Structural basis for stop codon recognition in eukaryotes

Abstract

Termination of protein synthesis occurs when a translating ribosome encounters one of three universally conserved stop codons: UGA, UAA, or UAG. Release factors recognise stop codons in the ribosomal A site to mediate release of the nascent chain and recycling of the ribosome. Bacteria decode stop codons using two separate release factors with differing specificities for the second and third bases¹. By contrast, eukaryotes rely on an evolutionarily unrelated omnipotent release factor (eRF1) to recognise all three stop codons². The molecular basis of eRF1 discrimination for stop codons over sense codons is not known. Here, we present electron cryo-microscopy (cryo-EM) structures at 3.5 – 3.8 Å resolution of mammalian ribosomal complexes containing eRF1 interacting with each of the three stop codons in the A site. Binding of eRF1 flips nucleotide A1825 of 18S rRNA so that it stacks on the second and third stop codon bases. This configuration pulls the fourth position base into the A site, where it is stabilised by stacking against G626 of 18S rRNA. Thus, eRF1 exploits two rRNA nucleotides also used during tRNA selection to drive mRNA compaction. Stop codons are favoured in this compacted mRNA conformation by a hydrogen-bonding network with essential eRF1 residues that constrains the identity of the bases. These results provide a molecular framework for eukaryotic stop codon recognition and have implications for future studies on the mechanisms of canonical and premature translation termination^3,4.

Termination of translation in eukaryotes is initiated when a ternary complex of eRF1-eRF3-GTP binds to a stop codon in the ribosomal A site^5,6. GTP hydrolysis by eRF3 induces a conformational change that leads to its dissociation, permitting eRF1 to accommodate fully in the A site. This change is thought to bring a universally conserved GGQ motif close to the ester bond between the nascent polypeptide and the tRNA, stimulating its hydrolysis. Concomitant with these events, the ATPase ABCE1 is recruited to the ribosome after eRF3 dissociation and, together with eRF1, catalyses splitting of the ribosomal subunits to recycle post-termination complexes^3,4,7,8.

We reasoned that a catalytically inactive eRF1 mutant may trap a pre-hydrolysis termination complex with two key features. First, eRF1 would be in complex with the stop codon it had recognised. Second, the unreleased nascent polypeptide would provide a unique affinity handle to enrich this species for structural analysis. Therefore, we substituted the glycines of the GGQ motif with alanines (eRF1^AAQ)⁹ and added this mutant to in vitro translation reactions in rabbit reticulocyte lysate. Peptide release at all three stop codons was inhibited by eRF1^AAQ as judged by persistence of a peptidyl-tRNA (Extended Data Fig. 1a, b). Affinity purification of these ribosome-nascent chain complexes (RNCs) via the nascent chain recovered both eRF1^AAQ and ABCE1 (Extended Data Fig. 1c, d), suggesting that eRF1^AAQ was trapped on the RNCs in its accommodated state. Association of ABCE1 was enhanced with eRF1^AAQ-stalled RNCs relative to RNCs stalled with a truncated mRNA (Extended Data Fig. 1c), consistent with a report that the function of ABCE1 in post-termination recycling requires peptidyl-tRNA hydrolysis⁷.

Purified RNCs stalled with eRF1^AAQ at each stop codon (Extended Data Fig. 1d) were directly utilised for cryo-EM. Multiple rounds of three-dimensional classification in silico revealed that ~10% of the particles contained eRF1^AAQ-ABCE1 (Extended Data Fig. 2). Datasets of between 20,000 and 50,000 particles for the three stop codons yielded maps with overall resolutions of 3.45 Å (UAG), 3.65 Å (UAA) and 3.83 Å (UGA), against which the models were refined (Fig. 1, Extended Data Fig. 3, and Extended Data Table 1).

Figure 1. Overall structure of a eukaryotic translation termination complex.

a, Overview of the structure of an eRF1^AAQ-stalled mammalian ribosome-nascent chain complex containing the UAG stop codon showing the 40S and 60S ribosomal subunits, E- (yellow) and P-site (green) tRNAs, eRF1^AAQ (purple) occupying the A site, and ABCE1 (blue) occupying the GTPase centre. b, Close-up view of eRF1^AAQ coloured by domain (N, M, C) with the GGQ, NIKS and YxCxxxF motifs highlighted (pink). Also shown are the P-site tRNA (green), nascent polypeptide (teal), the mRNA containing the UAG stop codon (slate), and ABCE1 (blue) with its iron-sulfur clusters (orange/yellow) and nucleotide binding sites (gray) indicated.

In each reconstruction, eRF1 is in its extended conformation¹⁰, and ABCE1 occupies the GTPase centre (Fig. 1a). The three domains of eRF1 (N, M and C) have moved relative to one another compared to the crystal structure¹¹ (Extended Data Fig. 4a) and are each well resolved (Extended Data Fig. 3b). Direct interactions of the N domain with the codon deep in the decoding centre argues against an earlier suggestion that eRF1 disengages from the stop codon in the presence of ABCE1¹⁰.

The N and M domains of eRF1 independently contact the P-site tRNA and together structurally resemble a tRNA in the A site (Fig. 1b). Helix α2 of the N domain runs parallel to, and interacts with, the anticodon stem-loop of the P-site tRNA (Extended Data Fig. 4b). The M domain is functionally analogous to the tRNA acceptor stem¹¹, and positions the GGQ motif⁹ in the peptidyl transferase centre (Fig. 1b). To occupy a similar position as the 3’ end of an A-site tRNA, the GGQ-loop is shifted by 10 Å compared to the crystal structure¹¹ (Extended Data Fig. 4c). This positions the alanine residues of the mutated GGQ motif directly opposite the terminal 3’ adenosine of the P-site tRNA and the glutamine (Gln185) in proximity of the ester bond between the nascent polypeptide and the tRNA. This conformation closely resembles that of the GGQ-loop in bacterial release factors^12-14 (Fig. 2a).

Figure 2. Conformation of essential eRF1 motifs.

a, Conformation of the GGQ-loop (teal) of eRF1^AAQ (purple) within the peptidyl transferase centre. The AAQ tripeptide, positioned next to the CCA end of the P-site tRNA (green), closely resembles the conformation adopted by GGQ of a bacterial release factor (RF1, grey) bound to the ribosome. b, Positions of NIKS, YxCxxxF, and GTS (teal) motifs in the N domain of eRF1^AAQ (purple) relative to the mRNA (slate). The positions of the stop codon (+1 to +3) and the following base (+4) are indicated.

At the decoding centre, the interactions between the N domain of eRF1 and the stop codon are well resolved (Fig. 2b and Extended Data Fig. 5). The most striking feature is a compact configuration of mRNA that accommodates four nucleotides in the A site instead of three (Fig. 3a, b). mRNA compaction depends crucially on the ‘flipping out’ of A1825 (A1493 in bacteria) in helix 44 (h44) of 18S rRNA, forming a stacking interaction with the +2 nucleotide of the stop codon, on which the +3 nucleotide stacks. This configuration allows the +4 nucleotide to stack with base G626 (G530 in bacteria) of 18S rRNA (Fig. 3b). Stacking with G626 would be more stable for purines, explaining their statistical bias at the +4 position in eukaryotes¹⁵. Compaction of mRNA in the A site probably results in pulling downstream mRNA further into the mRNA channel. This is consistent with protection of two additional nucleotides of 3’ mRNA upon eRF1 binding to the ribosome^16,17.

Figure 3. Stop codon configuration in the eukaryotic decoding centre.

a, EM map densities of the mRNA in the termination complexes containing the UAA, UAG and UGA stop codons reveal that they adopt the same compacted conformation. The Val^GUU codon in the P site and the stop codon (+1 to +3) and following (+4) bases in the A site (purple) are indicated. b, The core termination signal recognised by eRF1^AAQ (purple) is formed by four mRNA bases (+1 to +4, slate) that occupy the A site. Bases +2 and +3 stack on A1825, which is flipped out of helix 44 (h44), and base +4 on G626 of 18S rRNA (yellow). c, In bacteria, RF1 (grey) recognises a more extended stop codon configuration where the +3 base stacks on G530 (the equivalent of G626) of 16S rRNA.

The structures of the bacterial release factor complexes are qualitatively distinct (Fig. 3c). Instead of A1493, the adjacent A1492 is flipped out^12-14. However, this flipped base does not stack with any of the stop codon bases and does not lead to mRNA compaction. G530 therefore stacks with the +3 base of the stop codon^12-14 instead of the +4 base. While the +4 base also impacts stop codon recognition in bacteria, the greatest preference appears to be for uridine rather than for purines¹⁸. Thus, eukaryotes appear to exploit the +4 base to stabilise mRNA compaction. The unique stacking of the +2 and +3 bases in this compacted state is an important element for stop codon recognition by eRF1 (see below).

The conserved NIKS, YxCxxxF, and GTS motifs in eRF1 have been crosslinked to stop codon bases^19,20. In the structures here, the essential NIKS sequence (residues 61-64)²¹, located at the end of helix α2, imposes a requirement for uridine in the first (+1) position via interactions with its Watson-Crick edge. A local distortion in helix α2 and the subsequent loop helps to optimise hydrogen bond interactions with the nucleotide (Fig. 4a). Notably, the side chains of Asn61 and Lys63 form hydrogen bonds with the uracil carbonyl groups. A further hydrogen bond forms with the main chain carbonyl of Asn61. Hydroxylation of Lys63 (at the Cδ position) reduces stop codon read-through and promotes peptide release in vivo²². The hydroxyl group may allow an additional interaction with the phosphate backbone of the mRNA and help position the ε-amino group for optimal hydrogen-bonding.

Figure 4. Molecular basis of stop codon recognition by eRF1.

a, The conformation of the NIKS motif (purple) at the end of helix α2 of ribosome-bound eRF1^AAQ compared to the eRF1 crystal structure (PDB ID: 1DT9, light grey) allows it to form hydrogen bonds with the uridine in the +1 position of stop codons (slate; see also Extended Data Fig. 6). The hydroxyl groups of Ser64 and Cδ–hydroxylated (*) Lys63 help to position the NIKS motif by interacting with the mRNA phosphate backbone. b, Detailed interactions between the UAG stop codon (slate), eRF1^AAQ (purple), and A1825 (yellow), depicting (i) stacking of the +2 and +3 bases with A1825, (ii) stabilisation of the flipped out position of A1825 by hydrogen bonds with the main chain of Cys127, (iii) a hydrogen bonding network involving Thr32, Glu55 and Tyr125 that specifies stop codon selectivity, and (iv) coordination of a Mg²⁺ ion (green) by the +2 adenosine. c, Model for stop codon discrimination by eRF1. Stacking of the +2 and +3 bases (slate) and possible hydrogen bonding interactions with Glu55 of eRF1 (purple) are shown for the three stop codons (UAA, UAG, UGA) and the sense codon UGG, which codes for tryptophan. d, The UGA stop codon induces a conformational change in the YxCxxxF and GTS motifs of eRF1^AAQ (purple) compared to UAG- (white) and UAA-bound eRF1^AAQ and eRF1 crystal structures.

Purines would be disfavoured at the +1 position by steric hindrance with eRF1 while cytidine would be incompatible with the hydrogen bonding requirements of the NIKS motif (Extended Data Fig. 6). Thus, the universality of uridine in the first position of stop codons is defined by extensive hydrogen bonding with eRF1. In bacteria, RF1 and RF2 also utilise hydrogen-bonding networks to specify uridine in the +1 position, but with different interactions¹², highlighting how independent solutions have evolved for the same specificity problem.

Interactions of the YxCxxxF motif and Glu55 of eRF1 with the +2 and +3 purines provide a basis for stop codon discrimination from sense codons (Fig. 4b, c). The main chain of Cys127 from the YxCxxxF motif forms two hydrogen bonds with the Watson-Crick edge of A1825 to stabilise its flipped orientation, facilitating stacking of the +2 and +3 bases (Fig. 4b). The invariant residues Glu55 from helix α2 and Tyr125 act as a pair²³ to position the glutamate side chain so that it can hydrogen bond with the N6 atoms of adenosines at the +2 and/or +3 positions. These interactions are only possible with purines, excluding pyrimidines from the +2 and +3 positions in stop codons. In the case of UGG, which codes for tryptophan, the two O6 atoms of consecutive stacked guanosines would not satisfy the hydrogen-bonding requirements, and their increased repulsion with each other and with the negatively-charged Glu55 would also disfavor the formation of the conformation that allows these interactions (Fig. 4c).

The conserved GTS motif, located towards the sugar edge of the +3 base, adopts two different conformations that are interdependent with the position of the YxCxxxF motif²⁴. With adenosine in the +2 position, Thr32 faces the +3 base and can hydrogen bond with the N2 atom of the guanosine in UAG (Fig. 4b). A guanosine at the +2 position is accommodated by a movement of the YxCxxxF motif towards the stacked pair by ~1 Å, which propagates a 4 Å movement of the GTS motif so that Thr32 now faces away from the stop codon (Fig. 4d). Consistent with these observations, perturbation of any residue contributing to this network has substantial effects on stop codon recognition and specificity^17,23,25,26.

In conclusion, these structures show how stop codons are specifically selected by eRF1. At the +1 position, only uridine can form the network of interactions with the NIKS motif. The flipping of A1825 results in its stacking onto the +2 and +3 bases of a distorted mRNA so that they are decoded as a single unit (Fig. 4b). This solves the puzzle of how guanosine can occur at either the +2 or +3 position but not at both: two successive guanosines would lead to repulsion between their O6 atoms and with Glu55 (Fig. 4c). By requiring two purines while specifically excluding consecutive guanosines, the selectivity of eRF1 at the +2 and +3 positions is logically equivalent to a NAND gate.

This mechanism for stop codon recognition by eRF1 is distinct from that used by the evolutionarily and structurally unrelated bacterial release factors, even though they share some common strategies including the extensive use of hydrogen bonding for decoding specificity and an invariant GGQ motif to catalyse peptide hydrolysis. The work here paves the way for studies on the mechanism of termination and ribosome recycling, including the roles of eRF3 and ABCE1, as well as the interaction of termination factors with quality control pathways such as nonsense-mediated mRNA decay²⁷. Finally, insight into stop codon recognition can provide a framework for the development of inhibitors of termination that may be useful to treat the ~11% of hereditary diseases caused by premature termination^28,29.

Methods

Plasmids and antibodies

An SP64-based plasmid encoding 3× Flag-tagged Sec61β containing the autonomously-folding villin headpiece (VHP) domain³⁰ was modified by individually inserting each stop codon (TAA, TAG, TGA) after the valine at position 68 of unmodified Sec61β. The remaining C-terminal portion of the protein was deleted using Phusion mutagenesis (Thermo Scientific). In vitro transcription reactions were performed using PCR products generated with primers flanking the SP6 promoter and the 3’UTR of the SP64 vector as previously described³¹. The cDNA encoding human eRF1 (Origene) was subcloned into a pRSETA expression vector after an N-terminal 6× His tag and TEV cleavage site using standard procedures. The eRF1^AAQ mutant was generated via Phusion mutagenesis. Antibodies against Hbs1 and ABCE1 have been described³²; the eRF1 antibody was from New England Biolabs.

Release factor purification

Wild-type and mutant eRF1 (eRF1^AAQ) were expressed in and purified from Escherichia coli BL21(DE3) cells grown under antibiotic selection in LB. Transformed cells were induced at A₆₀₀ = 0.4-0.6 with 0.2 mM IPTG for 2 hr at 37°C and lysed with a microfluidiser in lysis buffer (1× PBS, pH 7.5, 250 mM NaCl, 10 mM imidazole, 1 mM DTT) containing 1× protease inhibitor cocktail (Roche). Lysates were clarified by centrifugation and the supernatant passed over a NiNTA column. After washing with 25 column volumes of lysis buffer, elutions were carried out with 250 mM imidazole in lysis buffer. Peak fractions were pooled, dialysed overnight in the presence of TEV protease against 50 mM Hepes, pH 7.4, 150 mM KAc, 5 mM MgAc₂, 10 mM imidazole, 10% glycerol, 1 mM DTT. The TEV protease and cleaved His tag were removed by passage over a NiNTA column.

In vitro translations and sample preparation

In vitro translations in a rabbit reticulocyte (RRL) system were for 25 min at 32°C as before^30,31. Where indicated, 0.5 μM eRF1^AAQ was included to trap termination complexes. 4 mL translation reactions were directly incubated with 100 μL (packed volume) of anti-Flag M2 beads (Sigma) for 1-1.5 hr at 4°C with gentle mixing. The beads were washed sequentially with 6 mL 50 mM Hepes, pH 7.4, 100 mM KAc, 5 mM MgAc₂, 0.1% Triton X-100, 1 mM DTT, 6 mL 50 mM Hepes, pH 7.4, 250 mM KAc, 5 mM MgAc₂, 0.5% Triton X-100, 1 mM DTT and 6 mL RNC buffer (50 mM Hepes, pH 7.4, 100 mM KAc, 5 mM MgAc₂, 1 mM DTT). Two sequential elutions were carried out with 100 μL 0.1 mg/mL 3× Flag peptide (Sigma) in RNC buffer at room temperature for 25 min. The elutions were combined and centrifuged at 100,000 rpm at 4°C for 40 min in a TLA120.2 rotor (Beckman Coulter) before resuspension of the ribosomal pellet in RNC buffer.

Electron microscopy

3 μL aliquots of purified ribosome complexes at a concentration of 120 nM were applied onto Quantifoil R2/2 cryo-EM grids covered with continuous carbon (estimated to be 50 Å thick) at 4°C and 100% ambient humidity. After 30 s incubation, the grids were blotted for 3 s and vitrified in liquid ethane using a Vitrobot MKIII (FEI).

Automated data collection (EPU software, FEI) was conducted on a Titan Krios microscope equipped with a XFEG electron source using 300 kV acceleration voltage. For each 1.1 s exposure, 17 movie frames were recorded on a Falcon II direct electron detector (FEI) at a calibrated magnification of 104,478, resulting in a pixel size of 1.34 ų³. A dose rate of ~30 electrons per Ų per second was used. Defocus values ranged from −1.1 to −5.9 μm for the UAA-eRF1^AAQ dataset, −0.7 to −4.1 μm for the UAG-eRF1^AAQ dataset, and −0.7 to −3.8 μm for the UGA-eRF1^AAQ dataset (Extended Data Table 1), as more images were collected closer to focus on the latter two datasets.

Image Processing

The 17 frames were aligned using whole-image motion correction³⁴ to reduce beam-induced blurring of the images. Parameters of the contrast transfer function for each motion-corrected micrograph were obtained using Gctf (K. Zhang, MRC-LMB, in development). Ribosome particles were selected using semi-automated particle picking implemented in RELION³⁵. All two- and three-dimensional classifications and refinements were performed using RELION³⁶. We used reference-free two-dimensional class averaging to discard non-ribosomal particles and three-dimensional classification to sort different compositions and conformations of the ribosome complexes. After two-dimensional classification, the UAA-eRF1^AAQ dataset contained 556,994 particles from two independent data acquisitions, while the UAG-eRF1^AAQ and UGA-eRF1^AAQ datasets contained 216,276 and 250,705 particles, respectively, both from a single data acquisition.

A 40 Å low-pass filtered cryo-EM reconstruction of the rabbit 80S ribosome (Electron Microscopy Data Bank (EMDB) accession 2704)³² was used as an initial model for the three-dimensional refinement. In a subsequent three-dimensional classification run with 12 classes, an angular sampling of 1.8° was combined with local angular searches, and the refined model from the first refinement was used as a starting model. A further round of refinement and three-dimensional classification into 6 classes using an angular sampling of 0.9° combined with local angular searches was necessary to isolate a homogeneous class of ribosomes bound with eRF1-ABCE1 (see Extended Data Fig. 2). This resulted in 49,979 particles for the UAA-eRF1^AAQ dataset, 20,515 particles for the UAG-eRF1^AAQ dataset, and 22,058 particles for the UGA-eRF1^AAQ dataset.

Prior to a final round of refinement and classification with classes combined from different datasets, statistical particle-based movie correction was performed in RELION-1.4. For these calculations we used running averages of five movie frames, and a standard deviation of one pixel for the translational alignment. In addition, we used a resolution and dose-dependent model for the radiation damage, in which each frame is B-factor weighted as estimated from single-frame reconstructions³⁷.

Reported resolutions are based on the Fourier shell correlation (FSC) 0.143 criterion. High-resolution noise substitution was used to correct for the effects of a soft mask on FSC curves³⁸. Before visualization, density maps were corrected for the modulation transfer function of the Falcon II detector and then sharpened by applying a negative B-factor that was estimated using automated procedures³⁹ (Extended Data Table 1). Local resolution was quantified using ResMap⁴⁰.

Model building

The reconstruction was initially interpreted by docking the large and small subunits of the mammalian 80S ribosome, with respective Protein Data Bank (PDB) accession codes 3J92 and 3J7P^41,42, into the map using Chimera⁴³. The crystal structures of human eRF1 (PDB accession code 1DT9)¹¹ and Pyrococcus abysii ABCE1 (PDB accession code 3BK7)⁴⁴ were docked into the A site and GTPase center, respectively, before being subjected to a Jiggle Fit in Coot⁴⁵. To model the bound tRNAs, bacterial tRNA (PDB accession code 4V51)⁴⁶ was used as a template and modified to the sequence of the most prevalent tRNA for the particular codon from the genomic tRNA database⁴⁷. The atomic models were then modified in Coot v0.8 to agree with the rabbit sequences and optimised for fit to density.

Model refinement and validation

Reciprocal space refinement was carried out in REFMAC v5.8 optimised for EM maps utilizing external restraints generated by ProSMART and LIBG⁴⁵. The model was refined against amplitudes and phases from the experimental map that were unchanged during the course of refinement. FSC_average was monitored during refinement to follow the fit-to-density, and the final model was validated using MolProbity⁴⁸ (Extended Data Table 1). The Ramachandran statistics for eRF1 are 93.0% favoured, 1.5% outliers and for ABCE1, 91.1% favoured, 1.9 % outliers. Cross-validation against over-fitting was calculated as previously described^45,49 (Extended Data Figure 3c).

Extended Data

Extended Data Figure 1. eRF1^AAQ stalls ribosomes at stop codons.

a, Line diagrams of mRNA encoding nascent chain (NC) substrates used in this study. The cytosolic portion of human Sec61β (residues 1-68, orange) was modified to contain an N-terminal 3× Flag tag (green) for affinity purification and the autonomously-folding villin headpiece (VHP, blue) domain. The three stop codons were individually inserted after Val68 of Sec61β to generate substrates for eRF1^AAQ-mediated stalling, or the mRNA was truncated after the same residue to generate an independently-stalling substrate. b, In vitro translation reactions of NC-stop substrates containing the indicated stop codon (see panel a) in the presence of ³⁵S-methionine without or with excess eRF1^WT or eRF1^AAQ. Reactions were for 25 min at 32°C and directly analyzed by SDS-PAGE and auto-radiography. The terminated (NC) and tRNA-associated (NC-tRNA) nascent chain products are indicated. Addition of eRF1^AAQ selectively prevents peptide hydrolysis when the stop codon is reached. c, Anti-Flag affinity purifications of ribosome-nascent chains (RNCs) stalled either by mRNA truncation or at the UAA stop codon with eRF1^AAQ (see panel a) were immunoblotted for the splitting factors Hbs1 and ABCE1. The different amounts of Hbs1 and ABCE1 co-purified despite identical nascent chain sequences in each RNC complex suggest that eRF1^AAQ selectively traps ABCE1 on pre-termination complexes. d, SDS-PAGE and Coomassie staining of affinity-purified eRF1^AAQ-stalled ribosome-nascent chains containing the UGA stop codon utilised for cryo-EM analysis. Bands corresponding to ribosomal proteins, ABCE1, and eRF1^AAQ, which were verified by immunoblotting and mass spectrometry (data not shown), are indicated.

Extended Data Figure 2. In silico 3D classification scheme for cryo-EM datasets.

Particles extracted from automated particle picking in Relion were subjected to 2D classification. Non-ribosomal particles were discarded and the remaining particles were combined for a 3D refinement. The resulting map was used as a reference for 3D classification, which typically isolated 5 distinct classes of ribosomal complexes with the indicated distributions. Classes containing 80S ribosomes with canonical P- and E-site tRNAs and weak factor density in the A site (~40%) were combined and subjected to another round of 3D classification for A site occupancy. Approximately 1/3 of this population contained strong density for eRF1^AAQ and ABCE1. These particles were combined for subsequent 3D refinement and movie processing. All four datasets (two for the UAA stop codon and one each for the UAG and UGA stop codons) were processed similarly. The eRF1^AAQ-ABCE1-containing particles of the two UAA datasets after the two rounds of classification were combined for refinement to yield the final map.

Extended Data Figure 3. Quality of maps and models.

a, Fourier shell correlation (FSC) curves for the EM maps of each termination complex containing the indicated stop codon. b, Isolated eRF1^AAQ-ABCE1 density from the UAA termination complex map coloured by local resolution. c, Fit of models to maps. FSC curves calculated between the refined model and the final map (black), and with the self- (blue) and cross-validated (magenta) correlations for each stop codon complex. The EM map of each termination complex coloured by local resolution (as in panel b) is displayed next to the corresponding curves.

Extended Data Figure 4. eRF1^AAQ interactions within the termination complex.

a, Comparison of ribosome-bound eRF1^AAQ (coloured by domain) with the crystal structure of eRF1 (PDB ID: 1DT9, grey) superposed on the C domain. Both the N and M domains of eRF1 rotate upon stop codon recognition on the ribosome. The P-site tRNA (green) and nascent chain (teal) are shown for orientation. b, Interaction of helix α2 of the N domain of eRF1^AAQ (purple) with the anticodon stem loop (ASL) of the P-site tRNA (green). c, Superposition of the eRF1^AAQ M domain (purple) with the eRF1 crystal structure (PDB ID: 1DT9) showing a 10 Å movement of the GGQ-loop to accommodate within the peptidyl transferase centre.

Extended Data Figure 5. Examples of map densities.

a, Density (from the UAG-containing termination complex) for the nascent chain (teal) attached to the CCA end of the P-site tRNA (green) is of sufficient resolution to model the defined sequence of the C-terminal end of the programmed nascent chain. This provides additional verification that the termination complexes are stalled at Val68 of Sec61β (human numbering) with the stop codon in the A site (see also Extended Data Fig. 1a). A stacking interaction between an aromatic residue of the nascent chain and U4555 (blue) lining the ribosomal exit tunnel can also be observed. b, Densities for the interactions between the UAG stop codon (grey), a portion of h44 of 18S rRNA (yellow) and the YxCxxxF and NIKS motifs of eRF1^AAQ (purple). The invariant isoleucine of the NIKS motif provides a hydrophobic base for the stacking of the +2 and +3 bases of the stop codon with A1825. Unlike the tyrosine and cysteine residues of the YxCxxxF motif, the phenylalanine does not contribute to stop codon recognition, but to the hydrophobic packing of the eRF1 N domain.

Extended Data Figure 6. Hydrogen bonds specify for uridine at the +1 position.

Chemical diagrams of uridine and cytidine with hydrogen bond donors (blue) and acceptors (magenta) indicated. Two of the three hydrogen bonds that uridine forms with Asn61 and Lys63 of the NIKS motif of eRF1^AAQ (purple) are not possible with cytidine (see also Fig. 4a).

Extended Data Table 1.

Refinement and model statistics.

	UAA	UAG	UGA
Data Collection
Particles	49,979	20,515	22,058
Pixel size (Å)	1.34	1.34	1.34
Defocus range (μm)	1.1-5.9	0.7-4.1	0.7-3.8
Defocus mean (μm)	3.2	2.4	2.3
Voltage (kV)	300	300	300
Electron dose (e- Å−2)	30	30	30
Model composition
Non-hydrogen atoms	226,532	226,533	226,533
Protein residues	12,676	12,676	12,676
RNA bases	5,820	5,820	5,820
Ligands (Zn2+/Mg2+/ADP)	8/197/2	8/197/2	8/197/2
Refinement
Resolution (Å)	3.65	3.45	3.83
Map sharpening B-factor (Å2)	−81.7	−50.6	−82.7
Average B factor (Å2)	105.8	87.4	93.5
FSCaverage	0.85	0.84	0.88
FSCaverage (eRF1)	0.70	0.64	0.74
FSCaverage (ABCE1)	0.71	0.62	0.75
R.m.s. deviations
Bond lengths (Å)	0.006	0.006	0.008
Bond angles (°)	1.19	1.22	1.40
Validation
Molprobity score	2.7 (93rd percentile)	2.8 (89th percentile)	3.0 (89th percentile)
Clashscore, all atoms	5.2 (100th percentile)	6.2 (97th percentile)	8.2 (97th percentile)
Good rotamers (%)	88.2	87.5	86.4
Ramachandran plot
Favored (%)	87.0	85.8	83.5
Outliers (%)	3.3	3.5	4.2
Validation (RNA)
Correct sugar puckers (%)	96.5	93.4	96.0
Good backbone conformations (%)	68.2	66.8	65.8