Structural basis of mRNA decay by the human exosome–ribosome supercomplex

Cloning, protein expression and purification

The complete open reading frames of SKI2 (SKIV2L, UniProt: Q6PGP7), SKI3 (TTC37, UniProt: Q15477) and SKI8 (WDR61, UniProt: Q9GZS3) were obtained from a human cDNA library (MegaMan Human Transcriptome Library, Agilent Technologies) and cloned in separate expression cassettes on a single plasmid (pACEBac1)³⁵. SKI3 was designed with a cleavable 10×His tag at the N terminus. The complex was expressed in baculovirus-infected Hi5 insect cells and purified by nickel affinity and ion-exchange chromatography. All derivatives, that is, SKI2_N38, SKI2_Δarch38 and SKI2_Δwedge38, were expressed and purified similarly. A detailed protocol can be found in a previous study¹⁷.

The complete open reading frame of SKI7 (HBS1L isoform 3, UniProt: Q9Y450; also described as SKI7_FL here) was commercially synthesized (Eurofins Genomics) and cloned with N-terminal 6×His-TRX-3C and C-terminal GSG-eGFP-TwinStrep tags under an IPTG-inducible promotor for expression in Escherichia coli (pEC-A-6xHis-TrxA-3C-SKI7-eGFP-TwinStrepII construct). Transformed E. coli BL21 STAR pRARE cells were grown in TB medium at 37 °C under antibiotic selection to an optical density at 600 nm (OD₆₀₀) of 2. The temperature was reduced to 18 °C and protein expression was induced by the addition of 500 μM IPTG. Cells were collected after 16 h by centrifugation at 8,500g and resuspended in lysis buffer containing 20 mM HEPES pH 7.5, 300 mM NaCl, 25 mM imidazole and 2 mM β-mercaptoethanol (β-ME), and supplemented with 200 U ml^−1 benzonase (Merck), 500 μM AEBSF protease inhibitor and cOmplete EDTA-free protease inhibitor cocktail (Roche). The cells were lysed by ultrasonication (Bandelin, Sonopuls basic). The recombinant protein was kept strictly at 4 °C and the purification procedure was performed quickly to avoid degradation. The lysate was cleared by centrifugation and loaded on a HisTrap HP 5-ml column (Cytiva) for nickel affinity chromatography. The column was washed with 15 column volumes of 20 mM HEPES pH 7.5, 1,000 mM NaCl, 200 mM KCl, 10 mM MgCl₂, 25 mM imidazole and 2 mM β-ME, followed by washing with 5 column volumes of lysis buffer. The protein was eluted with lysis buffer supplemented with 350 mM imidazole into a StrepTrap HP 5-ml column (Cytiva) for a second affinity step. The bound protein was washed with 10 column volumes of buffer containing 20 mM HEPES pH 7.5, 300 mM NaCl and 2 mM DTT, and the addition of 5 mM desthiobiotin (DTB) eluted the protein. The protein was concentrated in 10% glycerol (v/v), flash frozen in liquid nitrogen (LN₂) and stored at −80 °C.

The SKI7_SKI-EXO construct (residues 369–632) was cloned with a N-terminal 6×His-TRX-3C tag and also with a C-terminal 3C-eGFP-TwinStrep tag. The construct behaved similarly to SKI7_FL and was quickly purified with the same protocol to prevent degradation.

The shorter SKI7 constructs (residues 1–144 (here also SKI7_40S), 136–277, 267–386, 372–546 (here also SKI7_SKI) and 540–632 (here also SKI7_EXO)) were cloned similarly with N-terminal 6×His-TRX-3C and C-terminal GSG-eGFP-TwinStrep tags. The constructs expressed well and had no tendency for degradation. An initial affinity chromatography step followed by cleavage of the N-terminal 6×His-TRX tag during overnight dialysis combined with a second nickel affinity step to remove the cleaved 6×His-TRX tag resulted in sufficiently pure constructs as judged by Coomassie-stained SDS–PAGE. The N-terminal 6×His-TRX tag of SKI7_EXO (residues 540–632) was not cleaved, because its presence helped to avoid protein precipitation during the final concentration step. For cryo-EM sample reconstitution, TRX–SKI7_SKI was also purified without cleavage of the TRX tag.

The complete open reading frame of DIS3L (UniProt: Q8T64F) was cloned with a cleavable TwinStrep tag at the N terminus into the piggyBac vector and expressed in a HEK 293T stable cell line adapted to grow in suspension. In brief, 400–800 ml of stably transfected cells (10⁶ cells per ml) were induced with 1 μg ml^−1 doxycycline for 48 h and collected by centrifugation at 800g. Cells were resuspended in lysis buffer containing 20 mM HEPES pH 7.5, 300 mM NaCl and 2 mM DTT, and lysed using a dounce homogenizer. After clearing the lysate by centrifugation, it was loaded on a 5-ml StrepTrap XT column (Cytiva) and washed with high salt (20 mM HEPES pH 7.5, 1,000 mM NaCl, 200 mM KCl, 10 mM MgCl₂ and 2 mM DTT) and lysis buffer. The protein was eluted in buffer containing 20 mM Tris-HCl pH 8.0, 150 mM NaCl, 50 mM biotin and 2 mM DTT. A detailed description, including generation of the HEK 293T stable cell line, can be found in a previous study¹⁷.

Reconstitution of cytoplasmic exosome

Human cytoplasmic exosome (EXO10) was reconstituted from equimolar amounts of pre-assembled EXO9 (purified as previously described^22,28 and freshly purified DIS3L or DIS3L(D486N). After incubation for 30 min on ice, the mixture was concentrated by centrifugation at 3,000g using an Amicon Ultra MWCO100 centricon (Millipore) and purified by size-exclusion chromatography (Superdex 200 Increase 10/300 GL, Cytiva) in buffer containing 20 mM HEPES pH 7.5, 150 mM NaCl and 2 mM DTT.

Purification of ribosomal subunits

Human ribosomal 40S and 60S subunits were obtained from HEK 293T cells (using adapted protocols^36,37). In brief, ribosomes were pelleted by ultracentrifugation from cleared HEK 293T cell lysate and resuspended in buffer containing 2 mM puromycin and 500 mM KCl to release the nascent peptide chain and dissociate the ribosomal subunits. The 40S and 60S ribosomal subunits were subsequently separated by sucrose gradient centrifugation and the corresponding 40S and 60S fractions concentrated in buffer containing 20 mM HEPES pH 7.5, 50 mM KCl, 4 mM MgCl₂ and 2 mM DTT. A detailed description can be found in a previous study¹⁷.

Co-precipitation assays

A total of 100 pmol GFP-tagged SKI7_FL or derivates (SKI7_40S 1–144, SKI7 136–277, SKI7 267–386, SKI7_SKI 372–546 and SKI7_EXO 540–632) were mixed with 130 pmol SKI238, SKI2_N38 or EXO10 in buffer containing 20 mM HEPES pH 7.5, 150 mM NaCl, 0.05% v/v NP40 and 2 mM DTT. After incubation with GFP-binder resin (ChromoTek GFP-Trap Agarose) at 4 °C for 45 min and three washes in the same buffer, the resin was dried and taken up in SDS buffer. Input and precipitate fractions were analysed by Coomassie-stained SDS–PAGE on a NuPAGE 4–12% Bis-Tris gel (Thermo Fisher Scientific). Co-precipitation of 40S and 60S ribosomal subunits was done accordingly, but in buffer containing 20 mM HEPES pH 7.5, 50 mM KCl, 4 mM MgCl₂, 0.05% v/v NP40 and 2 mM DTT.

Intensity-based absolute quantification analysis of proteomic data

Intensity-based absolute quantification (iBAQ) values were collected across five cell lines (GM12878, hESC, HeLaS3, HepG2 and K562) using five different proteases (AspN, GluC, LysC, LysN and trypsin) from a previous proteome study²⁶. The data were plotted as box plots in Extended Data Fig. 1b. Here, quartiles are represented by boxes with whiskers extending to the highest or lowest value within 1.5 times the interquartile range. Median values are shown as thick horizontal bars inside the boxes. Values falling further than 1.5 times the interquartile distance are considered outliers and are shown as dots. iBAQ values are plotted in log scale.

In vitro transcription of RNA substrates

The genomic sequence of the CrPV-IRES was taken from the intergenic region of the CrPV genome (NCBI GeneBank: 6025–6232 nt, NC_003924.1) and fused to a portion of the firefly luciferase open reading frame (modified from a previous report³⁸). The used CrPV-IRES construct was amplified by PCR, gel-purified and transcribed in vitro with T7 RNA polymerase by run-off transcription. The transcription reaction was performed in 40 mM Tris-HCl pH 8.0, 28 mM MgCl₂, 0.01% Triton X-100, 1 mM spermidine and 5 mM DTT in the presence of 25 mM of each ribonucleotide (ATP, CTP, GTP and UTP; Jena Bioscience) and 100 U μl^−1 T7 RNA Polymerase (MPI Biochemistry Core Facility) at 37 °C for 4 h. The DNA template was digested with 1 U μl^−1 DNase I (Roche) and the CrPV-IRES purified by LiCl₂ precipitation. A detailed description can be found in a previous study¹⁷.

The transcription template for the Brc-CrPV-IRES used in the exosome degradation assays was amplified by PCR, fusing the sequence of the broccoli light-up aptamer (GCGGAGACGGTCGGGTCCAGATATTCGTATCTGTCGAGTAGAGTGTGGGCTCCGC)³⁹ to the 5′ end of the CrPV-IRES. Otherwise, the Brc-CrPV-IRES was transcribed as described above.

Exosome RNA-degradation assays

The 80S-Brc-CrPV-IRES substrate used in the assays was reconstituted by mixing 100 pmol purified human 40S with 120 pmol Brc-CrPV-IRES. After incubation at 37 °C for 5 min, 120 pmol purified human 60S was added, and the incubation was allowed to proceed for another 10 min. The 80S-Brc-CrPV-IRES complex was subsequently purified by sucrose gradient centrifugation in a SW 40 Ti rotor (Beckman Coulter) at 4 °C. Sucrose gradients (15–40%) (w/v) in buffer containing 20 mM HEPES pH 7.5, 50 mM KCl, 4 mM MgCl₂ and 2 mM DTT were mixed using the Gradient Master (Biocomp). After 17 h centrifugation at 22,800 rpm, the gradients were fractionated using a Piston Gradient Fractionator (Biocomp). Fractions corresponding to the 80S-Brc-CrPV-IRES complex were concentrated in buffer containing 20 mM HEPES pH 7.5, 50 mM KCl, 4 mM MgCl₂ and 2 mM DTT by centrifugation at 3,000g using an Amicon Ultra MWCO100 centricon (Millipore). The concentrated 80S-Brc-CrPV-IRES was estimated to be 0.24 μM and used immediately.

Exosome degradation assays were performed in technical triplicates, over a time course of 45 min at 37 °C in buffer containing 20 mM HEPES pH 7.5, 50 mM KCl, 4 mM MgCl₂ and 2 mM DTT, supplemented with 0.5 mM ADP or ATP as indicated in the condition. A single reaction contained 2 pmol 80S-Brc-CrPV-IRES substrate, 2 pmol SKI238, 4 pmol SKI7 and 4 pmol EXO10 in a total volume of 30 μl. Reactions without the addition of EXO10 were included as negative controls for every condition. Time-point samples were taken at 0 min, 5 min, 15 min and 45 min, and the reaction was stopped by adding 10× excess buffer containing 100 mM Tris-HCl pH 7.5, 150 mM NaCl, 300 mM NaOAc pH.5.2, 10 mM EDTA and 1% SDS. The reactions were purified by phenol extraction and ethanol precipitation and separated on a denaturing 12% polyacrylamide gel containing 7 M urea. After incubation of the gel in water to remove urea, the Brc-CrPV-IRES was refolded for 20 min under agitation in the presence of 3,5-difluoro-4-hydroxybenzylidene imidazolinone (Sigma-Aldrich) in buffer containing 40 mM HEPES pH 7.5, 125 mM KCl and 10 mM MgCl₂, and subsequently visualized by fluorescence at a 501-nm wavelength (Typhoon FLA7000, Cytiva). Gel images were quantified by densitometry in ImageJ and plotted with R.

Cryo-EM sample preparation

The exosome–ribosome supercomplex sample (EXO10-SKI-80S) was prepared by sequentially mixing the individual components, similar to the ribosome-bound samples that we previously described¹⁷. Sucrose gradient centrifugation and concentration steps resulted in reduced amounts of 80S-bound exosomes according to SDS–PAGE and cryo-EM analysis, and were therefore omitted from the sample preparation procedure. In brief, 100 pmol of purified human 40S was mixed with 100 pmol CrPV-IRES carrying a 29-nt 3′ overhang (referred to throughout the text simply as CrPV-IRES) in buffer containing 20 mM HEPES pH 7.5, 50 mM KCl, 4 mM MgCl₂ and 2 mM DTT. After incubation at 37 °C for 5 min, 110 pmol human 60S was added and incubation proceeded for another 10 min. Next, 100 pmol of purified wild-type SKI238 was added to the 80S-CrPV-IRES complex and incubated for 5 min at 37 °C before adjusting the reaction to 1 mM ATP. After incubation at 37 °C for 15 min, 100 pmol of preincubated SKI7_SKI-EXO–EXO10(D486N) (equimolar amounts of EXO10(D486N) and SKI7_SKI-EXO incubated for 15 min at 37 °C) was added to the reaction. After 5 min at 37 °C the reaction was adjusted to 2 mM ADP-BeF to stop the ATPase reaction. The incubation was allowed to proceed for another 15 min before adjusting the final volume to 130 μl with the buffer described above and storing the sample at 4 °C. The concentrations of SKI238, SKI7_SKI-EXO and EXO10(D486N) were between 12 and 20 μM, leading to only minor alterations in volume and salt concentration in the reaction, and were not compensated for. Aggregated protein was removed by centrifugation at 10,000g and the sample supplemented with 0.04% (v/v) n-octyl-β-d-glucoside before grid preparation. Two times 4-5 μl of the sample were incubated for 1 min on holey carbon grids with carbon support (R2/1, 200 mesh, Quantifoil), that were previously glow-discharged with negative polarity at 20 mA for 30 s using an EMS GloQube (MiTeGen). The sample was plunge-frozen in a liquid ethane–propane mix using a Vitrobot Mark IV (Thermo Fisher Scientific) operated at 4 °C and 95% humidity.

The SKI7-SKI238 sample was prepared by mixing 600 pmol SKI2_Δarch38 with approximately 900 pmol of TRX–SKI7_SKI and incubating at 37 °C for 20 min in the presence of 3C protease to cleave the TRX tag. The SKI7_SKI fragment (residues 372–546) was difficult to quantify because it does not contain amino acids that would absorb at 280 nm. Therefore, the TRX tag was kept and only cleaved while reconstituting the complex. The sample was then concentrated in buffer containing 20 mM HEPES pH 7.5, 100 mM NaCl and 2 mM DTT to approximately 30 μl. After centrifugation at 18,000g for 10 min to pellet larger aggregates, the sample was injected onto a Superose 6 Increase 3.2/300 column (Cytiva) for size-exclusion chromatography by an Aekta micro (Cytiva). A single peak fraction, containing approximately 500 nM, was collected, and mixed with 0.04% (v/v) n-octyl-β-d-glucoside. Four to five microlitres of the sample was applied to holey carbon grids (R2/1, 200 mesh, Quantifoil), that had been glow-discharged with negative polarity at 20 mA for 30 s using an EMS GloQube (MiTeGen) beforehand. The sample was plunge-frozen in a liquid ethane–propane mix using a Vitrobot Mark IV (Thermo Fisher Scientific) operated at 4 °C and 95% humidity.

Cryo-EM data collection and processing

The cryo-EM data of the exosome–ribosome supercomplex were collected on a FEI Titan Krios GIII microscope (Thermo Fisher Scientific) at 300 kV, equipped with a Gatan K3 direct electron detector operating in electron-counting mode. The microscope is equipped with a post column energy filter (Gatan GIF Quantum LS) set to a slit width of 10 eV. Images were collected by underfocused acquisition (target range of −0.6 μm and −2.4 μm) at a nominal magnification of 105,000× set up in SerialEM⁴⁰ using a beam-tilt-based multi-shot acquisition scheme for faster imaging. This resulted in 48,004 micrograph movies (30 movie frames each) acquired at a pixel size of 0.85 Å per pixel with a total exposure of 64.14 e per Ų spread over 3 s.

On-the-fly micrograph movie processing was assisted by Focus⁴¹, which ran MotionCor2 (ref. ⁴²), GCTF⁴³ and Gautomatch (https://www2.mrc-lmb.cam.ac.uk/download/gautomatch-053/) on individual images while the data were being collected. Subsequent particle processing was performed in RELION v.3.1 (ref. ⁴⁴). Eightfold downscaled particles were extracted from the aligned, exposure-weighted micrographs and classified in 2D and 3D to discard non-ribosomal and low-resolution particles from the data. These data were separated on the basis of the results of the 2D classification into one particle stack displaying features of the EXO10-SKI-80S assembly and a second one showing density for the EXO10-SKI-40S assembly only.

The cleaned EXO10-SKI-80S particles (1,189,927) were extracted at the original pixel size of 0.85 Å per pixel and aligned in 3D auto-refinement with a spherical mask using a 40-Å downfiltered starting model based on PDB 4UG0 (ref. ⁴⁵). The refined particles were then used to subtract a large portion of the 80S ribosome signal from the corresponding particle images. To improve the alignment precision and the quality of the reconstructions in the subsequent steps, the particle images were recentred on the remaining EXO10-SKI signal and re-extracted in a smaller box in RELION v.3.1 (ref. ⁴⁴). Subtracted particle images that did not show SKI, EXO10 or leftover ribosome signal were removed in 3D classification and resulted in a set of 565,489 cleaner particles for downstream processing. Further classification in 3D showed a small range of different particles which aligned in classes with SKI2_H complexed by a prominent EXO10 density, with SKI238 without visible exosome density, and in classes that only showed leftover ribosome signal without visible EXO10-SKI density.

In a final step, all EXO10-SKI-containing particles (229,580) were 3D-classified into 6 classes, which resulted in 79,353 well-aligning particles. The ribosome-signal-subtracted particles were aligned by 3D auto-refinement and resulted in a reconstruction of EXO10-SKI at an overall resolution of 3.7 Å (according to the Fourier shell correlation (FSC) cut-off criterion of 0.143)⁴⁶. This map is referred to as the ‘EXO10-SKI-80S (80S subtracted)’ map, or map1. For masking and map sharpening in the RELION v.3.1 post-processing procedure⁴⁴, an ad hoc b-factor of −30 was applied. The automatically determined b-factor of −110.8 resulted in an oversharpened map with a loss of connectivity particularly in the area of DIS3L. Next, the signal subtraction was reverted and the corresponding EXO10-SKI-80S particles refined to 3.2-Å global resolution (according to the FSC cut-off criterion of 0.143)⁴⁶. An ad hoc b-factor of −20 was preferred over the automatically estimated b-factor of −71.7 in the RELION post-processing routine⁴⁴ to better visualize the 40S ribosomal subunit. This reconstruction is referred to as the ‘EXO10-SKI-80S (full map)’ map, or map3. The quality of the map in areas of the CrPV-IRES and EXO10-SKI, however was not satisfactory. Therefore, we subtracted the signal of the 60S ribosomal subunit from the reconstruction and aligned the resulting EXO10-SKI-40S particles by 3D auto-refinement, which yielded a reconstruction at 3.3-Å global resolution (according to the FSC cut-off criterion of 0.143)⁴⁶. Masking and map sharpening in RELION post-processing⁴⁴ was performed using an ad hoc b-factor of −20 (automatically determined b-factor −73.6). Although the map quality for EXO10-SKI in this reconstruction improved only marginally, the resolution and volume connectivity for the CrPV-IRES in the intersubunit space improved substantially. This EXO10-SKI-40S reconstruction is referred to as the ‘EXO10-SKI-80S (60S subtracted)’ map, or map2.

For display and visualization purposes, a composite map of the EXO10-SKI-80S assembly was calculated using map1, map2 and map3 and is shown in Fig. 1. The EXO10-SKI map detailed in Fig. 2 and Extended Data Fig. 3 was produced using DeepEMhancer⁴⁷ with the wide target setting.

As described above, the ribosome-bound EXO10-SKI supercomplex data contained a fraction of particles that lacked the entire 60S ribosomal subunit (described as EXO10-SKI-40S here). Two- and three-dimensional classification was performed in RELION v.3.1 (ref. ⁴⁴) and resulted in 119,144 clean 40S ribosomal particles, which were aligned in 3D auto-refinement for subsequent subtraction of the majority of the 40S ribosome signal. The subtracted particle images were 3D-classified into four classes using a wide mask and a 30-Å low-pass-filtered starting model of the SKI2_N38 gatekeeping module (PDB 7QDS)¹⁷. The classification showed one class with 53,460 particles that shows well-aligning density for the SKI2_N38 gatekeeping module. Three-dimensional auto-refinement of the subset gave a reconstruction at 4.1-Å global resolution (according to the FSC cut-off criterion of 0.143)⁴⁶. An ad hoc b-factor of −30 was used in the RELION v.3.1 post-processing procedure (automatically determined b-factor −134.4)⁴⁴. This reconstruction is referred to as ‘EXO10-SKI-40S (EXO10-SKI2_H + 40S subtracted)’, or map5. For the corresponding ribosome-bound reconstruction, signal reversion procedures similar to the EXO10-SKI-80S particles above were applied, which led to a EXO10-SKI-40S reconstruction at 3.4-Å global resolution using a manually determined b-factor of −20 (automatically determined b-factor −68.9). This reconstruction is labelled ‘EXO10-SKI-40S (full map)’, or map6. Although the reconstruction shows clear density for the SKI2_N38 and SKI2_H helicase modules interacting at two different positions of the 40S subunit, the signal for EXO10 was less coherent. To improve the EXO10 density, repeated signal subtraction of the 40S ribosomal subunit combined with 3D classification on the exosome and signal reversion resulted in 40S ribosome signal-subtracted and signal-reverted 3D reconstructions at 6.5 Å (ad hoc b-factor −30, automatically determined b-factor −128.3) and 4.7 Å (ad hoc b-factor −20, automatically determined b-factor −92.5) global resolution, respectively. The reconstruction focused on EXO10-SKI is labelled ‘EXO10-SKI-40S (40S + SKI2_N38 gatekeeping module subtracted)’, or map7, and the signal-reverted reconstruction is named ‘Control EXO10-SKI-40S’, or map8. The main purpose of the latter was to clarify that in the particle stack showing clear presence for the EXO10-SKI assembly, the SKI2_N38 gatekeeping module is also bound to the 40S subunit (see also Extended Data Fig. 7g).

For display and visualization purposes, a composite map of the EXO10-SKI-40S assembly was calculated using map5, map6 and map7 and is shown in Fig. 3.

Data processing of the ribosome-bound human EXO10-SKI supercomplex dataset required signal subtraction of the ribosome to yield reconstructions of comparable interpretability and quality to those described in the manuscript. Classical focused classification and refinement procedures did not yield results of comparable quality.

The EXO10-SKI-80S particles show additional density at the location where the SKI2_N38 gatekeeping module binds to the 40S ribosomal subunits in the particles that lack 60S (see Extended Data Fig. 6b). However, the signal is rather weak in comparison, owing presumably to the presence of the 60S. To improve the additional unaccounted density at the 40S head (Extended Data Fig. 6b, left) focused 3D classification using a spherical mask located around the area of interest was performed. The resulting particle stack was used for 3D refinement, resulting in a 3D reconstruction of the 80S ribosome. A low-pass-filtered map of this assembly was used to place the model of SKI2_N38 by rigid-body fitting (Extended Data Fig. 6b, right).

The cryo-EM data of the human SKI7-SKI238 complex were collected similarly to the ribosome-bound dataset described above. A total of 20,224 micrograph movies were acquired at a pixel size of 0.85 Å per pixel with a total exposure of 67.8 e per Ų spread over 6 s and 40 movie frames. Beam-induced motion correction, CTF estimation, particle picking and processing were done similarly to what was done for the ribosome-bound data above. Fourfold downscaled particles were first extracted from the aligned, exposure-weighted micrographs, and particles that seemed to be non-SKI238 were discarded by 2D and 3D classification. The resulting subset of 422,913 SKI238 particles was extracted at the original pixel size of 0.85 Å per pixel. Further 3D classification into six classes, using a starting model based on a previous cryo-EM structure of the SKI238 complex (PDB 7QDS)¹⁷, gave a final subset of 151,860 particles. Initial 3D auto-refinement gave a 3D reconstruction at an estimated overall resolution of 3.6 Å. The quality of the reconstruction and the level of resolved detail were further improved by Bayesian polishing to correct for per-particle motion and by iterative refinement of the per-particle CTF (taking into account the beam-tilted data-acquisition scheme). This resulted in a final reconstruction with an overall global resolution of 3.4 Å according to the gold-standard FSC (0.143) criterion⁴⁶. Masking and map sharpening using an automatically determined global b-factor of −99.4 were performed in the RELION post-processing routine⁴⁴. This map is referred to here as ‘SKI7-SKI238’, or map4.

Density interpretation, model building and refinement

The 80S-bound EXO10-SKI or map1 reconstruction was interpreted by rigid-body fitting pre-existing models of SKI2 (PDB 7QE0)¹⁷, the nuclear human exosome (PDB 6D6Q)²⁷ and AlphaFold multimer predictions of either SKI7_EXO or DIS3L with different human exosome subcomplexes⁴⁸. Initial rigid-body fitting was performed in UCSF Chimera⁴⁹. In several areas of the map (the EXO9 barrel and SKI2_H), the resolution and quality of the reconstruction allowed us to manually adjust the fit of the models in Coot⁵⁰, followed by real-space refinement from within the PHENIX suite⁵¹. In peripheral areas of the reconstruction (the DIS3L exonuclease), the resolution and quality of the map was not as good, and required us to use a focused refined map with improved map quality and connectivity, which allowed for rigid-body docking of the AlphaFold model. The electron density for the RNA path inside SKI2_H, between SKI2_H and EXO9, and within DIS3L was sufficient to model individual nucleotides. However, the precise sequence of the RNA inside the barrel could not be determined from the reconstruction. Inside the EXO9 barrel and between EXO9 and DIS3L, a previous model (PDB 6D6Q)²⁷ was used to inform structure building. The final model of the 80S-bound EXO10-SKI complex was validated using MolProbity⁵².

The complete model of 80S-bound EXO10-SKI was subsequently assembled by rigid-body fitting of the individual 80S, CrPV-IRES RNA and the above-described EXO10-SKI model into the full reconstruction of the EXO10-SKI-80S (map3) in PHENIX. As a starting model for the human 80S-CrPV-IRES, we used PDB 4UG0 (ref. ⁴⁵) and PDB 4V92 (ref. ³⁶) as already described in detail previously¹⁷. After placing the models into the map, the CrPV-IRES was manually adjusted and refined in PHENIX to reflect the lack of PK-1, the connection between the RNA exit of the ribosome and entry into SKI2_H and the channelling through the EXO10-SKI complex.

The model for the reconstruction of the 40S-bound SKI2_N38 gatekeeping module from which the signal of the 40S subunit and EXO10-SKI had been subtracted (map5) was interpreted with a pre-existing model of the SKI2_N38 gatekeeping module (PDB 7QDS)¹⁷. Initial rigid-body fitting was performed in UCSF Chimera⁴⁹. Inspection of the fit in Coot⁵⁰ showed insufficient density for several TPRs towards the N terminus of SKI3, which were consequently not modelled. Otherwise, the model is in good agreement with the experimental density. The final model was real-space-refined within the PHENIX suite⁵¹ and subsequently validated using MolProbity⁵².

The model for the reconstruction of the 40S-bound EXO10-SKI from which the signal of the 40S and the SKI2_N38 gatekeeping module had been subtracted (map7) was interpreted by rigid-body fitting of the above-described 80S-bound EXO10-SKI model, real-space-refined within the PHENIX suite⁵¹ and validated using MolProbity⁵².

The complete model of the 40S-bound EXO10-SKI was subsequently assembled by rigid-body fitting of the individual models of the 40S subunit, the adjusted model of the CrPV-IRES similar to that described above, the model of the SKI2_N38 gatekeeping module and the EXO10-SKI model into the full reconstruction of the 40S-EXO10-SKI (map6) in PHENIX. As a starting model for the human 40S, we used the model of the 40S ribosomal subunit from PDB 4UG0 (ref. ⁴⁵).

The reconstruction of SKI7-SKI238 (map4) was interpreted with pre-existing models of the SKI2_N38 gatekeeping module (PDB 7QDS)¹⁷ and an AlphaFold multimer prediction of a part of the SKI2_N38 gatekeeping module in complex with SKI7_SKI. Initial rigid-body fitting was done in UCSF Chimera⁴⁹. The resolution and quality of the map allowed us to manually modify and refine the models in Coot⁵⁰ and subject them to subsequent rounds of real-space refinement from within the PHENIX suite⁵¹. The final model of human SKI7-SKI238 was validated using MolProbity⁵².

Isolation of stalled RNCs and polysome profiling

The 5′ capped XBP1-XTEN mRNA was obtained by an in vitro transcription reaction from the corresponding DNA template using the mMESSAGE mMACHINE T7 transcription Kit (Thermo Fisher Scientific) as described previously³³. In vitro translation reactions were performed using the rabbit reticulocyte lysate (RRL) system (nuclease treated, Promega). Translation of the 5′ capped XBP1-XTEN mRNA was performed at 32 °C for 25 min using 0.05 μM XBP1-XTEN template and 0.8 U μl^−1 RNasin Plus ribonuclease inhibitor (Promega) in a total reaction volume of 1 ml. The stalled RNCs on the 5′ capped XBP1-XTEN mRNA were affinity-purified through the 8×His tag on the nascent polypeptide chain and magnetic beads, as described³³. The amounts of isolated ribosomes were estimated by UV absorbance at 260 nm. Sucrose density gradients (10–50% sucrose in 50 mM HEPES pH 7.5, 100 mM KOAc, 5 mM Mg(OAc)₂, 2 mM DTT, 0.1% octaethylene glycol monodedecyl ether (TCI) and 20 U ml^−1 recombinant RNasin ribonuclease inhibitor (Promega)) were prepared using a Gradient Master (Biocomp). The purified RNCs were incubated with 1.2 × molar excess of SKI2_N38 or SKI2_Δwedge38 for 40 min at 4 °C, subsequently applied on a sucrose gradient, and the ribosomal fractions were separated by centrifugation for 3 h at 172,000g and 4 °C in an SW 40 Ti rotor (Beckman Coulter). Polysome profiles were generated by continuous absorbance measurement using an UV absorbance reader (Gilson) and simultaneously fractionated (around 3.39 mm per fraction) with a Piston Gradient Fractionator (Biocomp). For display purposes, the sucrose gradient profiles were superimposed using the highest A260 maximum of the disome peak as a reference.

Western blotting

Fractions obtained by gradient fractionation were precipitated with trichloroacetic acid and pellets were resuspended in SDS loading buffer. All samples were analysed on 15% SDS–PAGE followed by western blotting onto Immobilon-P^SQ transfer membrane (Merck Millipore). Transfer efficiency was monitored by stain-free imaging using the 2,2,2-trichloroethanol in-gel technique. Western blots were probed against monoclonal anti-Flag antibody (Sigma-Aldrich, F3165) in a dilution of 1:5,000 and polyclonal anti-mouse HRP-coupled antibody (Bio-Rad, 172-1011) in a dilution of 1:10,000 in 5% milk, made from 5% w/v milk powder (Roth) in 1× Dulbecco’s PBS buffer (DPBS, Thermo Fisher Scientific) and 0.001% Tween20 (Bio-Rad). The chemiluminescence signal was detected using Amersham ECL prime western blotting detection reagent (Cytiva) on a ImageQuant LAS 4000 (GE Healthcare; exposed for 140 s at standard sensitivity).

Statistics and reproducibility

All in vitro assays (protein co-precipitation (Fig. 1b and Extended Data Figs. 1c,d and 4h); protein purifications and complex reconstitutions (Extended Data Figs. 2a and 4a)), in part or in whole, were successfully reproduced at least twice. The nuclease assays (Extended Data Fig. 1f) were performed as triplicates. For each cryo-EM sample (Extended Data Figs. 2b and 4b), multiple grids were screened before final data collection. The final datasets presented in this article were collected once. The sucrose gradient centrifugations and corresponding western blots (Fig. 4b and Extended Data Fig. 9b) were reproduced twice.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.