The ribosome termination complex remodels release factor RF3 and ejects GDP

Abstract

Translation termination involves release factors RF1, RF2 and the GTPase RF3 that recycles RF1 and RF2 from the ribosome. RF3 dissociates from the ribosome in the GDP-bound form and must then exchange GDP for GTP. The 70S ribosome termination complex (70S-TC) accelerates GDP exchange in RF3, suggesting that the 70S-TC can function as the guanine nucleotide exchange factor for RF3. Here, we use cryogenic-electron microscopy to elucidate the mechanism of GDP dissociation from RF3 catalyzed by the Escherichia coli 70S-TC. The non-rotated ribosome bound to RF1 remodels RF3 and induces a peptide flip in the phosphate-binding loop, efficiently ejecting GDP. Binding of GTP allows RF3 to dock at the GTPase center, promoting the dissociation of RF1 from the ribosome. The structures recapitulate the functional cycle of RF3 on the ribosome and uncover the mechanism by which the 70S-TC allosterically dismantles the phosphate-binding groove in RF3, a previously overlooked function of the ribosome.

Main text

Translation termination occurs when the ribosome encounters a stop codon¹. In bacteria, the class-I release factors RF1 and RF2 decode the stop codons—UAA, UAG and UGA—and catalyze the hydrolysis of the peptidyl-tRNA². Subsequently, the class-II release factor RF3 binds to the 70S-TC, which contains RF1 or RF2. Then, in a GTP-dependent manner, it induces the rotated conformation of the ribosome, promoting dissociation of class-I RFs^{3,4,5,6,7,8,9,10}. Docking of RF3 to the sarcin-ricin loop (SRL) induces hydrolysis of GTP and release of inorganic phosphate (P_i) by RF3, leading to the dissociation of RF3 from the ribosome in its GDP-bound form¹¹.

Despite decades of kinetic and structural studies^{3,4,5,6,7,8,9,10,12,13,14,15}, the molecular basis by which RF3 exchanges GDP for GTP in the cell remains unclear. The conundrum with RF3 is that its binding affinity for GDP and GTP are about the same^4,5, and kinetic assays have shown that the spontaneous dissociation of GDP and GTP from RF3 can be relatively fast (5–10 s)^4,14 or much slower, ranging from 30 s¹³ to several minutes⁵. The intracellular GTP/GDP ratio is ~7 (ref. 16) and the rate of GDP dissociation from RF3 (k_off-GDP) is ~0.13 s⁻¹ (meaning that it takes ~7 s for GDP to dissociate), suggesting that RF3 would be predominantly bound to GTP in the cell^4,14, akin to other translational GTPases such as the initiation factor 2 (IF2) and elongation factor G (EF-G). These GTPases do not require a guanine nucleotide exchange factor (GEF) because their affinity for GTP is higher than for GDP, and GDP has a fast spontaneous dissociation rate^17,18. Alternatively, studies that have reported slower exchange rates (k_off-GDP = ~0.005–0.036 s⁻¹) have proposed that RF3 could require a GEF to function catalytically during translation termination^5,13. These slow GDP dissociation rates are comparable to that of elongation factor thermo-unstable (EF-Tu) in the absence of its guanine nucleotide exchange factor (GEF), EF-Ts¹⁹, and would not account for the catalytic effects of RF3 on the recycling of RF1 and RF2 from the 70S-TC. Nevertheless, all studies agree that the physiological substrate of RF3, the ribosome bound to RF1 or RF2, accelerates the dissociation rate of GDP (k_off-GDP) in RF3 by up to >250-fold^{4,10,12,13,20,21}, suggesting that the 70S-TC can function as the GEF for RF3. Although the molecular mechanism of GDP exchange in EF-Tu by its GEF, EF-Ts, has been elucidated^19,22,23, the structural basis by which the 70S-TC triggers GDP exchange in RF3 has remained elusive.

Crystal and cryogenic-electron microscopy (cryo-EM) structures of RF3 bound to the ribosome have been reported, showing that the GTP-bound form of RF3 promotes the counter-clockwise rotation of the 30S subunit relative to the 50S subunit^3,6,7,8,10. Binding of RF3–GTP and RF1 or RF2 to the ribosome seem to be mutually exclusive^5,10,12,13, suggesting that the rotation of the ribosome induced by RF3 promotes dissociation of RF1 and RF2. Yet, on the basis of single-molecule fluorescence resonance energy transfer (smFRET), biochemical and kinetic studies^4,5,14,15,20, it has been proposed that RF3–GTP and RF1 bind simultaneously to the non-rotated ribosome. Despite this, there are only a few ribosome structures in complex with RF3 and RF1 or RF2. One study used the peptide antibiotic apidaecin to trap RF1 on the ribosome and visualized the complex it forms with GTP-bound RF3 (ref. 3). However, the absence of interaction between RF3 and RF1 in this structure has kept the mechanism by which RF3 recognizes the 70S-TC obscure. The low-resolution cryo-EM reconstruction of the 70S ribosome bound to RF1 and RF3, obtained in the absence of an exogenous nucleotide, suggested that there was a direct contact between the two factors on the non-rotated ribosome¹², hinting at a possible ribosome state that is specifically recognized by RF3.

To understand the molecular mechanism of the guanine nucleotide exchange in RF3 that is catalyzed by the 70S-TC, it is necessary to visualize how the complex modulates RF3 conformation and facilitates GDP release. We used cryo-EM to visualize RF3 bound to the Escherichia coli 70S ribosome–RF1 termination complex, uncovering the structural basis by which the 70S-TC promotes nucleotide exchange in RF3 (Fig. 1a). Our results show that the GEF activity of the ribosome destabilizes bound GDP in RF3 through a conformational change of the phosphate-binding loop (P-loop), revealing commonalities with other GEFs, notably EF-Ts. The reported structures recapitulate the complete cycle of RF3 on the ribosome, including the long-sought on-pathway GTP-activated state, and the inactive RF3 bound to the ‘magic spot’ ppGpp, showing how it stalls protein synthesis during the stringent response in bacteria.

Fig. 1: Structure of the 70S ribosome termination complex bound to apo-RF3.

a, Overview of the GEF-mediated GDP exchange 70S ribosome termination complex bound to apo-RF3 (structure I-B). b, Cryo-EM map (gray mesh) of the nucleotide-binding pocket in the G-domain, showing the absence of density for the nucleotide. The conformation of the P-loop is not compatible with a bound nucleotide. c, Apo-RF3 interacts with the 30S subunit, and the closest distance to the sarcin-ricin loop (SRL) in the 50S subunit is observed with domain III (yellow). Residue H311 in domain II (maroon) stacks with the A55-U368 base pair in helix h5 of 16S rRNA and acts as a pivot point for domain movements in RF3. d, The tip of helix α10 in domain III locates in the cleft formed between helix h5 and protein uS12, with residue N434 forming a salt bridge with Q75 in uS12. e, Model and density of the main interface between helix α13 of domain III in RF3 (yellow) and helix α1 in RF1 (magenta). The salt bridge between E14 in RF1 and R517 in RF3 is shown as a dashed line.

Results

Cryo-EM structures of the ribosome termination complex

Purified RF3 contains a stoichiometric amount of GDP, and in the absence of an exogenous nucleotide, RF3 binds to the ribosome only in the presence of RF1 (Supplementary Fig. 1a), indicating that RF3 specifically recognizes the 70S-TC, as reported previously^12,20,24. We therefore assembled the ribosome termination complex under similar conditions and subjected the sample to cryo-EM (Table 1). Three-dimensional (3D) classification sorted the selected particles into three class averages (Extended Data Fig. 1). Class average I represents the non-rotated 70S ribosome bound to RF3, RF1 and tRNAs in the P and E sites. The sample also contains two additional populations of particles: one featuring rotated 70S ribosomes bound to RF3 and a tRNA in the p/E hybrid state (class average II), and the other consisting of discarded 50S subunits (class average III). Focused 3D variability sorted the non-rotated 70S ribosomal particles with clear density for RF3 and RF1 (class I) into two main 3D volumes containing RF1 with P- and E-site tRNAs (structure I-A) or RF3, RF1 and P- and E-site tRNAs (structure I-B) (Fig. 1a and Extended Data Fig. 1). Similarly, particles in class average II were further sorted to discard empty ribosomes, which yielded one 3D volume containing the rotated 70S ribosome bound to RF3 and a tRNA in the p/E hybrid state (structure I-C) (Extended Data Fig. 1). Refinement of each reconstruction yielded density maps with a nominal resolution of between 3.0 and 3.2 Å. The combination of signal subtraction and local refinement with a mask around RF3 generated density maps for RF3–RF1 bound to the non-rotated ribosome I-B, and for RF3 bound to the rotated ribosome I-C, with an overall resolution of 3.7 Å and 3.6 Å, respectively (Extended Data Fig. 2).

Table 1.

Cryo-EM data collection, refinement and validation statistics (dataset I; no exogenous nucleotide)

	Structure I-A (70S-RF1) (EMD-29621) (PDB 8FZE)	Composite I-B a (70S-apoRF3-RF1) (EMD-29620) (PDB 8FZD)	Composite I-C a (70S-RF3-ppGpp) (EMD-29624) (PDB 8FZF)
Data collection and processing
Magnification	96,000x	96,000x	96,000x
Voltage (kV)	300	300	300
Electron exposure (e–/Å2)	40	40	40
Defocus range (μm)	−1 to −2.5	−1 to −2.5	−1 to −2.5
Pixel size (Å)	0.85	0.85	0.85
Symmetry imposed	C1	C1	C1
Initial particle images (no.)	492,815	492,815	492,815
Final particle images (no.)	127,567	80,766	33,348
Map resolution (Å)	3.0	3.1	3.2
FSC threshold	0.143	0.143	0.143
Refinement
Initial model used (PDB code)	7K00	7K00	7K00
Model resolution (Å)	3.1	3.2	3.4
FSC threshold	0.5	0.5	0.5
Map sharpening B factor (Å2)	−63.1	-	-
Model composition
Non-hydrogen atoms	150,031	153,934	148,897
Protein residues	6,169	6,662	6,262
RNA residues	4,708	4,708	4,629
Ligands: Mg2+/Zn2+	570/2	581/2	384/2
Waters	159	155	96
B factors (Å2)
Protein	64.1	46.5	88.3
RNA	56.8	53.9	91.9
Ions	37.3	34.0	49.7
Waters	33.4	32.3	44.9
CCmask	0.85	0.82	0.82
R.m.s. deviations
Bond lengths (Å)	0.002	0.003	0.004
Bond angles (°)	0.549	0.518	0.571
Validation
MolProbity score	1.6	1.6	1.8
Clashscore	5.4	5.2	6.8
Rotamer outliers (%)	0	0	0
Cβ outliers (%)	0	0	0
Ramachandran plot
Favored (%)	96.0	95.4	93.8
Allowed (%)	3.9	4.6	6.1
Disallowed (%)	0.1	0	0.1

^aThe individual maps used to generate the composite maps are listed in the Supplementary Table 4.

3D classification of the selected particles from the sample containing the non-hydrolysable GTP analog GDPCP (Table 2) produced three class averages (Extended Data Fig. 3). Similar to the previous RF3–GDP sample prepared in the absence of an exogenous nucleotide, class averages I and II contained non-rotated and rotated 70S ribosomes, respectively, and class average III represented the 50S subunits, which were not analyzed. Despite the presence of GDPCP in the sample, the non-rotated ribosomal particles (class I) contain RF3, RF1 and P- and E-site tRNAs. The rotated ribosomal particles (class II) contain RF3 and tRNA in the p/E hybrid state of binding, with no RF1, as expected from earlier studies^6,7,8,10,25. Using focused 3D classification, we sorted particles from classes I and II harboring solid density for RF3 and/or RF1. This strategy generated two non-rotated ribosome volumes in class I, structure II-A containing RF3, RF1 and P- and E-site tRNAs at 3.1-Å resolution, and structure II-D with RF1 and P- and E-site tRNAs at 2.9-Å resolution (Extended Data Fig. 4). Although the position of RF3 in structure II-A differs from that seen in structure I-B, structures I-A and II-D represent the same ribosome complex bound to RF1 after release of the nascent polypeptide chain. Similarly, two volumes emerged from class average II that produced structures II-B and II-C at 3.1-Å and 3.0-Å resolutions, respectively, representing rotated 70S ribosomes with no RF1 and differing by the binding position of RF3 (Extended Data Fig. 3). Signal subtraction combined with local refinement with a mask around RF3 generated density maps for RF3–RF1 bound to the non-rotated ribosome II-A, and for RF3 bound to the rotated ribosomes II-B and II-C, with overall resolutions of 3.8 Å, 3.6 Å and 3.4 Å, respectively (Extended Data Fig. 4).

Table 2.

Cryo-EM data collection, refinement and validation statistics (dataset II; GDPCP)

	Composite II-A a (70S-RF3-GDPCP-RF1) (EMD-29627) (PDB 8FZG)	Composite II-B a (70S-RF3-GDPCP) (EMD-29631) (PDB 8FZI)	Composite II-C a (70S-RF3-GDPCP) (EMD-29634) (PDB 8FZJ)	Structure II-D (70S-RF1) (EMD-29628) (PDB 8FZH)
Data collection and processing
Magnification	96,000x	96,000x	96,000x	96,000x
Voltage (kV)	300	300	300	300
Electron exposure (e–/Å2)	40.44	40.44	40.44	40.44
Defocus range (μm)	−1 to −2.5	−1 to −2.5	−1 to −2.5	−1 to −2.5
Pixel size (Å)	0.85	0.85	0.85	0.85
Symmetry imposed	C1	C1	C1	C1
Initial particle images (no.)	585,303	585,303	585,303	585,303
Final particle images (no.)	31,256	56,202	116,992	146,778
Map resolution (Å)	3.1	3.1	3.0	2.9
FSC threshold	0.143	0.143	0.143	0.143
Refinement
Initial model used (PDB code)	7K00	7K00	7K00	7K00
Model resolution (Å)	3.3	3.2	3.1	3.0
FSC threshold	0.5	0.5	0.5	0.5
Map sharpening B factor (Å2)	-	-	-	−55.7
Model composition
Non-hydrogen atoms	153,589	149,604	149,584	150,294
Protein residues	6,656	6,294	6,277	6,184
RNA residues	4,708	4,629	4,629	4,708
Ligands: Mg2+/Zn2+	428/2	737/2	757/2	595/2
Waters	144	252	303	262
B factors (Å2)
Protein	39.8	68.6	56.7	63.1
RNA	46.8	72.0	67.2	54.9
Ions	24.5	41.4	38.4	36.6
Waters	24.7	37.6	33.5	32.5
CCmask	0.83	0.83	0.80	0.87
R.m.s. deviations
Bond lengths (Å)	0.004	0.006	0.005	0.006
Bond angles (°)	0.594	0.655	0.618	0.652
Validation
MolProbity score	1.7	1.8	1.8	1.6
Clashscore	6.1	6.6	6.2	4.9
Rotamer outliers (%)	0	0	0	0
Cβ outliers (%)	0	0	0	0
Ramachandran plot
Favored (%)	94.0	93.5	93.9	95.2
Allowed (%)	5.9	6.4	6.0	4.7
Disallowed (%)	0.1	0.1	0.1	0.1

^aThe individual maps used to generate the composite maps are listed in the Supplementary Table 4.

The 70S-TC allosterically ejects GDP from RF3

In the sample containing RF3 and GDP, more than 65% of the selected particles are non-rotated ribosomes bound to RF3 and/or RF1 (Extended Data Fig. 1). To improve the density of RF3, we used focused classification combined with particle subtraction and focused refinement around RF3. This procedure showed unambiguously that RF3 is nucleotide-free in structure I-B (Fig. 1a,b), and given that RF3 co-purifies with GDP^4,5,10 and the alarmone guanosine 3’,5’-(bis)diphosphate (ppGpp)²⁶ in a ~1:1 stoichiometry (Supplementary Fig. 2), our results suggest that apo-RF3 in this structure represents a state that follows nucleotide dissociation catalyzed by the 70S-TC, as previously suspected¹².

Relative to GDP-bound RF3 that is not associated with the ribosome^10,26, domain III of 70S-TC-bound apo-RF3 rotates by 24° relative to domains I (G-domain) and II (Fig. 2a and Supplementary Video 1). The position of domain III, which is presumably maintained by its interactions with domain I of RF1, induces long-range allosteric rearrangements that propagate to the nucleotide-binding pocket in the G-domain. Rotation of domain III induces a shift of the switch 2 (sw2)-helix by 12 Å toward helix α4 in the G-domain, forming a new interface (Fig. 2b,c). To yield space for the new position of H92, the side chain of R122 moves together with helix α4, which in turn displaces H21 toward P22 in the P-loop (Fig. 2c). The domino effect of side chain displacements induces the 180° flip of the carbonyl oxygen at P22 (Fig. 2d and Supplementary Video 1). The flip of the P22 main chain carbonyl oxygen invades into the nucleotide-binding pocket and would collide with the β-phosphate of GDP, thereby catalyzing the release of GDP from RF3 (Fig. 2d). The model of the P-loop in apo-RF3, including the flipped carbonyl oxygen at residue P22, fits the EM density well (Fig. 2e) and, remarkably, superimposes almost perfectly with the P-loop of EF-Tu in complex with EF-Ts (Fig. 2f)^22,23. In the ribosome-free conformation, the conformation of the P-loop in RF3–GDP closely matches that in EF-Tu–GDP (Fig. 2f)^10,27, indicating that the 70S-TC and EF-Ts exert their GEF activity through a common mechanism.

Fig. 2: Orientation of domain III in apo-RF3 triggers dissociation of GDP.

a, Comparison of apo-RF3 (colored domain) with RF3 in its GDP-bound form (PDB 2H5E)¹⁰. Alignment of domains I and II reveals the 24° rotation of domain III in apo-RF3 that is required to form the interface with RF1. b, Rotation of domain III leads to the 12-Å displacement and 20° rotation of the sw2-helix (shown as a cylinder). c, The displacement of sw2 induces a domino effect of side chain movements that remodel the conformation of the P-loop. d, The rearrangements of the P-loop include the flip of the carbonyl oxygen at residue P22, which would collide with the β-phosphate of GDP (white), causing its release from RF3. e, Cryo-EM map (gray mesh) showing the fit of the P-loop region in apo-RF3, including the carbonyl oxygen of residue P22. f, The alignment of the P-loop region in apo-RF3 (magenta) with that of EF-Tu (black) in complex with its GEF, EF-Ts (PDB 1AIP)²², shows the similarity of the conformational rearrangements, causing dissociation of GDP. For reference, the conformation of P22 in RF3–GDP (PDB 2H5E, light magenta)¹⁰ is the same as in EF-Tu–GDP (PDB 1TUI, gray)²⁷. g,h, Effects of substitutions in RF1 (E18A) and RF3 (all other mutations) on the rate of GDP exchange, assessed by the decrease of fluorescence at 340 nm as mant-GDP was chased by unlabeled GDP in the presence of 70S ribosomes bound to RF1. WT, wild type.

In the structure, helix α13 in domain III of apo-RF3 forms the main interface with helix α1 in domain I of RF1 (Fig. 1e). The relative orientation of helices α13 in RF3 and α1 in RF1 is such that V509, and the longer side chains of L513 and R517 in RF3, form the complementary surface with RF1. The EM map suggests that a salt bridge forms between R517 in RF3 and E14 in RF1 (Fig. 1e and Extended Data Fig. 5a). The main chain Cα and side chain Cβ atoms of residue N510 contribute to the surface against which domain I of RF1 docks, and residues E18 and R15 form an intramolecular salt bridge lining one side of helix α1 in RF1 (Extended Data Fig. 5b). The interface formed by domain I of RF1 keeps domain III of RF3 in a position that promotes rearrangements of the sw2-region and the nucleotide-binding pocket in the G-domain.

To test this model, we used rapid kinetics to determine the rates of N-methylanthraniloyl (mant)-labeled GDP exchange for native GDP. In a stopped-flow apparatus, we rapidly mixed variants of RF3 bound to mant-GDP with a solution containing the RF1-bound ribosome termination complex. Under our experimental conditions, the apparent mant-GDP exchange rate (k_off-GDP) for ribosome-free RF3 is ~0.13 s⁻¹, as previously reported4. It is stimulated ~75-fold by the 70S-TC, reaching a rate of 10.16 s⁻¹, similar to previous measurements (Fig. 2g,h and Supplementary Table 1)^4,10. We find that the substitutions V509A, L513A and R517A in helix α13 and K488A in helix α12, lining the interface with helix α1 in domain I of RF1, reduce the GDP exchange rate by ~10- to 20-fold (Fig. 2g and Extended Data Fig. 5a,b). Similarly, E18A, which disrupts the intramolecular salt bridge within helix α1 of RF1, reduces the rate of GDP exchange by ~14-fold (Fig. 2g and Extended Data Fig. 5b), in line with the reported threefold-slower RF3-mediated dissociation of RF1-E18A from the ribosome¹². Conversely, N510A and E516A in RF3 essentially do not affect GDP exchange, with apparent dissociation rates of ~6.8 and 9.2 s⁻¹, respectively (Fig. 2h and Supplementary Table 1), in agreement with the absence of interaction between these residues and RF1 (Extended Data Fig. 5a,b). The mutation of H21, one of the residues involved in the allosteric network of conformational changes, to alanine reduces GDP exchange by ~15-fold (Fig. 2g), whereas P22A had essentially no effect, with a k_off-GDP of 7.9 s⁻¹ (Fig. 2h). This corroborates the involvement of the main chain carbonyl oxygen of P22 that flips and sterically interferes with bound GDP.

In structure I-B, apo-RF3 does not contact the 50S subunit; the closest distance, ~14 Å, is observed between the SRL and domain III (Fig. 1c). Apo-RF3 is anchored to the 30S subunit through domains II and III (Fig. 1c,d). One contact point is mediated by the π–π-stacking interaction between H311 in domain II and the nucleobase A55 in helix h5 of the 30S subunit body domain (Fig. 1c). The importance of this interaction is substantiated by findings that substitution of A55 in the 16S rRNA confers a lethal phenotype in E. coli by interfering with the translation elongation cycle²⁸. Similarly, H311A in RF3 impairs the rate of 70S ribosome termination complex-mediated GDP exchange by 50-fold and severely reduces the ability of RF3 to recycle RF1 and RF2 (ref. 10). In domain III, helix α10 runs along helix h5 of the 30S subunit and docks in the cleft formed by h5 and protein uS12 (Fig. 1d). Domain III further contacts uS12 through β-hairpin 429–436, with residue N434 in RF3 located within interaction distance of Q75 in uS12 (Fig. 1d). These interactions place domain III of RF3 in the optimal orientation to interact with domain I of RF1.

Relative to structure I-A, which represents a state of the ribosome after release of the nascent polypeptide chain before RF3 binding, the tips of 23S rRNA helices H43 and H44 in the ribosomal stalk of structure I-B move toward domain I of RF1 by more than 3 Å, stabilizing this region of RF1 and the uL11-stalk domain (Extended Data Fig. 6). Correspondingly, we observe clear density for proteins uL10 and uL11 (Extended Data Fig. 6a,b). The α1-turn-α2 region in domain I of RF1 forms stacking interactions with the proline-rich 3₁₀-helix in the N terminus of protein uL11, while nucleotides A1067 (H43) and A1095 (H44) stack against helix α2 of RF1 (Extended Data Fig. 6e). We observe additional density proximal to the beak region of the 30S subunit head and near the G-domain of apo-RF3, which appears to be that of the C-terminal domain (CTD) of ribosomal protein bL12 interacting with the α2-turn-α3 in RF1 and the G′ subdomain in RF3, as observed previously (Extended Data Fig. 6d)^3,10,12. The structure is consistent with reports that the CTD of bL12 contributes to the recruitment of RF1, EF-Tu and EF-G to the ribosome3^,29^,30, and accelerates GTP hydrolysis by RF3 (ref. 31), IF2 (refs. 25^,31), EF-Tu and EF-G^29,32.

GTP binding promotes domain rotation in RF3

To visualize RF3 in the activated state on the 70S-TC, we assembled the complex, as described above, and included the non-hydrolysable GTP analog GDPCP. Despite the presence of GDPCP, the reconstruction of structure II-A yielded a non-rotated ribosome containing RF3, RF1 and P- and E-site tRNAs (Fig. 3a and Extended Data Fig. 3), representing a state that precedes the dissociation of RF1. The map of structure II-A, with a nominal resolution of 3.1 Å (Extended Data Fig. 4), was used to locally refine RF3, which allowed us to unambiguously visualize GDPCP bound to the G-domain (Fig. 3d). Relative to apo-RF3 in structure I-B, domain III of RF3 in structure II-A tilts by ~12° in response to the movements of domains I (G) and II (Fig. 3a and Supplementary Table 2). The density associated with the side chain of R517 disappears; correspondingly, the salt bridge between R517 in RF3 and E14 in RF1 seen in the apo-complex (I-B) no longer forms, suggesting that the interface with RF1 is weakened (Fig. 3f). The other two domains, II and G, undergo larger movements, rotating by ~32° with an effective displacement of ~17 Å, carrying the outskirt of the G-domain by ~29 Å toward the 50S subunit and bringing the catalytic H92 within 9 Å of A2662 in the SRL (Fig. 3a,b,e and Supplementary Video 1). Although the contacts between domain II and the 30S subunit are mostly dissolved, the stacking mediated by residue H311 remains largely unchanged, acting as the pivot point for the rotation of domain II (Fig. 3c and Supplementary Video 1).

Fig. 3: Binding of GTP initiates domain rearrangements in RF3.

a, Comparison of RF3–GDPCP bound to the non-rotated 70S-TC (colored domains, structure II-A) with apo-RF3 (structure I-B, gray). The structures were aligned by the 23S rRNA. Domains I and II in RF3–GDPCP rotate by 32°, whereas domain III rotates by 12°. b, The rotation brings the G-domain (blue) closer to the SRL (green) in the 50S subunit. c, The pivot point of the rotation is the interaction between H311 and the 30S subunit. d, Density map of the nucleotide-binding pocket, with that of GDPCP shown as teal mesh. e, The catalytic histidine 92 is located 9 Å from the non-bridging phosphate of A2662 in the SRL. f, Map of the interface between RF3 domain III (yellow) and domain I of RF1 (magenta). The absence of density for R517 suggests that the salt bridge that is observed in structure I-B (apo-RF3) is dissolved in structure II-A, weakening the interaction between RF3 and RF1.

The simultaneous binding of RF3–GTP and RF1 to the non-rotated ribosome was inferred from smFRET, biochemical and kinetic studies^4,5,14,15,20. Here, the overall conformation of RF3–GDPCP bound to the non-rotated 70S-TC is reminiscent of that reported in the apidaecin-stalled 70S-TC (Extended Data Fig. 7a)³. In the previous study, however, domain III of RF3 was located 9 Å away from helix α1 of RF1, excluding the possibility of interaction between the two. In our structure, domains of RF3–GDPCP further rotate inward by ~17° (I and II) and ~23° (III) toward RF1, placing domain III within interaction distance of RF1 (Extended Data Fig. 7a and Supplementary Table 3).

Docking of RF3–GTP to the SRL on the rotated ribosome

The coordinated conformational changes in RF3 and the ribosome progressively dock the G-domain of RF3 to the SRL. The 3D volumes were generated by sorting the particles on the basis of the binding position of RF3, revealing two conformations of RF3–GDPCP on the rotated ribosome (Extended Data Fig. 3). In structures II-B and II-C, the 30S subunit is rotated by 8.5° relative to the 50S subunit (Figs. 1a and 4a), in line with previous reports^3,6,7,8,10. The absence of RF1 in structures II-B and II-C is consistent with the notion that the rotation of the 30S subunit promotes the dissociation of class-I RFs from the ribosome (Figs. 4 and 5)^5,14,24.

Fig. 4: Rotation of the 30S subunit positions the G-domain of RF3 near the SRL.

a, Comparison of structures II-A (gray) and II-B (colored domains) aligned by the 23S rRNA. b, The rotation of the 30S subunit by 8.5° in structure II-B is accompanied by the rotation of domains I and II in RF3 toward the 50S subunit, while domain III rotates in the opposite direction toward the 30S subunit. c, The rotation of the 30S subunit is transmitted to domains II and III of RF3 through the two contact points—H311–A55 and L407–uS12—leading to the rotation of domains in RF3. d, The rotated conformation of the ribosome stabilizes switch 1 (sw1) in RF3 through the stacking between residue Q41 and nucleotide A344 in helix h14 in the 30S subunit. e, The 20° rotation of domain III toward the 30S subunit creates a space for RF1 to dissociate from the ribosome. For reference, RF1 from structure II-A (gray) is shown. f, Despite the G-domain getting closer to the 50S subunit, the catalytic histidine 92 remains 11 Å away from A2662 in the SRL.

Fig. 5: Rolling-like motion of the 30S subunit activates RF3 for GTP hydrolysis.

a, Relative to structure II-B, the shoulder domain of the 30S subunit in structure II-C undergoes a rolling-like movement toward the intersubunit space. b, This movement causes the whole RF3 to rotate by 27°, bringing the G-domain 5 Å closer to the 50S subunit. c, The induced rotation of RF3 occurs around the two pivot points, H311–A55 and L407–uS12. d, In structure II-C, H92 inserts through the hydrophobic gate formed by residues I68 in sw1 and P22 in the P-loop, to locate within interaction distance of A2662 in the SRL, representing the GTP-activated state of RF3 on the ribosome.

Relative to structure II-A, rotation of the 30S subunit in structure II-B moves domains II and III of RF3 by ~9 Å through the H311–A55 nucleobase π–π-stacking interaction and the L407–uS12 contact point, which is accompanied by pivoting rotations of ~11° (domain II) and ~20° (domain III) (Fig. 4a–c and Supplementary Table 2). Concomitantly, the G-domain moves by ~13 Å, with a rotation of ~13° toward the SRL in the 50S subunit (Fig. 4b and Supplementary Video 2). The switch 1 (sw1)-loop is ordered in structure II-B, and residue Q41 stacks with the nucleobase of A344 in helix h14 of 16S rRNA (Fig. 4d), a feature that is observed with EF-G and has been proposed to stabilize the active conformation of the GTPase center^33,34. The rotation of domain III presumably breaks the contacts between RF3 and RF1, facilitating the dissociation of RF1 from the 70S-TC (Fig. 4e and Supplementary Video 2). The overall conformation of RF3–GDPCP on the rotated ribosome in structure II-B is similar to the previously reported one (Extended Data Fig. 8a,b)^3,6,7 wherein, despite the additional stabilization of sw1 provided by helix h14 (Fig. 4d) and in contrast to the EF-Tu-ribosome complex^35,36, nucleotide A2662 of the SRL does not contact GDPCP or RF3 (Fig. 4f, Extended Data Fig. 8c–f and Supplementary Video 2). This observation led to the suggestion that ribosome-dependent GTP hydrolysis in RF3 is indirectly triggered by the ordering of RF3 and positioning of a critical Mg²⁺ ion at the β-γ phosphate linkage of GTP upon binding to the ribosome6.

In the GTP-activated EF-Tu–ribosome complex^35,36,37, the catalytic histidine interacts with A2662 of the SRL and is proposed to optimally orient a water molecule for an inline nucleophilic attack on the γ-phosphate of GTP^38,39. Although H92 is directed away from the γ-phosphate in structure II-B and in that of the E. coli RF3–GDPNP bound to the rotated ribosome (Fig. 4f and Extended Data Fig. 8c)⁶, previous studies have shown that H92 is directed toward GTP (Supplementary Fig. 9d,e)^3,7. Nonetheless, no interaction is seen with A2662 in the SRL, indicating that RF3 is not fully accommodated on the ribosome (Fig. 4f and Extended Data Fig. 8c–f).

The hydrolysis of GTP by RF3 is ribosome-dependent¹¹, and the universal histidine in the G-domain of RF3 suggests that its GTP-hydrolysis mechanism is similar to that of other translational GTPases. It has therefore been proposed that these structures, such as II-B, represent an intermediate state along the activation pathway, and that subsequent conformational rearrangements of RF3 could affect the orientation of H92 for catalysis⁴⁰. Structure II-C seemingly represents the catalytically competent state of RF3 on the rotated ribosome (Fig. 5). Relative to structure II-B, the shoulder domain of the 30S subunit moves by ~3 Å toward RF3, tightening the intersubunit space (Fig. 5a,b). This displacement, which is reminiscent of the rolling motion of the small ribosomal subunit in prokaryotes^41,42,43,44 and eukaryotes^44,45,46, causes the whole RF3 to further rotate by ~27° around the H311–A55 and L407–uS12 hinge points (Fig. 5c, Supplementary Table 2 and Supplementary Video 2). These rearrangements bring the G-domain ~5 Å closer to the SRL (Fig. 5b), and H92 is directed toward GDPCP through the hydrophobic gate formed by residues P22 and I68, located within interaction distance of the non-bridging phosphate oxygen of A2662 (Fig. 5d and Supplementary Video 2). The interactions observed here surrounding GDPCP are similar to those reported in catalytically competent EF-Tu^35,36,37, EF-G^33,34,47,48 and IF2 (refs. 49^,50) bound to the ribosome, suggesting that we captured a similar state of RF3 and that the mechanism of GTP hydrolysis is likely to be universal among all translational GTPases.

ppGpp stalls RF3 on the rotated ribosome

The RF3–GDP sample prepared in the absence of an exogenous nucleotide contained a second class of particles which yielded structure I-C, a rotated 70S ribosome bound to RF3 and lacking RF1 (Extended Data Figs. 1 and 9a,c). Local refinement produced a map of RF3 with an overall resolution of 3.6 Å (Extended Data Fig. 2) that shows additional density emerging from the nucleotide-binding pocket toward the SRL (Extended Data Fig. 9d). The EM density of the bound nucleotide is consistent with that of ppGpp (Supplementary Fig. 2), the bacterial stringent stress response mediator (Extended Data Fig. 9d). The α- and β-phosphates of ppGpp locate along the nucleotide-binding groove in the G-domain, whereas the δ- and ε-phosphates protrude out of RF3 toward the SRL with the δ-phosphate forming a Mg²⁺-mediated interaction with the non-bridging phosphate oxygen of A2662 in the SRL (Extended Data Fig. 9d,e).

The conformation of RF3–ppGpp bound to the ribosome in structure I-C is identical to that of ribosome-free RF3–GDP and RF3–ppGpp (Extended Data Fig. 9b)^10,26. Given that GDP-bound RF3 has low affinity for the ribosome (Supplementary Fig. 1b–d)¹², it is probable that the ability of ppGpp to establish an ion-mediated interaction with the SRL stabilizes the binding of RF3 to the 50S subunit. This, in turn, drives the rotated conformation of the ribosome, facilitating the dissociation of RF1, as observed previously²⁶. By locking RF3 in a non-active conformation on the rotated ribosome, the structure provides a basis by which the alarmone ppGpp inhibits translation during the stringent stress response in bacteria.

Discussion

All GTPases function as molecular switches that cycle between two conformations, the ‘active’ one induced by GTP and the ‘inactive’ one bound to GDP. The return of GTPases from the inactive GDP-bound conformation to the active GTP-bound form is often catalyzed by a GEF^51,52. Among the GTPases involved in translation, EF-Tu requires EF-Ts to rapidly exchange GDP for GTP^{19,53,54,55,56}. Although IF2 and EF-G do not require the action of a GEF, the fate of RF3 has remained unclear. For many years, there has been controversy regarding the exchange of GDP in RF3, fueled by data suggesting that RF3 does not require a GEF in vivo to exchange GDP for GTP^4,14, that it might use a GEF⁵ and that a GEF is needed for RF3 to function catalytically in translation termination^10,12,13,15. Our results and previous⁴ kinetic data show that it takes ~7 s for RF3 to exchange GDP (k_off-GDP = ~0.13-0.15 s⁻¹). In theory, this should be fast enough for RF3 to exchange GDP for GTP and participate in translation termination in E. coli, because it takes ~20 s to complete the synthesis of an average polypeptide chain (~300 amino acids at ~15 amino acids per s)^4,14. However, the 70S-TC substantially accelerates GDP exchange in RF3, ranging from ~75-fold (this study) to more than 250-fold^4,10,13,21. Therefore, it is plausible that, in vivo, RF3 uses two pathways to catalyze recycling of class-I release factors from the ribosome: (1) RF3–GDP encounters the 70S-TC and rapidly exchanges GDP for GTP, and (2) RF3–GTP directly participates in termination (Fig. 6).

Fig. 6: Model for the functional cycle of RF3 during translation termination.

Following recognition of the stop codon by RF1 or RF2, peptidyl-tRNA is hydrolyzed and the nascent peptide chain released (structure I-A). In step 1 of scenario 1, RF3–GDP binds to the non-rotated ribosome in complex with RF1 or RF2 following release of the nascent polypeptide chain, forming the GDP exchange complex (structure I-B). In this complex, the interaction between RF3 and RF1 remodels the P-loop in the G-domain of RF3, causing the rapid dissociation of GDP. Step 2 depicts the rapid binding of GTP to ribosome-bound RF3 (~130 s⁻¹)⁴. The GTP-bound RF3 can co-exist on the non-rotated ribosome together with RF1 or RF2 (structure II-A). In scenario 2, RF3 that has exchanged GDP for GTP in the cell can associate with the 70S-TC^4,14, resulting in the same structure II-A. In step 3, RF1 is released from the ribosome as RF3–GTP promotes the rotation of the 30S subunit (structure II-B). In step 4, the rolling-like motion of the 30S subunit and further rotation of RF3–GTP bring the G-domain proximal to the sarcin-ricin loop (SRL) of the 50S subunit, activating RF3 for GTP hydrolysis (structure II-C) which is followed by the dissociation of RF3–GDP.

The structures reported here provide the basis on which RF3 singles out ribosomes undergoing translation termination and reveal the molecular mechanism of GDP-catalyzed dissociation from RF3, illustrating how the 70S-TC can function as the GEF for RF3. We show that perturbations of the interface between RF3 and RF1 lead to reduced rates of GDP exchange in RF3 (Fig. 2g and Supplementary Table 1), supporting previous findings suggesting that the truncation of domain I in RF1 and RF2 abolishes the exchange of GDP and the RF3-catalyzed recycling of the mutant factors⁵⁷. Although these alterations do not appear to affect the binding of RF3 to the 70S-TC (Supplementary Fig. 1a), it remains unclear whether they affect only the orientation of domain III or the overall binding position of RF3 in the 70S-TC. Subsequent binding of GTP initiates domain rearrangements in RF3 on the non-rotated 70S-TC, and the rotation of the 30S subunit promotes dissociation of RF1 from the ribosome. Further rotation and displacement of domains in RF3 in structures II-B and II-C bring the G-domain proximal to the SRL in the 50S subunit. The rolling-like motion of the 30S shoulder domain assists in the final positioning of RF3, docking the G-domain onto the SRL and bringing the catalytic H92 within interaction distance of the non-bridging phosphate oxygen of A2662, thereby activating RF3 for GTP hydrolysis.

On the basis of our structures, we propose that when RF3–GDP encounters the ribosome termination complex (Fig. 6, scenario 1, step 1), the 70S-TC triggers remodeling of the P-loop in the G-domain of RF3, leading to the flip of the main chain carbonyl oxygen at P22. The P-loop intrudes sterically and electrostatically upon the nucleotide-binding pocket, catalyzing the dissociation of GDP from RF3 (Fig. 6, step 2). In E. coli, the concentration of GTP is estimated to be ~1,660 μM, and that of GDP is ~230 μM¹⁶. The fact that GDP and GTP bind to RF3 with similar affinities4^,5 and that GTP binds to RF3 in complex with the 70S-TC at a rate of ~130 s⁻¹ (ref. 4) suggests that once GDP is ejected from RF3, it is rapidly replaced with GTP. Subsequently, stepwise movement of RF3 toward the SRL and rotation of the 30S subunit promote the dissociation of RF1 from the ribosome (Fig. 6, step 3). Final docking of the G-domain to the SRL orients H92 for GTP hydrolysis, ending with the dissociation of RF3–GDP from the ribosome (Fig. 6, step 4). Alternatively, GTP-bound RF3 could specifically recognize the 70S-TC, exemplified by structure II-A, and induce rotation of the 30S subunit and dissociation of RF1 from the ribosome (Fig. 6, scenario 2). It is worth pointing out that kinetic and smFRET assays suggest that there are a multitude of potentially productive and unproductive pathways during translation termination^14,15.

Our structures show that GTP-bound RF3 can adopt at least two conformations on the rotated ribosome, illustrated by complexes II-B and II-C. Because domain III of RF3 would sterically clash with domain I of RF1 in structure II-C (Extended Data Fig. 7b), the final rotation of RF3 toward the SRL required for GTP hydrolysis is possible only following the dissociation of class-I RFs. This could function as a checkpoint that licenses RF3 for departure.

In structure I-B, we observe additional density proximal to the G′ domain of RF3 and in the vicinity of the turn between helices α2 and α3 of RF1 (Extended Data Fig. 6d). We attribute these density blobs to the CTD of protein bL12, as observed previously^3,10,12. In E. coli, protein uL10 binds to four copies of bL12, forming the pentameric uL10–stalk complex that stimulates GTP hydrolysis by RF3 (refs. 3^,31), IF2 (refs. 25^,31), EF-Tu and EF-G^29,31,32. Thus, structure I-B is consistent with reports that these interactions help recruit the decoding factors, such as RF1 (ref. 3), as well as the GTPases RF3, EF-G, IF2 (refs. 29^,31) and EF-Tu³⁰, to the ribosome.

It is worth noting that prfC, the gene encoding RF3, is not essential in E. coli and is found in only a subset of bacteria⁵⁸. In E. coli, RF3 has been reported to ensure quality control of protein synthesis, stimulating abortive termination on ribosomes with P-site mismatches^59,60,61, raising the possibility that the primary function of RF3 might not be to recycle release factors during normal termination. Consistent with this notion, a proteomic study revealed that RF3 mediates premature termination of misfolded nascent polypeptide chains by cooperating with RF2 (ref. 62). Similarly, kinetic assays have shown that, although RF1 needs RF3 for dissociation from the ribosome, recycling of RF2 can proceed independently of RF3 (ref. 14).

During the bacterial stringent stress response, the concentration of the alarmone ppGpp can increase above that of GDP. Reportedly, there is a sharp increase in the concentration of ppGpp in E. coli cells entering the stationary phase of growth¹⁶, and expression of recombinant proteins can reprogram metabolic processes⁶³, possibly explaining the presence of ppGpp in the pure sample of RF3 (Supplementary Fig. 2). Structure I-C reveals that RF3–ppGpp forms a bridge interaction with the SRL of the 50S subunit, promoting the rotated conformation of the ribosome and dissociation of RF1. The increased stabilization provided by the additional interaction between ppGpp and the SRL on the rotated ribosome might trap RF3 on the ribosome. In vivo concentrations of RF1, RF2 and RF3 are sub-stoichiometric relative to ribosomes⁶⁴, and ppGpp may deplete free RF3 and contribute to the reduced efficiency of RF3-mediated recycling of RF1 (ref. 26), similar to the action of apidaecin, which depletes free-RF1 and RF2 (ref. 65). However, given that RF3 is a non-essential protein, it is possible that the global inhibitory effects of ppGpp on translation may stem from, at least in part, ribosome-stalled RF3 causing a ‘traffic jam’ by interfering with the binding of EF-Tu and EF-G. As a key second messenger in bacterial cells, ppGpp not only interferes with RF3, but also with IF2 (refs. 66^,67), EF-G⁶⁸ and EF-Tu^67,68, along with the process of ribosome assembly^69,70.

This work draws parallels between the molecular mechanisms employed by the 70S-TC and EF-Ts catalyzing the exchange of GDP in RF3 and EF-Tu, respectively. In particular, it is remarkable how the peptide flip at P22 that ejects GDP from RF3 is reminiscent of the action of EF-Ts on EF-Tu. EF-Ts inserts an aromatic residue that initiates a chain reaction of conformational changes that involve H19 (H21 in RF3), ultimately leading to the main chain carbonyl oxygen flip at V20 (P22 in RF3)^22,23. On the basis of the similarity of the conformational changes in the G-domain of RF3 and EF-Tu, it is tempting to suggest that the ribosome could be the evolutionary prototype guanine exchange factor and that modern GEFs, such as EF-Ts, would have evolved through convergent evolution.

Methods

Preparation of 70S ribosomes, release factors, initiator tRNA^fMet and tRNA^Phe

The full-length RF1 sequence was PCR amplified from E. coli MRE600 genomic DNA and cloned into pET21a plasmid (Novagen) carrying a 6×His C-terminal tag. E. coli BL21 (DE3) Star (C601003, Invitrogen) cells transformed with this construct were grown in LB medium supplemented with 50 μg ml⁻¹ carbenicillin (C-103-25, Gold Biotechnology) to an absorbance of 0.6 at 600 nm before expression of C-His-RF1 was induced with 1 mM isopropyl-β-d-thiogalactoside (IPTG) (I2481C50, Gold Biotechnology) for 4 h at 37 °C. Cells were lysed in buffer A (50 mM Tris-HCl, pH 8.0, 10 mM MgCl₂, 500 mM KCl, 10 mM imidazole, 6 mM β-mercaptoethanol), supplemented with 1 tablet of complete EDTA-free protease inhibitor cocktail (Thermo Fisher Scientific). The resuspended cells were lysed by several passages through a high-pressure homogenizer LM20 (Microfluidics) operated at 15,000 psi. The cell debris was removed by centrifugation at 30,600g at 4 °C for 30 min. Filtered lysate (through a 0.22-μm filter, Millipore) was applied to an equilibrated 5 ml HisTrap HP (GE Healthcare) column. RF1 (C-His) was eluted with a linear gradient of imidazole to 500 mM. Fractions containing RF1 (C-His) were pooled, concentrated and further purified on a Superdex 200 HiLoad 16/60 (GE Healthcare) size-exclusion chromatography column equilibrated in buffer 10 mM Hepes-KOH, pH 7.5, 5 mM MgCl₂, 500 mM KCl. Pure RF1 (C-His) was concentrated to ~40 mg ml⁻¹, flash-frozen and stored at −80 °C.

The full-length RF3 sequence was PCR amplified from E. coli K-12 genomic DNA and cloned into pET28a plasmid with N-terminal 8×His-SUMO tag. E. coli BL21 (DE3) Star competent cells that received this construct were grown at 37 °C in LB medium with the addition of 50 μg ml⁻¹ kanamycin. Expression of RF3 was induced with 1 mM IPTG (Sangon Biotech) once the culture reached an absorbance of 0.5 at 600 nm. The cells were collected 3 h post-induction. The protein purification scheme was the same as that described for RF1, except that the 8×His-SUMO tag was removed with Ulp1 protease. RF3 was placed in a buffer comprising 10 mM Tris-HCl, pH 7.4, 10 mM magnesium acetate, 200 mM KCl and 2 mM β-mercaptoethanol at a concentration of 16 mg ml⁻¹, flash-frozen and stored at −80 °C.

The presence of nucleotide in purified RF3 was analyzed using a reverse-phase high-performance liquid chromatography (HPLC) assay, as described previously⁷¹ but with minor modifications. In brief, 15 nmol purified RF3 or 15 nmol NTP (Sigma) or (p)ppGpp (Jena Bioscience) in 150 μl of buffer (10 mM Tris-HCl pH 7.66, 200 mM KCl, 10 mM magnesium acetate, 1 mM β-mercaptoethanol) was heated at 95 °C for 5 min to denature proteins and release their bound GNPs. The samples were centrifuged for 30 min (14,000g) to pellet the proteins, which was repeated twice. The supernatants were injected onto the C18 column (Boston, 5 μm, 4.6 × 250 mm) and equilibrated in buffer 150 mM KH₂PO₄ and K₂HPO₄ buffer pH 6.0 and 3% methanol. The elution was performed isocratically over four column volumes (Supplementary Fig. 2).

70S ribosomes lacking ribosomal protein bL9 were purified from E. coli KC6 (ΔssrA, ΔsmpB, rplI_Δ41-149, rpsA_C-term-His:kan^r), essentially as has been reported for the Thermus thermophilus HB8 (rplI_Δ58-148:htk)^72,73,74,75 but with minor modifications. In brief, the cells were lysed in buffer A (20 mM Tris-HCl pH 7.4, 100 mM NH₄Cl, 10 mM magnesium acetate, 0.5 mM EDTA-Na pH 8.0, 6 mM β-mercaptoethanol, 150 mM sucrose), and the lysate was clarified by centrifugation. Following ultracentrifugation for 16 h (100,000g) at 4 °C in the Type 45Ti rotor (Beckman Coulter), the ribosome pellets were resuspended in the HIC buffer A (20 mM Tris-HCl pH 7.4, 200 mM KCl, 10 mM magnesium acetate, 1.5 M (NH₄)₂SO₄, 6 mM β-mercaptoethanol) and applied onto a Toyopearl HIC column to remove ribosomal protein bS1. Further purification to ensure complete removal of bS1-C-term-His was accomplished by passing the eluate through a Ni-NTA column. Subsequently, the ribosomes were separated by centrifugation through 10–40% sucrose gradients in the SW32 rotor at 90,000g for 14 h. Pure 70S:ΔL9 ribosomes were collected, pooled and buffer exchanged into 5 mM Tris-HCl pH 7.4, 60 mM NH₄Cl, 10 mM MgCl₂ and 6 mM β-mercaptoethanol, flash-frozen in liquid nitrogen and stored at −80 °C.

E. coli initiator tRNA^fMet was purified using previously established procedures⁷⁶. E. coli tRNA^Phe was cloned into the pBSTNAV plasmid (45801, Addgene) and expressed under the constitutive lpp promoter overnight in E. coli JM109 cells in 2× YT medium. Total tRNA was extracted using phenol-chloroform, followed by 96% ethanol precipitation. tRNA^Phe was purified by anion exchange on source 15Q (HR 16/10) (Cytiva) and reverse-phase chromatography PROTO 300 C4 HPLC (10 × 250 mm) (Higgins Analytical) columns. The unacylated-tRNA^Phe was then aminoacylated with E. coli phenylalanyl-tRNA synthetase (50 μM tRNA^Phe, 2.5 μM PheRS in aminoacylation buffer supplemented with 2.5 mM ATP and 2.5 mM l-Phe for 30 min at 37 °C) and purified by reverse-phase HPLC by isolating the shifted peak on the chromatogram corresponding to aminoacylated Phe-tRNA^Phe on the C4 column. Phe-tRNA^Phe was finally deacylated in 100 mM Tris pH 8.0 and re-purified on the C4 column.

The 21-nucleotide mRNA, containing a Shine-Dalgarno sequence, phenylalanyl (UUC) codon in the P-site (bold), and the stop codon (UAA) in the A site (underlined), with the sequence 5′-CAAGGAGGAAAAAUUCUAAUA-3′ was chemically synthesized by Integrated DNA Technologies.

Site-directed mutagenesis

RF1 and RF3 mutants were generated using the QuikChange II XL Site-Directed Mutagenesis kit (200251, Agilent Technologies), following the manufacturer’s instructions. The substitutions were confirmed by DNA sequencing, and the mutant proteins were purified using the same procedures as described for wild-type RF1 and RF3.

Ribosome binding assays

E. coli 70S:ΔL9 ribosomes (final concentration, 0.5 μM) were programmed with Phe-Stop mRNA (1 μM) and deacylated-tRNA^Phe (1 μM) by incubation at 37 °C for 10 min in 5 mM Tris-HCl pH 7.4, 60 mM NH₄Cl, 10 mM MgCl₂ and 6 mM β-mercaptoethanol. Then, RF1 (WT or E18A) was added to a final concentration of 2 μM and incubated for 5 min. In parallel, 6 μM RF3 was pre-heated for 15 min at 37 °C and added to the release complex mixture, followed by 5 min of incubation at 37 °C. The mixture was passed through a cold sucrose cushion buffer (5 mM Tris-HCl pH 7.4, 60 mM NH₄Cl, 10 mM MgCl₂, 6 mM β-mercaptoethanol, 30% sucrose), by centrifugation in the TLS55 rotor (Beckman Coulter) for 1 h at 213,630g at 4 °C. The top 50 μl sample solution and the bottom 20 μl were precipitated with 100% TCA (trichloroacetic acid), pelleted by centrifugation, washed with ice-cold acetone and dissolved in 1× protein-loading SDS-buffer. Results were analyzed using 4–12% SDS–PAGE (SurePAGE Gel, GenScript), stained by Coomassie. The gels are shown in the Supplementary Figure 1.

Stopped-flow measurements of mant-GDP dissociation from RF3

Kinetic experiments were performed with a SX20-LED stopped-flow spectrometer (Applied Photophysics) using an excitation wavelength of 340 nm, and the fluorescence emission was measured after the sample was passed through a 400-nm cutoff filter. To prepare RF3-mant–GDP complexes, 50 μM RF3 (wild-type and its mutants) was preincubated with 500 μM N-methylanthraniloyl (mant)-GDP (JBS-NU-204S, Axxora) at 37 °C for 10 min (in a 1:10 ratio, as used previously)⁷⁷. The mixture was passed through MicroSpin G-25 column to remove the excess of mant-GDP. To measure mant-GDP dissociation from RF3, equal volumes (60 μl each) of the reactants (0.6 μM mant-RF3 in one syringe, and 2 mM ’cold’ GDP with or without release complex in the second syringe) in the previously used buffer⁷⁷, containing 20 mM Hepes-KOH pH 7.5, 100 mM KCl, 7 mM MgCl₂ and 1 mM β-mercaptoethanol, were mixed rapidly. Release complexes were formed as follows: 1 μM E. coli 70S ribosomes, 4 μM 21M-UAA (Met-Stop) mRNA and 4 μM tRNA^fMet were incubated in 2× buffer for 10 min at 37 °C, then 10 μM RF1 (WT or the E18A mutant) was added and incubated for 5 min. The final concentrations of all components were: 0.3 μM mant-RF3 and 1 mM GDP (first syringe); and 0.5 μM 70S, 2 μM 21M-UAA mRNA, 2 μM tRNA^fMet, and 5 μM RF1 (second syringe). The rate of GDP exchange was monitored for 100 s at 37 °C by measuring mant-GDP fluorescence after passing through a 400 nm cutoff filter. The fluorescence traces (Fig. 2g,h) were fitted with a single exponential function using GraphPad Prism 8 and represent the average of five to seven traces over two independent experiments. The apparent rates of GDP dissociation from RF3 are summarized in the Supplementary Table 1.

Sample preparation and cryo-EM data acquisition

The ribosome release complex RC–apo-RF3–RF1 was prepared by incubating 2 μM E. coli 70S:ΔL9 ribosomes, 8 μM 21F-UAA (Phe-Stop) mRNA and 10 μM tRNA^Phe in 1× ribosome buffer (5 mM Tris-HCl pH 7.4, 60 mM NH₄Cl, 10 mM MgCl₂, 6 mM β-mercaptoethanol) at 37 °C for 10 min. Then, RF1 was added to a final concentration of 16 μM, and samples were incubated for 5 min. Finally, RF3 was added to a final concentration of 30 μM, and samples were incubated for an additional 5 min at 37 °C. For the release complex with β, γ-methyleneguanosine 5′-triphosphate (GDPCP, Sigma), RF3 was preincubated with 2 mM GDPCP for 15 min and added to the mixture and incubated for an additional 5 min at 37 °C to form RC–RF3-GDPCP–RF1.

Quantifoil R2/1 gold 200 mesh grids (Electron Microscopy Sciences) were pre-cleaned for 30 s in an H₂O₂ atmosphere using the Solarus 950 plasma cleaner (Gatan) to make them hydrophilic. The sample (4 μl) containing 2 μM E. coli 70S:ΔL9-RC was applied onto grids, blotted in 85% humidity at room temperature (22 °C) for 18 s and plunge-frozen in liquid-nitrogen-cooled ethane using a Leica EM GP2 cryo-plunger. Grids were transferred into a Titan Krios G3i electron microscope (Thermo Fisher Scientific) operating at 300 keV and equipped with a Falcon 3 direct electron detector camera (Thermo Fisher Scientific). The image stacks (movies) were acquired in the fast mode (linear) with a pixel size of 0.85 Å per pixel. Data were collected using the EPU software (Thermo Fisher Scientific) setup to record image stacks with 40 frames, with a total accumulated dose of 40 e⁻ per Ų per stack (RC–apo-RF3–RF1 dataset) and 40.44 e⁻ per Ų per stack (RC–RF3-GDPCP–RF1 dataset). A total of 10,284 (RC–apo-RF3–RF1) and 9,810 (RC–RF3-GDPCP–RF1) image stacks were collected, with a defocus range of −1 to −2.5 μm. The statistics of data acquisition are summarized in Tables 1 and 2.

Cryo-EM data processing

Data processing was done in cryoSPARC v.3.3.2 (ref. 78). The image stacks were imported into cryoSPARC and gain-corrected. Image frames (fractions) were motion-corrected with dose-weighting using the patch motion correction, and patch contrast transfer function (CTF) estimation was performed on the motion-corrected micrographs. On the basis of relative ice thickness, CTF fit, length and curvature of motion trajectories, 8,923 micrographs (apo-RF3 dataset) and 9,390 micrographs (RF3–GDPCP dataset) were selected for further processing (Extended Data Figs. 1 and 3).

For the dataset with no exogenous nucleotide (apo-RF3 dataset), the circular ‘blob’ picker in cryoSPARC picked 920,422 particles, which were filtered on the basis of defocus-adjusted power and pick scores to 871,635 particles. Particles were subjected to two rounds of reference-free two-dimensional (2D) classification. After discarding bad particles, 492,815 particles were selected from 2D classification and used to generate ab initio volumes with five groups. Four of the five groups generated maps resembling the ribosome. Using ‘heterogeneous refinement’ in cryoSPARC, the particles were further classified into four ribosome class averages. One class average did not yield anything meaningful and was considered not specimen-related (8,134 particles), and the other class average contained 50S subunits (83,554 particles; class average III), which were both discarded from further processing. The remaining two class averages represented non-rotated 70S ribosomes (326,675 particles; class average I) and rotated 70S ribosomes (74,452 particles; class average II). To further classify particles in class average I, we performed focused 3D variability analysis with a mask around RF3 and RF1. This process allowed the removal of 65,041 ‘bad’ particles, and further sorting of class average I particles into four subclasses. Two subclasses contained the rotated (12,965 particles) and non-rotated (40,336 particles) particles, with both classes lacking any bound factor. The remaining two subclasses represented non-rotated 70S ribosomes either bound to only RF1 (127,567 particles) or both RF3 and RF1 (80,766 particles). Particles from those two subclasses were re-extracted to full size (512 ×512 pixel box), and non-uniform and CTF refinement in cryoSPARC yielded reconstructions with nominal resolutions of 3.0 Å for structure I-A (RF1 only) and 3.1 Å for structure I-B (RF3-RF1) (Extended Data Fig. 1). Particles in class average II were similarly sorted using 3D variability analysis with a mask centered around RF3. After discarding 13,116 ‘bad’ particles and 28,078 particles representing rotated ribosomes with no bound factor, 33,348 particles remained, representing rotated ribosomes associated with RF3 and tRNA in the p/E hybrid state of binding. Re-extraction of the particles to full size (512 × 512 pixel box) followed by non-uniform and CTF refinement in cryoSPARC, yielded a ribosome reconstruction with an overall resolution of 3.2 Å for structure I-C (Extended Data Figs. 1 and 2).

The dataset with GDPCP was processed similarly. The circular ‘blob’ picker in cryoSPARC picked 953,653 particles, which were filtered on the basis of defocus-adjusted power and pick scores to 872,066 particles. Two rounds of reference-free two-dimensional (2D) classification selected 585,303 particles, which were used to generate ab initio volumes with five groups. We used ‘heterogeneous refinement’ in cryoSPARC to classify the particles into five ribosome class averages. This approach discarded 8,279 ‘bad’ particles, and 56,366 particles were 50S ribosomal subunits (class average III) and also discarded. The class average I, composed of two groups from the ‘heterogenous refinement’ job, represents non-rotated 70S ribosomes that seem to be bound to both RF3 and RF1 (334,113 particles). Class average II is from one ‘heterogenous refinement’ group in which the 70S ribosomes are rotated and bound to only RF3 (186,545 particles). The particles in both class averages I and II were further classified using focused 3D variability analysis with a mask around RF3 and RF1. In class average I, 30,189 particles were noisy and discarded. Similarly, 58,392 particles representing non-rotated 70S ribosomes with no bound factor, 8,460 particles representing non-rotated ribosomes bound to only RF3 and 59,038 particles representing non-rotated 70S ribosomes with bound RF1 and weak density for RF3 were not further processed. Particles from two subclasses, one containing non-rotated 70S ribosomes bound to RF1 and with solid density for RF3 (31,256 particles), and the other containing non-rotated 70S ribosomes bound to only RF1 (146,778 particles) were further re-extracted to full size (512 × 512 pixel box), and non-uniform and CTF refinement in cryoSPARC yielded reconstructions with nominal resolutions of 3.1 Å for structure II-A (RF3–GDPCP–RF1) and 2.9 Å for structure II-D (RF1 only) (Extended Data Fig. 3). Within class average II (rotated 70S), focused 3D variability analysis with a mask around RF3 yielded two subclasses, allowing 13,351 ‘bad’ particles to be discarded and visualization of RF3 bound to the rotated 70S ribosome in two conformations, in a state not fully docked to the SRL on the 50S subunit (56,202 particles) and in the activated state (116,992 particles). Re-extraction of the particles to full size (512 × 512 pixel box) and non-uniform and CTF refinement in cryoSPARC yielded reconstructions with nominal resolutions of 3.1 Å for structure II-B (RF3–GDPCP, not activated) and 3.0 Å for structure II-C (RF3–GDPCP, activated) (Extended Data Figs. 3 and 4).

For structures I-B, I-C, II-A, II-B and II-C, we performed local refinement with signal subtraction using a soft mask around RF3, which improved the overall resolution of apo-RF3 to 3.7 Å (I-B), of RF3–ppGpp on the rotated 70S ribosome to 3.6 Å (I-C) (Extended Data Fig. 2), of RF3–GDPCP on the non-rotated 70S ribosome to 3.8 Å (II-A) and of RF–GDPDP (not activated) (II-B) and RF3–GDPCP (activated) (II-C) to 3.6 Å and 3.4 Å, respectively (Extended Data Fig. 4).

Model building and refinement

The models were assembled from individual parts. The 30S and 50S subunits were taken from the high-resolution structure of the E. coli 70S ribosome (PDB 7K00)⁷⁹ and rigid-body docked into the 3.1-Å cryo-EM map of structure I-B using UCSF Chimera v.1.14 (ref. 80). The tRNA^Phe was taken from PDB: 6XHW ref. 81; RF3 was taken from PDB: 6GXP ref. 3; and RF1 was taken from PDB: 6GXN ref. 3. Structures were rigid-body fitted into the EM density and adjusted in COOT⁸². Regions of the ribosome, such as the uL1 and uL11-stalk elements, were individually adjusted and modeled into the density. A model of the E. coli 70S ribosome was then rigid-body fitted into the maps of structures I-A, II-A and II-D, and then into the maps of structures II-B and II-C by rotating the 30S subunit into the density. In structures II-B and II-C, tRNA^Phe was rigid-body fitted and adjusted in the p/E hybrid state.

For assistance during the modeling process, maps generated from the combination of signal subtraction and local refinement were used to model RF3 in structures I-B, I-C, II-A, II-B and II-C (Supplementary Table 4). Individual domains of RF3 were rigid-body fitted and manually adjusted into the density. Domain linkers and regions of visible discrepancies were manually built in COOT, guided by the sequence and density. The complete models of RF3 were real-space refined in PHENIX (with energy minimization, ADP refinement together with Ramachandran and secondary structure restraints) into the respective EM density map obtained after focused refinement. Composite maps combining the EM maps obtained after focused refinement of RF3 and the 70S ribosome were generated for structures I-B, I-C, II-A, II-B and II-C (Supplementary Table 4). The complete models of the 70S ribosome containing tRNA^Phe, RF1 and RF3 and ordered solvent were real-space refined into their respective composite EM map for five cycles in PHENIX v.1.19.2 (ref. 83), including global energy minimization and group ADP refinement strategies along with base pair restraints for rRNA and tRNA^Phe, together with Ramachandran and secondary structure restraints. The resulting models were validated using the comprehensive validation tool for cryo-EM and MolProbity⁸⁴ in PHENIX⁸⁵ (Tables 1 and 2). The angle and displacement measurements of RF3 domains were calculated in PyMOL (The PyMOL Molecular Graphics System, v.2.1.0 Schrödinger) and are listed in Supplementary Tables 2 and 3.

Extended Data

Extended Data Fig. 1. Cryo-EM data processing and particle classification workflow (dataset I with RF3-GDP and no exogenous nucleotide).

Scheme of the data processing workflow. All steps were performed in cryoSPARC 3.3.2⁷⁸. 10,284 micrographs were collected, of which 8,923 were selected for further processing. After two rounds of reference-free 2D classification, the selected particles were used to generate ab-initio volumes. Particles from the 3D volumes that appeared to be the 70S ribosome based on their shape and size were separated according to their ratcheted state using heterogeneous refinement, yielding three main classes of particles (class averages I, II, and III). Among the non-ratcheted 326,675 particles (class average I), focused 3D variability analysis around RF3 and RF1 was utilized to separate 80,766 particles for containing solid density for RF3 and RF1 or for RF1 only (127,567 particles). The process discarded 40,336 particles containing non-rotated ribosomes and 12,965 particles of rotated ribosomes, both of which were not bound to any factor. The remaining 65,041 particles yielded highly noisy reconstructions and were not further analyzed. Non-uniform and CTF refinement of particles with RF1 and RF3, or with RF1 only, produced structures I-B and I-A, respectively. Local refinement was combined with particle subtraction to produce a higher-quality map of RF3 and RF1 in structure I-B. Similarly, the rotated ribosome particles (74,452 particles in class average II) were separated by focused 3D variability analysis around RF3, producing two defined volumes, one containing rotated ribosome with no factor bound (28,078 particles) and one with rotated ribosome bound to RF3 (33,348 particles). Non-uniform and CTF refinement of the latter volume yielded structure I-C. Local refinement combined with particle subtraction produced a higher-quality map of RF3 in structure I-C.

Extended Data Fig. 2. Local resolution estimation and Fourier Shell Correlation (FSC) validation (dataset I with RF3-GDP and no exogenous nucleotide).

a, Local resolution heat maps on slices of density for structures I-A, I-B, and I-C shown in the range of 2.5 – 6.5 Å resolution, calculated with cryoSPARC 3.3.2 implementation of BlocRes⁸⁷. b, Gold-standard FSC curves of each half-map (red), using a ‘soft mask’ excluding solvent and model-map (blue), are plotted across resolution. Map and model validation was performed in PHENIX 1.19.2⁸⁵. For structures I-B and I-C, for which composite maps were generated from EM maps of RF3-RF1 (I-B) or RF3 (I-C) obtained after focused refinement and the 70S ribosome, FSC curves for the composite EM map and the model vs composite map fit are shown. c, Local resolution maps shown in the range of 3.0 – 7.0 Å resolution and FSC curves of half-maps (masked with a ‘soft mask’) for RF3-RF1 (I-B) and for RF3 (I-C).

Extended Data Fig. 3. Cryo-EM data processing and particle classification workflow (dataset II with GDPCP).

All steps were performed in cryoSPARC 3.3.2⁷⁸. 9,810 micrographs were collected, of which 9,390 were selected for further processing. After two rounds of reference-free 2D classification, the selected particles were used to generate ab-initio volumes. Particles from the 3D volumes that appeared to be the 70S ribosome based on their shape and size were separated according to their ratcheted state using heterogeneous refinement, yielding three main classes of particles (class averages I, II, and III). Among the non-ratcheted 334,113 particles (class average I), focused 3D variability analysis around RF3 and RF1 was utilized to separate 31,256 particles for containing solid density for RF3 and RF1 or for RF1 only (146,778 particles). The process discarded 59,038 particles containing non-rotated ribosomes with weak density for RF3 and RF1, 58,392 particles containing non-rotated ribosomes with no factor bound, 8,460 particles of non-rotated ribosomes with only RF3 bound, and 30,189 particles that yielded noisy reconstructions. Non-uniform and CTF refinement of particles with RF1 and RF3, or with RF1 only, produced structures II-A and II-D, respectively. Local refinement was combined with particle subtraction to produce a higher-quality map of RF3 and RF1 in structure II-A. Similarly, the rotated ribosome particles (186,545 particles in class average II) were separated by focused 3D variability analysis around RF3, producing two defined volumes (56,202 and 116,992 particles), both of which containing RF3 in a distinct binding position. Non-uniform and CTF refinement of the latter two volumes yielded structures II-B and II-C. Local refinement combined with particle subtraction produced higher-quality maps of RF3 in structures II-B and II-C.

Extended Data Fig. 4. Local resolution estimation and Fourier Shell Correlation (FSC) validation (dataset II with GDPCP).

a, Local resolution heat maps on slices of density for structures II-A, II-B, II-C, and II-D shown in the range of 2.5 – 6.5 Å resolution, calculated with cryoSPARC 3.3.2 implementation of BlocRes⁸⁷. b, Gold-standard FSC curves of each half-map (red), using a ‘soft mask’ excluding solvent and model-map (blue), are plotted across resolution. Map and model validation was performed in PHENIX 1.19.2⁸⁵. For structures II-A, II-B, and II-C, for which composite maps were generated from EM maps of RF3-RF1 (II-A), or RF3 (II-B and II-C) obtained after focused refinement and the 70S ribosome, FSC curves for the composite EM map and the model vs composite map fit are shown. c, Local resolution maps shown in the range of 3.0 – 7.0 Å resolution and FSC curves of half-maps (masked with a ‘soft mask’) for RF3-RF1 (II-A) and RF3 (II-B and II-C).

Extended Data Fig. 5. Interface between domain III of apo-RF3 and domain I of RF1 in structure I-B.

a, In apo-RF3, the residues for which mutation to alanine reduces the rate of GDP exchange are shown with yellow spheres. These residues form the interface with RF1. Mutation of the residues shown with light blue spheres has a mild effect on the rate of GDP exchange. Note that residue E516 does not interact with RF1. b, The C_α and C_β atoms of residue N510 contribute to the interface with RF1, while the hydrophilic groups of N510 are not interacting with RF1.

Extended Data Fig. 6. Binding of RF3 to the 70S ribosome termination complex stabilizes domain I of RF1.

a, EM density of structure I-B bound to apo-RF3 and RF1 relative to that of structure I-A bound to RF1 only. b, Clear density for ribosomal proteins uL10 and uL11 in structure I-B bound to apo-RF3 and RF1 (mesh). In structure I-A, these proteins are more flexible as seen from the lack of density (gray surface). c, EM maps of RF1 in structures I-B (mesh) and I-A (gray surface). The presence of apo-RF3 in structure I-B stabilizes domain I of RF1. d, Additional density for the C-terminal domain of ribosomal protein bL12 is observed proximal to the α2-turn-α3 in RF1 and the G’ sub-domain in RF3. e, Relative to structure I-A, the tips of helices H43-H44 of the uL11-ribosomal stalk in structure I-B move toward domain I of RF1 by more than 3 Å. The α1-turn-α2 region in domain I of RF1 forms π-π-stacking interactions with the proline-rich 3₁₀-helix in the N-terminus of protein uL11, while nucleotides A1067 (H43) and A1095 (H44) stack along helix α2 of RF1.

Extended Data Fig. 7. Relative orientation of RF3-GDPCP in structure II-A versus that from a prior study (PDB 6GWT)³.

(a), and potential steric clash between RF3-GDPCP in structure II-C with RF1 (taken from structure II-A) (b).

a, The structures are aligned based on the 16S rRNA. In the previous study, RF3-GDPCP (gray) is bound to a non-rotated ribosome with RF1 trapped by the antimicrobial peptide apidaecin (PDB 6GWT, state I)³. The orientation of domain III (gray RF3) is such that the closest distance to domain I of RF1 is ~9 Å, precluding the formation of any interaction³. b, The structures are aligned based on the 23S rRNA. On the rotated ribosome, RF3-GDPCP rotates by 27° as a whole between structures II-B (gray) and II-C (colored by domain) to dock against the sarcin-ricin loop (SRL) in the 50S subunit. In the GTP-activated state (II-C), domain III of RF3-GDPCP would clash with domain I of RF1 (red arrow), suggesting that complete accommodation of RF3 on the 50S subunit only occurs following dissociation of the class-I release factor.

Extended Data Fig. 8. Relative orientation of RF3 with GDPCP or GDPNP in previous structures versus that in structure II-B.

The structures are aligned based on the 23S rRNA. a, The conformation of RF3-GDPCP in structure II-B on the rotated ribosome (colored by domain) is highly similar to that of RF3-GDPCP reported in a previous study with the antimicrobial peptide apidaecin (state IV; rotated ribosome) (PDB 6GXO, gray)³. b, Same as in panel (a) relative to previous crystal structures of RF3-GDPNP bound to the rotated E. coli 70S ribosome (PDB 4V85)⁶ or RF3-GDPCP bound to the rotated Thermus thermophilus 70S ribosome (PDB 4V8O)⁷. c-f, Close-up views of the sarcin-ricin loop (SRL) and domain I of RF3 from (a) and (b). The catalytic histidine 92 in switch 2 (sw2) is located 9 to 13 Å away from the non-bridging oxygen of A2662 in the SRL, and RF3 is therefore in a GTP-inactive state on the ribosome. Switch 1 (sw1) is orange.

Extended Data Fig. 9. Structure of the ribosome bound to RF3-ppGpp.

a, Relative to the non-rotated ribosome in structure II-A (light gray), the ribosome has the rotated conformation in structure I-C bound to RF3-ppGpp (beige), similar to the ribosome bound to RF3-GDPCP in structure II-B (black). The structures are aligned by the 23S rRNA. b, Close-up view of ribosome-free RF3-ppGpp (PDB 3VR1, beige)²⁶ and RF3-GDP (PDB 3VQT ref. 26, black; PDB 2H5E ref. 10, gray) superimposed with ribosome-bound RF3-ppGpp in structure I-C (colored domains). c, Orthogonal view of (a), showing that RF3-ppGpp is bound proximal to the sarcin-ricin loop (SRL, light green) in the 50S subunit. d, Cryo-EM map of the nucleotide-binding pocket. The density of ppGpp is shown as teal mesh, and that of the Mg²⁺ ions is shown as black mesh. e, The catalytic histidine 92 in switch 2 (sw2) is located within interaction distance from the β-phosphate non-bridging oxygen. The δ-phosphate oxygen of ppGpp forms a Mg²⁺-mediated hydrogen bond with A2662 in the SRL and the ε-phosphate interacts with a Mg²⁺ ion.

Supplementary Material

Supplementary_Video_1

Supplementary Video 1

Structural basis for the GDP-to-GTP exchange in RF3 catalyzed by the ribosomal termination complex. The animation is a morph between structures I-A, I-B, and II-A. The comparison of apo-RF3 in structure I-B with the structure of ribosome-free RF3-GDP (PDB 2H5E)¹⁰ illustrates the conformational changes in RF3 that lead to remodeling of the sw2-helix and of the phosphate-binding loop (P-loop). The flip of the carbonyl oxygen at P22 encroaches into the nucleotide-binding pocket and would collide with the β-phosphate of GDP, promoting its dissociation from RF3. Further comparison between apo-RF3 (I-B) and RF3-GDPCP in structure II-A shows the rotation of domain I of RF3 toward the 50S subunit SRL. Note the absence of interaction between RF3 and the 50S subunit in structures I-B and II-A.

Supplementary_Video_2

Supplementary Video 2

Dissociation of RF1 catalyzed by RF3–GDPCP and activation of RF3 for GTP hydrolysis. The animation is a morph between structures II-A, II-B, and II-C. The movement of domain I in RF3 toward the SRL is facilitated by the ratchet-like rotation of the 30S subunit. In structure II-B, the catalytic histidine H92 in domain I of RF3 remains 11 Å away from the SRL, indicating that RF3–GDPCP in structure II-B is in a non-activated state. The rolling-like motion of the 30S subunit and the rotation of RF3–GDPCP in structure II-C bring domain I of RF3 proximal to the SRL. The catalytic H92 is within interaction distance from A2662 in the SRL, suggesting that RF3 is in the activated state primed for GTP hydrolysis.

Supplementary_Data_1

Supplementary Data 1

List and sequences of DNA oligonucleotides used in this study.

Supplementary_Data_2

Supplementary Data 2

Representative micrograph for apo-RF3 bound to the 70S ribosome termination complex.

Supplementary_Data_3

Supplementary Data 3

Representative micrograph for RF3–GDPCP bound to the 70S ribosome termination complex.

Supplementary_Data_4

Supplementary Data 4

Uncropped gels for Supplementary Fig. 1a,b,d.

Data availability

The atomic coordinates have been deposited in the RCSB Protein Data Bank (PDB) under accession codes 8FZE (structure I-A; 70S–RF1), 8FZD (composite I-B; 70S–apo-RF3–RF1), 8FZF (composite I-C; 70S–RF3–ppGpp), 8FZG (composite II-A; 70S–RF3–GDPCP–RF1), 8FZI (composite II-B; 70S–RF3–GDPCP, non-activated conformation), 8FZJ (composite II-C; 70S–RF3–GDPCP, GTP-activated conformation) and 8FZH (structure II-D; 70S–RF1). The cryo-EM maps have been deposited in the Electron Microscopy Data Bank (EMDB) under accession codes EMD-29621 (structure I-A; 70S–RF1), EMD-29618 (structure I-B; 70S–apo-RF3–RF1 initial map), EMD-29619 (I-B: apo-RF3; from focused refinement), EMD-29620 (structure I-B; 70S–apo-RF3–RF1 composite), EMD-29622 (structure I-C; 70S–RF3–ppGpp initial map), EMD-29623 (I-C: RF3–ppGpp; from focused refinement), EMD-29624 (structure I-C; 70S–RF3–ppGpp composite), EMD-29625 (structure II-A; 70S–RF3–GDPCP–RF1 initial map), EMD-29626 (II-A: RF3–GDPCP; from focused refinement), EMD-29627 (structure II-A; 70S–RF3–GDPCP–RF1 composite), EMD-29629 (structure II-B; 70S–RF3–GDPCP initial map, non-activated conformation), EMD-29630 (II-B: RF3–GDPCP; from focused refinement, non-activated conformation), EMD-29631 (structure II-B; 70S–RF3–GDPCP composite, non-activated conformation), EMD-29632 (structure II-C; 70S–RF3–GDPCP initial map, GTP-activated conformation), EMD-29633 (II-C: RF3–GDPCP; from focused refinement, GTP-activated conformation), EMD-29634 (structure II-C; 70S–RF3–GDPCP composite, GTP-activated conformation), and EMD-29628 (structure II-D; 70S–RF1).