

Discussion
We have previously proposed the existence of a strong causative link between cell activation by TLR ligands and eIF2a phosphorylation, notably by virtue of strong GADD34 expression in most transcriptomics studies performed on PAMPs-activated DCs (Claudio et al., 2013, Clavarino et al., 2012b. Our present work suggests that steady state DCs activate PERKmediated eIF2a phosphorylation to acquire their functional properties during differentiation (Sup Fig.   7), but distinctly from known ISR programs, normally induced upon acute or chronic ER stress (Han et al., 2013. To our knowledge, the level of p-eIF2a observed in primary DC both in vivo and in vitro are unique in their amplitude. Although, as judged comparatively from experiments performed with artificial induction of the different EIF2KAs, such p-eIF2a levels should be inhibitory for global protein synthesis (Dalet et al., 2017). DCs have acquired biochemical resistance, like high expression of eIF2B, to compensate for the consequences of this developmental PERK activation and to undergo high eIF2a phosphorylation while maintaining normal proteostasis. The role of ATF4 in controlling Ppp1r15a/GADD34 mRNA expression and eIF2a dephosphorylation to restore protein synthesis during stress has been extensively studied (Novoa et al., 2001), but the active translation observed in DCs is likely not permitting ATF4 synthesis and consequently the activation of a bona fide ISR despite high eIF2a phosphorylation. GADD34 is nevertheless functional in non-activated DCs, in which IKKϵ/TBK1 activity and potentially the transcription factor IRF3 are required for its mRNA transcription. This dependency of Ppp1r15a/GADD34 transcription on IKKϵ/TBK1 confirms that the PPP1R15a gene belongs to a group of genes directly induced by TLR or RLR signaling, as previously suggested by ChIP-seq and transcriptomics analysis of viral or poly (I:C)-stimulated cells (Freaney et al., 2013, Lazear et al., 2013, Dalet et al., 2017). GADD34 protein expression is undetectable in DCs, without prior treatment with proteasome inhibitors (Clavarino et al., 2012b), which is presumably indicative of an extremely tight equilibrium between its translation and active degradation. GADD34 mRNA translation, like that of ATF4, is controlled through 5' upstream ORFs regulation, which are normally bypassed upon global translation arrest to favor the synthesis of these specific ISR molecules (Palam et al., 2011).
Interestingly, GADD34 mRNA has been recently shown to be also actively translated in unstressed MEFs, albeit at much lower levels than upon ER stress (Reid et al., 2016). Thus, in steady state DCs, GADD34 synthesis likely occurs in a high eIF2a-phosphorylation context, despite relatively normal level of translation, while that of ATF4 does not. The difference in the 5' upstream ORFs organization of the mRNAs coding for these two molecules (Palam et al., 2011, Andreev et al., 2015, could explain this difference, and how GADD34/PP1c activity contributes to the maintenance of protein synthesis activity by counteracting PERK in steady state DCs. Importantly, the PERK/eIF2a/GADD34 molecular trio sets the physiological range for potential protein synthesis initiation available in the different DC activation stage, however the upward progression triggered by LPS stimulation, from one level of protein synthesis to the next, does not depend on this biochemical axis and is likely regulated by other protein synthesis regulation pathways, like the mTORC1 or Casein Kinase 2 pathways (Lelouard et al., 2007, Reverendo et al., 2019) (Sup Fig. 7).
Other mechanisms that contribute to escape PERK-mediated eIF2a phosphorylation, including the eIF2B-independent and eIF3-dependent pathway recently described to rescue translation during chronic ER stress , do not seem to be used in the DC context. These DC specific mechanisms prevent the induction of the ISR by the AB5 subtilase cytotoxin, which targets the ER chaperone BiP. Independently of demonstrating that induction of acute eIF2a phosphorylation by thapsigargin is not solely dependent on PERK activation, our observations suggest that DCs could escape EIF2KA-dependent translation arrest during exposure to different metabolic insults relevant to the immune context. These situations can include viral infection (Clavarino et al., 2012b), exposure to high levels of fatty acids during pathogenesis, oxidative stress during inflammation or amino acids starvation, mediated by amino acid degrading enzymes, like Arginase 1 or IDO, which are induced during infection or cancer development (Claudio et al., 2013, Munn et al., 2004. Importantly we could also show, that in contrast to macrophages, PERK and ISR induction are not necessary for DCs to drive the transcription of pro-inflammatory cytokines in response to TRIF or MAVs dependent-signaling (Abdel-Nour et al., 2019) nor the secretion of IL-1β (Chiritoiu et al., 2019).
PERK activation is therefore required to regulate mRNA translation during DC development and potentially also GADD34 synthesis, which not only provides a negative feed-back, but also is required for normal DC activation and cytokines expression (Clavarino et al., 2012b, Perego et al., 2018. Although a role for the ISR has been suggested to favor the survival of tissue associated DCs (     Student's t-test (* p < 0.05, ** p < 0.01, *** p < 0.001 and **** p < 0.0001).

Figure 5. GADD34 mediates eIF2α dephosphorylation and controls translation in DCs.
WT and GADD34ΔC BMDCs were stimulated with LPS (100ng/ml) for the indicated times. (A) Levels of p-eIF2α, total eIF2α, p-eIF2β, total eIF2β, p-eEF2, total eEF2, P-S6 and total S6 were detected by immunoblot (top left) and quantification is shown in the different panels. (B) The speed of translation elongation was measured by SunRISE after 3h of incubation with LPS. Harringtonine (2μg/ml) was added at different times up to 90 seconds prior incubation with puromycin during 10 min. Flow intracellular staining was performed in cDC2 using an α-puromycin antibody. The total decay of puromycin MFI between WT and GADD34ΔC in steady-state and upon activation indicates that translation initiation and elongation speed is decreased in GADD34-deficient DCs. All data are representative of n=3 independent experiments. Data in (B) represents MFI +/-SD of 3 independent experiments. Unpaired Student's t-test ( ** p < 0.01).    Cells were incubated with puromycin 10 min before harvesting. The graph shows total puromycin MFI levels. WT and PERKΔK Flt3-L BMDCs were treated with LPS (100ng/ml), Latrunc A and Jasp for 4h.
(D) Levels of p-eIF2α and total eIF2α monitored by immunoblot in Flt3-L BMDCs. Quantification is shown on the right. (E) Upregulation of surface MHC II and CD86 levels was measured by flow in cDC2. The graphs represent total MFIs. (F-G) mRNA levels and secretion of IL-6 (F) and IFN-β (G) were measured by RT-qPCR and by ELISA from Flt3-L BMDCs cultures. Raw data was normalized to the housekeeping gene (GAPDH). H) WT and PERKΔK GM-CSF BMDCs were treated with LPS (100ng/ml) during 30 min previous to 16h of migration. The graph represents instantaneous mean velocities of migration in 4-by 5µm fibronectin-coated microchannels of at least 100 cells per condition.