PPP1R15A-mediated dephosphorylation of eIF2α is unaffected by Sephin1 or Guanabenz

Dephosphorylation of translation initiation factor 2 (eIF2α) terminates signalling in the mammalian integrated stress response (ISR) and has emerged as a promising target for modifying the course of protein misfolding diseases. The [(o-chlorobenzylidene)amino]guanidines (Guanabenz and Sephin1) have been proposed to exert protective effects against misfolding by interfering with eIF2α-P dephosphorylation through selective disruption of a PP1-PPP1R15A holophosphatase complex. Surprisingly, they proved inert in vitro affecting neither stability of the PP1-PPP1R15A complex nor substrate-specific dephosphorylation. Furthermore, eIF2α-P dephosphorylation, assessed by a kinase shut-off experiment, progressed normally in Sephin1-treated cells. Consistent with its role in defending proteostasis, Sephin1 attenuated the IRE1 branch of the endoplasmic reticulum unfolded protein response. However, repression was noted in both wildtype and Ppp1r15a deleted cells and in cells rendered ISR-deficient by CRISPR editing of the Eif2s1 locus to encode a non-phosphorylatable eIF2α (eIF2αS51A). These findings challenge the view that [(o-chlorobenzylidene)amino]guanidines restore proteostasis by interfering with eIF2α-P dephosphorylation. DOI: http://dx.doi.org/10.7554/eLife.26109.001


Introduction
Protein folding homeostasis (proteostasis) is achieved by balancing the rate of production, folding and protein degradation. Proteostasis is strongly influenced by the phosphorylation state of serine 51 of the a subunit of eukaryotic translation initiation factor 2 (eIF2a) (Sonenberg and Hinnebusch, 2009). Diverse stress conditions activate kinases that phosphorylate eIF2a, resulting in attenuated rates of translation initiation of most mRNAs and increasing translation of a small group of mRNAs with special 5 0 untranslated regions (Hinnebusch, 2014). The latter encode potent transcription factors such as ATF4 that couple eIF2a phosphorylation to the Integrated Stress Response (ISR) , a transcriptional and translational program that adapts cells to stress and participates in diverse biological processes such as memory, immunity and metabolism (Baird and Wek, 2012).
Signalling in the ISR is terminated by eIF2a-P dephosphorylation. This requires the presence of a regulatory subunit, PPP1R15, to direct the catalytic, PP1 subunit, to its specific substrate. Two mammalian genes encode PPP1R15 regulatory subunits. Ppp1r15b (or CReP) encodes a constitutively expressed regulatory subunit (Jousse et al., 2003), whereas Ppp1r15a (or GADD34) encodes an ISR inducible regulatory subunit that contributes to a negative feed-back loop operative in the ISR (Brush et al., 2003;Ma and Hendershot, 2003;Novoa et al., 2001Novoa et al., , 2003. A minimal rate of eIF2a-P dephosphorylation is an essential cellular function, as reflected in the severe phenotypes of Ppp1r15b mutation or deletion and in the very early lethality of compound Ppp1r15a;b deficient mice (Abdulkarim et al., 2015;Harding et al., 2009).
Interestingly, whilst deletion of the inducible Ppp1r15a gene results in sluggish recovery of protein synthesis during the waning phase of stress (Kojima et al., 2003;Novoa et al., 2003), mice lacking any PPP1R15A-directed eIF2a-P dephosphorylation (homozygous Ppp1r15a tm1Dron ) are superficially indistinguishable from wildtype. Moreover, when challenged with tunicamycin, which causes unfolded protein stress in the endoplasmic reticulum by inhibiting N-linked glycosylation, homozygous Ppp1r15a tm1Dron mice and cultured cells derived from them are relatively resistant to the toxin's lethal effects (Marciniak et al., 2004;Reid et al., 2016). This feature is plausibly attributed to sustained activity of the ISR in the Ppp1r15a mutant mice, which favours proteostasis by limiting the production of unfolded proteins under stress conditions (Boyce et al., 2005;Han et al., 2013).
The proteostasis-promoting features of interfering with PPP1R15A-mediated eIF2a-P dephosphorylation are also played out in the context of certain disease models associated with protein misfolding and proteotoxicity. Both the neuropathic phenotype associated with Schwann cell expression of a mutant misfolding-prone myelin constituent, P0 S63D , and a mutant superoxide dismutase expressed in motor neurones are ameliorated by a concomitant dephosphorylation-defective Ppp1r15a tm1Dron mutation (D' Antonio et al., 2013;Wang et al., 2014), and similar amelioration of eLife digest Most drugs work by tweaking the way that cells are regulated. Adding or removing a phosphate group from proteins regulates many cellular decisions. There are known drugs that bind to and inhibit the enzymes that add phosphate to proteins, thereby controlling various aspects of cell behaviour. However, drug developers have been far less successful in finding drugs that inhibit phosphatases, the enzymes that remove phosphate from proteins.
Genetically modified mice can be used as 'models' to investigate human diseases. In 2015 a drug called Sephin1 was reported to suppress neurodegeneration in a group of these mice by inhibiting a particular phosphatase. The phosphatase is made of three component proteins that come together to create the active enzyme. Sephin1 was reported to disrupt the association between two of these three components. This discovery was met with excitement; both for its potential therapeutic implications in humans and as an important "first" in pharmacology.
To understand how Sephin1 and a related drug, Guanabenz, work at the molecular level, Crespillo-Casado et al. reconstructed in a test tube the phosphatase that Sephin1 and Guanabenz were reported to inhibit. To examine the effects the drugs have on the phosphatase, Crespillo-Casado et al. developed assays to measure the association between the components that make up the phosphatase. Further assays measured the removal of phosphate from the phosphatase's target, a protein called eIF2a.
The results of the assays show that Sephin1 did not affect the coming together of the components that make up the active phosphatase. The drug also did not inhibit the removal of phosphate from eIF2a in the test tube. To extend these findings Crespillo-Casado et al. exposed cells to Sephin1 and observed features that are consistent with the drug's reported ability to supress neurodegeneration. However, these features were also observed both in cells lacking the phosphatase that Sephin1 was reported to inhibit and in cells in which eIF2a never acquired a phosphate in the first place.
The findings presented by Crespillo-Casado et al. do not challenge Sephin1's role in supressing neurodegeneration, but do question its ability to do so by inhibiting the phosphatase that dephosphorylates eIF2a. This knowledge will be useful to drug developers and those interested in molecular mechanisms of drug action. For those researchers who are interested in Sephin1, further work is needed to discover alternative molecular mechanisms by which it suppresses neurodegeneration. And for those researchers who are interested in eIF2a dephosphorylation, there is a need to look further for inhibitors of this process, as Sephin1 is unlikely to serve in that role.
inflammatory-mediated central nervous system demyelination is observed in the Ppp1r15a tm1Dron mice (Lin et al., 2008).
These features have led to an interest in the therapeutic potential of targeting PPP1R15-mediated eIF2a-P dephosphorylation with small molecule inhibitors. Early work led to discovery of salubrinal, a small molecule that increases levels of eIF2a-P and retards its dephosphorylation. However, salubrinal is only known to work in vivo and its mechanism of action remains unclear (Boyce et al., 2005). Limitations of in vitro assays for substrate-specific PPP1R15-mediated eIF2a-P dephosphorylation (see below) have all but precluded a biochemical approach to the problem, but a cell based search for proteostasis regulators suggested that the a2 adrenergic blocker Guanabenz, [(o,o-dichlorobenzylidene)amino]guanidine, might exert its beneficial effects on proteotoxicity by interfering with eIF2a-P dephosphorylation (Tsaytler et al., 2011). This theme was extended further by the discovery of Sephin1, [(o-chlorobenzylidene)amino]guanidine, that had lost its a2 adrenergic blocking activity but retained its proteostasis promoting properties (Das et al., 2015). Importantly, biochemical characterization of Guanabenz and Sephin1 suggested that both disrupt the complex between PPP1R15A and PP1, providing strong support for a mechanism of action that involves interfering with PPP1R15A-mediated eIF2a-P dephosphorylation (Das et al., 2015), Figure 1C therein).
Genetic analysis reveals that a regulatory PPP1R15 subunit is essential for eIF2a-P dephosphorylation in vivo, and over-expression of either PPP1R15A or PPP1R15B or merely their conserved C-terminal portion, is sufficient to deregulate eIF2a-P dephosphorylation in vivo and inhibit the ISR (Brown et al., 1997;Brush et al., 2003;Jousse et al., 2003;Novoa et al., 2001). Though PPP1R15 regulatory subunits stably bind the catalytic subunit (PP1), the resulting binary complex is devoid of specificity towards eIF2a-P. However, G-actin joins the PPP1R15-PP1 binary complex as an ancillary subunit to form a ternary complex endowed with substrate-specific eIF2a-P directed phosphatase activity, both in cells (Chambers et al., 2015) and when constituted with pure components in vitro (Chen et al., 2015).
To explore in detail the mechanism of action of the [(o-chlorobenzylidene)amino]guanidines we reconstructed PPP1R15A-PP1-G-actin-mediated eIF2a-P dephosphorylation in vitro with pure components. Using mutants that interfere with complex formation and function we established the correlation of enzymatic activity with the kinetic parameters of complex formation to develop an assay responsive to the stability of the core PPP1R15-PP1 binary interaction; the proposed target of Guanabenz and Sephin1. The results of our inquiry, reported on below, question the role of destabilization of the eIF2a-P directed holophosphatase in the proteostatic effects of these compounds.

Results
In vitro assay for selective eIF2a-P dephosphorylation sensitive to the stability of the PPP1R15A-PP1 complex PPP1R15A/GADD34 is a protein of >600 residues, but only the C-terminal 70 residues are necessary for substrate-specific dephosphorylation of eIF2a-P ( Figure 1A). This active fragment is also the most conserved segment of the protein; both between homologues of PPP1R15A and with the paralogous PPP1R15B ( Figure 1B). This region of the protein is natively unfolded (Yu et al., 2004), attaining its structure upon binding the PP1 catalytic and the G-actin ancillary subunits (Chen et al., 2015;Choy et al., 2015).
For structural studies we found it convenient to co-express PPP1R15 and PP1 in bacteria (Chen et al., 2015). However, the fixed 1:1 stoichiometry of the two subunits imparted by co-expression is unsuited to a detailed examination of the bimolecular affinities involved in complex formation or to the design of an assay sensitive to the stability of PPP1R15A-PP1 complex. To circumvent this limitation, we incorporated a highly soluble maltose-binding protein (MBP) moiety C-terminal to the natively-unstructured human PP1R15A active fragment. When expressed in E. coli as a fusion protein with a cleavable N-terminal glutathione S-transferase (GST) tag, GST-PPP1R15A-MBP remained soluble and when added as a purified protein in vitro (after cleavage of the GST), imparted eIF2a-P dephosphorylation activity to reactions containing purified PP1 and G-actin ( Figure 1C, left panel). Moreover, the solubilizing MBP tag enabled recovery not only of a human PPP1R15A active fragment (residues 533-624) but also a much larger N-terminally extended fragment (residues 325-636). The minimal active fragment and the much longer N-terminally-extended PPP1R15A had similar The minimal C-terminal peptide required for eIF2a-P dephosphorylation is outlined ('active fragment') and key residues in the PP1 and G-actin binding regions are annotated. (B) Alignment of C-terminal active fragments of mammalian PPP1R15A and PPP1R15B (CREP) using ClustalX. Grey highlighted residues represent conserved or highly similar residues. Red asterisks highlight key residues that are analysed in further detail. (C) Image of Coomassie- Figure 1 continued on next page activity in this assay ( Figure 1C). Nonetheless the ability to purify a soluble, N-terminally extended PPP1R15A regulatory subunit expanded the possibilities to study more physiological models of PPP1R15A-mediated eIF2a-P dephosphorylation (a point we shall return to below).
We quantified the dependence of eIF2a-P dephosphorylation rates on both the concentration of the regulatory human PPP1R15A subunit (EC 50 = 7 nM) and on the ancillary G-actin subunit (EC 50 = 13 nM) ( Figure 2). The latter values agreed with our previous measurements of G-actin's stimulation of enzymatic activity (in an assay using the murine PPP1R15A) (Chen et al., 2015), whereas the EC 50 of human PPP1R15A was within an order of magnitude of the affinity of human PPP1R15A for PP1, as measured by isothermal titration calorimetry (Choy et al., 2015) (see below).
Previous studies have identified mutations in PPP1R15 that abolish substrate-specific dephosphorylation in vitro and block PPP1R15's ability to repress the ISR, when expressed in vivo. The human PPP1R15A V556E mutation alters a key residue, part of the RVxF motif involved in binding of diverse regulatory subunits to PP1 (Egloff et al., 1997); its presence abolished all PPP1R15A-mediated eIF2a-P dephosphorylation ( Figure 3A). Two previously-identified mutations in the C-terminal extension of PPP1R15A -the portion that interacts with the G-actin ancillary subunit (human PPP1R15A W582A and PPP1R15A F592A ) (Chen et al., 2015) -also abolished all PPP1R15A-mediated eIF2a-P dephosphorylation ( Figure 3B and C). These findings establish the dependence of the tripartite assay described above on features known to be important for PPP1R15 function.
A fourth mutation tested affects a residue whose counterpart in human PPP1R15B R658C results in a syndromatic form of diabetes mellitus. Consistent with the destabilizing effect of this mutation on PP1 binding (Abdulkarim et al., 2015), its presence in human PPP1R15A R578A resulted in a~4 fold increase in EC 50 for eIF2a-P dephosphorylation ( Figure 4A). The mutation also affected the maximal stimulation afforded by human PPP1R15A R578A , as even at saturating concentrations of regulatory subunit, eIF2a-P dephosphorylation reactions assembled with the mutant were three times slower than those assembled with the wildtype ( Figure 4B). The human PPP1R15A R578A mutation does not appear to have a major effect on the stability of the G-actin containing ternary complex, as the EC 50 for G-actin (20 nM) was relatively unaffected ( Figure 4C).

Bio-Layer Interferometry measurement of the affinity of the holophosphatase components for each other
The features of the human PPP1R15A R578A mutant noted above suggest that the tripartite assay is sensitive not only to the affinity of the three components for one another but also to subtle structural features of the holophosphatase. To explore this issue further, we used Bio-Layer Interferometry (BLI) (Abdiche et al., 2008) to measure directly the affinity of the three components of the holophosphatase for each other. PPP1R15A 533-624 was biotinylated on a single lysine residue of an AviTag (Fairhead and Howarth, 2015) added between the cleavable GST tag and PPP1R15A peptide ( Figure 5A) and the biotinylated protein was immobilized on a streptavidin-derivatized BLI biosensor tip. The biotinylated PPP1R15A 533-624 ligand showed a robust 1:1 bimolecular interaction, with pure PP1, yielding a k off = 0.21 ± 0.01 min À1 and a K d = 20 ± 0.61 nM ( Figure 5B). The higher affinity of PPP1R15 for PP1 observed here, compared to isothermal titration calorimetry (ITC) measurements of Choy et al. (2015) (K d = 62 ± 14 nM) might reflect the contribution of contacts made by residues C-terminal to PPP1R15A L567 , which are present in the construct used here, but absent from the one used in the ITC measurements (Choy et al., 2015). Cooperativity provided by G-actin (present in the enzymatic assay, but absent from the BLI experiment) and steric hindrance from probe components might have contributed to the 3-5 fold lower value of the PPP1R15A EC 50 for eIF2a-P dephosphorylation in the enzymatic assay (7 nM, Figure 2B) compared to the K d observed by BLI.
To gauge the affinity of the ancillary G-actin subunit for the complex, we first assembled a binary complex between the biotinylated PPP1R15A 533-624 ligand (described above) and a saturating Semi-log 10 plot of the initial velocity of eIF2a P dephosphorylation as a function of PPP1R15A 533-624 concentration derived from three repeats (one shown above). The EC 50 for PPP1R15A 533-624 was calculated using the agonist fitting function on Figure 2 continued on next page concentration of PP1 and then measured the BLI signal induced by addition of G-actin. A robust association-dissociation signal was observed with purified G-actin (k off = 2.84 ± 0.11 min À1 and K d = 151 ± 14.3 nM) ( Figure 5C). The human PPP1R15A V556E mutation, affecting the RVxF motif, abolished all measureable association with PP1, but had no effect on the kinetics of G-actin binding. Conversely, the human PPP1R15A F592A mutation markedly enfeebled G-actin binding but had no effect on PP1 binding ( Figure 6A and B). Together, these observations confirm the ability of PPP1R15A to engage PP1 and G-actin independently, via the N-and C-terminal parts of its active portion. Despite their strong detrimental effects on enzymatic activity ( Figures 3C and 4), neither the PPP1R15A R578A nor the PPP1R15A W582A mutations had a major effect on the kinetics of PP1 or G-actin binding ( Figure 6). Together, these observations suggest that the tripartite enzymatic assay is sensitive both to mutations that grossly interfere with complex stability (V556E and F592A) and to mutations that more subtly affect the structure of the complex (R578A and W582A).

No measureable effect of [(o-chlorobenzylidene)amino]guanidines on PPP1R15A-containing holophosphatases in vitro
Das and colleagues previously reported that addition of 50 mM Sephin1 to tissue culture media disrupts the PPP1R15A-PP1 complex recovered from cells (Das et al., 2015). To determine if these observations correlate with an effect of Sephin1 on the complex formed in vitro between PPP1R15A 533-624 and PP1, we sourced Sephin1 and confirmed its purity and identity by reverse phase HPLC and mass spectrometry ( Figure 7A). When added to the BLI assay at a concentration of 50 mM (before exposure to PP1), Sephin1 had no measureable effect on either the association or dissociation phase of the assay ( Figure 7B).
A biotinylated N-terminally extended PPP1R15A 325-636 , corresponding to the construct studied by Das and colleagues, proved unsuited as a ligand in the BLI experiment. To circumvent this problem we biotinylated PP1 and exploited it as a BLI ligand. Addition of either the minimal active fragment, human PPP1R15A 533-624 , or the longer human PPP1R15A 325-636 , gave rise to a robust BLI signal but addition of Sephin1 affected neither the association nor dissociation phase of the experiment ( Figure 7C and D). The kinetics of the bimolecular PP1-PPP1R15A interaction were reproducibly different when one or the other was used as a ligand (summarized in Figure 7E). These may reflect different distorting effect of other elements of the BLI biosensor on the kinetics of dissociation when PP1 or PPP1R15A were used as ligands, and/or a contribution of the N-terminal repeats of PPP1R15A to its interactions with PP1, as suggested previously (Brush and Shenolikar, 2008). However, reproducible inertness in all three assays lends confidence to the conclusion that in this experimental system 50 mM Sephin1 does not directly interfere with assembly or stability of the PPP1R15A-PP1 complex.
Next we sought to examine the effect of Sephin1 on the in vitro dephosphorylation activity of a tripartite eIF2a-P holophosphatase assembled with the N-terminally extended PPP1R15A 325-636 (corresponding to the construct studied by Das and colleagues). As Sephin1 is proposed to inhibit eIF2a-P dephosphorylation by disrupting the binding of PPP1R15A to PP1, we sought to incorporate the PPP1R15A 325-636 component at or below its EC 50 , thereby maximizing the prospects of detecting an inhibitory effect. Addition of purified PPP1R15A 325-636 to PP1 and G-actin accelerated eIF2a-P dephosphorylation with an EC 50 of 5-10 nM ( Figure 8A). However, Sephin1 had no effect on the dephosphorylation reaction ( Figure 8B), whilst tautomycin readily inhibited the reaction (IC 50 = 2.4 nM) confirming the sensitivity of the assay to a known inhibitor ( Figure 8C). These observations were also confirmed in an assay set up with the corresponding murine PPP1R15A 273-657 (Figure 8-figure supplement 1A and B). The related compound Guanabenz also proved inert, even when added to the enzymatic assay at the high concentration of 50 mM ( Figure 8D). Salubrinal, added at 12 mM (higher concentrations led to conspicuous precipitation) had a mild but highly reproducible inhibitory  Figure 8E, inhibition = 22% ± 2.045, unpaired t test, p<0.0001, n = 6). The weakness of salubrinal's inhibitory effect and the compound's tendency to precipitate at higher concentrations in the assay buffer frustrated our efforts to establish if inhibition was specific to the eIF2a-P directed ternary complex. Nonetheless these observations showcase the sensitivity of our assay to even weak inhibitors and strengthen the conclusion regarding Sephin1's inertness in the same assay.
Sephin1 exerts proteostatic effects in vivo independently of PPP1R15A or the eIF2a-P-dependent integrated stress response Sephin1's role as a proteostasis promoting agent was explored in cultured CHO-K1 cells containing reporters for both the ISR (CHOP::GFP) (Novoa et al., 2001) and the branch of the endoplasmic reticulum unfolded protein response (UPR) mediated by IRE1 (XBP1s::Turquoise) (Iwawaki et al., 2004;Sekine et al., 2016). Previous studies have emphasized the dominance of translational recovery in the physiological action of PPP1R15A, such that Ppp1r15a KO attenuates both the burden of protein misfolding (Marciniak et al., 2004) and the response to it (Reid et al., 2016). In keeping with these ideas and with the findings of Das and colleagues (Das et al., 2015), Sephin1 attenuated the activity of both UPR pathways in cells exposed to tunicamycin; an inhibitor of N-linked glycosylation, that promotes misfolding of newly-synthesized proteins ( Figure 9A). Though observed only over a narrow concentration range of tunicamycin ( Figure 9-figure supplement 1A) and at relatively high concentrations of the drug (Figure 9-figure supplement 1B), Sephin1's effects in this assay can be reconciled with a mechanism involving a net reduction of protein synthesis; as suggested by Das and colleagues. Sephin1 also inhibited induction of the ISR in response to histidinol ( Figure 9B), an agent that interferes with tRNA charging and thereby activates the eIF2a kinase GCN2 (Zhang et al., 2002) without affecting protein folding.
To probe deeper into this matter, we exploited an in vivo assay that monitors eIF2a-P dephosphorylation in cells. In this kinase shut-off experiment (Chambers et al., 2015)( Figure 10A), cultured cells are first exposed to a brief, 30 min pulse of thapsigargin, which rapidly activates the eIF2a kinase PERK and builds levels of eIF2a-P, and then exposed to a PERK kinase inhibitor (GSK260414A). The resulting decay in the eIF2a-P signal reflects its dephosphorylation. To minimize the contribution of other kinases to the eIF2a-P signal, the experiment was performed in cells lacking GCN2 (Chambers et al., 2015), which we inactivated in the CHO-K1 cells by CRISPR-Cas9 gene editing. Normally, the dephosphorylation of eIF2a-P is a rapid process, complete in 60 min ( Figure 10B, lanes 2-6). It was markedly delayed by inclusion of jasplakinolide ( Figure 10B lanes 7-10), which depletes the pool of G-actin (by promoting its oligomerization), thereby depriving the PPP1R15 subunits of an essential co-factor, as observed previously (Chambers et al., 2015).
Together, PPP1R15A and PPP1R15B account for the bulk of eIF2a-P dephosphorylation activity of mammalian cells (Harding et al., 2009), but their relative contribution to the process in any given circumstance is unknown. Therefore, to adapt this assay to measure Sephin1's effect on PPP1R15Amediated eIF2a-P dephosphorylation, it was essential to inactivate the gene encoding PPP1R15B, leaving PPP1R15A as the sole regulatory subunit of the eIF2a-P phosphatase. CRISPR-Cas9 mediated gene editing was used to create two different GCN2 KO ; Ppp1r15b KO compound-mutant CHO-K1 cells (Figure 10-figure supplement 1A and B). As expected, the GCN2 KO ; Ppp1r15b KO compound-mutant CHO-K1 cells retained their responsiveness to Sephin1 (Figure 10-figure supplement 1C). However, in this experimental system dependent solely on PPP1R15A, the timedependent decline of the eIF2a-P signal (fitted to an exponential decay curve) yielded a time constant of 0.23 min À1 for the untreated and 0.19 min À1 for the Sephin1 treated sample, an insignificant difference ( Figure 10B   . eIF2a-P dephosphorylation proceeded rapidly in tunicamycin-treated cells. However, Sephin1 had no inhibitory effect on the rate of dephosphorylation. This experiment reveals that under conditions in which Sephin1 exerts its proteostasis-promoting activities, it does not affect rates of eIF2a-P dephosphorylation. To follow up on this matter, both copies of the gene encoding PPP1R15A were inactivated by CRISPR-Cas9 in the reporter containing CHO-K1 cells ( Figure 11-figure supplement 1A and B). Inactivation was confirmed by loss of the PPP1R15A signal in immunoblot of lysates from stressed mutant cells ( Figure 11A). Sephin1 retained its ability to attenuate the response of both the XBP1s:: Turquoise and the CHOP::GFP reporter in tunicamycin treated PPP1R15A null cells ( Figure 11B). Similar observations were made in regard to the effect of Sephin1 in histidinol-treated cells ( Figure 11C). These observations suggest that Sephin1 also exerts its proteostatic effect(s) in cells lacking PPP1R15A.
Next, we used CRISPR-Cas9-mediated homologous recombination to introduce a site-specific mutation into the Eif2s1 locus, to encode an ISR-blocking eIF2a S51A mutation in the endogenous gene ( Figure 12A and B). Surprisingly, Sephin1 retained its ability to attenuate the XBP1s::Turquoise reporter in tunicamycin-treated eIF2a S51A mutant cells ( Figure 12C). As CHOP activation is highly dependent on the ISR (Harding et al., 2000), activity of the CHOP::GFP reporter was strongly attenuated in mutant eIF2a S51A cells. Nonetheless, it is notable that residual activation of the reporter by tunicamycin (likely a consequence of ATF6 action at the CHOP promoter [Yoshida et al., 2000]), was also attenuated by Sephin1 ( Figure 12C). These observations bring into question the primacy of the ISR in Sephin1's mechanism of action.

Discussion
The role of the eIF2a-P-dependent ISR in defending against unfolded protein stress is well supported by genetic and pharmacological experiments (Baird and Wek, 2012;Ron and Harding, 2007). By retarding its dephosphorylation, the primary consequence of eliminating PPP1R15A is to prolong the duration of the eIF2a-P signal in stress response scenarios (Kojima et al., 2003;Novoa et al., 2003) and to alter the repertoire of mRNA translation (Reid et al., 2016). Therefore, the finding that cells and mice lacking PPP1R15A are relatively resistant to pharmacological and genetic models associated with unfolded protein stress in the endoplasmic reticulum (ER stress) has engendered a specific interest in targeting the PPP1R15A-containing phosphatase complex for inhibition, as a means for accessing the therapeutic potential of enhanced ISR signalling.
Sephin1 and Guanabenz, compounds previously proposed to exert their proteostatic effects by disrupting the essential PP1-PPP1R15A complex and inhibiting eIF2a-P dephosphorylation (Das et al., 2015) are found here to have no effect in in vitro enzymatic assays dependent on the formation of a PP1-PPP1R15A complex. Sephin1 likewise proved inert in a Bio-Layer Interferometry assay that measured directly the affinity of PPP1R15A and PP1 for one another. Furthermore, we find that Sephin1 does not interfere with eIF2a-P dephosphorylation in cells (as measured by a kinase shut-off experiment) and that Sephin1 retains its ability to attenuate the impact of a challenge to proteostasis even in cells lacking PPP1R15A, or in ISR-defective Eifs1 S51A cells. These observations suggest that the previously-reported attenuation of the recovery of PP1 in complex with PPP1R15A, when both were purified from lysates of cells treated with Sephin1 was unlikely to be a direct consequence of Sephin1 interference with complex formation or of destabilization by Sephin1 of a preexisting complex and also call into question the importance of any indirect disruption of the PP1-PPP1R15A complex that may occur in vivo and remain undetected by our assays.
Our findings, questioning whether Sephin1 attains its proteostatic activity by inhibiting the PPP1R15A-containing eIF2a-P directed holophosphatase, and similar concerns raised by the Peti   PP1 association then dissociation G-actin association then dissociation and Shenolikar labs (Choy et al., 2015), do nothing to diminish the attractiveness of PPP1R15A inhibition as a potential means for defending proteostasis. Similarly, there is nothing in our study to question the beneficial effects reported for Sephin1 in mouse models of neurodegeneration (Das et al., 2015) nor we do not challenge the enhanced susceptibility of Ppp1r15b KO cells to the [(o-chlorobenzylidene)amino]guanidine, Guanabenz (Tsaytler et al., 2011). However, our findings that Sephin1 exerts its effects in CHO-K1 cells lacking PPP1R15A or in ISR-defective Eifs1 S51A cells raise doubts as to whether these phenomena were attained via inhibition of PPP1R15A or indeed modulation of the ISR. Crystal structures of the PPP1R15(A or B)-PP1 and the related PNUTS/PPP1R10-PP1 and spinophilin/PPP1R9B-PP1 complexes reveal that both the residues corresponding to human PPP1R15A V556 (of the RVxF motif) and the conserved arginine (human PPP1R15A R578 ) insert deeply into the surface of the PP1 subunit (Chen et al., 2015;Choy et al., 2014Choy et al., , 2015. The contrast between the dramatic effect of the PPP1R15A V556E mutation and the more modest effect of the human PPP1R15A R578A mutation on the PPP1R15A-PP1 complex may reflect a role for the former early in the pathway to complex assembly (a process that can be thought of as PPP1R15A folding on the surface of PP1). It is therefore possible that inhibitors of the eIF2a-P holophosphatase might disrupt complex assembly, without affecting the stability or activity of a preformed complex. However, Sephin1 is unlikely to belong to such a category, as it failed to exert an inhibitory effect on enzymatic activity or on the BLI signal even when added to pure PPP1R15A, before addition of PP1 and G-actin. But other compounds that remain to be found might selectively disrupt the assembly of the PPP1R15A-PP1 complex by binding to and stabilizing an intermediate step in its formation.
Similarly instructive is the human PPP1R15A W582A mutation, which eliminates all detectable selectivity of PPP1R15-holophosphatases for eIF2a-P (Chen et al., 2015)(and Figure 3B here) without affecting the kinetics of the bimolecular association of PPP1R15A with PP1 or G-actin. These features suggest that the side chain of W582 -a residue conserved throughout the PPP1R15 family -may have a special role in aligning PP1 and G-actin to form a composite surface with affinity for the substrate, without contributing measurably to the stability of the tripartite holophosphatase. That the side chain of single tryptophan residue can bias the complex towards activity (without affecting its stability), suggests the possibility that small molecules might access this allosteric mechanism and bias the tripartite holophosphatase towards or away from enzymatic activity without the need to disrupt an extensive protein-protein interface.
Specific cellular dephosphorylation events are notoriously difficult to target with small molecules (Sakoff and McCluskey, 2004). Here we presented evidence that the perception of Sephin1 as a milestone in overcoming that challenge may need re-thinking. However, features of the PPP1R15A-PP1-G-actin holophosphatase noted above suggest ways in which eIF2a-P dephosphorylation might indeed be selectively targeted by small molecules. In vitro assays for selective eIF2a-P dephosphorylation, such as the one described here, might prove useful in discovery of small molecules with such a mechanism of action.

Plasmid construction
Diverse cloning techniques were used to create the bacterial and mammalian expression vectors listed in Table 1. This table contains information about lab number, name, description and reference for each plasmid used.

Protein expression and purification
Actin was purified from rabbit muscle according to (Pardee and Spudich, 1982) as modified by (Chen et al., 2015).
Expression plasmids for PPP1R15A (GADD34) variants contained ampicillin resistance marker, N-terminal GST tag and C-terminal maltose binding protein (MBP) tag (UK1677, UK1920)( Table 1). They were transformed into BL21 T7 Express lysY/Iq E. coli (C3013, New England Biolabs) and colonies that grew in LB-ampicillin plates (100 mg/ml ampicillin) were used to create a saturated overnight culture. This saturated culture was used to inoculate 2-4 Litres of LB media supplemented with 100 mg/ml ampicillin. The cultures were incubated at 37˚C until OD600 = 0.6-0.8. At this point, they were induced with 1 mM Isopropyl b-D-1-thiogalactopyranoside (IPTG) and cultured for 20 more hours at 18˚C. It was followed by a centrifugation step to pellet bacteria and resuspension of the ice-cold pellets in 3-4 pellet volumes of lysis buffer (50 mM Tris pH 7.4, 500 mM NaCl, 1 mM MnCl 2 , 1 mM MgCl 2 , 1 mM tris(2-carboxyethyl)phosphine (TCEP), 100 mM phenylmethylsulfonyl fluoride (PMSF), 20 mTIU/ ml aprotinin, 2 mM leupeptin, and 2 mg/ml pepstatin in 10% glycerol). An Emulsi-Flex-C3 homogenizer (Avestin, Inc, Ottawa, Ontario) was used to lyse the bacteria, which were then clarified in a JA-25.50 rotor (Beckman Coulter) at 33,000Âg for 30 min at 4˚C. These suspensions were bound to pre-equilibrated glutathione sepharose 4B beads (17-0756-05, GE Healthcare) for 1-2 hr at 4˚C. Beads were transferred to a 10 mL column after being batch-washed with 20 bed volumes of lysis buffer. Proteins were eluted in glutathione elution buffer (50 mM Tris pH 7.4, 100 mM NaCl, 40 mM glutathione, 0.5 mM MnCl 2 , 0.5 mM TCEP, 10% glycerol), and cleaved with Tobacco Etch Virus protease (TEV) (12.5 mg TEV protease/mg protein) overnight at 4˚C to remove the N-terminal GST tag. Cleaved proteins were bound to amylose beads (E8021S, New England Biolabs) for 1-2 hr at 4˚C. Twenty/thirty bed volumes of lysis buffer were used to batch-wash the amylose beads, which were transferred to a 10 mL column and eluted with HEPES buffer (20 mM HEPES, 100 mM NaCl, 0.2 mM CaCl 2 , 0.2 mM ATP, 0.2 mM TCEP, 0.5 mM MnCl 2 , 100 mM PMSF, 20 mTIU/ ml aprotonin, 2 mM leupeptin, and 2 mg/ml pepstatin) and 10 mM maltose. PP1 (UK622) ( Table 1) was purified as above, with the following modifications: LB media cultures were supplemented with MnCl 2 , after TEV cleavage proteins were buffer exchanged using a 2 mL desalting column in HEPES buffer and re-bound to glutathione sepharose 4B beads to remove free GST tag. Phosphorylated eIF2a was encoded by an expression plasmid containing N-terminal His-Tag and kanamycin resistance marker (UK105) ( Table 1). BL21 T7 Express lysY/Iq E. coli were co-transformed with this plasmid and a GST-Tagged PERK plasmid carrying ampicillin resistance marker (UK168) ( Table 1). Colonies that grew in ampicillin (100 mg/ml) and kanamycin (50 mg/ml) LB-plates were used to create a saturated over-night cultured with which 2L of ampicillin and kanamycin LB were inoculated. Growth, induction and purification was as described for PPP1R15A, with the following changes: beads used were Ni-NTA (30230, Qiagen) to bind His-tag, lysis buffer contained 20 mM imidazole and elution buffer contained 500 mM imidazole instead of glutathione. This protein did not require TEV cleavage but an additional size exclusion chromatography step was included. A Superdex S200 (GE Healthcare) was used to gel filter the protein in 25 mM Tris, 100 mM NaCl, 0.1 mM EDTA, 1 mM DTT and 10% glycerol buffer.
All proteins were snap frozen and kept at À80˚C in small aliquots. Final concentration of proteins was calculated from UV absorbance at 280 nm measurements in Nanodrop (Thermo Scientific, UK) and based on their extinction coefficient values predicted by MacVector.

Figure 7 continued
Where indicated, the analyte was mixed with Sephin1 [50 mM) which was then present during the binding phase of the experiment. (C) As in 'B' above but with biotinylated PP1 as the ligand and PPP1R15A 533-624 as the analyte, in the absence or presence of Sephin1. (D) As in 'B' above but with biotinylated PP1 as the ligand and PPP1R15A 325-636 as the analyte, in the absence or presence of Sephin1. (E)

In vitro biotinylation reactions
Biotin (B1595, Thermo Scientific) was added to the AviTagged specified proteins (encoded by UK1897, 1920UK1897, , 1921UK1897, , 1992UK1897, , 1993UK1897, , 1994UK1897, , 1995 (Table 1) using BirA. BirA was amplified by PCR reaction from E. coli genomic DNA and inserted into an expression vector containing N-terminal GST (TEV cleavable) and ampicilin resistance marker (UK1881) ( Table 1). This protein was purified following standard GST-tagged protocols, eluted in glutathione elution buffer, aliquoted and stored in this buffer. All proteins were biotinylated and its biotinylation was checked as described (Fairhead and Howarth, 2015) Proteins in glutathione elution buffer were biotinylated and buffer exchanged into a HEPES buffer to remove excess of biotin that would interfere with the Bio-Layer Interferometry measurements.
Enzyme velocity, V was measured at substrate concentrations well below the enzyme's K m and in samples with less than 25% substrate depletion. Under these conditions, the instantaneous velocity (i.e. rate of substrate conversion to product per molecule of enzyme) is proportional to instantaneous substrate concentration and the equivalent velocity is obtained with the equation below, derived from the integrated rate equation for first order kinetics: Where Vi is the initial velocity (the instantaneous velocity at t = 0, with the dimensions of 1/t), ½s 0 and ½s f are, respectively, the substrate concentrations at the beginning and end of the reaction, Dt is the time interval of the reaction and ½ENZ is the concentration of enzyme. The 'agonist fitting' or 'inhibitor fitting' functions of GraphPad Prism V7 (RRID: SCR_002798) were used to analyze the effects of varying concentrations of reaction components on velocity.

Bio-Layer Interferometry (BLI) measurements
Proteins were diluted in HEPES buffer at the specified concentrations. Two-hundred microliters of each diluted protein preparation was placed in a 96 well plate (655209, Greiner). Streptavidin sensors (18-5019, ForteBio) were hydrated in this buffer for 2-5 min before the binding assay was performed. The plate was placed in the ForteBio Octet RED96 System for data acquisition which was performed at 25˚C at a constant orbital flow of 600 rpm. The binding assays consisted of the measurement of change in layer thickness (in nanometres) during a series of sequential steps. The sensor was equilibrated in the buffer (240 s) and the ligand (biotyninated protein) was loaded on the sensor. concentration derived from three repeats of the experiment is shown below. The EC 50 for PPP1R15A 325-636 was calculated using agonist fitting function on GraphPad Prism V7. (B) As in 'A' above, but in the presence of a fixed concentration of PPP1R15A 325-636 below the EC 50 [2 nM] and escalating concentrations of Sephin1. Shown is a representative of the two independent experiments performed. Plot contains data from the two repeats. (C) As in 'B' above, but in the presence of an escalating concentrations of the PP1 active site inhibitor tautomycin (Tau). Shown is a representative of the two independent experiments performed. Plot contains data from the two repeats. (D) As above, triplicate reactions of eIF2a-P dephosphorylation conducted in the absence or presence of Sephin1 or the related compound, Guanabenz. (E) As in 'D' using Sephin1, salubrinal or tautomycin. Shown is a representative experiment, (of two repeats). DOI: 10.7554/eLife.26109.010 The following figure supplement is available for figure 8: Ligand attachment to the sensor was checked by immersion of the sensor in buffer after loading (400-2000 s). Finally, association and dissociation of the proteins studied was analysed by soaking the sensor in analyte solutions and buffer, respectively. The duration of the ligand loading on the sensor was set to a specific time (600 s) or a specific value (2 nm displacement) depending on the experiment performed. The duration of the association and dissociation of the analyte to the ligand, . Sephin1 broadly attenuates the ER stress response in cultured CHO cells. (A) Two-dimensional plot of the fluorescence signals derived from CHO cells stably transduced with both a CHOP::GFP reporter (on the horizontal axis, Ex: 488 nm/ Em 530 ± 30 nm; reflecting mostly ISR activity) and a XBP1::Turquoise reporter (on the vertical axis, Ex: 405 nm/ Em 450 ± 50 nm; reflecting IRE1a activity) analysed by flow cytometry. Color-coded signals from untreated cells (blue) or cells exposed to a low concentration of tunicamycin (0.2 mg/mL; 20 hr) alone (green) or together with Sephin1 (50 mM, red) are superimposed. Histograms of the distribution of the two reporter signals in the three cell populations are plotted on the corresponding axis and the mean ± CV (coefficient of variation) of the fluorescence intensity of the two reporters is depicted in the bar diagram to the right. (B) As in 'A' above, but the cells were exposed to histidinol, an ISR inducer that does not promote unfolded protein stress in the ER and does not activate the XBP1:: Turquoise reporter. Shown is one of three independent experiments. DOI: 10.7554/eLife.26109.012 The following figure supplement is available for figure 9: was adjusted in order to capture bindings that had not reached equilibrium phase. Data analysis were performed using GraphPad V7 (RRID: SCR_002798) and curves were fitted to a receptor binding kinetics association then dissociation built-in model.

Gene editing Ppp1r15a mutant cells
Dual reporter CHOP::GFP, XBP1::Turquoise CHO-K1 cell line (clone S21 a derivative of RRID: CVCL_ 0214) (Sekine et al., 2015) were chosen to create Ppp1r15a knock out clones by CRISPR/Cas9 system (Ran et al., 2013). The identity of the S21 cells and their mutant derivatives has been confirmed by the persistence of the CHOP::GFP marker introduced into CHO-K1 cells (originally obtained from ATCC, catalogue number CCL-61) by the presence of proline auxotrophy and by genomic sequencing, which confirms them to be of Cricetulus griseus origin. Mycoplasma contamination is monitored frequently in our cell culture facility by cytoplasmic DAPI staining and by PCR CRISPy database (URL: http://staff.biosustain.dtu.dk/laeb/crispy/) was used to select single guide RNA sequences to target the PPP1R15A-encoding gene in exon 1 (upstream the PP1 binding motif). The two sequences selected were CRISPy Target ID 1668683 and 1671391 and duplex DNAs of the sequences were inserted into the pCas9-2A-GFP (UK1359) ( Table 1) plasmid to create CHO_PP-P1R15A_guide1_pSpCas9(BB)À2A-GFP (UK1599) and CHO_PPP1R15A_guide2_pSpCas9(BB)À2A-GFP (UK1600), (Table 1), respectively.
CHO-K1 cells were transfected with either plasmid (UK1599 or UK1600) ( Table 1) using Lipofectamine LTX (Invitrogen). Twenty-four hours later, cells were washed with PBS and resuspended in PBS containing 4 mM EDTA. A MoFlo Cell Sorter (Beckman Coulter) was used to individually sort GFPpositive cells (confirming plasmid transfection). Genomic analysis of the clones was performed using a PCR-based assay. Primers were designed to amplify the Ppp1r15a region targeted by the RNA guides. The reverse primer was labelled with 6-carboxyfluorescein (6-FAM) on the 5' end, to create fluorescently-labelled PCR products. The diluted PCR products were loaded on a 3130xl Genetic Analyzer (Applied Biosystems) and analysed using the Gene Mapper software (Applied Biosystems) to determine their length. Clones in which frame-shifting mutations were predicted by size of the PCR product, were sequenced. PPP1R15A KO (clone #1) (from Guide 1) was identified as compound heterozygous for two gene-disrupting alleles [1479_1492delGCTCAGGGTTGTCT/1491_1492ins (440n)] and PPP1R15A KO (clone #2) (from Guide 2) as homozygote [1588_1589insA]. All three alleles have a frame shift mutation 5' of the PP1 binding motif with no intervening AUG codon for in frame down-stream translation initiation of a fragment containing the PP1 binding motif.
A CRISPR guide was designed to target the region of exon 9 of Eif2ak4 (GCN2) that is upstream of the kinase domain. As previously described, a duplex DNA of CRISPy Target ID 1051489 was introduced in pCas9-2A-puro plasmid to create CHO_EIF2K4_guideA_pSpCas9(BB)À2A-Puro (UK1497) ( Table 1). Cell sorting was based on loss of the ISR (lost of CHOP::GFP signal) upon Histidinol treatment. The Eif2ak4 genomic region from sorted cells was sequenced. Figure 10 continued at t = 0 to visualize eIF2a P dephosphorylation at specified times. (B) Immunoblot of the time-dependent changes in the eIF2a P signal of compoundmutant Ppp1r15b KO ; Gcn2 KO CHO-K1 cells (clone #1) treated as in 'A'. Where indicated, the cells were additionally exposed to Sephin1 (50 mM) or the actin-polymerizing agent Jasplakinolide (1 mM), which inhibits eIF2a P -dephosphorylation by sequestering G-actin. The immunoblot of eIF2a (lower panel) serves as a loading control. Shown is a representative experiment repeated five times. (C) Plot of the eIF2a P signal (normalised to the value at t = 0 of the vehicle only (DMSO) sample) as a function of time derived from five independent experiments. The data have been fitted to an exponential decay curve (grey solid line for the vehicle and blue solid line for the Sephin1-treated sample). The exponential decay rate and the R 2 of the fit are indicated. DOI: 10.7554/eLife.26109.014 The following figure supplement is available for figure 10: The GCN2 KO clone selected was identified as homozygous for a 36019_36020insG InDel that encodes a truncated protein lacking the kinase domain.
The selected clone targeted by guide A (clone #1) was heterozygous for a 45_52del and 48_49insC. The selected clone targeted by guide B (clone #2) was heterozygous for a 1724del and 1722_1725del (residue numbering based on NCBI Reference Sequence: NW_003614184.1). Both clones encode only truncated PPP1R15B proteins, each lacking the KVxF containing catalytic domain.
(note: the existing anti-PPP1R15B sera do not recognize the hamster protein, hence confirmation of gene disruption was confined to genotypic analysis which revealed frame-shifting that precludes expression of the active C-terminal fragment).
Duplex DNAs of guide sequence [CRISPy Target ID: 1051485] was inserted into pCas9-2A-puro vector to create CHO_Eif2s1_guideC_pSpCas9(BB)À2A-Puro (UK1507) ( Table 1). CHO cells were transfected with this plasmid and a 190 bp single-stranded DNA oligonucleotide (ssODN) that carried the desired mutation (Ser51Ala) and a PAM mutation (to abolish the Cas9 cleavage site in recombinant alleles). Cells that were CHOP::GFP negative upon histidinol treatment were single-cell sorted. The genotypic analysis of the selected clone showed that it is heterozygous, one allele contains the desired mutation [5307_T>G (Serine), 5321_C>T (PAM)] and the other allele has an insertion that produces a truncated version of the protein (5326_5327insT), thus the only functional copy of eIF2a in this cell has the S51A mutation.

Flow cytometry analysis
CHO cells were plated in six well plate at 3Á10 5 cells/well density. Next day, they were treated for 20 hr with specified compounds. They were washed twice with PBS and suspended in PBS 4 mM EDTA to be evaluated by flow cytometry. Flow cytometry data were analyzed using FlowJo (FlowJo,LLC, RRID: SCR_008520) and GraphPad-Prism V7 (RRID: SCR_002798) was used to create bar graphs.
Replicates of flow cytometry experiments were analysed using Stata 14 StataCorp. (2015. Stata Statistical Software: Release 14. College Station, TX: StataCorp LP, RRID: SCR_012763). The interaction between treatments and genotypes in the different repeats was modeled using linear regression. The model was used to test whether the effect of different treatments differed between the genotypes, allowing for different mean values of CHOP::GFP and XBP1s::Turquoise on each repeat. The analysis showed non-significant differences between genotypes. However, for each cell type, there were significant differences (p<0.02) between untreated cells versus stressed cells and also between the latter and cells co-treated with Sephin1. Figure 11 continued fluorescent signal of the CHOP::GFP and XBP1s::Turquoise reporters in the two Ppp1r15a KO CHO-K1 clones. Where indicated, the cells were exposed to a low concentration of tunicamycin (0.2 mg/mL; 20 hr) alone or together with Sephin1 (50 mM). The mean ± CV (coefficient of variation) of the fluorescence intensity of the two reporters in each of the two clones is displayed in the bar diagram. Shown is one of three independent experiments. (C) As in 'C' above, but cells were exposed to histidinol. Shown is one of three independent experiments. DOI: 10.7554/eLife.26109.016 The following figure supplement is available for figure 11: Two-fold more cell lysate was loaded onto lanes 5 and 6 to compensate for the lower eIF2a content of the haploid mutant Eif2s1 S51A cells. (C) Two-dimensional plot and histograms of the fluorescent signal of the CHOP::GFP and XBP1s::Turquoise reporters in the Eif2s1 S51A CHO-K1 cells. Where indicated, the cells were exposed to a low concentration of tunicamycin (0.2 mg/mL; 20 hr) alone or together with Sephin1 (50 mM). The mean ± CV (coefficient of variation) of the fluorescence intensity of the two reporters in each of the two clones is displayed in the bar diagram. Shown is representative experiment of two independent experiments performed. Note the blunted expression of CHOP::GFP wrought by the ISR-defect imposed by the Eif2s1 S51A mutation. DOI: 10.7554/eLife.26109.018
Kinase shut-off experiment to assess eIF2a-P dephosphorylation in vivo The experimental procedure was adapted from (Chambers et al., 2015). Briefly, CHO cells (Gcn2 -/-; Ppp1r15b -/-) were plated in 10 cm dishes at 40% confluency. Sixteen-twenty hours later, fresh media was added and cells were incubated for 2 hr. Sephin1 (50 mM) or DMSO was added to the media for either 30 min or 5 hr before application of thapsigargin (300 nM for 30 min) or tunicamycin (2.5 mg/ mL for 2 hr) to induce stress by activation of PERK kinase. GSK2606414A [2 mM] was added to inhibit PERK. The PP1R15A-PP1-dependent decay of the eIF2a-P signal (by its dephosphorylation) was tracked by stopping the reaction at different time points by addition of ice-cold PBS. eIF2a-P and total eIF2a were detected by immunoblot. ImageJ (NIH) was used to quantify signal intensity and one phase decay model was used (GraphPad-Prism V7, RRID: SCR_002798) to analyse the rate decay of eIF2aP dephosphorylation.