Dynamics of release factor recycling during translation termination in bacteria

Abstract In bacteria, release of newly synthesized proteins from ribosomes during translation termination is catalyzed by class-I release factors (RFs) RF1 or RF2, reading UAA and UAG or UAA and UGA codons, respectively. Class-I RFs are recycled from the post-termination ribosome by a class-II RF, the GTPase RF3, which accelerates ribosome intersubunit rotation and class-I RF dissociation. How conformational states of the ribosome are coupled to the binding and dissociation of the RFs remains unclear and the importance of ribosome-catalyzed guanine nucleotide exchange on RF3 for RF3 recycling in vivo has been disputed. Here, we profile these molecular events using a single-molecule fluorescence assay to clarify the timings of RF3 binding and ribosome intersubunit rotation that trigger class-I RF dissociation, GTP hydrolysis, and RF3 dissociation. These findings in conjunction with quantitative modeling of intracellular termination flows reveal rapid ribosome-dependent guanine nucleotide exchange to be crucial for RF3 action in vivo.


INTRODUCTION
Termination of cellular protein synthesis begins when an mRNA stop codon is translocated into the aminoacyl-tRNA binding site (A site) of the ribosome. This triggers Asite binding of a class-I release factor (RF), which promotes hydrolytic release of the nascent polypeptide chain (1)(2)(3). In bacteria, class-I RFs RF1 and RF2 decode stop codons U AG or U AA and UGA or U AA, respecti v ely. Stop-codon recognition leads to a conformational change of the class-I RF, which brings its GGQ motif ( 4 ) into the ribosomal peptidyl tr ansfer ase center (PTC) ( 5 ) to promote r apid hydrolysis of the ester bond linking the nascent peptide chain to the tRNA at the peptidyl-tRNA binding site (P site) (6)(7)(8)(9)(10)(11)(12).
In vivo concentrations of RF1 and RF2 are substoichiometric to ribosome concentration ( 13 ), meaning that rapid class-I RF cycling may be important for hastening termination in the living cell ( 14 ). Dissociation of class-I RFs from the post-termination ribosome ( 15 , 16 ) is accelerated by a class-II RF, the GTPase RF3, in a GTP -dependent manner ( 14 ). RF3 action also minimizes the inhibitory effect of post-termination class-I RF rebinding on ribosomal recycling by RRF and EF-G ( 15 , 17 , 18 ). RF3 deficiency decreases growth rate, increases stop-codon r eadthrough fr equency and induces cold sensitivity of Esc heric hia coli cells ( 19 , 20 ). These phenotypes may arise from lower termination efficiency caused by slower recycling of RF1 and RF2 or lack of sense error correction by RF3-dependent post-peptidyl transfer control ( 21 ). Bacterial sensitivity to class-I RF dissocia tion ra te reduction makes this step a viable target of antimicrobial agents, such as Apidaecin 137 (Api137), which stabilize RF1 on the posttermination ribosome ( 22 ).
The time for exchange of GDP to GTP on RF3 free in solution is 30s and is dominated by GDP dissociation from RF3 ( 23 ). The exchange is much faster for RF3 in complex with the class-I RF-bound post-termination ribosome ( 16 , 23 ), making the ribosome a nucleotide exchange factor for RF3. From the lower binding affinity of GTP than GDP to RF3 in solution and the r equir ement for rapid RF3 recycling in bacterial termination, it was suggested that in the living cell it is RF3 ·GDP that normally enters the pretermination ribosome. Then GDP dissociates ra pidl y from RF3 which allows for rapid binding of solution-phase GTP into ribosome-bound apo-RF3 ( 12 , 23 ).
The proposition that the ribosome is a guanine nucleotide exchange factor for RF3 has been confirmed ( 16 , 24-26 ).
Howe v er, fluorescence e xperiments estimated just a fourfold higher affinity of RF3 to GDP than GTP, which, in conjunction with the estima ted [GTP] / [GDP] ra tio of 10 in vivo ( 27 ) and the assumption that RF3 ·GDP and RF3 ·GTP are equilibra ted of f the ribosome in vivo , led to the proposal that in the cell cytoplasm RF3 must be mainly in the GTP form and that ribosome-catalyzed guanine nucleotide exchange is redundant for RF3 function ( 16 ).
Structural and dynamic data suggest that the mechanism of class-I RF r ecycling between fr ee and ribosome-bound states involves changes in the global ribosome conformation. The 70S ribosome has two global conformations that differ by counterclockwise rotation of 6-12 degrees of the small with respect to the large subunit, here called the nonr otated and r otated ribosomal states, respecti v ely. Termination complexes bound to either RF1 or RF2 were observed in the non-rota ted sta te, wher eas complex es bound to RF3 ·GDPNP were observed in the rotated state ( 24 , 28 ), consistent with the hypothesis that RF3-induced intersubunit rotation accelerates the dissociation of class-I RFs, as further supported by cryo-EM structures of ribosomal termination complex with RF1 and RF3 ( 29 ). Thus, there is consensus regarding the importance of intersubunit rotation during termination, but its timing in relation to release factor binding and dissociation e v ents and GTP hydrolysis on RF3 remains disputed. From F örster Resonance Energy Transfer (FRET) studies of RF1 and RF3 dissociation kinetics it was proposed that GTP hydrolysis accelerates sequential dissociation of first RF3 and then RF1 ( 30 ). From another study using single-molecule FRET (sm-FRET) it was suggested that GTP hydrolysis is not r equir ed for either RF1 / RF2 or RF3 dissociation and that the order RF1 / RF2 and RF3 dissociation e v ents is stochastic rather than sequential ( 24 ).
To resolve these ambiguities regarding the interplay between nucleotide state and ribosomal rotation in the RF3 mechanism, we applied here bulk and real-time singlemolecule kinetics to determine the timings of 70S rotation, class-I RF and class-II RF dissociation e v ents during transla tion termina tion. We then used quantita ti v e modeling of protein synthesis in growing E. coli cells to answer the question if ribosome-dependent acceleration of guanine nucleotide exchange on RF3 ( 16 , 23 ) is crucial for RF3 action. The combined approach of detailed kinetics experiments with global termination modeling have pro vided no vel insights in the termination mechanism of bacteria. Our findings highlight its dynamic nature and the importance of using the cellular context to assess the functional meaning of in vitro data.

Single-molecule assays
Reagents and buffers for single-molecule experiment. Esc heric hia coli ribosomal subunits used in all single-molecule experiments were purified as described before ( 31 ). To label the ribosomal subunits specifically with fluorescent dyes, hairpin loop extensions were introduced into phylogenetically variab le, surface-accessib le loops of the E. coli 16S rRNA in helix 44 and 23S rRNA in helix 101 using previously described site-directed mutagenesis ( 32 ). The 70S ribosomes were purified from SQ380 cells expressing these mutant ribosomes, and the 30S and 50S subunits were prepared from dissociated 70S particles using previously described protocols ( 31 ).
IF2, EF-Tu, EF-G, EF-Ts, RRF and ribosomal protein S1 from E. coli were purified from ov ere xpressing strains as previously described ( 18 , 31 ). fMet-tRNA fMet , Lys-tRNA Lys and Phe-tRNA Phe were charged and purified according to published protocols ( 33 , 34 ). The mRNAs used, MF-UAA and 6FK-UAA, contains a 5 -biotin followed by a 5 -UTR and Shine-Dalgarno sequence deri v ed from gene 32 of the T4 phage upstream of the AUG start codon. For both mR-NAs, ther e ar e four spacer Phe codons downstr eam of the UAA stop codons. Both mRNAs were chemically synthesized by Dharmacon.
Cy5.5-labeled RF1 and RF2 were generated from singlecysteine variants of the proteins which were purified with previously described protocol ( 18 ). The purified proteins were incubated with 30-fold excess of the Cy5.5-maleimide dye (Lumiprobe) for 24 h at 4 • C. Removal of free dye and storage was done as previously described ( 18 ).
Wild-type RF3 was purified by ov ere xpressing them in E. coli BL21(DE3) cells transformed with pET-His6-MBP-Asn10-TEV-LIC cloning v ector (QB3-Ber keley Macrolab) containing an N-terminal six-histidine (6xHis) affinity tag, a maltose binding protein (MBP), linker of ten asparagines (N10), TEV protease cleavage site, followed by the RF3 gene from E. coli MG1655 K12 strain. Cells were lysed using sonication, and the lysate clarified by centrifugation was loaded onto a 5-ml HiTrap Ni 2+ column (GE Healthcare). The fractions containing the protein were dialyzed in the presence of TEV protease to cleave the N-terminal 6xHis-MBP-N10 tag. After flowing this cleaved protein through a column containing 1 ml Ni-NTA Agarose resin (Qiagen), the protein was then purified on a size-exclusion column (Super de x 200 26 / 60, GE Healthcare).
Cy5-RF3 was generated by enzymatically functionalizing Cy5-labeled Coenzyme A (CoA) on RF3 with an Nterminal ybbR peptide tag (DSLEFIASKLA) using Sfp synthase. Sfp synthase was purified using previously described protocol ( 35 ). Cy5-labeled CoA was generated using previously described protocol ( 36 ). ybbR-tagged RF3 was purified similarly to the wild-type RF3 with the insertion of the ybbR sequence upstream of the RF3 gene in pET-His6-MBP-Asn10-TEV-LIC vector. Purified ybbR-RF3 was incubated with 6-fold excess of Cy5-CoA and 5:4 molar ratio to Sfp synthase at 37 • C for 1 h in buffer containing 50 mM HEPES-KOH, pH 7.5, 100 mM NaCl, 10 mM MgCl 2 and 1 mM DTT. After the reaction, the mixture was passed through a column containing 1 ml Ni-NTA Agarose resin (Qiagen) to remove Sfp synthase and two 10DG desalting gravity columns (Bio-Rad) to remove free dye. This sample was then finally loaded onto a size-exclusion column (Super de x 200 10 / 300 GL, GE Healthcare) to remove any r esidual fr ee dye / enzyme. The labeled RF3 was stor ed in stora ge b uffer (20 mM Tris-HCl pH 7.5 a t 25 • C , 100 mM KCl, 2.5 mM MgCl 2 , 1 mM DTT, 50% w / w glycerol) a t −20 • C .
All single-molecule experiments were conducted in a Trisbased polymix buffer consisting of 50 mM Tris-acetate (pH 7.5), 100 mM potassium chloride, 5 mM ammonium aceta te, 0.5 mM calcium aceta te, 5 mM magnesium acetate, 0.5 mM EDTA, 5 mM putrescine-HCl and 1 mM spermidine. Prior to the single-molecule experiments, the purified 30S and 50S ribosomal subunits (final concentration 1 M) wer e mix ed in 1:1 ratio with the dye-labeled DNA oligonucleotides complementary to the mutant ribosome hairpin extensions ( 31 , 32 ) at 37 • C for 10 min and then at 30 • C for 20 min in the Tris-based polymix buffer system. The 30S subunit was labeled with 5'-Cy3B-labeled DNA and the 50S subunit was labeled with 3'-BHQ-2-labeled DNA.
Single-molecule intersubunit FRET / factor binding e xperiments . The 30S pre-initia tion complex es (PICs) wer e formed as described by incubating the following at 37 • C for 5 min: 0.25 M Cy3B-30S, pre-incubated with stoichiometric S1, 1 M IF2, 1 M fMet-tRNA fMet , 1 M mRNA, and 4 mM GTP to form 30S PICs in the polymix buffer ( 37 ). Before use, we pre-incubate a SMRT Cell v3 from Pacific Biosciences (Menlo Park, CA, USA), a zero-mode waveguide (ZMW) chip, with 0.2% w / w Tween 20 in 50 mM Trisacetate pH 7.5 and 50 mM KCl at room temperature for 10 min. After washing the chip in cell-washing buffer (50 mM Tris-acetate pH 7.5, 100 mM potassium chloride, 5 mM ammonium acetate, 0.5 mM calcium acetate, 5 mM magnesium acetate and 0.5 mM EDTA), the chip was incubated with a 1 mg / ml Neutravidin solution in 50 mM Tris-acetate pH 7.5 and 50 mM KCl at room temperature for 5 min. The cell was then washed with cell-washing buffer again. The formed 30S PICs were diluted with our Tris-based polymix buffer containing 4 mM GTP down to 10 nM PIC concentration. The diluted PICs were then loaded into the SMRT cell at room temperature for 3 min to immobilize the 30S PICs into the ZMW wells. Any e xcessi v e unbound material w as w ashed aw ay with our Tris-based polymix buffer containing 4 mM GTP. The immobilized 30S PICs were immersed with 20 l of our Tris-based polymix buffer containing 4 mM GTP, 5 mg / mL BSA, 1 M blocking dsDNA oligonucleotide, 2.5 mM Trolox, and a PCA / PCD oxygen scavenging system (2.5 mM 3,4-dihydroxybenzoic acid and 250 nM protoca techua te deoxygenase ( 38 ).
Ternary complex es (T Cs) wer e formed between charged Phe-tRN A Phe or Lys-tRN A L ys E. coli tRNAs and EF-Tu(GTP) by incubating in bulk the tRNA, EF-Tu, EF-Ts, GTP and energy regeneration system (phosphoenolpyruvate and pyruvate kinase). The TCs were added to the deliv- PacBio RSII instrumentation and data analysis. Singlemolecule intersubunit FRET and factor occupancy experiments were conducted using a commercial PacBio RSII sequencer that was modified to allow the collection of single-molecule fluorescence intensities from individual zero-mode waveguide (ZMW) wells about 130 nm in diameter in four different dye channels corresponding to Cy3, Cy3.5, Cy5 and Cy5.5 fluorescence ( 39 ). The RSII sequencer has two lasers for dye excitation at 532 nm and 632 nm. In all the Cy5.5-RF1 and Cy5.5-RF2 experiments, data was collected at 10 frames / s (100 ms exposure time) for 6 min using energy flux settings of the green laser at 0.72 mW / mm 2 and red laser at 0.10 mW / mm 2 .
Data analyses for all experiments were conducted with MATLAB (MathWorks) scripts written in-house ( 39 ). ZMW wells containing fluorescently-labeled ribosomes were initially selected via an automated process based on fluor escence intensity, fluor escence lifetime and the changes in intensity. ZMWs with single producti v e ribosome complex es wer e then manually selected based on the expected changes in Cy3B fluorescence signal due to ribosome conformational changes and in Cy5 and Cy5.5 fluorescence signal due to RF occupancy changes. For most of the experiments, unless otherwise noted, producti v e comple x was detected by the sequential changes in Cy3B and Cy5.5 fluorescence as shown in Figure 1 B. Producti v e comple x in the pre-elongated ribosome experiments described in the previous section is defined by the burst of Cy5.5 fluorescence signal due to RF1 binding. All conformational and compositional states defined by fluorescence were assigned as previously described ( 39 ) based on a hidden Markov model based a pproach and visuall y corr ected. All lifetimes wer e plotted as cumulati v e distributions and fitted to single-, doub le-, or triple-exponential functions using curve-fitting tool on MATLAB.

Determination of rate constants from single-molecule data.
Unless specified otherwise, all rate constants ( k ) were calculated by fitting the cumulati v e distribution of the corresponding dwell times as functions of time ( t ). The cumulati v e distribution corresponds to the probability that an e v ent has occurred between times 0 and t, where the probability goes to one when t goes to infinity. In the simplest case where dissociation is determined by a single rate constant the cumulati v e distribution f ( t ) is a single e xponential: In the presence of only GTP, GDP or GDPCP the cumulati v e distributions of RF3 occupancy times on the nonrotated ribosome were all biphasic, and fitted to double exponential functions: where a 1 and a 2 are the amplitudes of contribution of the two rate constants k 1 and k 2 . In the presence of 50% GTP and 50% GDP the cumulati v e distributions of RF3 occupancy times on the non-rotated ribosome were fitted to triple-exponential functions: where a 1 , a 2 and a 3 are the amplitudes of contribution of the rate constants k 1 , k 2 and k 3 . We model the RF3 concentration, [RF3], dependence of mean class-I RF (class-I RF1 and RF2 are denoted here as RF i , where i = 1 or 2) dwell time τ i BD τ i BD , defined as the mean time interval from RF i binding to the ribosome till RF i dissociation from it, with two kinetic schemes, one for first-time RF i bound ('first-bound') and the other for RF i -r ebound ('r ebound') ribosomes. In the 'first-bound' case there is an initial activation step from ribosome state R I P with RF3-independent rate constant k i IA that yields a rota tion-activa ted ribosome complex R A P that can then undergo intersubunit rotation with rate constant k − iAR in the absence and k + iAR in the presence of ribosome-bound RF3: After intersubunit rotation, RF i release occurs without detectable delay ( < 100 ms, as defined by the time-resolution of the instrument) in the absence and with detectable delay with mean time in the presence of RF3 (see Results). With experimental support for a sampling mode of RF3 binding (see Results) we assume initial equilibration with the same equilibrium dissociation constant K 3 between RF3free and RF3-containing complexes R I P · RF i and R A P · RF i (see the scheme Eq. 4 ). The [RF3] dependence of the mean duration time in Eq. ( 4 ) for subunit rotation, τ i BR , counted from RF i -binding to appearance of rotated ribosome has then the form: so that Analo gousl y, according to the scheme in Eq. ( 4 ) the [RF3] dependence of the mean duration time τ i BD from RF ibinding to RF i -dissociation is gi v en by where the post-rotation mean-dissociation time τ i RD is gi v en by 5778 Nucleic Acids Research, 2023, Vol. 51, No. 11 The latter two expressions follow from the experimental observa tion tha t τ − i RD = 0 (see Results). In the 'rebound' case the ribosome is already activated at time zero and the scheme in Eq. ( 4 ) is replaced by In Eqs. ( 5 ) and ( 7 ), we set τ i I A equal to zero while the forms of Eqs. ( 8 ) and ( 9 ) are unchanged.
The algebraic expression in Eq. ( 7 ) was fit to the mean ribosome rota tion da ta shown in Figure 2 F and Figure S2D, with the k + iAR fixed to 3 s −1 based on the rate constant of RF3-bound ribosome rotation (k 3BR ) measured from the single-molecule data collected in the presence of Cy5-RF3 (see Results).

Intracellular kinetic flow modeling
Relation between bacterial population doubling time T d and aver ag e tr anslation cy c le time R . The ra te dM / dt of increase of the number M of peptide bonds per time unit in the cell population is gi v en by: , respecti v ely; they include times for initia tion, elonga tion ( τ E ), termination and recy cling. Accor dingly, each ratio τ E /τ X R in Eq. ( 11 ) is the fraction of ribosomes in elongation mode translating mRNAs with stop codon X. Introducing fractions f X R of ribosomes translating mRNAs with stop codons X and normalizing by the current number of peptide bonds, M , Eq. ( 11 ) can be re-written as: Here, N R is the total number of ribosomes in the cell population; τ R is the average translation cycle time per ribosome gi v en by: In the stead y sta te, the ra tio r = N R /M between total number of ribosomes ( N R ) and peptide bonds ( M ) in the cell population is constant. The increase in M (which is proportional to cell population biomass) is gi v en by: where 14 it is seen that bacterial population doubling time T d is proportional to τ R and gi v en by:

Bulk kinetics assays
Components and buffer s. Buf fers and E. coli components for cell-free protein synthesis were prepared as described ( 7 ). Reaction buffer was polymixHEPES pH 7.5, containing 5 mM Mg(OAc) 2 , 95 mM KCl, 3 mM NH 4 Cl, 0.5 mM CaCl 2 , 1 mM spermidine, 8 mM putrescine, 1 mM dithioerythritol and 30 mM Hepes. It was supplemented with an energy regeneration system containing 2 mM GTP, 10 mM phosphoenolp yruvate, 50 g / ml p yruvate kinase and 2 g / ml myokinase. Purified ribosomal release complexes (RC) wer e pr epar ed as described ( 8 ). RF1 had methylated Glu in the catal ytic Gl y-Gl y-Glu (GGQ) motif and was purified as described ( 8 ). The total RF1 concentration was determined by Bradford assay and the concentration of acti v e RF1 was determined by the amount of peptide released during a single round of termination with RC in excess over RF1 ( 12 ). Both RC and RF1 samples were pre-incubated for 1 min at 37 • C prior to the reaction. 112 pmol RC were mixed with 0 -100 pmol RF1 for 15 s at 37 • C (reaction volume 70 l) and quenched with 17% (final concentration) formic acid. RF3 and energy regeneration system were not included to avoid recycling of RF1. The quenched samples were treated as described for quench flow experiments ( 8 ). RF3 was ov ere xpressed from an osmo-expression vector and further purified with ion exchange and gel filtration chromato gra phy ( 14 ).

RF3 accelerates ribosome intersubunit rotation rate for rapid class-I RF dissociation
To explore the dynamics of RF3-ribosome interaction in the presence of GTP, we measured the timing of ribosome intersubunit rotation during termination of protein synthesis in the context of the entire mRNA translation cycle. We applied single-molecule fluorescence assays using zero-mode waveguide (ZMW)-based instrumentation ( 39 ). The intersubunit rotational movements during initiation, elonga tion, termina tion and recycling stages of the ribosomal protein synthesis wer e monitor ed by site-specifically labeled 30S and 50S subunits with Cy3B and BHQ-2 (a non-fluor escent quencher), r especti v ely, allowing for FRET between the two dyes ( 40 ). The arrival and departure of fluor escently-labeled r elease factors during translation termination were monitored in real time ( 18 ).
Translation was started by deli v ering BHQ-50S subunit, elongation / recycling mix containing aa-tRNA, EF-Tu, EF-G, RRF, GTP and Cy5.5-labeled RF1 or RF2 to immobilized Cy3B-30S preinitiation complexes (30S subunit-mRN A-initiator tRN A) (Figure 1 A). Translation initiation was signaled by quenching of Cy3B-30S (initially high Cy3B intensity) by BHQ-50S upon subunit joining to form the 70S ribosome in non-rotated state (low Cy3B intensity). Transla tion elonga tion cycles were signaled by Cy3B fluorescence changes due to transitions between the nonrotated (low Cy3B intensity) and rotated (medium Cy3B intensity) states of the 70S ribosome. Eventually, the arrival of a stop codon into the A site was signaled by the Cy5.5 fluorescence increase due to A-site binding of a Cy5.5-labeled class-I RF. Departure of Cy5.5-labeled class-I RF was observed as Cy5.5 fluorescence decrease which, in the absence of RF3, coincides with Cy3B fluor escence incr ease due to ribosome intersubunit rotation ( Figure 1B). After the rotation and release factor dissociation, the 70S ribosome was split into subunits by RRF and EF-G ( 41 ), resulting in de-quenching of Cy3B signal. Kinetic scheme describing an activation step with rate constant k i IA that converts an inhibited RF1-or RF2-bound post-termination non-rotated ribosome R P ·RF i to an acti v e state R P ·RF i before RF3 binds and accelerates ribosome intersubunit rotation to the R P R ·RF i ·RF3 state for subsequent RF1 / RF2 dissociation with rate constant k + 1RD (also shown in Materials and Methods: Eq. ( 4 )).
Mean dwell time τ i BD for RF1 or RF2 (RF i , i = 1 or 2) on the terminating ribosome, defined as the mean of the time between RF i binding to and dissociation from the ribosome, was estimated as τ − 1 BD = 66 ± 3 s for RF1 and τ − 2 BD = 38 ± 2 s for RF2 in the absence of RF3 (specified with a minus) (see Materials and Methods for details). These values coincide with the mean ribosome rotation time τ i BR between RF i binding and ribosomal intersubunit rotation ( τ − i BD = τ − i BR ) since all RF1 and RF2 dissociation e v ents occur concurrently with ribosome rotation in the absence of RF3 (Figure  1 C). Increasing the unlabeled RF3 concentration decreased τ i BD for RF1 and RF2 (Figure 2 A), consistent with a previously observed RF3-dependent increase of RF i dissociation rate from A site in GTP presence ( 14 ). Addition of RF3 introduced a short post-rotation stochastic delay time conditional on that the ribosome is RF3-bound between ribosome rotation and RF i r elease, t + i RD , (Figur e 2 B) with mean value τ + i RD . For both RF1 and RF2, with increasing [RF3] the mean ribosome rotation time τ i BR decreased (Materials and Methods: Eq. 7 ) and mean post-rotation RF i dwell time τ i RD , defined as the probability that the ribosome is free from RF3 multiplied with τ − i BD ( = 0) plus the probability that the ribosome is RF3 bound multiplied with τ + i RD (Materials and Methods: Eq. 9 ) increased from zero toward its plateau value τ + i RD (Figure 2 C, D). From these observations we conclude that (i) in the GTP presence RF3 accelera ted RF i dissocia tion by inducing ribosome rota tion, and (ii) intersubunit rotation subsequently induced rapid RF i dissociation from the ribosome. From the subset of ribosomes that display the post-rotation stochastic delay t + i RD of RF i dissociation, we fitted the distribution of t + i RD to a single exponential function (Materials and Methods: Eq. 1 ) and estimated rate constants for RF1 and RF2 dissociation from the rotated ribosome as 1 / τ + 1 RD = k + 1 RD = 0.76 ± 0.04 s −1 and 1 / τ  Figure S1A). To investigate further the slow intersubunit rotation a t sa tura ting [RF3], we took advantage of the observation that in the absence of RRF a post-termination ribosome from which a 'first-bound' RF i has dissociated can r eadily r ebind another RF i ( 15 ). This brings the 'rebound' ribosome back to the non-rotated intersubunit state ( 18 ), as also observed here with Cy5.5-RF1 (Supplementary Figure  S2A). We found that the τ 1 BR times and their responses to RF3 addition are remar kab ly different for 'first-bound' and 'rebound' ribosomes (Supplementary Figure S2D). Fits of these data to Materials and Methods: Eq. ( 7 ) lend strong support to the hypothesis that 'first-bound' ribosome rotation is initially blocked and ther efor e occurs in two consecuti v e steps: an acti vation step which is insensiti v e to RF3 presence (k iIA ), a slow (k − iAR ) and a fast (k + iAR ) rotation step in absence and presence of RF3, respecti v ely ( Figure 2G; Materials and Methods: Eq. 4 ). In contrast, the 'rebound' ribosomes appear already activated, so that rotation occurs in a single step (Materials and Methods: Eq. 9 ). Accordingly, for 'first-bound' ribosomes we estimate k iIA and k − Further, RF1 dissociation times from 'first time bound' and 'rebound' ribosomes coincide with stochastic ribosome rotation times in absence ( t − 1 BD = t − 1 BR ) but not in the presence of RF3, where dissociation is delayed in rela tion to rota tion ( t + i BD = t + i BR + t + i RD ) ( Supplementary Figur e S2B). Her e, in the pr esence of RF3, these dwell times t + i RD of the 'first-bound' and 'rebound' RF1 are exponentially distributed with rate constants k + 1RD estimated as 0.77 ± 0.11 s −1 and 1.0 ± 0.2 s −1 , respecti v ely, similar to k + 1RD = 0.87 ± 0.04 s −1 for the single RF1 binding e v ent in the presence of RRF (Supplementary Figure S2C).
A ppl ying the same model ( Figure 2G; Materials and Methods: Eqs. 4 , 7 ) to 'first-bound' kinetics in the presence of RRF, we estimate the ribosome activation rate constant to be k 1IA = 0.034 s −1 at 20 • C and 1.3 s −1 at 30 • C (Table  1 ), r epr esenting the plateau values in Figure 2 F. The 38fold increase in k 1IA a t 30 • C indica tes a high enthalpic contribution to the standard free energy barrier for activation of intersubunit rota tion. Ra te constant k 1IA remains small (0.07 s −1 ) at 20 • C e v en for termination on longer peptides (13 aa), suggesting physiological relevance of this reaction barrier as a growth inhibitor at low but not high temperatures (Supplementary Figure S2E-F). Natural guesses of its origin would be hydrolysis of the peptidyl-tRNA ester bond or the subsequent peptide dissociation, suggested to precede subunit rotation and class-I RF dissociation ( 12 ).
Howe v er, our bulk kinetic data show that the ester bond hydrolysis induced by RF i and subsequent dissociation of fluorescently-labeled peptide from the ribosome occur too fast (2 s −1 ) to match k 1IA at 20 • C (Supplementary Figure  S1B), thus the origin of this rotation blockage remains unknown (Discussion).

RF3-catalyzed intersubunit rotation exhibits complex dynamics
To re v eal the temporal relation of the slow ribosome rota tion a t 20˚C (Figure 2 F, G) to the binding and dissociation of RF3, we next directly monitored RF3 occupancy sim ultaneousl y with RF1 binding and ribosome rotation during termination using Cy5-labeled RF3. In the presence of GTP, two distinct modes of Cy5-RF3 binding were observed on ribosomes bound to Cy5.5-RF1 during termination (Figure 3 A). The first mode, denoted the 'sampling mode', shows RF3 binding and dissociation e v ents with no change in RF1 occupancy or ribosome conformation. The second mode, denoted as 'producti v e mode' that followed the 'sampling mode' in most recorded traces (like in Figure 3 A), shows RF3 binding followed by a rapid ribosome rotation ( k 3BR = 3.2 ± 0.5 s −1 ) for 51 ± 2% of the ribosomes while the remaining ribosomes a pparentl y rotated concurrently with producti v e RF3 binding. The apparent 'concurrency' of the two e v ents is due, most probably, to the two e v ents occurring with a delay time less than the time resolution of the measurement (100 ms). Ribosomal intersubunit rotation was followed by Cy5.5-RF1 dissociation (k + 1RD = 0.69 ± 0.02 s −1 ) and then by Cy5-RF3 dissociation ( k 31D = 4 ± 1 s −1 ) for 68 ± 3% of the ribosomes (Figure 3 B). As much as 31 ± 1% of the ribosomes show a pparentl y sim ultaneous dissociation of RF1 and RF3, a result, we suggest, due to the limited time resolution of the measur ement. The r emaining small ( < 5%) fraction shows earl y disa ppearance of Cy5-RF3 signal before Cy5.5-RF1 dissociation. These results support a sequential RF dissociation model where ribosome rotation triggers RF1 dissociation followed by rapid RF3 dissociation. In line with this model, lowering the temperature further (to 12 • C) extends the fraction of ribosomes following this sequential dissociation pathway to 98%, accompanied by a 2.6-fold slo w-do wn Representati v e trace of ribosome terminating on MF-UAA mRNA with Cy5.5-RF1, Cy5-RF3 and GTP at 20 • C. After Cy5.5-RF1 binding, Cy5-RF3 binds to the ribosome either in sampling mode or producti v e mode, the latter coupled to ribosome rotation and subsequent RF dissociation e v ents. ( B ) Left: r epr esentati v e trace of the experiment from panel (A) zoomed in to highlight the producti v e Cy5-RF3 e v ent characterized by the sequential e v ents of Cy5-RF3 binding, ribosome rota tion, Cy5.5-RF1 dissocia tion, and Cy5-RF3 dissocia tion. The time intervals between these e v ents are defined by stochastic parameters t 3BR , t + 1RD and t 31D . Right: RF3-bound ribosome rotation ( k 3BR ), post-rotation RF1 dissociation ( k + 1 RD ) and post-RF1 RF3 dissociation ( k 31D ) rate constants, determined from single exponential fits to distributions of dwell times preceding the respecti v e steps ( t 3BR , t + 1RD and t 31D ) at differ ent temperatur es. Error bars are defined as 95% CI. ( C ) Representati v e trace of ribosome terminating on MF-UAA mRNA with Cy5.5-RF1, Cy5-RF3 and GTP a t 12 • C , with the producti v e Cy5-RF3 binding boxed and zoomed in to the right. The zoomed-in trace outlines the timing of intersubunit rota tion, Cy5.5-RF 1 dissocia tion, and Cy5-RF3 dissociation used to measure the post-rotation RF1 dwell time ( t + in RF1 dissociation and a 21-fold slo w-do wn in RF3 dissociation (Figure 3 C-E).
Termination in the presence of the non-hydrolysable GTP analogue GDPCP shows that Cy5-RF3 can still bind producti v ely to the RF1-ribosome complex and catalyze intersub unit rotation, b ut Cy5-RF3 ·GDPCP is afterward trapped on the rotated ribosome (Figure 3 F). The intersubunit rotation following Cy5-RF3 binding in GDPCP presence occurs through an interim ribosome conformational state r epr esented by an intermediate Cy3B-BHQ intensity (Supplementary Figure S3). This interim state likely r epr esents a partial rotation of the ribosome that was previously re v ealed in a cryo-EM structure of a partially rotated ribosome bound to RF1 and RF3 ·GDPCP ( 29 ). The Cy5.5-RF1 then dissociates from this intermediate state 4.6-fold more slowly than from the RF3 ·GTP-induced rotated state (Figure 3 G). This Cy5.5-RF1 dissociation e v ent is coupled to full ribosome rotation (Supplementary Figure S3). The rapid dissociation of Cy5-RF3 after RF1 departure in the presence of GTP is contrasted by the 270-fold slower dissociation of Cy5-RF3 in the presence of GDPCP (Figure 3 H). These observations suggest that the primary role of GTP hydrolysis on RF3 is to promote its rapid dissociation after intersubunit rotation and RF1 dissociation.
In presence of GTP, the addition of the antimicrobial peptide Api137 eliminated RF3-accelerated intersubunit rotation and extended the Cy5.5-RF1 occupancy time to the Cy5.5 photobleaching and movie-time limit (300 ± 7 s) (Supplementary Figure S4A-B), while the rates of elongation and Cy5.5-RF1 arrival were unaffected (Supplementary Figure S4C-F). This resulted in perpetual cycles of Cy5-RF3 sampling without intersubunit rotation throughout the entire movie (6 min) ( Supplementary Figure S4G, H), suggesting that Api137 inhibits RF3-mediated intersubunit rotation by thermodynamically and / or kinetically stabilizing the RF1-bound non-rotated ribosomal state.
We next determined the effects of different nucleotides GTP , GDP , GDPCP and their mixtures on RF3 binding dynamics. The producti v e Cy5-RF3 binding e v ents that are coupled to intersubunit rotation and RF1 dissociation --as observed in the presence of GTP or GDPCP (Figure 3 A, F) --were not observed in the presence of GDP (Figure 4 A), confirming that RF3-induced intersubunit rotation strictly r equir es GTP or one of its analogues. In contrast, the sampling Cy5-RF3 e v ents prior to the producti v e e v ent were observed in the presence of GTP, GDPCP, or GDP (Figure 3 A, F; Figure 4 A), showing this initial sampling phenomenon to be GTPase-independent. The time intervals measured between the Cy5.5-RF1 binding and Cy5-RF3 binding or between the Cy5-RF3 dissociation and subsequent Cy5-RF3 binding e v ents in the sampling mode, termed here arrival times (see Figure 4 B), are exponentially distributed (Figure 4 C,D). From the exponential factors k obs of these time distributions a bimolecular associa tion ra te constant ( k 3on ) can be obtained as k obs / [RF3] at any RF3 concentration. In the presence of GTP, k 3on was estimated as 2.3 ± 0.2 M −1 s −1 (Figure 4 E, Table 2 ), and very similar k 3on values were estimated from the arrival time distributions measured at 75 nM RF3 ·GDP or RF3 ·GDPCP concentration (Figure 4 D, E). This suggests similar and rapid association kinetics for the GDP, GDPCP and GTP forms of RF3, fully compatible with the previousl y observed, ra pid GDP to GTP exchange on RF3 in complex with the class-I RF bound ribosome ( 16 , 23 ).
The occupancy times defined as time intervals between binding and subsequent departure of Cy5-RF3 in sampling mode (Figure 4 B) re v eal a biphasic distribution (Figure 4 F), with a fast phase dissocia tion ra te ( k 3off,1 ) of 1.3 ± 0.1 s −1 and slow phase dissociation rate ( k 3off,2 ) of 0.30 ± 0.02 s −1 in the presence of GTP (distribution was fit to a double exponential, Materials and Methods: Eq. 2 ) (Figure 4 F, G). This suggests that in the presence of GTP, RF3 can bind to the non-rotated RF1-bound ribosome in two distinct states with dif ferent af finities.This biphasic behavior persists in the presence of GDPCP and GDP. With GDPCP the fast phase rate is very similar to that with GTP while with GDP the fast phase rate is 4.6 ±0.3 s -1 , i.e. 3.5-fold that with GTP (Figure 4 F, G). These distinct fast phases identified in the presence of 100% GTP and 100% GDP were both present when fitting a triple exponential (Materials and Methods: Eq. 3 ) to the Cy5-RF3 occupancy time distribution in the presence of 50% GDP (1:1 ratio of GTP:GDP) (Figure 4 G), with the slowest phase having a similar rate (about 0.3 s −1 ) as the slow phases of the 100% GTP and 100% GDP distributions. While the fast > 4 s −1 phase confirms the existence of a shorter-li v ed RF3 ·GDP state on the ribosome, the consistent occurrence of the slowest phase with a dissociation rate constant less than 0.4 s −1 suggests the existence also of a higher affinity RF3 state on the non-rotated ribosome that is not dependent on which guanine nucleotide is present. We propose that this high-affinity state r epr esents the apo-RF3 intermediate induced by the ribosome to facilitate the GDPto-GTP exchange mechanism (see SI for explanation of the observed states).

Kinetic modeling of intracellular termination flows
The rational for the kinetic modeling of intracellular termination flows in the present work ( Figure 5 ) is to answer pertinent and yet unanswered questions regarding RF3 function. One such question is if, in the living cell, the biochemically well-established guanine nucleotide exchange factor activity of the ribosome ( 16 , 23 , 42 ) is r equir ed for RF3 ·GDP conversion to its active RF3 ·GTP form ( 12 , 23 , 42 ) or, as suggested by Peske et al. ( 16 ) the spontaneous conversion of RF3 ·GDP to RF3 ·GTP in the cytoplasm is sufficient to assure the rate of RF3 ·GTP supply that matches the rate of demand of activated RF3 ·GTP in termination of protein synthesis. Another pertinent question is if rapid dissociation of RF2 in the absence of RF3 ( 24 ) explains the small growth rate reduction by RF3 deficiency in some bacterial strains ( 43 ).
Accordingly, we used kinetic modeling to decide if the formation of acti v e RF3 ·GTP occurs mainly through GDP to GTP exchange on free RF3 in bulk solution ( 16 ) or on ribosome-bound RF3 ( 23 ). We used in vitro estimates of kinetic parameters for the interactions of class-I RFs and RF3 to model translation termination embedded in the entire protein synthesis cycle ( Figure 5 ). We focused on the kinetic coupling between the action cycles of RF1 / RF2 and RF3 on one hand, and the action cycles of peptide release from The numbers reported are the RF3 dissociation rate constants ( k 3off1 , k 3off2 , k 3off3 ) extracted from the fits.  Figure 5. Scheme for ribosomal protein synthesis in the bacterial cell. In termination phase of protein synthesis class-I RFs (RF i ) bind to the stop codon programmed A site of pre-termination ribosome (R A ) leading to the formation of ribosomal complex R A ·RF i (upper left corner of Figure 5 ), upon which the RF i -induced peptide release from peptidyl-tRNA in P site leads to post-termination complex (R P ·RF i ). Subsequent RF i and RF3 release via three alternati v e pathways (upper right part of Figure 5 ) e v entually leads to post termination ribosome (R P ). Termination ends by irre v ersib le splitting of R P into ribosomal subunits by ribosome recycling factor RRF and elongation factor G (EF-G) ( 41 , 51 , 52 ). Subsequently, ribosomal subunits ar e r ecycled to initiation complex (R I ) ( 53 , 54 ) leading to elongation complex (R E ) with the global cycle of protein synthesis closed by the formation of pre-termination ribosome R A (see SI for details).
the ribosome on the other hand ( Figure 5 , Supplementary Figure S6A). Ther e ar e thr ee alternati v e pathways ('A', 'B' and 'C') for RF1 and RF2 (for brevity denoted as RF i where i = 1 or 2) release from post-translation ribosomal complex R P ·RF i (Figur e 5 ). Ther e ar e also two pathwa ys f or GDP to GTP exchange on RF3: one spontaneous 'off' the ribosome in solution ( Figure 5 , pathway 'D') and one 'on' the ribosome ( Figure 5 , pathway 'A'). In the latter pathway RF3 ·GDP binds to post-termination complex R P ·RF i to form an R P ·RF i ·RF3 ·GDP complex. Rapid GDP dissociation from this ribosome-bound RF3 ·GDP ( 16 ) results in an R P ·RF i ·RF3 · intermediate complex containing nucleotide-free, apo-RF3 (RF3 ·) which, upon GTP binding, becomes the activated ribosome-bound RF3 ·GTP in the R P ·RF i ·RF3 ·GTP complex ( Figure 5 ). This activa ted ribosome-bound RF3 ·GTP ca talyzes RF i dissociation, which is followed by rapid GTP hydrolysis on RF3 ·GTP leading to RF3 ·GDP formation and dissociation from the ribosome, resulting in post-termination complex R P with unoccupied A site ready to bind RRF ( 12 , 23 , 26 , 28 ). In pa thway B , RF3 ·GTP formed by spontaneous GDP to GTP exchange in solution binds to posttermination complex R P ·RF i directly leading to complex R P ·RF i ·RF3 ·GTP and from there the reactions are as in pa thway A ( 16 , 24 ). Importantly, pa thway B r equir es spontaneous GDP dissociation from RF3 ·GDP 'off' the ribosome ( Figure 5 , Rectangle 'D', arrow denoted q S 3D ), a slow reaction known to be rate limiting for the spontaneous GDP-to-GTP exchange on RF3 off the ribosome. In pathway C, release factor RF i dissociates from post-termination complex R P ·RF i in an RF3-independent manner as in RF3deficient E. coli mutants ( 19 , 43 ) and other RF3-lacking bacteria ( 44 ).
For actual modeling the detailed reaction scheme in Fig Figure S7 were deri v ed from ordinary differential equations (see SI for details). All kinetic parameters used in the modeling are compiled in Supplementary Table S1 along with their sources.