Structure of the mammalian ribosomal pre-termination complex associated with eRF1•eRF3•GDPNP

Eukaryotic translation termination results from the complex functional interplay between two release factors, eRF1 and eRF3, in which GTP hydrolysis by eRF3 couples codon recognition with peptidyl-tRNA hydrolysis by eRF1. Here, we present a cryo-electron microscopy structure of pre-termination complexes associated with eRF1•eRF3•GDPNP at 9.7 -Å resolution, which corresponds to the initial pre-GTP hydrolysis stage of factor attachment and stop codon recognition. It reveals the ribosomal positions of eRFs and provides insights into the mechanisms of stop codon recognition and triggering of eRF3’s GTPase activity.


INTRODUCTION
Termination occurs when a stop codon enters the ribosomal A site and consists of stop codon recognition followed by peptide release, which involves the nucleophilic attack of a water molecule on the P-site peptidyl-tRNA in the ribosomal peptidyl transferase center (PTC). Eukaryotic termination is mediated by two directly interacting release factors: eRF1, which is responsible for stop codon recognition and triggering peptide release, and eRF3, a GTPase that strongly stimulates peptide release by eRF1 in a GTPdependent manner (1)(2)(3). eRF1, in turn, stabilizes binding of GTP to eRF3 so that they form a stable ternary complex (4,5), and is required for eRF3's ribosomedependent GTPase activity (6). eRF1 has omnipotent decoding capacity and recognizes all three stop codons.
It comprises N-terminal (N), middle (M) and C-terminal (C) domains (7). The rigid core of domain C contains a flexible insertion forming a mini-domain (8). Domain N is involved in stop codon recognition. Although the mechanism by which eRF1 responds to all three stop codons is not clear, extensive mutational and genetic analyses identified the essential role in this process of GTS [31][32][33] (human numbering), TASNIKS 58-64 and YxCxxxF 125-131 motifs located at the apex of the Ndomain (7,(9)(10)(11)(12)(13)(14)(15)(16)(17)(18). eRF1's domain M contains the universal GGQ loop, whose placement into the PTC causes rearrangement of rRNA, allowing a water molecule to enter and induce peptide release (1,7,(19)(20)(21)(22). eRF3 consists of the essential C-terminal region comprising GTP-binding (G) domain and b-barrel domains 2 and 3, which are homologous to elongation factors EF-Tu and eEF1A (23), and a non-conserved Nterminal region that is not essential for termination (24). eRF1's C and M domains interact with eRF3 (15,25). eRF1 and eRF3 bind to the pre-termination complexes (pre-TCs) as an eRF1eRF3GTP ternary complex, but peptide release does not occur until eRF3 hydrolyzes GTP. GTP hydrolysis releases eRF1's domain M from eRF3, which enables its GGQ loop to enter the PTC and trigger peptide release.
Although the structures of individual eRFs have been determined (7,15,23), the structural basis for key steps in termination, such as stop codon recognition or triggering of eRF3's GTPase activity, remains unresolved. Our previous cryo-EM structure of the eRF1eRF3GDPNPbound pre-TC had a resolution of 18 Å (26), which did not allow accurate modeling of the ribosome-bound factors and therefore could not address such questions. The advent of a more powerful classification algorithm, RELION (27), prompted us to re-examine the original data set (26), resulting in a much higher resolution (9.7 Å ), which enabled us to propose a more detailed model of the interactions pivotal for peptide release.

Data processing
Mammalian pre-TCs were assembled in vitro on a derivative of b-globin messenger RNA (mRNA) encoding MVHL tetrapeptide followed by a UAA stop codon, using purified rabbit 40S and 60S ribosomal subunits, initiation (2,3,1,1A,4A,4B,4G,5 and 5B) and elongation (1A and 2) factors and aminoacylated tRNAs (26). eRF1eRF3GDPNP-bound pre-TCs were formed with full-length eRF1 and eRF3 lacking the non-essential N-terminal 138aa (26). A cryo-EM data set of the eRF1eRF3GDPNP-bound pre-TCs comprising 195 432 particles (26) was aligned and refined using the RELION autorefine procedure (27). It resulted in a 9.1-Å structure [gold standard Fourier shell correlation (FSC) = 0.143] showing fragmented densities in the intersubunit space, indicative of heterogeneity. The data set was then subjected to RELION classification (28) starting with 10 classes (k = 10). Of the 10 classes, class 8 ( Figure 1) was well populated (48 973 particles, 25.1%), with well-defined additional masses of density in the intersubunit space corresponding to eRF1, eRF3 and Psite tRNA as previously described (26). The particles from this class were isolated, and the structure was first refined with RELION autorefine procedure. This yielded a structure at 9.7 Å (gold standard FSC = 0.143), but the density exhibited strong stretching artifacts when filtered at the measured resolution, probably due to preferred orientations of the particles on the cryo-EM grid. The alignment parameters were further refined using RELION classification with k = 1 and a T factor of four for three iterations at an angular spacing of 1.8 , followed by three additional iterations at 0.9 . The resulting reconstruction displayed the same resolution of 9.7 Å (FSC = 0.143), but the quality of the structure was very much improved and the stretching artifacts were no longer visible. The statistics of alignment precision estimated by RELION were also significantly improved [improvement in terms of overall accuracy of rotations, 41%; overall accuracy of translations, 43% and average Pmax at 1.8 angular step, 180% (Supplementary Table S1)].
Modeling of eRF1, eRF3, mRNA and P-site tRNA Leu The cryo-EM map segmentation was performed using the Segger (29) plug-in in UCSF Chimera (30). Models of human eRF1 and eRF3 were built based on several experimental structures. Atomic coordinates of eRF1 and eRF3 domains 2 and 3 were taken from their crystal structure (15) (PDBID: 3E1Y), and the C-terminal mini-domain structure of eRF1 (residues 326-373) was modeled based on its nuclear magnetic resonance (NMR) structure (8) (PDBID: 2KTU). Most of eRF3 G-domain (residues 317-439, 207-216) was modeled by homology based on the crystal structure of Saccharomyces pombe eRF3GDP (23) (PDBID: 1R5B) because of its significantly higher resolution over the S. pombe eRF3GMPPNP crystal structure (23) (2.35 versus 3.20 Å , respectively). Only the GTP-binding pocket of the G-domain (residues 217-231) was modeled based on the crystal structure of eRF3GMPPNP (23) (PDBID: 1R5O). Missing residues and the switch region (residues 232-316) in eRF3 Gdomain were modeled by homology based on the archeal elongation factor 1a (aEF1a) taken from the eRF1GTPbound aEF1a crystal structure (31) (PDBID: 3VMF). The arrangement between eRF3's domains II and III and Gdomain was derived from the latter crystal structure as well. The P-site tRNA Leu was modeled from its crystal structure in complex with leucyl-tRNA synthetase from Pyrococcus horikoshii (32) (PDBID: 1WZ2). The mRNA was modeled based on the mRNA of the crystal structure of the Thermus thermophilus 70S ribosome bound with the Q253P mutant form of release factor 2 (22) (PDBID: 4KFK). All homology modeling was done using SWISS-MODEL (33,34).

Molecular dynamics flexible fitting
The derived model was first rigid body-fitted into its corresponding segmented cryo-EM density map using UCSF Chimera (30). Then, using the molecular graphics software VMD (35), the system was prepared for molecular dynamics flexible fitting (MDFF) (36) in explicit solvent, a procedure which applies the cryo-EM map as an additional potential to the system-in this study the eRF1eRF3tRNAmRNA cryo-EM segmented mapthus comprising only the molecules to be simulated: eRF1, eRF3, mRNA and P-site tRNA Leu . The MDFF system was embedded in a box of TIP3P water molecules with an extra 12-Å padding in each direction. The system was neutralized by potassium ions, and an excess of 0.2 M KCl was added. The simulated system was prepared using CHARMM force field parameters [Combined CHARMM All-hydrogen topology file for CHARMM22 proteins and CHARMM27 lipids (37)]. The system was energy-minimized in 600 steps in the molecular dynamics simulation package NAMD (38). After minimization, the fitting trajectories were run for 400 ps, once the root-mean-square deviation and crosscorrelation coefficient had stabilized.

RESULTS AND DISCUSSION
Structure determination of the mammalian eRF1eRF3GDPNP-bound pre-TC Classification of the cryo-EM data set comprising 195 432 particles of the mammalian eRF1eRF3GDPNP-bound pre-TC formed on mRNA encoding MVHL tetrapeptide, followed by a UAA stop codon, yielded 10 classes with different factor and tRNA occupancies ( Figure 1). One of these classes (25% of the particles) had P-site tRNA along with well-defined eRF1 and eRF3 densities. The characteristic long variable loop of tRNA Leu was apparent in the P-site tRNA, indicating that it was correctly programmed with an A-site stop codon. The eRF1eRF3GDPNPbound pre-TC class yielded a 9.7-Å reconstruction (Figure 2a), which allowed precise docking of each factor's domains. eRF1 and eRF3 were modeled (Figures 2b-d) based on the crystal structures of eRF1, eRF3, eRF1-eRF3 2-3 and aRF1-aEF1a (7,15,23,31) and the NMR structure of eRF1's C-domain (8). The model was refined using MDFF (36) with secondary structure constraints.

Interaction between eRF1-N and the stop codon
eRF1's N-domain reaches deep into the decoding center ( Figure 2), establishing multiple contacts with the 40S subunit, including h18, h30, h31, h34 and h44 of 18S rRNA and ribosomal proteins rpS30e and rpS31e. R65/ R68 in the long a3 helix are in proximity to nucleotides 1330-1331 of h34, consistent with these residues' strong influence on eRF1's ribosomal binding (11). In our structure, mRNA occupies its normal position in the decoding center, and eRF1's N-domain interacts directly with the stop codon (Figure 3), primarily via the TASNIKS 58-64 motif, consistent with ultraviolet cross-linking of the stop codon to this element (10). The positions of the mRNA bases and amino acid side-chains cannot be determined precisely at the present resolution. However, the observed proximity of particular residues and bases in the fitted structures makes it likely that they interact with one another. The TASN residues appear to be close to the first position of the stop codon, I to the second and KS to the third. The proximity of K to the nucleotide immediately downstream of the stop codon could potentially contribute to the mechanism by which this base influences termination efficiency (43,44). In addition to TASNIKS, three residues from GTS [31][32][33] and YxCxxxF 125-131 motifs are also positioned within interacting distance of the stop codon: G31 and T32 are close to the third and second base, respectively, and C127 to the bases in positions 1 and 2. The positions of stop codon nucleotides relative to these three residues are consistent with the experimentally determined influence of T32 and C127 on the specificity of recognition of the nucleotide in the second position of the stop codon (14,15,17). Y125 and F131 are farther away, and their influence on termination (9,12,15,17) may therefore reflect roles in stabilizing the domain structure. Additional contacts with the pre-TC may be important for proper placement of the N-domain into the decoding site. Thus, the b-sheet supporting the GTS and YCF motifs interacts with H69 of 28S rRNA, whereas the continuous density between helix a2 and the P-site tRNA (which is particularly strong between S46 at the tip of a2 and nucleotide 30 of the tRNA Leu anticodon stem-loop) indicates likely contact between eRF1 and P-site tRNA. Functional interaction between them has been proposed, and preferential binding to specific tRNAs could contribute to the observed bias for certain codons preceding stop codons (45,46). Interestingly, the characteristic long variable loop of the P-site tRNA Leu is also in proximity to the highly conserved P-site loop of rpL11 on the 60S subunit and to h30 and rpS18 of the 40S subunit (Supplementary Figure S2). The P-site loop of rpL11 (L5 in bacteria) contacts the T-loop of tRNA in the P/P, as well as in neighboring intermediate states, and was therefore proposed to escort tRNA through the P site during translocation (47)(48)(49)(50). Notably, residues in the P-site loop, which are seen in proximity to the variable loop of tRNA Leu in the present reconstruction, were shown to play a key role in binding of the P-site tRNA (51). The potential strengthening of the contact with the P-site loop of L11 and additional specific contacts with h30 and rpS18 may contribute to the known higher affinity of tRNA Leu to the P/P state, accounting for the increased stability of ribosomal complexes containing P-site tRNA Leu after peptide release (52). Additional stabilization of eRF1's interaction with the 40S subunit appears to be provided by its mini-domain, which protrudes toward the beak where it interacts via its flexible loop with h33/h34 of 18S rRNA, or possibly with the N-terminal tail of rpS31e (Figure 4).

Interaction between eRF1 and eRF3
The rigid core of eRF1's C-domain interacts with eRF3's domain 3 as observed in the eRF1-eRF3 2-3 crystal structure (15), and with H44 of the P stalk (Figures 2 and 4). rpL12 is in proximity to eRF1's C-domain and domain 3 of eRF3 and may contact them directly, but its scattered density suggests that they do not interact stably in this configuration. The inward rotation of the stalk base (26) is similar to that observed in Dom34/Hbs1-associated ribosomes (40).
The topology of eRF1's M domain is now clearly visible, folded back onto eRF3, with the GGQ loop close to the switch I region, similarly to the interaction between Dom34's central domain with Hbs1 (40) and aRF1 domain B with aEF1a (31) (Supplementary Figure S3), whereas a-helix 8 interacts with eRF3's domain 3, close to the conserved GRFTLRD 613-619 motif (23) (Figure 5a).   (15), whereas the G-domain is positioned similarly to that of aEF1a in the aRF1-aEF1a complex (31) and to G-domains in other GTPases, such as EF-Tu (39), and contacts the GAC in a similar manner ( Figure 5). The switch I region of eRF3's G-domain is well ordered and is tucked between the SRL, h14 of 18S rRNA and the GGQ loop of eRF1-M. The SRL nt 4599-4600 and 4601-4603 are in proximity to DK 273-274 in the switch I and GE 326-327 in helix a5, respectively ( Figure 5a).

GTPase-activation mechanism
There are numerous similarities in the functioning of eRF1eRF3GTP and aa-tRNAEF-TuGTP complexes that suggest likely conservation of the GTPase-activation mechanism: hydrolysis of EF-Tu/eRF3-bound GTP is triggered by binding to the A site of aa-tRNA or of eRF1, respectively, followed by dissociation of the factor and accommodation of either the aa-tRNA acceptor end or the apical loop of eRF1's M domain in the PTC. In the case of EF-Tu, binding of aa-tRNAEF-TuGTP to the ribosome induces a shift in domain 2, resulting in the interaction of a highly conserved b-turn with h5 and disruption of the interaction of the 3'-end of tRNA with the switch I loop, which leads to GTPase activation in a process that also involves direct participation of the SRL (39,53). A similar contact is established between h5 on the 40S subunit and the conserved KDxGT 449-453 b-turn in eRF3's domain 2 (Figure 5a), and a C3302U substitution in the Saccharomyces cerevisiae SRL (equivalent to human C4603U) caused a termination defect (54). One notable difference is that there is additional density between h14 and eRF1's M domain (Figure 5a), which involves the   conserved R192 that is important for GTP hydrolysis, but not for GTP binding or the eRF1/eRF3 interaction (15). A hypothetical mechanism of GTPase activation based on the presumed similarity with EF-Tu would be that off the ribosome, eRF1's M domain is too tightly packed onto the switch I region to allow GTP hydrolysis, but after eRF1eRF3GTP binds to the ribosome, the R192/h14 and b-turn/h5 interactions together pull the M-domain from eRF3, loosening the switch I region. The P stalk may act as a gatekeeper in this regard, impairing the eRF1-M/h14 and eRF3/h5 interactions until eRF1's binding to a bona fide stop codon is strong enough to pull the P stalk inward. Although eRF1 can stimulate ribosomedependent GTPase activity in the absence of an A-site stop codon (6), it is likely that GTP hydrolysis would be accelerated by the interaction between eRF1 and a stop codon.
In eRF1eRF3GDPNP-bound pre-TCs, the GGQ loop of eRF1 is positioned far away from the PTC, and its accommodation in the PTC would require substantial rearrangement of eRF1's domain M. Superimposing the structure of eRF1 in the eRF1eRF3GDPNP-bound pre-TC onto the crystal structures of individual or eRF3 2-3 -bound eRF1 (7,15) shows that domain M can undergo large hinge movements ( Figure 6). If eRF3bound eRF1 adopts a strained conformation on binding to the ribosome similar to that of the A/T state of EF-Tubound tRNA (55,56), then dissociation of eRF3 following GTP hydrolysis would initiate spontaneous relaxation of eRF1 resulting in accommodation of the GGQ loop in the PTC. Full entry of the GGQ loop into the PTC would then likely be facilitated by the high mobility of the apical GGQ loop in the thermal environment (57), the positive charge of this region promoting its retention in the PTC following entry.

CONCLUSION
Our structure revealed the topology of the interaction between eRF1's N-domain and the UAA stop codon, as well as the intricate network of interactions between eRF3's switch I region, eRF1's GGQ loop and the GAC. It suggests that GTP hydrolysis may be triggered once eRF1 is properly bound to the STOP codon in a manner similar to that of EF-Tu. Now, atomic resolution structures of this complex bound to the three different stop codons may be required to achieve complete understanding of how a single factor can recognize them all.

ACCESSION NUMBERS
The Cryo-EM map is deposited in the EMDataBank with accession code EMD-5801. The atomic model of eRF1-eRF3 bound to the pre-TC is deposited in the Protein Data Bank with the accession code 3J5Y.

SUPPLEMENTARY DATA
Supplementary Data are available at NAR Online.