To dissect the regulation of eIF2B assembly and activity, we purified human eIF2Bβδγε tetramers both with and without the βH160D mutation (Figure 1—figure supplement 1). We first performed sedimentation velocity experiments to assess the assembly state of eIF2B. Consistent with our previous observations (Tsai et al., 2018), WT eIF2B tetramers readily assembled into octamers in the presence of ISRIB, whereas βH160D tetramers did not (Figure 1A, B). In contrast, we found that assembly into the fully decameric holoenzyme by addition of the eIF2Bα₂ dimer was not compromised (Figure 1C, D).

To confirm that the βH160D mutation does not impair decamer assembly, we utilized an orthogonal, previously established Förster resonance energy transfer (FRET) assay to assess eIF2B’s assembly state (Schoof et al., 2021a). In this system, the C-terminus of eIF2Bβ is tagged with mNeonGreen as the FRET donor and the C-terminus of eIF2Bδ with mScarlet-i as the FRET acceptor. Both WT and βH160D tetramers were purified with these fluorescent tags (and hereafter are denoted eIF2Bβδγε-F). A titration of ISRIB readily assembled WT eIF2Bβδγε-F tetramers into octamers (EC₅₀ = 170 ± 25 nM) but did not promote βH160D eIF2Bβδγε-F assembly into octamers, even at the highest concentrations tested (Figure 1E). By contrast and in agreement with the analytical ultracentrifugation data in Figure 1A–D, titration of eIF2Bα₂ assembled both WT (EC₅₀ = 29 ± 3 nM) and βH160D (EC₅₀ = 33 ± 3 nM) eIF2Bβδγε-F tetramers into decamers with comparable efficiency (Figure 1F).

These properties are reminiscent of eIF2B’s behavior in the presence of its inhibitor eIF2-P. In the inhibited decameric conformation (I-State) induced by eIF2-P binding, ISRIB binding to eIF2B is impaired (Schoof et al., 2021a; Zyryanova et al., 2021). We next asked whether the βH160D mutation impacts eIF2B’s enzymatic activity. To this end, we monitored eIF2B’s GEF activity using a BODIPY-FL-GDP nucleotide exchange assay. Both WT and βH160D tetramers exhibited comparably low enzymatic activity. The activity was robustly enhanced in WT octamers assembled from tetramers with ISRIB but, as expected, ISRIB had no impact on βH160D tetramer activity (Figure 2A). Intriguingly, βH160D decamers were less active than WT decamers (t_1/2 = 23.6 ± 0.8 min vs. 9.3 ± 1.0 min, respectively) (Figure 2B). To understand how the βH160D mutation impaired eIF2B’s GEF activity, we next performed nucleotide exchange assays of WT and βH160D decamer activity at varying eIF2 concentrations. We measured the initial velocity of these reactions and fit the data to the Michaelis-Menten model of enzyme kinetics to determine the V_max and the K_M of the nucleotide loading reaction (Figure 2C, Figure 2—figure supplement 1). The V_max (and consequently also the k_cat) was significantly diminished by ~three fold for βH160D decamers when compared to WT decamers (WT V_max = 1.86 ± 0.13 pmol min^–1; βH160D V_max = 0.66 ± 0.03 pmol min^–1, two-sided t-test p=0.0045) suggesting that the βH160D mutation limits the intrinsic enzymatic activity of eIF2B (Figure 2D). In contrast, we could not detect a significant difference in measured K_M (WT K_M = 0.36 ± 0.06 μM, βH160D K_M = 0.19 ± 0.04 μM, two-sided t-test p=0.07).

The absence of a clear difference in K_M was puzzling, as we suspected the βH160D decamer to adopt an inhibited conformation reminiscent of the I-State, where both intrinsic enzymatic activity and binding of eIF2 are compromised (Schoof et al., 2021a). We therefore directly assessed binding affinities of eIF2B’s substrate (eIF2) and inhibitor (eIF2-P), using surface plasmon resonance (SPR) to measure binding to WT decamers, βH160D decamers, and WT tetramers. eIF2 association with WT and βH160D decamers was monophasic, but dissociation was notably biphasic irrespective of eIF2 concentration, with a fast phase and a slow phase (Figure 3A, B). Although the rate constants k_a, k_{d fast}, and k_{d slow} were broadly comparable, eIF2 binding to WT vs. βH160D decamers differed in the percentage of fast phase dissociation events (WT = 29%; βH160D = 67%) (Figure 3A, B, Table 1). Thus, a larger fraction of substrate molecules dissociates rapidly from βH160D compared to WT decamers. Since the K_M is only equal to the K_D when the dissociation rate constant k_d is much larger than the k_cat, this measurement can resolve the paradox of a similar K_M but different dissociation behavior.

In contrast to eIF2’s interaction with decameric eIF2B, binding to WT tetramers could be modeled using one phase association and dissociation. Indeed, eIF2 dissociation from tetrameric eIF2B can be thought of as being 100% fast phase as the dissociation constant was indistinguishable from the fast phase dissociation constant for both WT and βH160D (k_d = 0.12 s^–1) (Figure 3C). The fraction of eIF2 molecules that dissociate from decamers with fast phase kinetics may therefore only be engaging eIF2B through interfaces 1–3 (interfaces 1 and 2 in eIF2Bε and interface 3 in eIF2Bβ). In contrast, the eIF2 molecules that dissociate with slow phase kinetics may additionally contact interface 4 in eIF2Bδ’, reaching across the central symmetry interface (Schoof et al., 2021a). This explanation would be consistent with identical dissociation constants for tetramer dissociation and fast phase dissociation from the decamers. For eIF2-binding, the βH160D decamers can therefore be thought of as more like isolated tetramers. That is, eIF2 readily associates but then is likely to dissociate too rapidly for efficient catalysis.

We further interrogated the biphasic dissociation behavior of eIF2 from WT and βH160D decamers by varying the time allowed for eIF2 binding to eIF2B (Figure 3—figure supplement 1A, B). For both WT and βH160D we observed an exponential decrease in the percentage of fast phase dissociation, which within two minutes plateaued at ~11% fast phase dissociation for eIF2 binding to WT and at ~55% fast phase dissociation for eIF2 binding to βH160D decamers (Figure 3G). These data argue that at equilibrium the fast phase dissociation plays a small part in the engagement between eIF2 and WT eIF2B but plays a significant part in substrate engagement with βH160D decamers. This kinetic behavior can be explained by a model proposing stepwise engagement between eIF2 and eIF2B in a process that first entails engagement of 3 interaction interfaces (IF1-3), followed by a second, slower step that engages the fourth interaction interface (IF4; Figure 3H, I). In this model, the βH160D mutation does not affect the on/off rates of eIF2 engagement with eIF2B through interfaces 1–3, but slows the on-rate (k₂ in Figure 3H, I) of converting from 3 interface engagement to four interface engagement. Such a mechanism can explain the accumulation of the ‘intermediate’ fast phase dissociation species.

We next assessed eIF2-P binding to the immobilized eIF2B species. For both WT and βH160D decamer binding, this interaction could be modeled using one-phase association and dissociation kinetics. The overall affinity of eIF2-P for both species was largely comparable (WT K_D = 14 nM; βH160D K_D = 8.1 nM) (Figure 3D, E). As expected owing to the absence of the dimeric eIF2Bα subunit, which constitutes part of the eIF2-P binding site, we observed no noticeable eIF2-P binding to WT tetramers (Figure 3F).

From these results, we conclude that the βH160D decamer shares a number of properties with the eIF2-P-bound decamer: (1) reduced intrinsic GEF activity, (2) impaired substrate binding, and (3) insensitivity to ISRIB. Owing to these similarities, we wondered whether the βH160D mutation mimics eIF2-P binding and shifts eIF2B into an I-State or ‘I-State like’ conformation. To assess this notion, we determined the structure of the βH160D eIF2B decamer using single-particle cryo-EM.

We prepared the βH160D decamer by combining βH160D tetramers and eIF2Bα₂, and subjected the sample to cryo-EM imaging. After 2D and 3D classification, we generated a single consensus structure of the βH160D decamer at 2.8 Å resolution (Table 2, Figure 4—figure supplement 1) with most side chains clearly resolved (Figure 4A, Figure 4—figure supplement 1E, F). This map allowed us to build an atomic model of how the βH160D substitution alters the conformation of the eIF2B decamer. By superimposing the βH160D decamer structure and our previously published A-State structure (eIF2B-eIF2 complex, PDB ID: 6O81), we observed a significant difference in their overall architecture: the two tetramer halves of the βH160D decamer underwent a rocking motion that changed the angle between them by approximately 3.5° (Figure 4B). This rocking motion repositions the two tetramer halves in an orientation comparable to the I-State structure (eIF2B-eIF2αP complex, PDB ID: 6O9Z), although not reaching the 6° angle observed for the eIF2-P-inhibited decamer (Figure 4—figure supplement 2). To further understand how the βH160D mutation affects the conformation and dynamics of the decamer, we performed additional cryo-EM analysis of both the WT and the βH160D decamer particles (Figure 4—figure supplements 3–5). We found the following: (1) in both the WT and the mutant, the two tetrameric halves can undergo rocking motions around the central axis; (2) the βH160D mutation shifts the mean conformation of the decamer towards the I-State; and (3) the βH160D dataset likely represents particles that follow a continuous conformation distribution, rather than a mixture of distinct A and I-State populations. These observations validate our hypothesis that the βH160D mutation shifts eIF2B from the active conformation towards an inhibited conformation.

We next examined changes to the ISRIB-binding pocket. Comparing the βH160D decamer to A-State (eIF2-bound eIF2B) and I-State (eIF2α-P-bound eIF2B) structures, we noticed that its ISRIB binding pocket was 3.3 Å wider in its long dimension than that of the A-State (Figure 5A), again reminiscent of the I-State (Figure 5C). The widening of the binding pocket can explain why ISRIB fails to assemble βH160D tetramers into octamers or affect GEF activity.

Zooming in on the tetramer-tetramer interface, we examined the interactions in the WT eIF2B A-State decamer that stabilize the dimerization interface (Figure 5B). In the WT decamer, βH160 forms a π–π stacking interaction with δ’F452, which is lost in the βH160D eIF2B decamer and leads to the retraction of the short loop bearing this residue (Figure 5B and Figure 5—figure supplement 1). Other interactions in WT decamer include an ionic interaction between β’R228 and δ’D450, as well as a cation-π interaction between β’R228 andδ’F452. In the βH160D decamer, β’R228 repositions itself within the network of three negative charges (βE163, βD160, and δ’D450) and one aromatic amino acid (δ’F452) to reach a new stable state locally. The loop movement caused by the mutation propagates across the entire tetramer, resulting in the rocking motion observed in Figure 4B. This explains how the βH160D amino acid change in eIF2B remodels the dimerization interface to widen the ISRIB binding pocket and induce an I-State like conformation.

To further examine the long-range effect of this interface mutation, we looked at the critical interfaces for substrate (eIF2) binding provided by eIF2Bβ and eIF2Bδ. An overlay of the βH160D decamer structure with the eIF2B-eIF2 complex structure revealed that the substrate eIF2α binding pocket was widened by 2.9 Å (Figure 5F). As established before (Schoof et al., 2021a), a similar pocket widening is observed in the I-State structure of eIF2B (2.6 Å induced by eIF2α-P binding). This widening is predicted to prevent eIF2 from properly engaging the fourth binding site on eIF2Bδ’ and hence turns the decameric eIF2B into conjoined tetramers such that only three of the four eIF2-eIF2B binding interfaces remain readily accessible to eIF2 binding. Our structural observations, therefore, explain the decrease in eIF2 binding and reduction in GEF activity of the βH160D decamer. The remaining portion of slow phase dissociation of eIF2 from βH160D decamers, though, indicates that engagement with all four interfaces, while disfavored, is not impossible as is the case with the pure tetrameric species. By contrast, the inhibitor (eIF2α-P) binding site (Figure 5G) was not changed significantly compared to the eIF2B-eIF2α-P complex structure. This observation is consistent with the similar binding affinities measured for eIF2-P towards the βH160D decamer and the WT decamer. We conclude that the βH160D mutation shifts the eIF2B decamer into a conformation closely resembling the I-State.

Given that the eIF2B βH160D mutation biases eIF2B’s conformation toward an I-State like conformation, reducing its GEF activity, we predicted that expression of eIF2B βH160D in cells would lead to constitutive ISR activation. To test this notion, we introduced the βH160D mutation into the genome of HEK293FTR cells by editing the endogenous eIF2Bβ gene (EIF2B2) (Figure 6—figure supplement 1A). Using CRISPR/Cas9 technology, we obtained two such lines. One cell line yielded a homozygous clone in which all alleles were edited (line βH160D #1) (Figure 6—figure supplement 1B, C). The other was a heterozygous clone containing one edited allele while the remaining alleles were knocked out through CRISPR/Cas9-induced frameshift mutations (line βH160D #2). Critically, both βH160D cell lines showed eIF2Bβ and eIF2Bε protein levels comparable to the unedited parental cells, demonstrating that the mutation does not destabilize eIF2Bβ or other complex members and that compensatory mechanisms must normalize the gene dosage imbalance in clone #2 (Figure 6A; Wortham et al., 2016). We observed constitutive, low-level activation of the ISR in both clones, exemplified by elevated levels of ATF4 protein in the absence of stress (Figure 6A, lanes 5 and 9 vs lane 1). ATF4 induction was still responsive to induced stress with thapsigargin (lanes 7 and 11) but could not be alleviated by ISRIB treatment in the βH160D lines, both in the absence or presence of stressor (Figure 6A). ATF4 is translationally upregulated during the ISR and, accordingly, ATF4 mRNA levels remained unchanged between WT and the two βH160D clones (Figure 6B). However, as expected, key ATF4 transcriptional targets (such as DDIT3, ASNS, and CARS) were upregulated in βH160D cells, confirming that increased ATF4 mRNA translation leads to production of active ATF4, which in turn activates transcription of its downstream stress-responsive genes (Figure 6B).

The second hallmark of an active ISR is the general inhibition of translation initiation and, hence, a reduction in protein synthesis. To monitor protein synthesis, we treated WT and βH160D cells with puromycin and assessed puromycin incorporation in nascent polypeptide chains by immunoblotting. Both βH160D cell lines displayed significantly reduced levels of basal protein synthesis (βH160D #1 cells: 47 ± 9.0%; βH160D #2 cells: 69 ± 7.3%, both compared to WT), again consistent with constitutive activation of the ISR (Figure 6C, Figure 6—figure supplement 2). WT and βH160D cells did not differ in eIF2α phosphorylation levels, underlining the observation that the impairment of eIF2B GEF activity caused by this mutation is sufficient to induce a constitutive ISR (Figure 6C, Figure 6—figure supplement 3A-B).

Phenotypically, the constitutive ISR activation was accompanied by slow cell growth: cell doubling time increased from 25.7 ± 3.6 h for WT cells to 38.4 ± 3.5 h for βH160D (#1) cells and could not be rescued by ISRIB treatment (Figure 6D, Figure 6—figure supplement 3C).

Human eIF2Bα₂ (pJT075), Avi-tagged eIF2Bα₂ (pMS026), WT eIF2Bβγδε (pJT073 and pJT074 co-expression), eIF2Bβ^H160Dγδε (pJT102 and pJT074), Avi-tagged eIF2Bβγδε (pMS001 and pJT074 co-expression), WT eIF2Bβδγε-F tetramers (pMS029 and pJT074 co-expression), and βH160D eIF2Bβδγε-F tetramers (pMS114 and pJT074 co-expression) were purified as previously described (Tsai et al., 2018; Schoof et al., 2021a).

Human eIF2 was purified as previously described (Wong et al., 2018). This material was a generous gift of Calico Life Sciences LLC.

Analytical ultracentrifugation (sedimentation velocity) experiments were performed as previously described using the ProteomeLab XL-I system (Beckman Coulter) (Tsai et al., 2018). In brief, samples were loaded into cells in a buffer of 20 mM HEPES-KOH, pH 7.5, 150 mM KCl, 1 mM TCEP, and 5 mM MgCl₂. A buffer only reference control was also loaded. Samples were then centrifuged in an AN-50 Ti rotor at 40,000 rpm at 20°C and 280 nm absorbance was monitored. Subsequent data analysis was conducted with Sedfit using a non-model-based continuous c(s) distribution.

In vitro FRET assays were performed as previously described (Schoof et al., 2021a).

In vitro detection of GDP binding to eIF2 was performed as described previously (Schoof et al., 2021a). As before, we first monitored the loading of fluorescent BODIPY-FL-GDP to eIF2. Purified human eIF2 (100 nM) was incubated with 100 nM BODIPY-FL-GDP (Thermo Fisher Scientific) in assay buffer (20 mM HEPES-KOH, pH 7.5, 100 mM KCl, 5 mM MgCl₂, 1 mM TCEP, and 1 mg ml ^-1BSA) to a volume of 18 µl in 384 square-well black-walled, clear-bottom polystyrene assay plates (Corning). For the assay buffer, TCEP and BSA were always freshly added the day of the experiment. For the tetramer GEF assays, a 10 X GEF mix was prepared containing 1 µM eIF2Bβγδε tetramer (WT or βH160D), 2% N-methyl-2-pyrrolidone (NMP), and with or without 10 µM ISRIB, again in assay buffer. For the assay, 2 µl of the 10× GEF mix was spiked into the eIF2::BODIPY-FL-GDP mix, bringing the final concentrations to 100 nM tetramer, 0.2% NMP and with or without 1 µM ISRIB. Fluorescence intensity was recorded every 10 s for 40 s prior to the 10× GEF mix spike, and after the spike for 60 min, using a Clariostar PLUS (BMG LabTech) plate reader (excitation wavelength: 477 nm, bandwidth 14 nm; emission wavelength: 525 nm, bandwidth: 30 nm).

For assays with eIF2B decamers (WT or βH160D), decamers were first assembled by combining eIF2Bβγδε tetramer (WT or βH160D) with eIF2Bα₂ dimer in a 1:1 molar ratio (a twofold excess of eIF2Bα₂ dimer compared to the number of eIF2B(βγδε)₂ octamers) at room temperature for at least 30 min. The 10× GEF mix for decamer assays contained 100 nM eIF2B(αβγδε)₂ decamer (WT or βH160D) in assay buffer. The ensuing steps were performed as described for the GEF assays with tetramers. Immediately after the loading assay, in the same wells, we spiked in unlabeled GDP to 1 mM to measure unloading, again recording fluorescence intensities every 10 s for 60 min as before. These data were fit to a first-order exponential. For clarity, datapoints were averaged at 1 min intervals and then plotted as single datapoints in Figure 2.

The Michaelis-Menten kinetic analysis of eIF2B(αβγδε)₂ decamer (WT or βH160D) GEF activity was performed as described previously, with some minor modifications (Schoof et al., 2021a). Briefly, BODIPY-FL-GDP loading assays were performed as described above, keeping final decamer concentrations at 10 nM, but varying substrate concentration from 0 to 4 µM. BODIPY-FL-GDP concentration was kept at 2 µM final. The initial velocity was determined by a linear fit to time points acquired at 5 s intervals from 50 to 200 s after addition of decamer. To convert fluorescence intensities to pmol substrate, the gain in signal after 60 min was plotted against eIF2 concentration for the 31.5 nM – 1 µM concentrations. V_max and K_M were determined by fitting the initial velocities as a function of eIF2 concentration to the Michaelis–Menten equation in GraphPad Prism 9. For statistical comparisons of V_max and K_M, we used a two-sided t-test with α = 0.05, comparing V_max or K_M derived from the individual fit of each replicate experiment.

Affinity determination and variable association analysis by surface plasmon resonance eIF2 and eIF2-P affinity determination experiments were performed on a Biacore T200 instrument (Cytiva Life Sciences) by capturing the biotinylated WT eIF2B decamer, βH160D eIF2B decamer, and WT eIF2B tetramer at ~50 nM on a Biotin CAPture Series S sensor chip (Cytiva Life Sciences) to achieve maximum response (Rmax) of under ~150 response units (RUs) upon eIF2 or eIF2-P binding. eIF2-P was prepared by mixing 5 µM eIF2 in 50-fold excess of 100 nM PERK kinase and with 1 mM ATP. The mixture was incubated at room temperature for 60 min before incubation on ice until dilution into the titration series. 2-fold serial dilutions of purified eIF2 or eIF2-P were flowed over the captured eIF2B complexes at 30 µl min^-1 for 60 s followed by 600 s of dissociation flow. Following each cycle, the chip surface was regenerated with 3 M guanidine hydrochloride. A running buffer of 20 mM HEPES-KOH, pH 7.5, 100 mM KCl, 5 mM MgCl₂, and 1 mM TCEP was used throughout. The resulting sensorgrams were fit in GraphPad Prism 8.0. Association was fit for all species using the association then dissociation model. For eIF2-P binding this model was used to fit dissociation as well. For eIF2 binding, dissociation was fit using the two phase decay model. For eIF2 binding to WT tetramer the data could be modeled with one phase association, one phase dissociation kinetics by setting the percent fast phase dissociation to 100%. For variable association experiments, WT and βH160D eIF2B decamer was immobilized as described above. A solution containing 62.5 nM eIF2 was flowed over the captured eIF2B for 5–480 s at 30 µl min^-1 to reach the equilibrium of % fast phase dissociation vs % slow phase dissociation. Association was followed by 480 s of dissociation flow. The dissociation phase was then fit in GraphPad Prism 8.0 using the two phase decay model as described above.

Editing of the EIF2B2 gene to introduce the H160D mutation in HEK293Flp-In TRex (HEK293FTR) cells was performed using CRISPR-Cas9 according to a previously published protocol, with some minor modifications (Kwart et al., 2017). Cells were seeded at 250,000 cells well^-1 of a 12-well plate and grown for 24 hr prior to transfection with a PAGE-purified, phosphorothioate-protected single-stranded oligonucleotide donor (ssODN) for homologous recombination (C015) (Renaud et al., 2016), a plasmid containing Cas9-GFP, and a plasmid encoding the guide RNA (MLM3636-C005). The 100 nt ssODN was designed to simultaneously introduce the H160D missense mutation (CAC to GAC), to add a silent XbaI restriction site at L156 (TCTGGA to TCTAGA), and to block re-digestion by Cas9 after recombination. Transfection was done with Xtreme Gene9 reagent according to the manufacturer’s protocol, using a 3:1 ratio of reagent (µl) to DNA (µg). Reagent-only and pCas9-GFP controls were included. Two days post transfection, cells were trypsinized, washed twice in ice-cold filter-sterilized FACS buffer (25 mM HEPES pH 7.0, 2 mM EDTA, 0.5% v/v fetal bovine serum, in 1× PBS), and resuspended in FACS buffer with 400 ng ml^-1 7-AAD viability dye (Invitrogen) at around 1 million cells ml^-1 in filter-capped FACS tubes. Single GFP⁺, 7-AAD^- cells were sorted into recovery medium (a 1:1 mix of conditioned medium, and fresh medium with 20% fetal bovine serum, 2 mM L-Glutamine, 1 mM sodium pyruvate, and 1× non-essential amino acids) in single wells of 96-well plates using the Sony SH800 cell sorter. The survival rate was around 2% after 2–3 weeks. Surviving clones were expanded and first screened for correct editing by PCR and XbaI restriction digest. For this, genomic DNA was isolated using the PureLink Genomic DNA mini kit (Invitrogen), and a 473 bp fragment of the EIF2B2 gene was amplified by PCR using 300 nM forward and reverse primers (C001_F and C001_R), 300 µM dNTPs, 1× HF buffer, 100 ng genomic DNA 100 µl^-1 reaction and 2 U 100 µl^-1 reaction of KAPA HiFi polymerase for 3 min at 95°C; and 30 cycles of 98°C for 20 s, 68.9°C for 15 s, 72°C for 15 s, prior to cooling at 4°C. PCR reactions were purified using NucleoSpin Gel and PCR cleanup kit (Macherey Nagel), and HighPrep PCR Cleanup beads (MagBio Genomics) using the manufacturer’s instructions. Cleaned up products were digested using XbaI restriction enzyme (NEB) in 1× CutSmart buffer and run on a 1.5% agarose gel with 1× SYBR Safe (Invitrogen) and 100 bp DNA ladder (Promega). Clones with an XbaI restriction site were then deep sequenced to confirm correct editing and zygosity. For this, the EIF2B2 gene was amplified by PCR using 300 nM forward and reverse primers (C034_F and C034_R), 300 µM dNTPs, 1× HF buffer, 100 ng genomic DNA 100 µl^-1 reaction and 2 U 100 µl ^-1 reaction of KAPA HiFi polymerase for 3 min at 95°C; and 30 cycles of 98°C for 20 s, 64.9°C for 15 s, 72°C for 15 s, prior to cooling at 4°C. The 196 bp product was purified from a 1.5% agarose gel with 1× SYBR Safe using NucleoSpin Gel and PCR cleanup kit (Macherey Nagel), and HighPrep PCR Cleanup beads (MagBio Genomics) using the manufacturer’s instructions. A subsequent second PCR added the Illumina P5/P7 sequences and barcode for deep sequencing. For this, we used 15 ng purified PCR product per 100 µl reaction, 300 nM forward and reverse primer (C036_F_bcx, and C036_R), and 1× KAPA HiFi HotStart mix, for 3 min at 95°C, and 8 cycles of 20 s at 98°C, 15 s at 63.7°C, and 15 s at 72°C prior to cooling on ice. PCR reactions were purified using HighPrep beads (MagBio Genomics), and amplicon quality and size distribution was checked by chip electrophoresis (BioAnalyzer High Sensitivity kit, Agilent). Samples were then sequenced on an Illumina MiSeq (150 bp paired-end), and results were analyzed with CRISPResso (Pinello et al., 2016). All cell lines were negative for mycoplasma contamination. Amplicon sequencing data was deposited in NCBI’s Sequence Read Archive (SRA) under accession number PRJNA821864.

Cells were seeded at 100,000 cells well^-1 of a six-well plate and grown at 37°C and 5% CO₂. At confluency, cells were trypsinized, expanded into larger plates, and counted. This was repeated until the WT cells reached confluency in a T225 flask. For drug treatment conditions (Figure 6—figure supplement 3C), we used 500 nM ISRIB with DMSO at a final concentration of 0.1% across conditions.

Cells were seeded at 400,000 cells well^-1 of a six-well plate and grown at 37°C and 5% CO₂ for 24 hr. For drug treatment, we used 10 nM thapsigargin (Tg) (Invitrogen) and 200 nM ISRIB (made in-house) for 1 hr, ensuring the final DMSO concentration was 0.1% across all conditions. For the protein synthesis assay, puromycin was added to a final concentration of 10 µg ml^-1 for 10 min. Plates were put on ice, cells were washed once with ice-cold phosphate-buffered saline (PBS), and then lysed in 150 μl ice-cold lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% v/v Triton X-100, 10% v/v glycerol, 1× cOmplete protease inhibitor cocktail [Roche], and 1× PhosSTOP [Roche]). Cells were scraped off, collected in an eppendorf tube, and put on a rotator for 30 min at 4°C. Debris was pelleted at 12,000 g for 20 min at 4°C, and supernatant was removed to a new tube on ice. Protein concentration was measured using the bicinchonic acid (BCA) assay. Within an experiment, total protein concentration was normalized to the least concentrated sample (typically all values were within ~ 10%). A 5× Laemmli loading buffer (250 mM Tris-HCl pH 6.8, 30% glycerol, 0.25% bromophenol blue, 10% SDS, 5% β-mercaptoethanol) was added to each sample to 1×, and samples were denatured at 95°C for 12 min, then cooled on ice. Wells of AnyKd Mini-PROTEAN TGX precast protein gels (AnyKD, Bio-Rad) were loaded with equal amounts of total protein (around 10 µg), in between Precision Plus Dual Color protein ladder (BioRad). After electrophoresis, proteins were transferred onto a nitrocellulose membrane at 4°C, and then blocked for 2 hr at room temperature in PBS with 0.1% Tween (PBS-T) + 3% milk (blocking buffer) while rocking. Primary antibody staining was performed with gentle agitation at 4°C overnight using the conditions outlined in Table 3. After washing four times in blocking buffer, secondary antibody staining was performed for 1 hr at room temperature using anti-rabbit HRP or anti-mouse HRP secondary antibodies (Promega, 1:10,000) in blocking buffer. Membranes were washed 3× in blocking buffer and then 1× in PBS-T without milk. Chemiluminescent detection was performed using SuperSignal West Dura or Femto HRP substrate (Thermo Fisher Scientific), and membranes were imaged on a LI-COR Odyssey gel imager for 0.5–10 min depending on band intensity.

For the phospho-retention blots, equal amounts of total protein lysates (around 10 µg) were loaded on 12.5% Supersep Phos-tag gels (Wako Chemicals) in between Wide-view III protein ladder (Wako Chemicals). After electrophoresis, the gel was washed 3× in transfer buffer with 10 mM EDTA prior to transfer onto nitrocellulose. Blocking, antibody staining and detection was performed as described above.

Cells were seeded at 400,000 cells well^-1 of a 12-well plate and grown at 37°C and 5% CO₂ for 24 hr. The day of RNA extraction, medium was removed, and cells were lysed in 350 µl TriZOL reagent (Invitrogen). All further handling was done in a fume hood decontaminated for the presence of RNAses using RNAse ZAP (Invitrogen). Total RNA was isolated using the DirectZOL RNA miniprep kit (Zymo Research), including an on-column DNase digest, according to the manufacturer’s instructions. RNA concentration was measured using Nanodrop. cDNA was synthesized using 600 ng input total RNA per 40 µl reaction with the iScript cDNA Synthesis Kit (BioRad), cycling for 5 min at 25°C, 20 min at 46°C, and 1 min at 95°C. Samples were cooled and diluted 1/5 in Rnase-free water. qPCR reactions were set up with final 1× iQ SYBR Green supermix (BioRad), 400 nM each of Fw and Rev QPCR primers (see Table 4), 1/5 of the diluted cDNA reaction, and RNAse-free water. No-template and no-reverse transcription reactions were included as controls. Reactions were run in triplicates as 10 µl reactions in 384-well plates on a BioRad CFX384 Thermocycler, for 3 min at 95°C, and then 40 cycles of 95°C for 10 s and 60°C for 30 s, ending with heating from 55 to 95°C in 0.5°C increments for melting curve generation. C_qs and melting curves were calculated by the BioRad software. C_q values of technical replicates were averaged, and values were calculated with the ΔΔCt method using GAPDH for reference gene normalization. Graph points reflect fold changes compared to WT vehicle, with bars being the mean ± s.e.m. Statistical analysis was done using GraphPad Prism 9 on log-transformed values with ordinary one-way ANOVA and Dunnett’s post-hoc test.

Decameric eIF2Bβ^H160D was prepared by incubating 16 μM eIF2Bβ^H160Dγδε with 8.32 μM eIF2Bα₂ in a final solution containing 20 mM HEPES-KOH, 200 mM KCl, 5 mM MgCl₂, and 1 mM TCEP. This 8 μM eIF2B(αβ^H160Dγδε)₂ sample was further diluted to 750 nM. For grid freezing, a 3 μl aliquot of the sample was applied onto the Quantifoil R 1.2/1/3 400 mesh Gold grid and we waited for 30 s. A 0.5 μl aliquot of 0.1–0.2% Nonidet P-40 substitute was added immediately before blotting. The entire blotting procedure was performed using Vitrobot (FEI) at 10°C and 100% humidity.

Cryo-EM data was collected on a Titan Krios transmission electron microscope operating at 300 keV. Micrographs were acquired using a Gatan K3 direct electron detector. The total dose was 67 e^-/ Ų, and 117 frames were recorded during a 5.9 s exposure. Data was collected at 105,000× nominal magnification (0.835 Å/pixel at the specimen level), with a nominal defocus range of –0.6 to –2.0 μm.

The micrograph frames were aligned using MotionCor2 (Zheng et al., 2017). The contrast transfer function (CTF) parameters were estimated with GCTF (Zhang, 2016). For the decameric eIF2B^βH160D, Particles were picked in Cryosparc v2.15 using the apo eIF2B (EMDB: 23209) as a template (Punjani et al., 2017; Schoof et al., 2021a). Particles were extracted using an 80-pixel box size and classified in 2D. Classes that showed clear protein features were selected and extracted for ab initio reconstruction, followed by homogenous refinement. Particles belonging to the best class were then re-extracted with a pixel size of 2.09 Å, and then subjected to homogeneous refinement, yielding a reconstruction of 4.25 Å. These particles were subjected to another round of heterogeneous refinement followed by homogeneous refinement to generate a consensus reconstruction consisting of the best particles. These particles were re-extracted at a pixel size of 0.835 Å. Then, CTF refinement was performed to correct for the per-particle CTF as well as beam tilt. A final round of nonuniform refinement yielded the final structure of 2.8 Å.

For the decameric eIF2B^βH160D, the previously published apo eIF2B model (PDB ID: 7L70) was used as a starting model (Schoof et al., 2021a). Each subunit was docked into the EM density individually and then subjected to rigid body refinement in Phenix (Adams et al., 2010). The models were then manually adjusted in Coot and then refined in phenix.real_space_refine using global minimization, secondary structure restraints, Ramachandran restraints, and local grid search (Emsley and Cowtan, 2004). Then iterative cycles of manual rebuilding in Coot and phenix.real_space_refine were performed. The final model statistics were tabulated using Molprobity (Chen et al., 2010). Distances were calculated from the atomic models using UCSF Chimera (Pettersen et al., 2004). Molecular graphics and analyses were performed with the UCSF Chimera package. UCSF Chimera is developed by the Resource for Biocomputing, Visualization, and Informatics and is supported by NIGMS P41-GM103311. The atomic model is deposited into PDB under the accession code 7TRJ. The EM map is deposited into EMDB under the accession code EMD-26098.