Structure of Gcn1 bound to stalled and colliding 80S ribosomes

Significance There is growing evidence that collisions between ribosomes represent a cellular signal for activating multiple stress pathways, such as ribosome-associated quality control (RQC), the ribotoxic stress response, and the integrated stress response (ISR). Here we illustrate how a single protein can monitor both ribosomes within a disome, by presenting a cryo-electron microscopy structure of a native complex of the ISR protein Gcn1 interacting with both the leading stalled ribosome and the following colliding ribosome. The structure provides insight into the regulation of Gcn2 activation in yeast and has implications for the interplay between the RQC and ISR pathways in eukaryotic cells.


Preparation of cell extracts and sucrose gradient analysis
Yeast whole-cell extracts of the TAP-tagged GCN20 strain (SC0000; MATa; ura3-52; leu2-3,112; YFR009w::TAP-KlURA3; Euroscarf) were prepared from cultures grown to OD600 of 0.8-1.0 in either YPD or SC medium (plus supplements as required). Starvation conditions were induced by addition of 15 mM 3 amino-1,2,4-triazole (3-AT) (Sigma-Aldrich) to the SC culture 15 min before harvesting. Cells were harvested by centrifugation at 4,400 x g for 10 min at 4°C in a Sorvall SLC-6000 rotor (Marshall Scientific). The cell pellets were washed with ice cold water and lysis buffer (20 mM HEPES (pH 7.4), 100 mM KOAc, 10 mM Mg(OAc)2, 1 mM DTT, Complete EDTA-free Protease Inhibitor cocktail (Roche)), transferred to a 50 ml falcon, resuspended in lysis buffer and disrupted using glass beads. Glass beads were removed by centrifugation for 5 min at 4,400 x g at 4°C in the Rotanta 460R falcon centrifuge (Hettich). The cell debris were further pelleted by centrifugation at 17,600 x g for 15 min at 4°C in a Sorvall RC 6 SS-34 rotor. The cleared lysate was layered onto 12 ml 10-60% sucrose gradient prepared in lysis buffer and centrifuged in an SW-41Ti rotor (Beckman Coulter) for 5 h at 99,000 x g at 4°C. The ribosomal fractions were analysed using a Gradient Station (Biocomp) with an Econo UV Monitor (Bio-Rad) (Fig. S1A).

Tandem affinity purification for cryo-EM analysis and Immunoblotting
The native TAP-Tag pull-out was performed using the GCN20 (and GCN1) TAP-tagged strain essentially as described before (1). Yeast whole-cell extracts of TAP-tagged GCN20 strain (SC0000; MATa; ura3-52; leu2-3,112; YFR009w::TAP-KlURA3; Euroscarf), were prepared from cultures grown in YPD or SC media (plus supplements as required). Cells grown in YPD media were harvested at the mid log phase at an OD600 of 2.5. Cells cultured in the SC media were grown until OD600 of 0.8 and starvation was induced for 15 min by addition of 15 mM 3-AT (Sigma-Aldrich) before harvesting. Both cell pellets were lysed via glass bead disruption and the cleared lysates were incubated with IgG-coated magnetic DynabeadsR©M-270 Epoxy (Invitrogen) for 1 h at 4°C with slow tilt rotation. The elution was performed by addition of AcTEV Protease (Invitrogen) for 2 h at 17°C in elution buffer containing 20 mM HEPES (pH 7.4). 100 mM KOAc, 10 mM Mg(OAc)2, 1 mM DTT and 1 mM ADPNP (Sigma-Aldrich). The pull-out samples were analyzed by 4-12% NuPAGE gel (Thermofisher) and immunoblot analysis (Fig. S1B,C). For the latter one, the membrane with the transferred protein bands was shortly washed with phosphate-buffered saline (PBS) (pH 7.4) and blocked 3 x 15 min with 2% milk/PBS at RT. The blocked membrane was incubated for 1 h with the first TAP-tag polyclonal antibody (CAB1001, Rabbit/IgG, Thermofisher #CAB1001), washed 3x15 min with PBS and incubated with the second anti-rabbit HRP antibody (Goat/IgG, Thermofisher #31460) ON at 4°C. The next day, the membrane was shortly washed with PBS and protein bands were detected using Pierce TM ECL Western Blotting Substrate (Thermofisher #32109) and the Amersham Imager600 (GE Healthcare). Control purifications were also performed using an untagged wildtype strain, where no purification of ribosomal proteins was observed, as reported previously (2), suggesting that the ribosomal complexes observed in the TAP-tagged strains co-purified with the Gcn20. Moreover, purifications performed using eEF3-TAP strains did not enrich for Gcn1, Rbg2, Gir2 or Mbf1 (2), as we observed here for Gcn20-TAP, suggesting that the detection of these proteins is specific for the Gcn20 pull-out complex.

Detection of eIF2 phosphorylation
The yeast cell extract of the GCN20-TAP strain was prepared from culture grown to log-phase (OD600 of 0.8-1.0) in SC medium and starvation was induced by addition of 1, 5 or 15 mM 3-AT for 15 min before harvesting. 1 ml of cells with an OD600 of 1 was transferred into a 1.5 ml Eppendorf tube and centrifuged in a 5417R Microcentrifuge (Eppendorf) for 5 min at 20,000 x g at 4°C. The cell pellet was resuspended in 1 ml of ice-cold water and 150 µL of NaOH/β-Me buffer (2 M NaOH and 1.4 M β-Mercaptoethanol) was added to the sample and incubated on ice for 15 min with occasional vortexing. Subsequently, 150 µL 72% (w/v) TCA was added to each sample and further incubated on ice for 15 min and then centrifuged at 4°C, 25,000 x g for 20 min. After removal of the supernatant, the dried pellets were resuspended in 50 µL HU buffer (200 mM PBS (pH 7.4), 8 mM Urea, 5% (w/v) SDS, 1 mM EDTA, 100 mM freshly added DTT), denatured by heating 15 min at 65°C, separated on 10% SDS polyacrylamide gel and subjected to immunoblot analysis as described before. Phosphorylation level of eIF2α was detected using eIF2α antibody (Rabbit, Cell Signalling #3597) while PGK1, which was used as loading control, was detected using the commercially available PGK1 antibody (Mouse, Abacam #113687) (Fig. S1D). amino acids long (with up to 2 missed cleavages) were accepted. Intensity-based absolute quantification (iBAQ) (4) feature of MaxQuant was enabled. This normalizes protein intensities by the number of theoretically observable peptides and enables rough intra-sample estimation of protein abundance. Peptide-spectrum match, peptide and protein false discovery rate was kept below 1% using a target-decoy approach (5). All other parameters were default. Data are available via ProteomeXchange with identifier PXD021365.

Sample and grid preparation
0.02% glutaraldehyde were added to the freshly eluted TAP-Tag pull-out complex and incubated for 20 min on ice. The crosslinking reaction was quenched by addition of 25 mM Tris-HCl (pH 7.5) and n-dodecyl-D-maltoside (DDM) was added to a final concentration of 0.01% (v/v). 5 µL (8 A260/mL) of the freshly purified and crosslinked complex was applied to 2 nm precoated Quantifoil R3/3 holey carbon supported grids and vitrified using a Vitrobot Mark IV (FEI, Netherlands).

Low resolution data collection and image processing
The TAP-Tag in vivo Gcn20 pull-out sample was initially checked by generating a low resolution cryo-EM reconstruction with a dataset consisting of 264 micrographs collected on a 120 kV Tecnai G2 Spirit (FEI) transmission electron microscope (TEM) equipped with a TemCam-F816 camera (TVIPS) at a pixel size of 2.55 Å with a defocus range of −3.5 to −1.5 μm. Particle picking was performed automatically using Gautomatch (http://www.mrc-lmb.cam.ac.uk/kzhang/) resulting in 40,012 particles (Fig. S2). An initial 3D reconstruction was obtained using a vacant S. cerevisiae 80S ribosome as a reference, followed by 3D classification into 5 classes. When compared to the vacant 80S ribosome, class 1 (5%; 2,033 particles) displayed additional density within the A-site factor binding site as well as a tube-like extra density spanning the head and central protuberance of the 80S ribosome. In addition, the class 1 ribosome displayed clear density for a neighbouring ribosome, suggesting the presence of a disome. Class 2 (14%; 5,589 particles) revealed a vacant 80S ribosome whereas class 3 (17%; 6,809 particles) had some density within the A-site region.
Finally, class 4 and 5 were attributed to ´junk´ classes. The particles from class 1 were selected and re-extracted with bigger box size (increased from 420 to 700) in order to encompass the full neighbouring ribosome and subsequently 3D refined at around 21 Å resolution, revealing the structure of disome with the extra ´worm-like´ density for Gcn1 spanning both ribosomes within the disome (Fig. S2).

High resolution cryo-EM data collection and image processing
To obtain high resolution of the Gcn1-disome complex, a total of 16,823 micrographs with a total dose of 25 e -/Å 2 at a nominal pixel size of 1.084 Å and with a defocus ranging from −2.8 to −1.3 μm were collected using the EPU software (Thermo Fisher) on a FEI Titan Krios TEM (Thermo Fisher) operating at 300 kV equipped with a Falcon II direct electron detector. Each micrograph, consisting of a series of 10 frames, was summed and corrected for drift and beam-induced motion using MotionCor2 (6). The power spectra, defocus values, astigmatism and estimation of micrograph resolution were determined using Gctf (7). Automated particle picking was then performed using Gautomatch (http://www.mrc-lmb.cam.ac.uk/kzhang/), yielding initially 949,522 particles that were subjected to 2D classification using the RELION-3.0 software package (8) (Fig. S3). After 2D classification, 616,079 particles were selected and subsequently subjected to 3D refinement using a vacant S. cerevisiae 80S ribosome as initial reference. The initially refined particles were further 3D classified into 15 classes (Fig. S3). Class 1 (8.6%, 52,933 particles) was identified as the leading stalled ribosome containing densities for A-and P-site tRNAs, eIF5A, Gcn1, Rbg2 and Gir2, whereas class 2 (6.5%, 39,575 particles) was identified as the collided ribosome with densities for Gcn1, Mbf1, A/P-and P/E-site tRNAs. Class 3 (24.2%, 149,887 particles) and class 4 (24.4%, 151,167 particles) represent the major classes, both containing tRNAs. Class 3 had rotated ribosomes with hybrid A/P-and P/E-site tRNAs, analogous to that observed previously (9, 10), whereas class 4 was non-rotated with A-and P-site tRNAs. In addition, class 4 had additional density in the A-site that would be consistent with an open conformation of eEF1A that occurs after tRNA release and GTP hydrolysis (11). Class 5 (10.6%, 65,173 particles) contained non-rotated ribosomes with P-and E-site tRNAs and additional density for eRF1 in the A-site and eEF3 bound to the head and central protuberance of the 40S and 60S, respectively. Further sub-sorting of this class revealed subpopulations containing eEF3 and eRF1, eRF1 and eIF5A, as well as eRF1, substoichiometric eRF3 and eIF5A, however, these subpopulations could not be refined to high resolution due to the low particle number. Class 6 (7.7%, 47,171 particles) contained hibernating 80S ribosomes with the presence of eEF2 and Stm1, as reported previously (12,13 In order to ensure that Gcn1-containing particles were not lost during the initial 3D classification, all classes (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15) were pooled together and sub-sorted again but using a mask to focus sorting on Gcn1. This resulting in a major class with 88,453 particles. Sub-sorting of this class produced a high-resolution class (class 1 with 30,016 particles) as well as two low resolution or orientation biased classes (classes 2 and 3, totally 58,437 particles). The particles from class 1 were selected and re-extracted with bigger box size (increased from 420 to 700) in order to encompass the full neighbouring ribosome. Following 3D refinement, CTF refinement yielded a structure of the full disome (Fig. S3) with an average resolution of 4.0 Å for the leading stalled and 8.4 Å for the colliding ribosome ( Fig. S4A-C). Local refinement of the of the individual ribosomes yielded average resolutions of 3.9 Å (Fig. S4D-G) and 4.4 Å, respectively (Fig. S4H-K). Local refinement of aligned particles was also performed on the 40S head/Gcn1 region of the leading stalled ribosome (Fig. S3), which improved the local resolution of Gcn1 (Fig. S4L-O).

Molecular Modelling
A homology model of the Gcn1 eEF3-like HEAT repeat region (predicted residue 1330-1641) was created using the 80S-bound eEF3 model (this study) as a template for SWISS-MODELLER (14).
Comparison of the two cryo-EM maps of the ribosome bound by eEF3 and Gcn1 revealed an identical overall conformation and ribosomal binding site of eEF3-HEATs and Gcn1 eEF3-like HEATs. Based on that similarity, the created Gcn1 model (8 HEAT repeats; residues 1324-1638) was fitted (cross-correlation (CC) of 0.75 between model and map) into the appropriate 7 Å lowpass filtered cryo-EM map using the command ´fit in map´ in UCSF Chimera 1.13.1 (15) and manually adjusted with Coot version 0.8.9.2 (16). Beyond the Gcn1 EF3-like HEAT repeat region, three poly-alanine HEAT repeats were de novo modelled in the N-terminal direction (residues 1216-1323; (CC of 0.63 between model and map) and eight C-terminal HEAT repeats (residues 1639-1922; cross-correlation of 0.65 between model and map) using Coot version 0.8.9.2 (16). The peripheral N-and C-terminal regions could not be modelled due high flexibility of these regions in the cryo-EM map, however, the tube-like features of the density is consistent with secondary structure predictions of additional HEAT repeats, therefore, we tentatively fitted HEAT repeats from residues ~600-1100 (CC of 0.50 between model and map) and ~1950-2600 (CC of 0.57 between model and map). The very N-and C-terminal regions fuse with the stalk proteins and could not be modelled. Although Gcn20 was not well resolved, the location of the density and interaction region with Gcn1 would be consistent with the N-terminal region of Gcn20 based on available literature.
The extra density at the A-site entry factor of the stalled-leading ribosome was identified and modelled as Rbg2-Gir2. The crystal structure of the Rbg1 protein (PDB ID 4A9A) (17) was rigid body fitted into the isolated density using Chimera and the structure of Rgb2 was then generated by homology modelling within Coot (16) and Isolde (18) and subsequently refined in Phenix (19). Gir2-DFRP model was obtained using a homology model based on the Tma46 template from PDB ID 4A9A. However, due to the lack of side chain information, Gir2-DFRP has been modelled only as poly-Ala. Finally, to generate a full molecular model for the stalled leading Gcn1-Rbg2-Gir2-80S complex, existing models for the translating S. cerevisiae ribosome (PDB ID 6I7O), Phe-tRNA for A-tRNA and P-tRNA (PDB ID 1EVV), eIF5A (PDB ID 5GAK) (20) were used and combined with the Rbg2-Gir2 and Gcn1 model. The molecular model was then refined using Phenix (19).
The extra density located between the head and body of the colliding ribosome was

Figure preparation
Figures showing atomic models and electron densities were generated using either UCSF Chimera (15) or Chimera X (21) and assembled with Inkscape (https://inkscape.org/) and Adobe Illustrator.

Data Availability
The cryo-EM maps of the leading Gcn1(-Gcn20)-Rbg2-Gir2-80S) and the colliding ribosome (Gcn1-    S2. Processing of the low resolution Gcn20-TAP dataset. After automatic acquisition of 264 micrographs, 40,012 particles were selected for the initial reconstruction with an initial alignment using a vacant S. cerevisiae 80S ribosome. The particles were then subjected to 3D classification, sorting the particles into 5 classes. Class 2 and 3 contain vacant ribosomes without any extra density at the nominal resolution of 21 Å. Class 1 with the extra densities was then refined using a bigger box size and shows a disome with an extra density connecting the two ribosomes (orange) as well as an extra density at the A-site entry. Fig. S3. Processing of the high resolution Gcn20-TAP dataset. After manual inspection (AMI; before manual inspection, BMI) of 16,823 micrographs, 15,945 were selected for the initial particle picking, resulting in 949,522 particles. Following 2D classification, 616,079 particles were initially aligned against a vacant S. cerevisiae 80S ribosome and subjected to 3D classification, sorting the particles into 15 initial classes as depicted. Subsequently, all classes (1-15) were re-combined and subjected to local sorting using a mask for Gcn1. Class 1 (14.3%, 88,453 particles) containing Gcn1 bound particles was subsequently refined and 3D classified into three classes. High resolution particles in Class1 (4.9%, 30,016 particles) were again 3D and CTF refined with an enlarged box size and resulted in a final disome reconstruction at 4.0 Å for the leading ribosome (LR) and 8.4 Å for the colliding ribosome (CR). The LR (blue) and CR (grey) were masked and individually local refined resulting to 3.9 Å and 4.4 Å, respectively. To increase the resolution of Gcn1 and Mbf1, the LR and CR were subjected to an further local refinement with an individual mask encompassing the important regions: the 40S head and Gcn1 (yellow mask) for the LR and 40S (purple mask) for the CR, which resulted in final reconstructions of the masked regions at a final resolution of 4.3 Å and 3.8 Å, respectively.