We performed maximum-likelihood classification of a large cryo-EM dataset (~1.5 million particles) collected from 70S ribosomes that were purified from E coli at the exponential growth phase and then incubated with defined mRNA, tRNA^Phe, tRNA^fMet and stringent factor RelA (Supplementary Figure S1; Methods). Remarkably, we found that ~14,000 of the 1.15 million 70S ribosome particles contain tmRNA (~1.2%). Because neither tmRNA nor tRNA^Ala were added to the 70S sample, the tmRNA-bound ribosomes must represent endogenous E. coli trans-translation complexes. Since the tmRNA-bound ribosomes formed prior to the addition of RelA (Methods), they likely represent a homeostatic trans-translation complex. Indeed, the 1.2% recovery of trans-translation ribosomes comports with the cellular estimates of rescue-complex abundance (Moore and Sauer, 2005; Ito et al., 2011).

The predominant 3.7 Å cryo-EM reconstruction with circularized tmRNA density features a non-rotated ribosome with strong densities for TLD•SmpB in the P site and tRNA in the A site (Figure 1A; Supplementary Figure S1). Our extensive classification did not identify tmRNA in other ribosome sites, similarly to the recent study of M. smegmatis ribosomes (Mishra et al., 2018), suggesting that this complex represents an intermediate accumulating during trans-translation. The ribosome and the endogenous tmRNA, wrapped around the head of the 30S subunit (Figure 1B), are overall similar to those in post-translocation 70S complexes assembled from in vitro transcribed tmRNA constructs (Rae et al., 2019; Guyomar et al., 2021). The ribosomal intersubunit rotation state during elongation correlates with stages of decoding and translocation (Cornish et al., 2008; Ermolenko and Noller, 2011; Rodnina, 2018). The non-rotated ribosome with A-site tRNA corresponds to a post-decoding stage, preceding the translocation of the tRNA into the P site that requires intersubunit rotation. In addition, translocation involves a “swiveling” motion of the head domain of the 30S subunit (Ratje et al., 2010). Another mode of the head movement, known as “tilt,” normally occurs during initiation, when the free 30S subunit interacts with the initiator tRNA sampling the P site (Hussain et al., 2016; Jahagirdar et al., 2020). While mechanistically similar to the elongation ribosomes with P- and A-site tRNAs, the tmRNA-bound ribosome is reorganized via a head tilt to accommodate the bulky tmRNA. As tmRNA helix 2 and pseudoknot 1 are placed between the 30S head and the 50S central protuberance (Figures 1B,C), the head is tilted 7° away from the large subunit placing uS19 ~ 15 Å farther than in canonical elongation complexes (measured at Gly25; Figures 2A,B). The A-site finger (ASF) of the large subunit, involved in tRNA accommodation (Sanbonmatsu et al., 2005; Loveland et al., 2020), is shifted by ~20 Å (measured at the U887 tip) to dock onto PK1 (Figure 2C).

The structure brings insight into tRNA positioning in the A site of the tmRNA-bound ribosome. On the 50S subunit, despite the large shift of the 50S ASF, the A-site tRNA elbow is stabilized by packing against the ASF (Figure 2B). Here, the C19-G56 pair likely stacks on the bulged A896 of the ASF (Supplementary Figure S2A), similarly to canonical tRNA-bound complexes. Accordingly, the position of the A-site tRNA relative to the 50S subunit is nearly identical to those in tRNA-bound structures, emphasizing the invariant mechanism of tRNA accommodation for peptidyl transfer. The acceptor arm with the 3′ terminal CCA is inserted into the A site next to the CCA end of the tmRNA (Figure 1D). In the polypeptide tunnel, scattered density suggests compositional and conformational heterogeneity of peptides in the endogenous rescue complexes on different mRNAs.

Due to the 30S head tilt, interactions between the A-site tRNA and the 30S subunit slightly differ from those in canonical tRNA-bound complexes. Whereas helix 30 of 16S rRNA normally binds near the anticodon stem of tRNA (at nt 42), helix 30 is retracted by ~9 Å (measured at U956) to accommodate the tmRNA linker connecting PK1 and MLD. Universally conserved C1054, which bulges from h34, normally buttresses the anticodon by packing on the ribose of nt 34. But in the tmRNA-bound complex, the 30S head tilt shifts C1054 by ~4 Å, detaching it from tRNA^Ala (Figure 2B).

The loss of interactions between tRNA^Ala and the 30S head is partially compensated by interactions with tmRNA and SmpB (see below), firmly positioning tRNA^Ala in the 30S decoding center. Local density confirms tRNA^Ala-specific nucleotides and Watson-Crick base pairing of the tRNA UGC anticodon with the corresponding GCA codon of tmRNA (Figure 3C; Supplementary Figures S2B,C). The codon-anticodon helix is stabilized by interactions with ribosomal decoding-center nucleotides G530, A1492, and A1493 (Figure 3D). The G530 loop of the shoulder domain is disengaged from the h34 of the shifted head, unlike in canonical tRNA-bound ribosomes where A532 packs on G1207. Despite this difference, G530 stabilizes the tRNA anticodon by hydrogen bonding with the ribose of G35, and both nucleotides are placed nearly identical to those in canonical tRNA-ribosome complexes. The adenosines A1492 and A1493 of the body domain stabilize the opposite side of the codon-anticodon helix by hydrogen-bonding with the riboses of tmRNA codon nucleotides G90 and C91. Thus, the ribosome recognizes and stabilizes the tmRNA-tRNA^Ala codon-anticodon helix via the universally conserved G530 and A-minor interactions, as in canonical elongation complexes (Ogle et al., 2001; Demeshkina et al., 2012; Loveland et al., 2017).

To accommodate tRNA^Ala in the A site, the adenosine-rich linker of tmRNA (⁸⁰AAAAAU⁸⁵) shifts away from its position in pre-decoding structures (Rae et al., 2019; Guyomar et al., 2021), where the linker traverses the A site (Figures 3A,B). Adenosines A82 and A83 support the minor groove of the tRNA^Ala anticodon stem at nucleotides C40 and A41 (Figures 3E,G). This interaction appears similar to the A-minor-like interaction in the 30S P site, where conserved 16S nucleotides G1338 and A1339 pack at the minor groove of initiator tRNA^fMet (Figure 3F) to assist translation initiation and perhaps other translation stages (Lancaster and Noller, 2005; Korostelev et al., 2006; Hussain et al., 2016). A-minor-like interactions are a unique tertiary structure that plays critical roles in RNA stabilization, including tetraloop-receptor recognition (Cate et al., 1996; Doherty et al., 2001; Nissen et al., 2001; Battle and Doudna, 2002) and mRNA decoding described above. Furthermore, their modest interaction surface and thermodynamic stability (Doherty et al., 2001) allow for local structural rearrangements, such as tRNA dynamics during mRNA decoding and translocation (Ogle et al., 2001; Demeshkina et al., 2012; Loveland et al., 2017; Carbone et al., 2021). Thus, the tmRNA linker not only replaces the tRNA interactions with the 30S head during decoding but may also transiently stabilize tRNA^Ala and/or disengage from the tRNA^Ala in the subsequent—highly dynamic—translocation step. The functional role of this A-minor interaction is supported by the conservation of adenosines positioned 6–10 nucleotides upstream of the first codon of tmRNA in most bacterial species (Zwieb et al., 1999). In species without consecutive adenosines in this position (e.g., Thermus thermophilus tmRNA), however, it remains to be seen how tmRNA interacts with tRNA^Ala (Op De Bekke et al., 1998; Kaur et al., 2006).

SmpB adopts the same overall conformation as in the structures without A-site tRNA (Rae et al., 2019; Guyomar et al., 2021). Here, the protein’s globular domain binds the 30S P site, with the His79 loop sandwiched between TLD and the elbow of tRNA^Ala. To accommodate the tRNA, the loop is slightly rearranged, bringing His79 into contact with the tRNA^Ala backbone at nucleotide 17. The C-terminal helix occupies the mRNA binding pocket in the E site (Figure 1D).

In our tmRNA-containing cryo-EM reconstructions, low-density features in the E site suggested sub-stoichiometric tRNA. To better resolve this density, we subclassified the cryo-EM maps, using a mask covering the E site, yielding a 3.9-Å resolution class with tRNA near the canonical E site (Figure 4A; Supplementary Figures S1, S2D). Other complex constituents, including the A-site tRNA, are similar to the complex described above.

Interactions of the tRNA with the 30S E site differ from canonical ribosome structures with an untilted head domain. In canonical complexes, E-site tRNA binds in the cleft between the head (near ribosomal protein uS7) and body (near G693 of h23 of 16S rRNA). In the tmRNA-containing complexes, however, the tRNA is shifted ~8 Å away from this binding pocket, despite ample space near the C-terminal helix of SmpB (Figures 4B–D). Furthermore, the tRNA is partially retracted from the E site of the head, formed by h29 of 16S rRNA and uS7. Here, the anticodon is shifted ~9 Å away from the tip of the β-hairpin of uS7 (at Gly80), where the tRNA anticodon resides in elongation complexes (Figures 4C,D). In this position, the anticodon stem loop is loosely held near h29 (at A1339) and the β-sheet (at Arg78) and C-terminal α-helix (near Arg142) of uS7. On the 50S subunit, the tRNA interacts with the L1 stalk (elbow) and H88 of 23S rRNA, where the terminal nucleotide A76 is inserted into. These contacts are nearly identical to those observed in numerous structural studies of tRNA-bound complexes (Figure 4B).

Previous structural studies of the P-site bound tmRNA did not report E-site tRNA, suggesting that tRNA bound to the preceding truncated mRNA readily dissociates upon tmRNA translocation. The tRNA in our map likely results from the addition of tRNA^Phe and tRNA^fMet to the ribosome sample (Methods). Nevertheless, this observation may report a transient tRNA binding state sampled during or immediately after tmRNA translocation. The structure underlines that unlike the A and P site, where the tRNA interacts closely with both the 30S and 50S subunits, E-site positioning is driven primarily by interactions with the 50S subunit. Indeed, these interactions are established during the initial stages of translocation, in which the acceptor arm of deacyl-tRNA is spontaneously transferred from the P site to the E site on the 50S subunit, while the anticodon stem loop remains bound to the 30S subunit (Agirrezabala et al., 2008; Fei et al., 2008; Julian et al., 2008; Cornish et al., 2009). In addition, non-canonical E-like tRNAs were found in other studies of bacterial ribosomes (E-out; Zhang et al., 2018) and eukaryotic ribosomes (Z-site; Brown et al., 2018). While retaining the invariant interactions with the large-subunit L1 stalk and the CCA-binging pocket, these structures feature different tRNA interactions with the small subunit. Distinct binding modes allow for increased tRNA dynamics in the E site, underlying the tRNA-dissociation function of the E site (Brown et al., 2018; Zhang et al., 2018). Our findings expand the repertoire of possible tRNA rearrangements during tRNA departure from the ribosome.

In conclusion, our cryo-EM analyses visualize how the first tmRNA codon is decoded and how tmRNA rearranges to accommodate tRNA^Ala. Interactions with tmRNA stabilize tRNA^Ala, which binds nearly identically to canonical A-site tRNA despite a substantial tilt of the 30S head. Future work will detail whether A-minor-like interactions of tmRNA with the tRNA anticodon stem occur in bacterial species, whose tmRNA linker does not contain continuous adenosines. Further, rearrangements of these and other interactions of tmRNA•SmpB with tRNA^Ala and the ribosome during translocation of tRNA^Ala to the P site remain to be visualized. Such structural studies may inform the development of new drugs that target trans-translation, a promising target for antibacterial therapeutics (Campos-Silva et al., 2021; Marathe et al., 2023).

70S ribosomes were prepared from MRE600 E. coli essentially as described (Moazed and Noller, 1986, 1989) and stored in the ribosome-storage buffer A (20 mM HEPES (pH 7.5), 100 mM KCl, 10.5 mM MgCl₂, 0.5 mM EDTA, 5 mM β-mercaptoethanol) at −80°C. In short, MRE600 E. coli stock was grown on LB agar plates at 37°C. Then, one colony from the plate was inoculated in 100 mL LB media and grown at 37°C overnight in an incubated shaker at ~220 rpm. Forty eight milliliters of overnight culture was inoculated into 6 L LB media and the culture was incubated at 37°C to mid-log phase (OD₆₀₀ 0.5–1.0).

Escherichia coli cells obtained from a 6 L culture were suspended in 50 mL cold buffer A and lysed using a microfluidizer (Microfluidics, United States) at 18 k psi. The lysate was clarified using a JA-20 rotor at 39,200 × g, 4°C, for 20 min. The clarified supernatant was layered onto 35 mL (per tube) of 37.7% sucrose in buffer A. Ribosomes were sedimented onto the sucrose cushion by ultracentrifugation in a 45 Ti rotor at 185,677 × g (40,000 rpm), 4°C, for 20 h. The ribosome pellet was dissolved in 2 mL of buffer A. The ribosome solution was transferred to 1 mL microcentrifuge tubes and spun at 13 K rpm at 4°C for 10 min. Supernatant was transferred to a 50 mL tube, the volume was adjusted to 40 mL using cold buffer B (70 mM Tris–HCl (pH 7.0), 500 mM NH₄Cl, 15 mM MgCl₂, 0.5 mM EDTA, 5 mM β-mercaptoethanol), and ribosomes were sedimented in 70 Ti tubes at 310,801 × g (55,000 rpm), 4°C, for 2 h. The ribosome pellet was resuspended in 1 mL buffer A, aliquoted and stored at −80°C.

The ribosome complex for cryo-EM analyses was prepared as described (Loveland et al., 2016) with some modifications. 70S ribosomes at a final concentration of 0.4 μM were activated in 20 mM HEPES-KOH (pH 7.5), 120 mM KCl, 15 mM MgCl₂, 2 mM spermidine, and 0.05 mM spermine at 42°C for 15 min. Activated ribosomes were incubated with 0.8 μM mRNA (5′ GGCAAGGAGGUAAAAAUGUUCAAAAAA 3′), 0.8 μM tRNA^fMet, and 1 μM tRNA^Phe (all final concentrations) at 37°C for 30 min. The sample was then incubated with 4 μM RelA, 15 μM Adenosine-5′-[(α,β)-methyleno] triphosphate APCPP; Thermo scientific and 15 μM Guanosine triphosphate (GTP; Thermo scientific) to assemble a stringent-response 70S complex, for 15 min at room temperature. The final volume of the sample was 30 μL.

Carbon-coated EM grids (Ultrathin Carbon on Quantifoil®, 2 μm Diameter Holes, 1 μm Separation, mounted on a 200 M Cu grid coated with a 2-nm thin layer of carbon; TedPella) were glow discharged at 20 mA with a negative polarity setting for 30 s in a PELCO easiGlow glow discharge unit. 3 μL of the 70S sample was applied to the grid. Grids were blotted for 4 s with a blotting force of 7 and plunged into liquid ethane using a Vitrobot Mark IV (ThermoFisher Scientific), whose chamber was pre-equilibrated to 4°C and 95% humidity.

Two data sets were collected on a UMass Chan Cryo-EM Facility Talos Arctica electron microscope (ThermoFisher Scientific) operating at 200 kV and equipped with a K3 direct electron detector (Gatan Inc.) targeting 0.55–1.1-μm underfocus. Data were collected using SerialEM (Mastronarde, 2005), with beam tilts to record several movies at each stage position. The datasets contain 4,016 movies (total dose of 29.9 e⁻/Ų on the sample), yielding 474,262 particles and 8,021 movies (30.4 e⁻/Ų on the sample), yielding 1,039,487 particles. Movies were aligned during data collection using IMOD (Kremer et al., 1996) to decompress frames, apply the gain reference, and to correct for image drift and particle damage and bin the super-resolution pixel by 2.

CTF parameter determination, reference-free particle picking, and stack creation were carried out in cisTEM (v1.0-beta; Grant et al., 2018). Particle alignment and refinement were carried out in FREALIGNX (Lyumkis et al., 2013). Data processing was initially performed independently for each dataset (Supplementary Figure S1A). To speed up the processing, 2×− and 4×−image stacks were prepared using resample.exe, which is part of the FREALIGN distribution (Lyumkis et al., 2013). The initial model for particle alignment of 70S maps was the 11.5 Å resolution EMDB-1003 (Gabashvili et al., 2000), sampled to match 4×−binned image stack using resample.exe. Three rounds of mode 3 search alignment to 25 Å were run using the 4×−binned stack. Next, 25–30 rounds of mode 1 refinement were run with the 4×−binned, 2×−binned, and eventually unbinned stacks until the resolutions stopped improving, to the final resolutions of 2.8 Å and 2.7 Å of the overall maps. 3D maximum-likelihood classification into 20 classes was performed in FREALIGN v9.11 to separate 70S conformations, 50S subunits, and junk (poorly aligned or damaged) particles. An unexpected class emerged in each stack, featuring density near the 30S head, which connects the mRNA tunnel with the A-site finger and P site. The 70S classes with different tRNA occupancies and ribosome conformations (including the tmRNA class) were extracted into a stack per data set, using merge_classes.exe from FREALIGN distribution. Two 70S stacks were merged using IMOD 4.7 (Kremer et al., 1996).

The merged 70S stack was refined as described above, yielding a final average 70S reconstruction at 2.8 Å resolution. The refined parameters were used to run a 3D maximum-likelihood classification into 32 classes without a mask, with an ASF-covering mask, or with the A-site-covering mask. All masks were “spherical,” also known as “2D” masks on micrographs (Grigorieff, 2016), as opposed to specifically shaped “3D” masks (Supplementary Figure S1A). The tmRNA class was found in the no-mask and ASF-mask classifications. Particles with tmRNA density resulting from the ASF-mask classification were extracted into a substack. The tmRNA substack was classified at 4× with a P- and A-site covering mask or the E-site mask to further purify the tmRNA-containing density (P-A mask) or the E-tRNA-containing density (E mask). The unbinned stack was used to yield the resulting cryo-EM reconstructions with tmRNA (with A-tRNA and partial E-tRNA) and with tmRNA (with full-occupancy A-tRNA and E-tRNA) at resolutions of 3.7 Å and 3.9 Å, respectively (Supplementary Figure S1A).

Fourier Shell Correlation (FSC) curves were calculated by FREALIGNX for even and odd particle half-sets (Supplementary Figure S1B). The maps used for structure refinements was B-factor sharpened using the B factor of −100 Ų up to 3.4 Å (tmRNA with A-tRNA) and −50 Ų to 3.8 Å (tmRNA with A-tRNA and E-tRNA), using bfactor.exe (included with the FREALIGN distribution; Lyumkis et al., 2013).

Structure of the non-rotated 70S•tRNA₃ complex (V-B; PDB 6WDE; Loveland et al., 2020) and structures of 70S•tmRNA complexes (PDB: 6Q98 and 7ACJ; Rae et al., 2019; D’Urso et al., 2023) were used as starting models for 70S ribosome and tmRNA fitting, respectively. The model of tRNA^Ala (GGC anticodon) from PDB:6OF6 (Nguyen et al., 2020) was modified to fit the cryo-EM map and match the nucleotide sequence of tRNA^Ala (UGC). The 50S, 30S and tmRNA domains were fitted using UCSF Chimera 1.6 (Goddard et al., 2018; Pettersen et al., 2021) and locally modeled in Pymol1.2r1 (DeLano, 2002). The fitted structures were refined conservatively, using secondary-structure restraints and low simulated-annealing temperatures (100 K, 300 K or 500 K), against cryo-EM maps using phenix.real_space_refine v1.19.2 (Adams et al., 2010). Refinement parameters, such as the relative weighting of stereochemical restraints and experimental energy term, were optimized to produce the optimal structure stereochemistry and real-space correlation coefficients (Supplementary Table S1). B-factors of the models were refined at the final stages using phenix.real_space_refine. Structure stereochemistry validation was performed using phenix.molprobity.

Structure superpositions and distance calculations were performed in PyMOL. To calculate the angles of the 30S rotation and head tilt, 23S rRNAs of corresponding structures were aligned using PyMOL, and the angle between 16S domains were measured in Chimera. Figures were prepared in PyMOL and Chimera.