Higher-order phosphatase–substrate contacts terminate the integrated stress response

Many regulatory PPP1R subunits join few catalytic PP1c subunits to mediate phosphoserine and phosphothreonine dephosphorylation in metazoans. Regulatory subunits engage the surface of PP1c, locally affecting flexible access of the phosphopeptide to the active site. However, catalytic efficiency of holophosphatases towards their phosphoprotein substrates remains unexplained. Here we present a cryo-EM structure of the tripartite PP1c–PPP1R15A–G-actin holophosphatase that terminates signaling in the mammalian integrated stress response (ISR) in the pre-dephosphorylation complex with its substrate, translation initiation factor 2α (eIF2α). G-actin, whose essential role in eIF2α dephosphorylation is supported crystallographically, biochemically and genetically, aligns the catalytic and regulatory subunits, creating a composite surface that engages the N-terminal domain of eIF2α to position the distant phosphoserine-51 at the active site. Substrate residues that mediate affinity for the holophosphatase also make critical contacts with eIF2α kinases. Thus, a convergent process of higher-order substrate recognition specifies functionally antagonistic phosphorylation and dephosphorylation in the ISR.

. the PP1-and G-actin-binding segments are marked. beneath is an alignment of human PPP1R15A (huR15A), human PPP1R15b (huR15b) and the regulatory subunit of herpes simplex virus (HSV-1). Conserved residues are in bold. the 'RVxF' motif, common to many PPP1Rs, is noted. Numbering refers to human PPP1R15A. b, Ribbon diagram of the overall crystallographic structure of the PPP1R15A-G-actin-DNase I complex at a resolution of 2.55 Å. Actin domains are numbered (I-IV). c, Close-up view of hydrogen-bonding interactions between PPP1R15A (red) and G-actin (green). the density map for PPPR15A is shown in extended Data Fig. 1. d, Hydrophobic interactions of G-actin (blue surface) with the indicated PPP1R15A residues (side chains as turquoise sticks). e, Model of a tripartite holophosphatase constructed by aligning the PPP1R15A-G-actin-DNase I complex (above) and the binary PP1G-PPP1R15b complex (PDb 4V0X) to the low-resolution PP1G-PPP1R15b-G-actin complex (PDb 4V0U); the former via G-actin and the latter via PP1. Shown is the PP1G-PPP1R15b from PDb 4V0X (with a root mean squared deviation (r.m.s.d.) of 0.318 Å between the 290 Cα pairs of PP1c) and the PPP1R15A-G-actin from the PPP1R15A-G-actin-DNase I complex (with r.m.s.d. of 0.626 Å between 343 Cα pairs of actin). DNase I is omitted for clarity. Note the proximity of the C terminus of PPP1R15b and the N terminus of PPP1R15A (consisting of the same residue of both orthologs: PPP1R15b trp 662 and PPP1R15A trp 582) in the two binary complexes. the previously unaccounted density in the barbed end of G-actin (displayed as an average difference electron density from PDb 4V0U and shown as a blue mesh) accommodates the actin-binding helices of PPP1R15A (red) from the aligned PPP1R15A-G-actin-DNase I complex.
portion of PPP1R15A, which crystallized and diffracted to a resolution of 2.55 Å (Fig. 1b-d, Extended Data Fig. 1a and Table 1).
Residues 582-619 of PPP1R15A form two helices separated by a sharp turn that wraps around G-actin, engaging the groove between subdomains I and III. DNase I binds the opposite end of G-actin, whereas D-loop insertion in F-actin and G-actin binding by small molecules, such as cytochalasin D, overlaps the PPP1R15A site (Extended Data Fig. 1b,c). These findings explain the ability of cytochalasin to antagonize the stimulatory effect of G-actin on eIF2α P dephosphorylation in vitro 25 , and the inhibitory effect of drug-enforced actin polymerization on dephosphorylation in cells 28 .
The location of PPP1R15A on the surface of G-actin corresponds to a site of electron density observed in a crystal structure of the PP1G-PPP1R15B-G-actin complex (PDB 4V0U) 25 . Due to the low resolution of the map (7.8 Å), the density remained unassigned. However, aligning G-actin in the ternary PP1G-PPP1R15B-G-actin (PDB 4V0U) and binary PPP1R15A-G-actin-DNase I complex (as here) reveals that the PPP1R15A from the high-resolution binary complex fits nicely in the unassigned density in the groove of G-actin in PDB 4V0U. Furthermore, the C-terminal PPP1R15B Trp662 from a high-resolution binary PP1G-PPP1R15B complex (PDB 4V0X, aligned by its PP1 to PDB 4V0U) is close enough to the N-terminal PPP1R15A orthologous residue (Trp582) in its complex with G-actin (as here) to complete a PPP1R15 peptide chain in the ternary complex and provide a view of a composite PPP1R15B/A-containing tripartite holoenzyme (Fig. 1e).

Validation of the role of actin in eIF2α P dephosphorylation.
Residues lining one face of the PPP1R15A helical extension form hydrophobic interactions and hydrogen bonds with G-actin (Fig.  1c,d). Human PPP1R15A Phe592 inserts into a hydrophobic cavity on the surface of actin, whereas Arg595 forms a salt bridge with actin Asp334 and PPP1R15A Arg612 forms hydrogen bonds with actin Tyr169 and Met355. These PPP1R15A residues are conserved across species and were selected for functional studies by mutagenesis.
Recombinant wild type (WT) or mutant human PPP1R15A core fragment (residues 553-624) in complex with PP1A and G-actin were compared in dephosphorylating the N-terminal lobe of eIF2α P . A kinetic defect was observed in all four mutants tested (Fig. 2a,b). The defect was selective for actin-containing mutant holophosphatases-the low baseline activity of the apoenzymes (lacking G-actin) was barely affected by the mutations. The strongest defect was observed in double mutants compromising both PPP1R15A Phe592 and Arg595; their catalytic efficiency (k cat /K m ) was nearly 50-fold less than the wild type, approaching that of the apoenzyme (lacking G-actin, Fig. 2c). The more C-terminal contact with G-actin, mediated by Arg612, proved less important.

Fig. 2 | actin-facing mutations disrupt PPP1R15a action in vitro and in cells. a,
Coomassie-stained Phostag gels resolving phosphorylated from nonphosphorylated eIF2α following eIF2α P dephosphorylation in vitro with Wt or mutant PPP1R15A holophosphatases. the concentration of binary PPP1R15A-PP1A and G-actin is indicated, as is the genotype of the PPP1R15A component. Shown is a representative example of experiments reproduced at least three times. b, time-dependent progression of dephosphorylation reactions described in a (showing all replicates). the dotted line is fit to a first-order decay. c, best fit of three independent experiments in panel b ±95% confidence intervals of the k cat /K m of the indicated PPP1R15A-PP1A pairs in reactions lacking (apo) and containing (holo) G-actin. d, Plot of the association-phase plateau bLI signal arising from a probe consisting of immobilized Wt or mutant PPP1R15A derivatives interacting with the indicated concentration of G-actin in solution. e, two-dimensional flow cytometry contour plots (with outliers) of untreated and thapsigargin (tg)-treated CHOP::GFP transgenic CHO-K1 cells transfected with mCherry alone or full-length Wt or mutant mouse PPP1R15A fused to mCherry at their C termini. the gate of the mCherry + cells analyzed in panels f and g is indicated in gray. Note the lack of effect of mCherry on the CHOP::GFP ISR marker, the attenuation of CHOP::GFP induction in cells transfected with wild-type PPP1R15A::mCherry and defective attenuation by mutations in G-actin-facing residues in PPP1R15A. f, box (25th and 75th percentile) and whiskers plots (cut-off at the 1-99th percentiles) of the CHOP::GFP signal in the mCherry + experimental populations indicated in gray (cell numbers shown in e The lack of response of the mutants to G-actin was confirmed in titration: whilst binary complexes of PP1A and wild-type PPP1R15A responded to G-actin with a vigorous increase in activity, PPP1R15A R595E alone, and even more so in conjunction with a PPP1R15A F592A mutation, markedly attenuated the responsiveness of the binary complex to G-actin (Extended Data Fig. 2). The attenuated response of the mutant binary complexes in the enzymatic assay correlated with defective G-actin binding, as assessed by biolayer interferometry (BLI) (Fig. 2d).
This new information on functionally important contacts between PPP1R15A and G-actin motivated us to revisit the role of actin in eIF2α P dephosphorylation in cells. PPP1R15     expression-mediated eIF2α P dephosphorylation attenuates the ISR 15,18 . This feature was exploited to examine the effect of the actin-facing PPP1R15A mutations in cells. We used a cell-based assay in which ISR activity was monitored by a CHOP::GFP fluorescent reporter 15 . Treatment with thapsigargin, an agent that activates the ISR by triggering PERK-dependent eIF2α phosphorylation, activated the CHOP::GFP reporter, as reflected by a shift to the right in the distribution of the fluorescent signal detected by flow cytometry. Coexpression of a control plasmid encoding mCherry alone had no effect on the CHOP::GFP signal. Expression of wild-type full-length mouse PPP1R15A (fused to mCherry at its C terminus) attenuated the thapsigargin-mediated CHOP::GFP signal 25 . Mutant mouse PPP1R15A R588E and PPP1R15A F585A (counterparts to human PPP1R15A R595E and PPP1R15A F592A ) had a weaker attenuating effect on the ISR marker, a defect that was most conspicuous in the mouse PPP1R15A F585A;R588E double mutant ( Fig. 2e-g). The mouse PPP1R15A R605A mutation had no observable effect in this assay, consistent with the weaker effect of the counterpart human PPP1R15A R612A mutation in vitro (Fig. 2a). Stress granule formation is a convenient orthogonal marker of ISR activation 32 . Consistent with this notion, the abundance of stress granules in stressed cells was attenuated by enforced expression of wild-type full-length mouse PPP1R15A, but less so by expression of PPP1R15A with G-actin-facing mutations (Fig. 2h,i).
The effects of the actin-facing mutations are consistent with G-actin contributing to the in vivo activity of a holophosphatase consisting of full-length PPP1R15A. To explore this further we compared β-actin that was wild type at its PPP1R15A-contacting surface with a charge-reversal mutant β-actin E334R (a residue contacting human PPP1R15A Arg595, Fig. 1c) for their ability to restore ISR inhibition when coexpressed with the severe, charge-reversed full-length mutant mouse PPP1R15A F585A;R588E . To favor retention of the exogenous β-actin in its G-form, a polymerization-deficient variant was used (β-actin A204E;P243K ) 33 . Neither enforced expression of otherwise wild type, nor mutant β-actin E334R affected the ability of wild-type full-length PPP1R15A to attenuate the ISR. However, β-actin E334R consistently (albeit partially) reversed the defect in mouse PPP1R15A F585A;R588E -mediated suppression of the ISR (red arrow, Fig. 3a-c).
Genetic rescue by the charge-substituted β-actin E334R was PPP1R15A allele-specific: both wild type and β-actin E334R modestly but similarly enhanced the activity of the mouse PPP1R15A F585A;R588A mutant (green arrows, Fig. 3a), whereas neither reversed the defect in the mouse PPP1R15A W575A mutant (the counterpart to human PPP1R15A W582 ). These differences in response of the PPP1R15A mutants to actin are consistent with charge complementarity of the reciprocal β-actin E334R /PPP1R15A R588E pairing, and charge exclusion of the wild-type β-actin Asp334/PPP1R15A R588E pairing accounted for the selective (albeit partial) genetic rescue in the case of the

Fig. 3 | allele-specific suppression of a PPP1R15a surface-charge mutation by a reciprocal surface-charge mutation in
β-actin. a, twodimensional flow cytometry dot plots of thapsigargin-treated cells transfected with plasmids encoding either blue fluorescent protein (bFP) alone, bFP linked (in-trans) to an otherwise wild-type (polymerization-deficient A204e;P243K) β-actin or bFP linked to polymerization-deficient β-actin e334R surface-charge reversal mutant (affecting a residue that forms a salt bridge with mouse PPP1R15A R588 ). Cells were cotransfected with plasmids encoding mCherry alone or the indicated mouse PPP1R15A::mCherry fusion proteins (as in Fig. 2e-g). Shown are the mCherry and GFP channels of the β-actin-expressing (bFP + ) populations in each dataset. β-Actin e334R selectively restores the ability of the PPP1R15A F585A;R588e mutation to attenuate the ISR (red arrow), whereas both the wild type and β-actin e334R enhance ISR attenuation by PPP1R15A F585A;R588A (green arrows). b, box (25th and 75th percentile) and whiskers plots (cut-off at the 1-99th percentiles) of the CHOP::GFP signal in untreated (-) and thapsigargin-treated (+) from the bFP + , mCherry + cells marked by gray shading in a above (cell numbers shown in a). the signal in the critical sample coexpressing PPP1R15A F585A;R588e and β-actin e334R is denoted by a horizontal red arrow. c, Individual data points and mean ± s.d. of the ISR attenuation factor of the indicated subset of PPP1R15A::mCherry fusions from replicate experiments as in a (P values for two-tailed t-test comparisons shown, n = 4 independent experiments for mCherry, Wt and F585A;R588e and n = 3 for W575A and F585A;R588A). Source data for a-c are available online.
former. The mouse PPP1R15A R588A mutation was indifferent to the surface charge of the exogenous β-actin, accounting for the similar ability of increased concentrations of wild type or β-actin E334R to partially restore its function by mass action. The mouse PPP1R15A W575A mutation, which affects PPP1R15A by a different mechanism (see below) and binds G-actin with wild-type affinity 27 , was indifferent to an increase in concentration of β-actin, wild type or mutant.

Structure of the pre-dephosphorylation complex.
To better understand the catalytically efficient tripartite holophosphatase, we sought to trap a pre-dephosphorylation complex for structural analysis. To this end, we exploited observations that a PP1A D64A mutation in the active site attenuates dephosphorylation and facilitates trapping of enzyme-substrate pre-dephosphorylation complexes 34 .
The substrate (the N-terminal lobe of eIF2α P ) coeluted with a catalytically deficient PP1A D64A -PPP1R15A-G-actin tripartite holophosphatase in size-exclusion chromatography, and the complex gave rise to well-defined particles in cryo-EM (Extended Data Fig. 3a,b and Table 2).
A cryo-EM map at an overall resolution of 3.96 Å was reconstructed and one copy of each component, PP1A D64A -PPP1R15A-G-actin-DNase I-eIF2α P , was well resolved in the map (Fig. 4a). PP1A is bound by the conserved N-terminal segment of PPP1R15A (residues 560-582 are resolved). High-resolution crystal structures of PP1A-PPP1R15A (PDB 4XPN) and PP1G-PPP1R15B (PDB 4V0X) readily dock in the cryo-EM map. The PPP1R15A chain, truncated in the crystal structure at Ala568, can be completed in the cryo-EM map and is seen to follow the known trajectory of PPP1R1B on the surface of PP1 (Extended Data Fig. 3d). The crystal structure of the binary PPP1R15A-G-actin (with the bound DNase I) also docks comfortably in the cryo-EM density, and the continuity of a composite PPP1R15B>A chain is readily established in the consecutively docked crystallographic structures (Extended Data Although present in the crystallized constructs, residues that are C-terminal to PPP1R15B Trp662 (the counterpart of PPP1R15A Trp582) were invariably disordered in binary complexes with PP1c 24,25 . This suggests an important role for G-actin binding in stabilizing the helical conformation of this segment. Engagement of the β-barrel of eIF2α by this helical assembly and the attendant positioning of the substrate loop in the active site of the holoenzyme, provide a structural explanation for the role of G-actin as a cofactor in eIF2α P dephosphorylation.
Alignment of PP1G-PPP1R15B (PDB 4V0X) to the PP1c of the pre-dephosphorylation complex reveals a conspicuous difference in the location of an invariant PPP1R15 Trp residue: in the binary complex, PPP1R15B Trp662 engages a pocket on the surface of PP1 (also bound by Phactr1 Trp542 in the PP1-Phactr1 complex 7 , PDB 6ZEE). In the pre-dephosphorylation complex the corresponding residue, PPP1R15A Trp582, has flipped into a different position on the surface of PP1c (Fig. 4d). Enforcing the 'Trp up' position, observed in the binary PP1G-PPP1R15B complex on PPP1R15A in the holoenzyme, would swing the helical assembly upwards to clash with the substrate (Fig. 4d). The 'Trp down' position, observed in the pre-dephosphorylation complex, is stabilized locally by hydrogen bonds between the Nε atom of Trp582, the PP1A Glu139 side chain and the carbonyl oxygens of Gly97 and Lys98, bringing the helical assembly into a conformation compatible with substrate recruitment and catalysis. These findings provide a structural explanation for observation that, although it makes no measurable contribution to PP1's or G-actin's binding kinetics to PPP1R15 (ref. 27 ), this invariant Trp residue has an important role in substrate dephosphorylation (see ref. 25

and below).
Plasticity of the holoenzyme is also supported by comparing the crystal structure of the PP1G-PPP1R15B-G-actin holophosphatase (PDB 4V0U) with the cryo-EM structure of the pre-dephosphorylation complex. When aligned by their PP1c regions, the G-actin-stabilized helical assembly of the crystallized holoenzyme is tilted ~18° towards the substrate (as observed in the pre-dephosphorylation complex), thus clashing with it (Supplementary Video 1). The pivot of the tilt corresponds to the region of PPP1R15B Trp662, although the resolution of the crystal structure is too low to assign a specific 'up' or 'down' position to its side chain. These observations show the potential for the holoenzyme to assume both catalytically competent and incompetent conformations. Disallowed (%) 0 eIF2α P binding in the competent conformation positions the pSer51 substrate loop in the PP1 D64A mutant active site. Density consistent with a phosphate is attached to Ser51, marking this as a pre-dephosphorylation complex. The Ser51 phosphoryl group, which occupies the position of a phosphate ion commonly observed in other PP1c structures, interacts with surrounding PP1A residues (His66, Arg96, Asn124, His125, Arg221, His248) (Fig. 4c). As expected, the M1 metal that is coordinated by PP1 Asp64 (and aligns the attacking water molecule in the wild-type enzyme 34,35 ) is missing in this PP1A D64A mutant complex. Instead, the side chain of PP1A His248 moves from its more distal position, found in bimetallic PP1s, to contact a phosphate oxygen of eIF2α P here.
There are no known structures of isolated eIF2α P . The substrate loop is disordered in the pre-phosphorylation complex with PKR (PDB 2A19). In complexes of eIF2(α P ) with its nucleotide exchange factor eIF2B, pSer51 faces the interior of the loop, a conformation that may be stabilized by interactions with surrounding eIF2B residues (PDB 6O9Z 36 , 6I3M 37 , 6JLZ 38 , 7D43 39 ). Ser51 is also inward facing in NMR solution structures of isolated nonphosphorylated eIF2α (PDB 1Q8K) 40 , suggesting this to be the favored conformation of the substrate loop. Contacts observed here between the loop backbone and PP1 residues Tyr134, Arg221 and Arg96 ( Fig. 4c and Extended Data Fig. 3e,f) may play a role in stabilizing its extended, catalytically competent conformation, with pSer51 facing outwards: but in this, there seems to be no direct role for PPP1R15.
The role of PPP1R15 in promoting catalysis is played out by enabling distant enzyme-substrate interactions. These are contingent on G-actin-mediated stabilization of the helical assembly of PPP1R15A. Binding of G-actin to one face of the helix presents its other face to the conserved β-barrel of the substrate, enabling several interactions: PPP1R15A Arg591 is buried between eIF2α Met44 and Tyr81 and forms a salt bridge with Asp83. PPP1R15A Arg594 forms hydrogen bonds with eIF2α Tyr32 and a salt bridge with Asp42. PPP1R15A Arg587 forms a salt bridge with eIF2α Asp83 and eIF2α Lys79 is within range of salt bridges with PPP1R15A Asp588 and G-actin Asp25 (Fig. 4b). These substrate interactions are limited to the N-terminal half of the kinked helical assembly of PPP1R15A, which may explain the greater impact of mutations affecting N-terminal contacts with G-actin (PPP1R15A F592A and PPP1R15A R595E/A ) compared with perturbations affecting C-terminal contacts (PPP1R15A R612A or cytochalasin D) ( Fig. 2 and ref. 25 ). Although absent from the complexes assembled here, the C-terminal lobe of eIF2α and the attached eIF2β and γ subunits are readily accommodated in the structure (Extended Data Fig. 4) Validation of substrate recognition by the holophosphatase. To examine the functional importance of the enzyme-substrate contacts noted above, we focused on three mutations: PPP1R15A R591A and PPP1R15A R594A , predicted on structural grounds to disrupt contacts with the β-barrel of eIF2α, and PPP1R15A W582A , predicted to destabilize the active conformation of the holophosphatase. Mutation of any one of these three residues to alanine weakened the holophosphatase in vitro, with PPP1R15A R591A and PPP1R15A W582A having the strongest effect (Fig. 5a-c). In cells also, the corresponding mutations, mouse PPP1R15A R584A and PPP1R15 W575A , attenuated the ability of the expressed mouse PPP1R15A to block the ISR, or interfere with stress granule formation (Fig. 5d-g).
The formation of a pre-dephosphorylation complex gave rise to a robust binding signal in BLI, with biotinylated eIF2α P (immobilized on the probe) interacting with a mutant PP1A D64A -PPP1R15A-G-actin holophosphatase (in solution) (Fig. 5h). Complex formation depended on G-actin, consistent with the latter's role in stabilizing the substrate-binding conformation of the holophosphatase. When used as a probe, biotinylated nonphosphorylated eIF2α 0 also formed a complex with the holophosphatase, but it was less stable. Importantly, the PPP1R15A W582A and PPP1R15A R591A mutations, which interfered with the enzyme's catalytic efficiency, also interfered with substrate binding in BLI (Fig. 5i).
Contacts present in the pre-dephosphorylation complex involve eIF2α residues previously observed to interact with the kinase PKR in a prephosphorylation enzyme/substrate complex 29 . Notably, the side chain of PKR Phe489 is inserted into the cleft formed by eIF2α Met44 and Tyr81 (a counterpart to PPP1R15A Arg591, whose mutation blocked dephosphorylation, Fig. 5a). eIF2α Lys79 forms a salt bridge with PKR Glu490 in the prephosphorylation complex and is within range of salt bridge formation with PPP1R15A Asp588 and actin Asp25, in the pre-dephosphorylation complex, whereas eIF2α Asp83 hydrogen bonds with PPP1R15A Arg591 in the pre-dephosphorylation complex (Fig. 6a) and with the backbone nitrogen of PKR Ala488 in the latter's substrate-binding helix G 29 .
In keeping with the functional importance of these contacts, we observed that eIF2α M44A and eIF2α Y81A mutations interfered with the ability to serve as a substrate of both the PPP1R15A-containing holophosphatase and the ISR-inducing kinase PERK, in vitro. The eIF2α K79A mutation also affected both reactions: dephosphorylation more than phosphorylation (Fig. 6b,c). The effect of the eIF2a D83A mutation on dephosphorylation could not be tested, as it blocked all phosphorylation (Fig. 6c). Together, these observations point to convergence of higher-order enzyme-substrate contacts in opposing reactions affecting eIF2α.

Discussion
The structure of the PPP1R15A-based pre-dephosphorylation complex presented here rationalizes the role of G-actin as an essential component of the holophosphatase in vitro. Corroborating biochemical studies and allele-specific suppression of a defective PPP1R15A by a rationally designed compensatory mutation in β-actin, also links G-actin to holophosphatase function in vivo. Actin-dependent contacts between the holophosphatase and a distant surface of eIF2α position the pSer51 at the enzyme active site, Fig. 4 | a cryo-eM structure of the eiF2α P pre-dephosphorylation complex. a, Overview of the cryo-eM structure of the pre-dephosphorylation complex (rendered as a GSFSC map in cryoSPARC). the threshold was set to keep the enclosed volume continuous in UCSF Chimera. Constituent proteins labeled. eIF2α P -NtD refers to its phosphorylated N-terminal domain (NtD). b, Close-up view of contacts between the β-barrel of eIF2α P and the non-actin-facing surface of PPP1R15A from the cryo-eM structure. Dotted lines mark hydrogen-bonding interactions. the density map for PPP1R15A is shown in extended Data Fig. 3d. c, Close-up view of the eIF2α P substrate loop (blue) in the PP1A active site (cyan). Substrate-contacting active site residues are indicated, as are eIF2α P pS51 and the mutated PP1A D64A . the gray ball denotes the M2 metal ion. the density map for the eIF2α P substrate loop (residues 48-53) is shown in extended Data Fig. 3e. d, Surface view of PP1A (colored by charge for orientation) with PPP1R15A (in red), G-actin and eIF2α P in ribbon diagram (all from the cryo-eM structure). PPP1R15b from the binary PP1G-PPP1R15b complex (PDb 4V0X, aligned to the cryo-eM structure by PP1G) in yellow and PPP1R15A from the crystal structure of its binary complex with G-actin-DNase I (as here, aligned to the cryo-eM structure by G-actin) in tan. Note the different disposition of PPP1R15b trp662 (from the binary PP1G-PPP1R15b complex) and its counterpart, PPP1R15A trp582 (from the cryo-eM structure of the pre-dephosphorylation complex). the 'trp up' disposition (exemplified by PPP1R15b trp662, yellow) favors the PPP1R15 helix (C-terminal to it) as found in the binary PPP1R15A-G-actin-DNase I complex (shown in tan), where it would clash with eIF2α P . e, Superposition of the cryo-eM pre-dephosphorylation complex (colored as in a) and the crystal structure of the PP1G-PPP1R15b-G-actin complex (PDb 4V0U, in ivory and gray) aligned by PP1c. Note the ~18˚ rotation of the G-actin-bound PPP1R15b helical assembly, which would lead it to clash with eIF2α P .
paralleling distal contacts conserved in the pre-phosphorylation complex. Thus, the kinases that initiate, and the phosphatases that terminate, signaling in the ISR appear to have converged on a common solution for catalytic efficiency, one that relies on recognizing features of their folded globular substrate, distinct from the substrate loop.
The enzyme studied here contains the conserved core common to PPP1R15s from all phyla. Likewise, the independently folded N-terminal domain of eIF2α, which extends flexibly from the eIF2 trimer, serves as a minimal specific substrate for kinases and phosphatases 40 . Cofactors other than G-actin may contribute to PPP1R15A-mediated dephosphorylation in cells. Interactions between the extended N termini of PPP1R15A (or B) and the β or γ subunits of the eIF2(α P ) trimer (missing here) may contribute further to catalysis. For now, however, these are unproven. Furthermore, the biochemical properties of the minimal system studied here in vitro are adequate to explain the kinetics of eIF2α P dephosphorylation observed in cultured cells (Methods). Therefore, the enzyme described here is a valid starting point to study the dephosphorylation event that terminates signaling in the ISR. The observation that higher-order contacts distant from both the substrate loop and enzyme active site promote dephosphorylation presents a challenge: whilst affinity of an enzyme for its substrate favors catalysis, residual affinity for the product is often anticatalytic. This conundrum could be settled by dephosphorylation-dependent structural changes to eIF2α that favor product dissociation. Allosteric changes have been noted in eIF2α upon Ser51 phosphorylation/dephosphorylation 36,38 , but they do not extend to the region involved in the higher-order contacts identified here. Furthermore, given the similarity in contacts with the kinases that phosphorylate eIF2α 0 and the PPP1R15 holophosphatase that dephosphorylates eIF2α P , a coherent allosteric change that promotes product release in both antagonistic reactions seems unlikely.
In an alternative scenario, cooperation between local contacts at the active site and higher-order contacts is important in formation of the pre-reaction complex. Enhanced binding of the holophosphatase to immobilized eIF2α P , compared with its binding to nonphosphorylated eIF2α (in the BLI assay), speaks to this point. The PP1A D64A mutation, used to stall the reaction, destabilizes the M1 metal ion of the active site, which can no longer coordinate an eIF2α pSer51 oxygen 35 . Due to loss of these contacts the difference in affinity of the substrate (eIF2α P ) and product (eIF2α 0 ) for the wild-type holoenzyme may be greater than that reported by the BLI experiment; possible large enough to dominate the kinetics of product release. The higher-order contacts are clearly important, as both binding in the BLI assay and the 50-fold spread in catalytic efficiency between the apo and holophosphatase are sensitive to their disruption. However, it stands to reason that successful product dissociation requires that the affinity imparted by these distal contacts be tuned to differences in binding kinetics of the active site for the substrate versus the product. The higher 'off ' rates of the product observed in BLI likely play into this. Similar considerations apply to eIF2α phosphorylation, in which higher-order contacts likely cooperate with contacts at the active site to stabilize a pre-phosphorylation complex 29,41 . We speculate that the relatively modest surface of eIF2α buried in both complexes (~1,200 Å 2 ) reflects a constraint imposed on substrate and product affinity by catalytic efficiency. Similar considerations apply to the inherent competition between eIF2(α P ) binding to eIF2B and its dephosphorylation by the PPP1R15 holophosphatases, as residues on eIF2α L 4,5 (residues 77-80) and β 5 (residues 81-85) are important to both interactions [36][37][38] . Here, the relatively high turnover of the eIF2B-eIF2α P complex 39 likely favors the one-way flow of substrate from eIF2B to the phosphatase.
PPP1R15's positive role in eIF2α P dephosphorylation appears limited to scaffolding the enzyme to favor the aforementioned higher-order contacts-PPP1R15A makes no contacts with the eIF2α P substrate loop. Nonetheless, local sculpting of the surface of PP1c by regulatory subunits is likely to affect eIF2α P dephosphorylation. For example, insertion of the eIF2α P substrate loop in the active site of PP1c is compatible with neither Phactr1, nor spinophilin, binding 5,7 . Thus, the process described here for selecting a globular domain of a phosphoprotein as substrate likely operates alongside the well-established mechanism for locally biasing access to the PP1c active site by regulatory subunits engaging its surface 9 .
In eIF2α phosphorylation, formation of higher-order contacts with the substrate is dependent on allosteric coupling of back-to-back dimerization of the kinase to alignment of its eIF2α-binding helix G 42 . Precise alignment of components is also required for higher-order contacts in the tripartite holophosphatase. This fine tuning seems to have arisen solely by refinement of the PPP1R15 component, as features that enable it to function together with PP1c and G-actin are present in simpler eukaryotes that have no counterpart to PPP1R15. For example, mammalian actin Asp334 that contacts PPP1R15A Arg595 is conserved in yeast. Similarly conserved is the actin hydrophobic pocket that accommodates PPP1R15A Phe592. Indeed, PPP1R15A is functional as an eIF2α P phosphatase in yeast and its activity depends on the integrity of residues involved in what we now understand to be G-actin contacts 43 . It thus appears that in metazoans the holophosphatase was cobbled together from one rapidly evolving component (PPP1R15) and two off-the-shelf pre-existing blocks (PP1 and G-actin).
Actin is a highly dynamic protein and its partitioning between filamentous, F-actin, and monomeric, G-actin, is responsive to upstream signals. G-actin is a limiting ligand in some physiological reactions, subordinating them to the F/G-actin ratio 44 . Actin dynamics may have been co-opted to regulate the ISR in some circumstances, but these have yet to be identified. Alternatively, G-actin incorporation into the holophosphatase may have arisen simply by its availability as a convenient building block. The genetic evidence provided here for G-actin's role as a cofactor in eIF2α P dephosphorylation in cells renders these questions pertinent.
Potent inhibitors of PP1c exist, but they nonselectively block dephosphorylation. Mechanism-based pharmacological targeting Shown is a representative of experiments reproduced at least three times. b, As in a but in the absence of G-actin. c, Display of the best fit of three independent experiments in panels a and b ±95% confidence intervals of the k cat /K m of the indicated PPP1R15A-PP1A pairs in reactions lacking (apo) and containing (holo) G-actin. d, two-dimensional flow cytometry plots of untreated and thapsigargin (tg)-treated CHOP::GFP transgenic CHO-K1 cells transfected with mCherry alone or full-length (wild type or mutant) mouse PPP1R15A fused to mCherry at its C termini (as in Fig. 2e). e, box (25th and 75th percentile) and whiskers plots (cut-off at the 99th percentile) of the CHOP::GFP signal in untreated and thapsigargin-treated cells from the gray region in d (cell numbers shown in d). f, Individual data points and mean ± s.d. of the ISR attenuation factor of the indicated subset of PPP1R15A::mCherry fusions from replicate experiments as in d (P values for two-tailed t-test comparisons shown, n = 3 independent experiments). g, Quantitation of stress granules in sodium arsenite-treated cells expressing the proteins as in d (all data, mean ± s.d., P values for two-tailed t-test comparisons shown, n = 8 biological replicates for mCherry, 16 for Wt, 10 for W575 and 11 for R584A). Results are representative of experiments reproduced three times. h, time-dependent change in the bLI signal arising from phosphorylated eIF2α P and nonphosphorylated eIF2α 0 immobilized on a bLI probe and introduced in a 4 µM solution of a ternary complexes (tC, PP1A D64A -PPP1R15A-G-actin), binary complexes of the same lacking G-actin (bC) or G-actin alone. At 900 s into the association phase the probe was shifted to a solution lacking proteins, to record the dissociation phase. i, bLI signal arising from immobilized eIF2α P and ternary complexes of PP1A D64A , G-actin and wild type or the indicated mutant PPP1R15A (as in h). Shown are representative traces of experiments reproduced at least three times. Source data for panels a-c and e-i are available online. of substrate-specific holophosphatases is a long-sought goal, but we believe previous reports of success 45 to be in error 26,27 . Dependence of the eIF2α P dephosphorylation reaction on the precise alignment of the three components of the holophosphatase hints at a path towards this goal. As noted above, the catalytic and substrate-binding lobes of the holophosphatase pivot about a hinge located between the PP1-facing and G-actin-facing portions of PPP1R15. In the catalytically favored conformation, the conserved PPP1R15A Trp582 may transition from its conventional pocket on the surface of PP1A (observed in the binary PP1A-PPP1R15B and PP1A-Phactr1     7,25 ) to the novel pocket observed here. Ligands that bind the holophosphatase and stabilize the inactive conformation of Trp582 are predicted to selectively inhibit eIF2α P dephosphorylation by interfering with the higher-order contacts necessary for terminating signaling in the ISR. Time will tell if such insights, derived from these snapshots of the eIF2α P phosphatase in action, advance targeting of dephosphorylation reactions to beneficial ends.

online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41594-021-00666-7. the indicated concentrations (2-100 nM) in a 200-µl PCR tube maintained at 20 °C. G-actin was added (to a final concentration 330 nM, unless indicated otherwise) and the sample was incubated at 20 °C for 5 min to allow ternary complex assembly. The reaction was initiated by adding phosphorylated human eIF2α 2-187 (purified from E. coli and phosphorylated in vitro with GST-PERK, as indicated above) to a final concentration of 2 µM.
The intensity of the phosphorylated and nonphosphorylated species in each lane was quantified using NIH image 65 .
Biotinylated ligands, either wild type or mutant PPP1R15A 583-621 (Fig. 2d) or phosphorylated or nonphosphorylated eIF2α   (Fig. 5h,i) both expressed in and purified from E. coli as C-terminally AviTag-His6-tagged proteins were loaded onto biosensor at a concentration of 150 nM to a binding signal of 1-2 nm, followed by baseline equilibration in buffer. Association reactions with analyte: G-actin (Fig. 2d) or wild-type or mutant complexes of PPP1R15A 553-624 (fused to MBP at its C terminus) with PP1A D64A , with or without G-actin (Fig. 5h,i), in the assay buffer described above, were conducted with a reaction volume of 200 μl in 96-well microplates (Greiner Bio-One). Experiments tracking formation of stress granules were adapted from ref. 66 : CHO-K1 cells were transfected on poly(lysine)-coated multiwell slides. DNA complexes composed of 100 ng each of G3BP-GFP and R15A-mCherry expression vector, and 450 ng of empty plasmid carrier, were formed in a ratio of 3 μl ng −1 DNA and 1 μl of Plus reagent per ng of DNA in the subsequent complex formation in OPT-MEM according to the manufacturer's instructions. After 15 min the complexes were diluted into 0.5 ml CHO medium and 125 μl was aliquoted into four wells containing cells. Medium was changed at 16 h post-transfection and 4 h later was replaced with medium containing 0.5 mM sodium arsenite. Cells were then washed and fixed in PBS-4% paraformaldehyde, followed by washing and mounting for microscopy. Micrographs of the red and green channels were taken on an EVOS (M5000 1.0.466.664) inverted photomicroscope using the ×20 magnification lens.
Quantification and statistical analysis. Enzyme kinetics. At concentrations well below the substrate K m the conversion to product (that is, substrate consumption) in an enzymatic reaction follows first-order kinetics and can be described ideally as a monophasic exponential decay.
When The experimental k obs of the conversion of substrate to product (fit to a monophasic exponential decay) corresponds to the term: At 2 µM, eIF2α is well below the substrate K m of both the PPP1R15 holophosphatase and GST-PERK 25 . Therefore, the time-dependent substrate depletion (S) in the phosphorylation and dephosphorylation reactions was fitted to a one-phase decay function in GraphPad Prism v.9 using the model below: S = (S0 − S Plateau ) exp (−Kt) + S Plateau (S 0 was constrained to the concentration of substrate at t = 0, S Plateau was set to zero, as at t = ∞, S = 0).
Due to practical limitations in the time range over which one can extend the dephosphorylation assay, or accurately obtain soundings of its progression, the range over which we can measure k obs is limited. To measure reaction progression accurately with the faster holophosphatase(s) and the much slower apoenzymes, we designed the assays to have similar k obs by varying [ENZ]: in assays of the various holophosphatases [ENZ] = 2 nM, whereas in assays of the apoenzyme [ENZ] was ~25 times higher. Therefore, k cat /K m (which is simply k obs /[ENZ]) emerges as a useful metric to compare catalytic efficiency of the holo and apoenzymes measured in assays that have very different [ENZ]. However, as the experiments most critical to the conclusions drawn here-namely, those comparing dephosphorylation by different holophosphatases-were conducted with the same enzyme concentrations [ENZ] and the same initial substrate concentrations [S] 0 , and as k cat /K m = k obs /[ENZ], the fold difference in k cat /K m of such reactions (reported in Figs. 2c, 5c and 6b,c) is identical to the fold difference of their directly measured k obs .
Computational analysis of in vivo versus in vitro kinetics of eIF2α P dephosphorylation. Following inactivation of the cognate eIF2α kinase, the eIF2α P signal in the immunoblot of CHO-K1 cell lysates decays exponentially, with an observed rate constant (k obs ) of 0.0013 s −1 (ref. 28 ) to 0.0039 s −1 (ref. 27 ). Modeling the time-dependent change in the eIF2α P signal (that is, the dephosphorylation process) to a first-order decay is justified by the observation that the concentration of eIF2α in CHO-K1 cells, ~1 µM, is well below the K m of the holophosphatase for its substrate, >20 µM (refs. 25,67 ). Working at the catalytic efficiency of the PP1A-PPP1R15A-G-actin holoenzyme observed here in vitro (k cat /K m ~0.4 × 10 6 s −1 M), a similar enzyme present in the cell at a concentration of 4-10 nM could account for the k obs of eIF2α P dephosphorylation.
Cellular concentrations of PP1A vary by tissue (70 nM−3 µM) 68 , but at the median concentration (435 nM) recruitment of just 2.5% of that pool into a PPP1R15A-containing complex would account for the observed rate of eIF2α P dephosphorylation in cells.
These crude estimates suggest that, at reasonable concentrations, an enzyme consisting of the conserved C-terminal portion of PPP1R15A could account for the activity observed in cells.

FACS (ISR quantitation).
As the effector protein in these assays is a PPP1R15A::mCherry fusion, the intensity of the fluorescence in the red channel (the y axis of the two-dimensional FACS scan) reports on the level of effector protein in each cell and facilitates gating on populations expressing similar levels of effector in comparing the median and standard error of the CHOP::GFP signal (ISR channel).
The ISR attenuation factor of wild-type and mutant PPP1R15A proteins (used in Figs. 2g, 3c and 5f) was derived as follows: GFP, mean of the median GFP fluorescence in replicate experiments; TG, thapsigargin-treated cells; mCherry, expressing mCherry alone; PPP1R15A, expressing the cognate wild-type or mutant PPP1R15A:: mCherry fusion protein.
Because it does not take into account the effect of the expressed PPP1R15A on the basal level of the ISR marker, this method of calculation underestimates the attenuation of the ISR, especially by the more potent PPP1R15A derivatives. However, as this tends to underestimate differences in the effect of various PPP1R15A derivatives on the ISR, it strengthens the reliability of such differences when detected.
Quantification of stress granules. Micrographs were imported into NIH-Fuji and transfected cells expressing both mCherry and G3BP-GFP were marked as double positive and counted using the cell counter plugin. These same cells were then assigned to stress granule-positive (two or more granules) or stress granule-negative (no detectable granules) and marked and counted. The counts for each micrograph were exported into Excel and the percentage of stress granule-positive cells over the total number of double-positive transfected cells was calculated. A total of 8-16 images containing an average of greater than 50 transfected cells were counted for each condition. A representative of three experiments is shown.