We determined the structure of the E. coli 70S ribosome in the classical (non-rotated) state with mRNA and tRNAs bound in the aminoacyl-tRNA and peptidyl-tRNA sites (A site and P site, respectively). Partial density for exit-site (E-site) tRNA is also visible in the maps, particularly for the 3’-terminal C75 and A76 nucleotides. Our final maps were generated from two 70S ribosome complexes that were formed separately using P-site tRNA^fMet that differed only by being charged with two different non-amino acid monomers (see Materials and methods). In both complexes, we used the same mRNA and A-site Val-tRNA^Val. Both complexes yielded structures in the same functional state, with similar occupancy of the tRNAs. As neither A-site nor P-site tRNA 3’-CCA-ends were resolved in the individual cryo-EM maps, we merged the two datasets for the final reconstructions (Figure 1—figure supplement 1). After Ewald sphere correction in RELION (Zivanov et al., 2018), global resolution of the entire complex reached 1.98 Å resolution by GS-FSC, with local resolution reaching 1.8 Å (Figure 1A, Figure 1—figure supplement 2, Table 1). The global resolution reached 2.04 Å using the map-to-model FSC criterion (Figure 1—figure supplement 2). We also used focused refinement of the large (50S) and small (30S) ribosomal subunits, and further focused refinements of smaller regions that are known to be conformationally flexible, to enhance their resolution (Figure 1—figure supplement 3; von Loeffelholz et al., 2017). In particular, focused refinement of the 30S subunit improved its map quality substantially, along with its immediate contacts to the 50S subunit and the mRNA and tRNA anticodon stem-loops. Additional focused refinement of the central protuberance (CP) in the 50S subunit, the 30S head domain, and the 30S platform aided in model building and refinement. In the following descriptions, maps of specific ribosomal subunits or domains refer to the focused-refined maps. Details of the resolutions obtained are given in Figure 1—figure supplements 2–3 and Tables 2–3.

The high resolution indicated by the FSC curves is supported by several visual features observable in the maps, including holes in many aromatic rings and riboses, as well as ring puckers, the directionality of non-bridging phosphate oxygens in ribosomal RNA, and numerous well-resolved ions, water molecules, and small molecules (Figure 1B–D, Figure 1—figure supplement 4). The maps also reveal known post-transcriptional and post-translational modifications of ribosomal RNA and proteins in detail (Figure 1—figure supplement 5, Figure 1—figure supplement 6), many of which are as previously described (Fischer et al., 2015; Noeske et al., 2015; Polikanov et al., 2015; Stojković et al., 2020). As seen in other ribosome structures, elements of the ribosome at the periphery are less ordered, including the uL1 arm, the GTPase activating center, bL12 proteins, and the central portion of the A-site finger in the 50S subunit (23S rRNA helix H38), as well as the periphery of the 30S subunit head domain and spur (16S rRNA helix h6). The resolution of the maps for the elbow and acceptor ends of the P-site and A-site tRNAs is also relatively low, likely as a result of poor accommodation of the unnatural substrates used.

The 30S ribosomal subunit is highly dynamic in carrying out its role in the translation cycle (Munro et al., 2009). Previously published structures demonstrate that even in defined conformational states of the ribosome, the 30S subunit exhibits more flexibility than the core of the 50S subunit. Furthermore, the mass of the 50S subunit dominates alignments in cryo-EM reconstructions of the 70S ribosome. Solvation of the 30S subunit has not been extensively modeled, again owing to the fact that it is generally more flexible and less well-resolved than the 50S subunit, even in available high-resolution structures (Noeske et al., 2015; Polikanov et al., 2015). Using focused-refined maps of the 30S subunit, we achieved the resolution necessary for in-depth chemical analysis of key contacts to mRNA and tRNAs, the 50S subunit, and to generate more complete models of 30S subunit components, Mg²⁺ ion positions, and solvation (Figure 2—figure supplement 1).

The 30S ribosomal subunit controls interactions of tRNAs with mRNA and helps maintain the mRNA in the proper reading frame. The present structure reveals the interactions of the 30S subunit with mRNA and tRNA in the A and P sites including solvation. The tight binding of tRNA to the P site (Lill et al., 1986) is reflected in extensive direct contacts between the tRNA anticodon stem-loop (ASL) and both the 30S subunit head and platform domains (Figure 2A–D). The C-terminal residues Lys129 and Arg130 of protein uS9, which are important for translational fidelity (Arora et al., 2013), form ionic and hydrogen bonding interactions with the nucleotides in the U-turn motif of the P-site ASL, and Arg130 stacks with the base of U33 (Figure 2D). However, the C-terminal tails of ribosomal proteins uS13 and uS19, which come from the 30S subunit head domain and are lysine-rich, are not visible in the map, suggesting they do not make specific contacts with the tRNA when the ribosome is in the unrotated state. Direct interactions between A-site tRNA and the 30S subunit are highly localized to the top of helices h44 and h18 in 16S rRNA and to helix H69 in 23S rRNA (i.e. 16S rRNA nucleotides G530, A1492, and A1493, and 23S rRNA nucleotide A1913). By contrast, contacts between the A-site tRNA and 30S subunit head domain are entirely solvent mediated (Figure 2E) apart from nucleotide C1054, possibly reflecting the weaker binding of tRNA to the A site (Lill et al., 1986) and the need for A-site tRNA conformational dynamics during mRNA decoding (Rodnina et al., 2017).

Protein bS21, which resides near the path of mRNA on the 30S subunit platform (Held et al., 1973; Marzi et al., 2007; Sashital et al., 2014), is essential in E. coli (Bubunenko et al., 2007; Goodall et al., 2018). Although partial structural models of bS21 have been determined (Fischer et al., 2015; Noeske et al., 2015), structural disorder in this region has precluded modeling of the C-terminus. In the present maps, low-pass filtering provides clear evidence for the conformation of the entire protein chain, including 13 amino acids at the C-terminus that extend to the base of the 30S subunit head domain (Figure 3A). Most of the C-terminal residues are found in an alpha-helical conformation near the Shine-Dalgarno helix formed between the 3’ end of 16S rRNA and the mRNA ribosome binding site (Shine and Dalgarno, 1975), while the C-terminal arginine-leucine-tyrosine (RLY) motif makes close contacts with 16S rRNA helix h37 and nucleotide A1167 (Figure 3B). The arginine and leucine residues pack in the minor groove of helix h37, and the terminal tyrosine stacks on A1167. Multiple sequence alignment of bS21 sequences from distinct bacterial phyla revealed that the RLY C-terminal motif is conserved in bS21 sequences of the Gammaproteobacteria phylum while in Betaproteobacteria, the sibling group of Gammaproteobacteria, bS21 sequences possess a lysine-leucine-tyrosine (KLY) C-terminal motif instead (Figure 3C). Interestingly, such C-terminal extensions (RLY or KLY) are absent in other bacterial phyla (Figure 3—figure supplement 1). Recently, putative homologs of bS21 were identified in huge bacteriophages, which were shown to harbor genes encoding components of the translational machinery (Al-Shayeb et al., 2020). Inspection of sequence alignments reveals that some phage S21 homologs also contain KLY-like motifs (Figure 3C). The presence of phage S21 homologs containing KLY-like motifs is consistent with the host range of these phages (Figure 3C, Supplementary file 1). Other phages with predicted hosts lacking the C-terminal motif in bS21 encode S21 homologs that also lack the motif (Supplementary file 1). Although the C-terminal region of bS21 resides near the Shine-Dalgarno helix, the examination of ribosome binding site consensus sequences in these bacterial clades and the predicted ribosome binding sites in the associated phages do not reveal obvious similarities (Supplementary file 1).

Ribosomal protein uS11 is a central component of the 30S subunit platform domain and assembles cooperatively with ribosomal proteins uS6 and uS18 (Stern et al., 1988), preceding the binding of bS21 late in 30S subunit maturation (Held et al., 1973; Sashital et al., 2014). Protein uS11 makes intimate contact with 16S rRNA residues in a 3-helix junction that forms part of the 30S subunit E site and stabilizes the 16S rRNA that forms the platform component of the P site (Stern et al., 1988). Remarkably, an inspection of the 30S subunit map uncovered a previously unmodeled isoaspartyl residue in protein uS11, at the encoded residue N119 (Figure 4; see Materials and methods), marking the first identified protein backbone modification in the ribosome. While it has been known that this modification can exist in uS11 (David et al., 1999), its functional significance has remained unclear, and prior structures did not have the resolution to pinpoint its exact location. Conversion of asparagine to isoaspartate inserts an additional methylene group into the backbone (the Cβ position) and generates a methylcarboxylate side chain (Reissner and Aswad, 2003). In the present map, the shapes of the backbone density and proximal residues in the chain reveal the presence of the additional methylene, allowing it the flexibility to pack closely with the contacting rRNA nucleotides (Figure 4A and B). A 30S-focused reconstruction using only early movie frames, in which damage to carboxylates would not be as severe (Marques et al., 2019), shows improved density for the IAS sidechain, albeit at slightly lower resolution overall (Figure 4—figure supplement 1).

Investigation of residues flanking the isoaspartate in uS11 reveals near-universal conservation in bacteria, chloroplasts, and mitochondria (Figure 4C, Figure 4—figure supplement 2), suggesting that the isoaspartate may contribute to 30S subunit assembly or stability. Consistent with this idea, the isoaspartate allows high shape complementarity including van der Waals contacts and hydrogen bonds between this region of uS11 and the 16S rRNA it contacts, which involves three consecutive purine-purine base pairs in bacteria (Supplementary file 2), and a change in rRNA helical direction that is capped by stacking of histidine 118 in uS11 on a conserved purine (A718 in E. coli; Figure 4B, Supplementary file 2). Strikingly, the sequence motif in bacterial uS11 is also conserved in a domain-specific manner in archaea and eukaryotes (Figure 4C, Figure 4—figure supplement 1), as are the rRNA residues near the predicted isoAsp (Supplementary file 2). Remodeling the isoAsp motifs in maps from recently-published cryo-EM reconstructions of an archaeal 30S ribosomal subunit complex at 2.8 Å resolution (Nürenberg-Goloub et al., 2020) and a yeast 80S ribosome complex at 2.6 Å resolution (Tesina et al., 2020) shows that the isoaspartate also seems to be present in these organisms based on the residue-level correlation between map and model (Figure 4—figure supplement 2). Taken together, the phylogenetic data and structural data indicate that the isoaspartate in uS11 is nearly universally conserved, highlighting its likely important role in ribosome assembly and function.

The E. coli 30S ribosomal subunit has eleven post-transcriptionally modified nucleotides in 16S rRNA, all of which can be seen in the present maps or inferred from hydrogen bonding patterns in the cases of many pseudouridines. Interestingly, two methylated nucleotides–m⁷G527 and m⁶₂A1519–appear not to be fully modified, based on the density at ~2.1 Å resolution (Figure 1—figure supplement 5). In the map, m⁷G527 appears partially methylated, and m⁶₂A1519 lacks one of the two methyl groups. Loss of methylation at m⁷G527, which is located near the mRNA decoding site, has been shown to confer low-level streptomycin resistance (Okamoto et al., 2007), and possibly neomycin resistance in some cases (Mikheil et al., 2012). The position of the methyl group is located in a pocket formed with ribosomal protein uS12, adjacent to the post-translationally modified Asp89, β-methylthio-Asp (Anton et al., 2008; Kowalak and Walsh, 1996). The β-methylthio-Asp also has weak density for the β-methylthio group suggesting it is also hypomodified in the present structure (Figure 1—figure supplement 6). Notably, loss of m⁷G527 methylation is synergistic with mutations in uS12 that lead to high-level streptomycin resistance (Benítez-Páez et al., 2014; Okamoto et al., 2007). Loss of m⁷G527 methylation would remove a positive charge and open a cavity adjacent to uS12, which may contribute to resistance by shifting the equilibrium of 30S subunit conformational states to an ‘open’ form that is thought to be hyperaccurate with respect to mRNA decoding (Loveland et al., 2020; Ogle et al., 2002; Zaher and Green, 2010).

Within the 30S subunit platform near the P site, the two dimethylated adenosines–m⁶₂A1518 and m⁶₂A1519–have also been connected to antibiotic resistance. Although impacting the assembly of the 30S subunit (Connolly et al., 2008) and ribosome function (Sharma and Anand, 2019), loss of methylation of these nucleotides also leads to kasugamycin resistance (Ochi et al., 2009). By contrast, bacteria lacking KsgA, the methyltransferase responsible for dimethylation of both nucleotides, become highly susceptible to other antibiotics including aminoglycosides and macrolides (O’Farrell and Rife, 2012; Phunpruch et al., 2013; Zou et al., 2018). In the present structure, m⁶₂A1519 is singly-methylated whereas m⁶₂A1518 is fully methylated (Figure 1—figure supplement 5). KsgA fully methylates both nucleotides in in vitro biochemical conditions (O'Farrell et al., 2012), but the methylation status of fully-assembled 30S subunits in vivo has not been determined. The loss of a single methylation of m⁶₂A1519, observed here for the first time, could be a mechanism for conferring low-level antibiotic resistance to some antibiotics without appreciably affecting assembly of the 30S subunit or leading to sensitivity to other classes of antibiotics, a hypothesis that could be tested in the future.

Aminoglycoside antibiotics (AGAs) are a widely-used class of drugs targeting the mRNA decoding site (A site) of the ribosome, making them an important focus for continued development against antibiotic resistance (Sati et al., 2019). Paromomycin, a 4,6-disubstituted 2-deoxystreptamine AGA (Figure 5A), is one of the best-studied structurally. Structures include paromomycin bound to an oligoribonucleotide analog of the A site at 2.5 Å resolution (Vicens and Westhof, 2001), to the small subunit of the T. thermophilus ribosome at 2.5 Å resolution (Kurata et al., 2008), and to the full 70S T. thermophilus ribosome at 2.8 Å resolution (Selmer et al., 2006). While the overall conformation of rings I–III of paromomycin is modeled in largely the same way (Figure 5B), the resolution of previous structures did not allow for unambiguous interpretation of ring IV, and only the oligoribonucleotide structural model of the decoding site includes some water molecules in the drug’s vicinity (Vicens and Westhof, 2001). Although ring IV remains the least ordered of the four rings in the present structure, the cryo-EM map of the focus-refined 30S subunit allows high-resolution modeling of the entire molecule and the surrounding solvation for the first time (Figure 5C and D). The conformation of paromomycin in the oligonucleotide structure (Vicens and Westhof, 2001) agrees most closely with the current structure, with ring IV adopting a chair conformation with the same axial and equatorial positioning of exocyclic functional groups. However, the N6’’’ group in the present structure points in the opposite direction and forms multiple contacts with the backbone phosphate groups of G1489 and U1490 (Figure 5D). Tilting of ring IV in the present model also positions N2’’’ and O3’’’ to make contacts with the G1405 and A1406 phosphate groups, respectively. The paromomycin models in the previous structures of the 30S subunit and 70S ribosome differ further by modeling ring IV in the alternative chair conformation, which also breaks the contacts observed here. We do not see paromomycin bound in H69 of the 50S ribosomal subunit, a second known AGA binding site, consistent with prior work indicating that binding of aminoglycosides to H69 may be favored in intermediate states of ribosomal subunit rotation (Wang et al., 2012; Wasserman et al., 2015).

The ribosome undergoes large conformational changes within and between the ribosomal subunits during translation, necessitating a complex set of interactions that maintain ribosome function. Contacts at the periphery of the subunit interface have been less resolved in many structures, likely due to motions within the ribosome populations. Additionally, some key regions involved in these contacts are too conformationally flexible to resolve in structures of the isolated subunits. In the present structure of the unrotated state of the ribosome, with tRNAs positioned in the A site and P site, improvement of maps of the individual ribosomal subunits and smaller domains within the subunits help to define these contacts more clearly. Helix H69 of the 50S subunit, which is mostly disordered in the isolated subunit, becomes better defined once the intact ribosome is formed. The 23S rRNA stem-loop closed by H69 is intimately connected to the 30S subunit at the end of 16S rRNA helix h44 near the mRNA decoding site and tRNA binding sites in the ribosome. During mRNA decoding, the RNA loop closing H69 rearranges to form specific interactions with the A-site tRNA (Selmer et al., 2006). The stem of H69 also compresses as the 30S subunit rotates during mRNA and tRNA translocation, thereby maintaining contacts between the 30S and 50S subunits (Dunkle et al., 2011). In the present reconstructions, helix H69 seems conformationally more aligned to the 30S subunit than the 50S subunit, as the cryo-EM density for H69 is much better defined in the map of the 30S subunit compared to the map of the 50S subunit.

The loop comprising 23S rRNA nucleotides G713-A718 closing helix H34 forms an additional bridge between the 50S and 30S subunits and is also known to be dynamic in its position (Dunkle et al., 2011). In the present reconstructions, this bridge is also better defined in the map of the 30S subunit compared to the 50S subunit, placing the more highly conserved arginine Arg88 in uS15 in direct contact with the RNA backbone of the H34 stem-loop, rather than the less conserved Arg89 (Figure 6A). An additional conformationally dynamic contact between the 30S and 50S subunits involves the A-site finger (ASF, helix H38 in 23S rRNA), which is known to modulate mRNA and tRNA translocation (Komoda et al., 2006). In the present model, although the central helical region of the ASF is only visible in low-resolution maps, loop nucleotide C888 which stacks on uS13 residues Met81 and Arg82 in the 30S subunit head domain is clearly defined (Figure 6B). Maps of the 30S subunit head domain and central protuberance of the 50S subunit also reveal clearer density defining the unrotated-state contacts between uS13 in the small subunit, uL5 in the large subunit, and bL31 (bL31A in the present structure) which spans the two ribosomal subunits.

The core of the 50S subunit is the most rigid part of the ribosome, which has enabled it to be modeled to a higher resolution than the 30S subunit, historically and in the present structure. In the present 70S ribosome and 50S subunit reconstructions, which have global map-to-model resolutions of 2.04 Å and 1.90 Å, respectively, the resolution of the core of the 50S subunit reaches 1.8 Å (Figure 1, Figure 1—figure supplement 3A, Table 2), revealing unprecedented structural details of 23S rRNA, ribosomal proteins, ions, and solvation (Figure 1—figure supplement 4, Figure 1—figure supplement 5, Figure 7—figure supplement 2). The resulting maps are also superior in the level of detail when compared to maps previously obtained by X-ray crystallography (Noeske et al., 2015; Polikanov et al., 2015), which aided in improving models of high-resolution chemical features like backbone dihedrals in much of the rRNA, non-canonical base pairs and triples, arginine side-chain rotamers, and glycines in conformationally constrained RNA-protein contacts. The density also enabled modeling of thousands of water molecules, dozens of magnesium ions, and polyamines (Figure 1—figure supplement 4, Figure 7—figure supplement 2, Table 3) The present model now even allows for comparison of ribosome phylogenetic conservation to the level of solvent positioning. For example, water molecules and ions with conserved positions in the peptidyl transferase center (PTC) can be seen in comparisons of different bacterial and archaeal ribosome structures, even when solvation was not included in the deposited models (Halfon et al., 2019; Polikanov et al., 2015; Schmeing et al., 2005; Figure 7). The central protuberance (CP) of the 50S subunit, which contacts the P-site tRNA and the head of the 30S subunit, is dynamic, however, is well-resolved with focused refinement, here reaching a resolution of 2.13 Å by GS-FSC and 2.26 Å in map-to-model FSC comparisons (Figure 1—figure supplement 3). The improved resolution of the CP aided in modeling ribosomal proteins uL5 and bL31A, as well as the CP contact to P-site tRNA.

Despite the overall high resolution of the 50S subunit core, there are a number of regions where the RNA backbone density shows relatively poor connectivity (Figure 1—figure supplement 7). In some cases, linkages between the ribose and phosphate groups become weakly visible or broken, with the phosphate group having an overall more rounded appearance than in cases where density is strong throughout the backbone. In some cases, breaks in the ribose ring are observed. Our initial impression was that this may be indicative of damage from the electron beam. However, reconstructions using the first two or three frames (corresponding to the first ~2–3 e^-/Ų in the exposure) show similar patterns of weak or broken density in these regions (Figure 1—figure supplement 7). This suggests that RNA conformational flexibility rather than radiation damage may be responsible for the broken density.

Details of post-transcriptional and post-translational modifications are also clear in the 50S subunit maps (Figure 1—figure supplement 5). Surprisingly, the post-translationally modified β-hydroxyarginine at position 81 in uL16 (Ge et al., 2012) is followed by unexplained density consistent with a thiopeptide bond between Met82 and Gly83 (Figure 8A). Adjusting the contour level of the map shows map density for the modified atom similar to that of the sulfur in the adjacent methionine and nearby phosphorus atoms in the RNA backbone, in contrast to neighboring peptide oxygen atoms. In the other cryo-EM maps of the 50S subunit (Pichkur et al., 2020; Stojković et al., 2020), the density for the sulfur in the thioamide is not visible or barely visible (Figure 8—figure supplement 1). Notably, the mass for E. coli uL16 has been shown to be 15328.1 Da and drops to 15312.1 Da with loss of Arg81 hydroxylation (Ge et al., 2012). However, this mass is still +30.9 more than the encoded sequence (15281.2 Da, Uniprot P0ADY7). In E. coli uL16 is also N-terminally methylated (Brosius and Chen, 1976), leaving 16 mass units unaccounted for, consistent with the thiopeptide we observe in the cryo-EM map. We examined a high-resolution mass spectrometry bottom-up proteomics dataset (Dai et al., 2017) to find additional evidence supporting the interpretation of the cryo-EM map as a thiopeptide. Several uL16 peptides were found across multiple experiments that matched the expected mass shift closer to that of a thiopeptide’s O to S conversion (+15.9772 Da) rather than oxidation (+15.9949), a common modification with a similar mass shift (Figure 8B). Fragmentation spectra localized the mass shift near the Met82-Gly83 bond, further supporting the presence of a thiopeptide (Figure 8C). Taken together, the cryo-EM map of the 50S subunit and mass spectrometry data support the model of a thiopeptide between Met82 and Gly83 in E. coli uL16.

The enzymes that might be responsible for the insertion of the thioamide in uL16 remain to be identified. E. coli encodes the prototypical YcaO enzyme, which can form thiopeptides but for which no substrate is known (Burkhart et al., 2017). A phylogenetic tree of YcaO family members shows a clear break separating YcaO proteins associated with secondary metabolism into a major branch (Figure 8—figure supplement 2A). A sub-grouping in the other major branch includes YcaO family members within Gammaproteobacteria (Figure 8—figure supplement 2A). The examination of genes in close proximity to YcaO across Gammaproteobacteria reveals three genes that form the focA-pfl operon involved in the anaerobic metabolism of E. coli (Figure 8—figure supplement 2B; Sawers and Suppmann, 1992). The combination of its unknown substrate in E. coli, the ability to catalyze thioamidiation in other species, and syntenic conservation in Gammaproteobacteria identify YcaO as a primary candidate for uL16 thioamidation.

E. coli 70S ribosome purification (Travin et al., 2019) and tRNA synthesis, purification, and charging (Ad et al., 2019) were performed as previously described. Briefly, 70S ribosomes were purified from E. coli MRE600 cells using sucrose gradients to isolate 30S and 50S ribosomal subunits, followed by subunit reassociation and a second round of sucrose gradient purification. Transfer RNAs were transcribed from PCR DNA templates using T7 RNA polymerase and purified by phenol-chloroform extraction, ethanol precipitation, and column desalting. Flexizyme ribozymes were used to charge the P-site tRNA^fMet with either pentafluorobenzoic acid or malonate methyl ester and the A-site tRNA^Val with valine (Goto et al., 2011). Ribosome-mRNA-tRNA complexes were formed non-enzymatically by incubating 10 µM P-site tRNA, 10 µM mRNA, and 100 µM paromomycin with 1 µM ribosomes for 15 min at 37°C in buffer AC (20 mM Tris pH 7.5, 100 mM NH₄Cl, 15 MgCl₂, 0.5 mM EDTA, 2 mM DTT, 2 mM spermidine, 0.05 mM spermine). Then, 10 µM A-site tRNA was added and the sample was incubated for an additional 15 min at 37°C. Complexes were held at 4°C and diluted to 100 nM ribosome concentration in the same buffer immediately before grid preparation. The mRNA of sequence 5’-GUAUAAGGAGGUAAAAAUGGUAUAACUA-3’ was chemically synthesized (IDT) and was resuspended in water without further purification. The Shine-Dalgarno sequence is shown in bold, and the Met-Val codons are in italics.

300 mesh R1.2/1.3 UltraAuFoil grids from Quantifoil with an additional amorphous carbon support layer were glow discharged in a Pelco sputter coater. About 4 µL of each sample was deposited onto grids and incubated for 1 min, then washed in buffer AC with 20 mM NH₄Cl rather than 100 mM NH₄Cl. Grids were plunge-frozen in liquid ethane using a Vitrobot Mark IV with settings: 4°C, 100% humidity, blot force 6, blot time 3.

Movies were collected on a 300-kV Titan Krios microscope with a GIF energy filter and Gatan K3 camera. Super-resolution pixel size was 0.355 Å, for a physical pixel size of 0.71 Å. SerialEM (Schorb et al., 2019) was used to correct astigmatism, perform coma-free alignment, and automate data collection. Movies were collected with the defocus range −0.6 to −1.5 µm and the total dose was 39.89 e^-/Ų split over 40 frames. One movie was collected for each hole, with image shift used to collect a series of 3 × 3 holes for faster data collection (Cheng et al., 2018), and stage shift used to move to the center hole. Based on the 1.2/1.3 grid hole specification, this should correspond to a maximum image shift of ~1.8 μm, although the true image shift used was not measured. The beam size was chosen such that its diameter was slightly larger than that of the hole, that is, >1.2 μm, although we have observed variation in the actual hole size compared to the manufacturer specifications.

Datasets of 70S ribosome complexes with the two differently charged P-site tRNAs were initially processed separately. Movies were motion-corrected with dose weighting and binned to the recorded physical pixel size (0.71 Å) within RELION 3.0 (Scheres, 2012) using MotionCor2 (Zheng et al., 2017). CTF estimation was done with CTFFind4 (Rohou and Grigorieff, 2015), and micrographs with poor CTF fit as determined by visual inspection were rejected. Particles were auto-picked with RELION’s Laplacian-of-Gaussian method. The 2D classification of particles was performed in RELION, and 4× binned particles were used for all classification steps. Particles were separated into 3D classes in cryoSPARC heterogeneous refinement (Punjani et al., 2017), using an initial model generated from PDB 1VY4 with A-site and P-site tRNAs (Polikanov et al., 2014) low-pass filtered to the default 20 Å resolution, and keeping particles that were sorted into well-resolved 70S ribosome classes. Particles were migrated back to RELION to generate an initial 3D-refined volume (reference low-pass filtered to 60 Å) on which to perform masked 3D classification without alignment to further sort particles based on A-site tRNA occupancy. CTF Refinement and Bayesian polishing were performed in RELION 3.1 before pooling the two datasets together, with nine optics groups defined based on the 3 × 3 groups for image shift-based data collection. The resulting 70S ribosome reconstruction was used as input for focused refinements of the 50S and 30S subunits. We used rigid-body docked coordinates for the 70S ribosome, individual ribosomal subunits or domains (30S subunit head, 50S subunit central protuberance) to define the boundaries of the map regions to be used in the focused refinements. Focused refinement of the central protuberance was performed starting from the 50S subunit-focused refinement reconstruction, and head- and platform-focused refinements started from the 30S subunit focused refinement reconstruction. Ewald sphere correction, as implemented in RELION 3.1 with the single side-band correction (Russo and Henderson, 2018; Zivanov et al., 2018), provided some additional improvements in resolution (Tables 1–2).

In addition to using the 40-frame movies, we used the first three frames corresponding to a ~3 electron/Ų dose to calculate 3D reconstructions, including focused refinements of the 30S and 50S subunits. The focused-refined map of the 30S subunit had a resolution of 2.45 Å by the map-to-model FSC metric. These maps were used to examine the density for the isoAsp in uS11, which lacked clear density for the side chain in maps reconstructed from the full 40-frame movies. We also used the maps from the initial three frames to examine connectivity in ribose density, to determine if there is visual evidence for the impact of electron damage.

The previous high-resolution structure of the E. coli 70S ribosome (Noeske et al., 2015) was used as a starting model. We used the ‘Fit to Map’ function in Chimera (Pettersen et al., 2004) to calibrate the magnification of the cryo-EM map of the 50S ribosomal subunit generated here to maximize correlation, resulting in a pixel size of 0.7118 Å rather than the recorded 0.71 Å. Focused-refined maps were transformed into the frame of reference of the 70S ribosome for modeling and refinement, using the ‘Fit to Map’ function in Chimera, and resampling the maps on the 70S ribosome grid. The 50S and 30S subunits were refined separately into their respective focused-refined maps using PHENIX real-space refinement (RSR; Liebschner et al., 2019). Protein and rRNA chains were visually inspected in Coot (Casañal et al., 2020) and manually adjusted where residues did not fit well into the density, making use of B-factor blurred maps where needed to interpret regions of lower resolution. Focused-refined maps on smaller regions were used to make further manual adjustments to the model, alternating with PHENIX RSR. Some parts of the 50S subunit, including H69, H34, and the tip of the A-site finger, were modeled based on the 30S subunit focused-refined map. The A-site and P-site tRNAs were modeled as follows: anticodon stem-loops, 30S subunit focused-refined map; P-site tRNA body, 50S subunit focused-refined map, with a B factor of 20 Ų applied; A-site tRNA body, 30S subunit focused-refined map and 50S subunit focused-refined map with B factors of 20 Ų applied; tRNA-ACCA 3’ ends, 50S subunit focused-refined map with B factors of 20–30 Ų applied. Alignments of uS15 were generated using BLAST (Altschul et al., 1997) with the E. coli sequence as reference. The model for bL31A (E. coli gene rpmE) was manually built into the CP and 30S subunit head domain focused-refined maps before refinement in PHENIX.

A model for paromomycin was manually docked into the 30S subunit focused-refined map, followed by real-space refinement in Coot and PHENIX. Comparisons to prior paromomycin structural models (PDB codes 1J7T, 2VQE, and 4V51; Kurata et al., 2008; Selmer et al., 2006; Vicens and Westhof, 2001) used least-squares superposition of paromomycin in Coot. Although ring IV is in different conformations in the various paromomycin models, the least-squares superposition is dominated by rings I–III, which are in nearly identical conformations across models.

Ribosome solvation including water molecules, magnesium ions, and polyamines was modeled using a combination of PHENIX (phenix.douse) and manual inspection. The phenix.douse feature was run separately on individual focused-refined maps, and the resulting solvent models were combined into the final 30S and 50S subunit models. Due to the fact that the solvent conditions used here contained ammonium ions and no potassium, no effort was made to systematically identify monovalent ion positions. The numbers of various solvent molecules are given in .

Along with the individual maps used for model building and refinement, we have also generated a composite map of the 70S ribosome from the focused-refined maps for deposition to the PDB and EMDB for ease of use (however, experimental maps are recommended for the examination of high-resolution features). We made the composite map using the ‘Fit in Map’ and vop commands in Chimera. First, we aligned the unmasked focus-refined maps with the 70S ribosome map using the ‘Fit in Map’ tool. We then used the ‘vop resample’ command to transform these aligned maps to the 70S ribosome grid. After the resampling step, we recorded the map standard deviations as reported in the ‘Volume Mean, SD, RMS’ tool. Then, we added the maps sequentially using ‘vop add’ followed by rescaling the intermediate maps to the starting standard deviation using the ‘vop scale’ command.

Initial real-space refinement of the 30S subunit against the focused-refined map using PHENIX resulted in a single chiral volume inversion involving the backbone of N119 in ribosomal protein uS11, indicating that the L-amino acid was being forced into a D-amino acid chirality, as reported by phenix.real_space_refine. Of the 10,564 chiral centers in the 30S subunit model, the Cα of N119 had an energy residual nearly two orders of magnitude larger than the next highest deviation. Inspection of the map in this region revealed clear placement for carbonyl oxygens in the backbone, and extra density consistent with an inserted methylene group, as expected for isoAsp. The model of isoAsp at this position was refined into the cryo-EM map using PHENIX RSR, which resolved the stereochemical problem with the Cα chiral center. IsoAsp was also built and refined into models of archaeal and eukaryotic uS11 based on cryo-EM maps of an archaeal 30S ribosomal subunit complex (PDB 6TMF; Nürenberg-Goloub et al., 2020) and a yeast 80S ribosome complex (PDB 6T4Q; Tesina et al., 2020). These models were refined using PHENIX RSR, and real-space correlations by residue calculated using phenix.model_map_cc.

All archaeal genomes were downloaded from the NCBI genome database (2618 archaeal genomes, last accessed September 2018). Due to the high number of bacterial genomes available in the NCBI genome database, only one bacterial genome per genus (2552 bacterial genomes) was randomly chosen based on the taxonomy provided by the NCBI (last accessed in December 2017). The eukaryotic dataset comprises nuclear, mitochondrial, and chloroplast genomes of 10 organisms (Homo sapiens, Drosophila melanogaster, Saccharomyces cerevisiae, Acanthamoeba castellanii, Arabidopsis thaliana, Chlamydomonas reinhardtii, Phaeodactylum tricornutum, Emiliania huxleyi, Paramecium aurelia, and Naegleria gruberi).

Genome completeness and contamination were estimated based on the presence of single-copy genes (SCGs) as described in Anantharaman et al., 2016. Only genomes with completeness >70% and contamination <10% (based on duplicated copies of the SCGs) were kept and were further de-replicated using dRep at 95% average nucleotide identity (version v2.0.5; Olm et al., 2017). The most complete genome per cluster was used in downstream analyses.

Ribosomal uS11 genes were detected based on matches to the uS11 Pfam domain (PF00411; Punta et al., 2012) using hmmsearch with an E-value below 0.001 (Eddy, 1998). Amino acid sequences were aligned using the MAFFT software (version v7.453; Katoh and Standley, 2016). The alignment was further trimmed using Trimal (version 1.4.22; --gappyout option; Capella-Gutiérrez et al., 2009). Tree reconstruction was performed using IQ-TREE (version 1.6.12; Nguyen et al., 2015), using ModelFinder (Kalyaanamoorthy et al., 2017) to select the best model of evolution, and with 1000 ultrafast bootstrap (Hoang et al., 2018). The tree was visualized with iTol (version 4; Letunic and Bork, 2019) and logos were made using the weblogo server (Crooks et al., 2004).

16S and 18S rRNA genes were identified from the prokaryotic and eukaryotic genomes using the method based on hidden Markov model (HMM) searches using the cmsearch program from the Infernal package (Nawrocki et al., 2009) and fully described in Brown et al., 2015. The sequences were aligned using the MAFFT software.

S21 sequences were retrieved from the huge phage database described in Al-Shayeb et al., 2020. Cd-hit was run on the set of S21 sequences to reduce the redundancies (Fu et al., 2012; default parameters; version 4.8.1). Non redundant sequences were used as a query against the database of prokaryotic genomes used for uS11 above using BLASTP (version 2.10.0+; e-value 1e-20; Altschul et al., 1997). Alignment and tree reconstruction were performed as described for uS11 except that we did not perform the alignment trimming step.

Similarly to uS11, the YcaO sequences were identified in prokaryotic genomes based on its PFAM accession (PF02624; Punta et al., 2012) using hmmsearch with an E-value below 0.001 (Eddy, 1998). Amino acid sequences were aligned using the MAFFT software (version v7.453; Katoh and Standley, 2016). Alignment was further trimmed using Trimal (version 1.4.22; --gappyout option; Capella-Gutiérrez et al., 2009). Tree reconstruction was performed using IQ-TREE (version 1.6.12; Nguyen et al., 2015), using ModelFinder (Kalyaanamoorthy et al., 2017) to select the best model of evolution, and with 1000 ultrafast bootstraps (Hoang et al., 2018). The tree was visualized with iTol (version 4; Letunic and Bork, 2019). The three genes downstream and upstream of each YcaO gene were identified and annotated using the PFAM (Punta et al., 2012) and the Kegg (Kanehisa et al., 2016) databases.

Masks for each map were generated in two ways. First, to calculate the map-to-model FSC curves for comparisons of the present models with the cryo-EM maps generated here, we used masked maps generated by RELION during postprocessing (Zivanov et al., 2018 ). The effective global resolution of a given map is given at the FSC cutoff of 0.5 in Table 2 and Figure 1—figure supplements 2–3. Second, we used refined PDB coordinates for the 70S ribosome, individual ribosomal subunits or domains (30S subunit head, 50S subunit central protuberance) for comparisons to the 70S ribosome map or focused-refined maps, and to previously published maps and structural models. Masks for each map were generated in Chimera (Pettersen et al., 2004) using the relevant PDB coordinates as follows. A 10 Å resolution map from the coordinates was calculated using molmap, and the surface defined at one standard deviation was used to mask the high-resolution map. For the present models and maps, the effective global resolution of a given map using this second approach was similar or slightly lower than that using the approach in RELION (within a few hundredths of an Å).

For the recent E. coli 50S subunit structure (Stojković et al., 2020), we used Chimera to resize the deposited map (emd_20353) to match the dimensions of the maps presented here. Briefly, our atomic coordinates for the 50S subunit were used with the ‘Fit to Map’ function and the voxel size of the deposited map was calibrated to maximize correlation. The resulting voxel size changed from 0.822 Å to 0.8275 Å in linear dimension. After rescaling the deposited map, we used phenix.model_map_cc to compare the map with rescaled atomic coordinates deposited in the PDB (6PJ6) or to the present 50S model, yielding a map-to-model FSC of 0.5 at ~2.5 Å. Similar comparisons of the structure of the Staphylococcus aureus 50S subunit to the deposited map (PDB 6S0Z, emd_10077; Halfon et al., 2019) yielded a map-to-model FSC of 0.5 at 2.43 Å, accounting for a change in voxel linear dimension from 1.067 Å to 1.052 Å. For comparisons to the map and model deposited by Pichkur et al., 2020 (EMD-10655 and PDB 6XZ7), we removed tRNA bodies, the L1 arm, and the GTPase-associated-center coordinates from 6XZ7 since these regions are disordered or missing in the deposited 50S subunit reconstruction.

Previously published E. coli tryptic peptide mass spectrometry (MS/MS) raw data was used for the analysis (Dai et al., 2017; MassIVE accession: MSV000081144). Peptide searches were performed with MSFragger (Kong et al., 2017) using the default parameters for a closed search with the following exceptions: additional variable modifications were specified on residues R (hydroxylation, Δ mass: 15.9949) and M (thioamide, Δ mass: 15.9772), maximum modifications per peptide set to four, and multiple modifications on a residue were allowed. Spectra were searched against a database of all E. coli proteins plus common contaminants concatenated to a decoy database with all original sequences reversed. Results were analyzed using TPP (Deutsch et al., 2015) and Skyline (Pino et al., 2020).

Cryo-EM maps were supersampled in Coot for smoothness. Figure panels showing structural models were prepared using Pymol (Schrödinger) and ChimeraX (Goddard et al., 2018). Sequence logo figures were made with WebLogo 3.7.4 (Crooks et al., 2004). Phylogenetic trees were visualized with iTol (version 4; Letunic and Bork, 2019) and multiple alignments were visualized with geneious 9.0.5 (https://www.geneious.com).

Ribosome coordinates have been deposited in the Protein Data Bank (entry 7K00), maps in the EM Database (entries EMD-22586, EMD-22607, EMD-22614, EMD-22632, EMD-22635, EMD-22636, and EMD-22637 for the 70S ribosome composite map, 70S ribosome, 50S subunit, 30S subunit, 30S subunit head, 30S subunit platform, and 50S subunit CP maps, respectively), and raw movies in EMPIAR (entry EMPIAR-10509).