We previously used forward genetic screens to identify strains of Mycobacterium smegmatis with extremely high specific rates of mistranslation – i.e. mistranslation of Asn→Asp. These screens identified mutations in the essential amidotransferase genes gatCAB that resulted in high rates of mistranslation of Gln/Asn, specifically [12, 36]. We hypothesized that the mycobacterial genome may encode for other ‘fidelity factors’ that could modulate mistranslation rates in the background of a compromised indirect tRNA aminoacylation pathway. We, therefore, designed a suppressor screen strategy in the strain HWS19 [12], which encodes a three amino acid deletion in gatA and, consequently, has extremely high rates of translational error of Gln→Glu and Asn→Asp specifically (Fig 1B). We wondered whether HWS19, with a high background mistranslation rate of Gln/Asn codons, is more susceptible to aminoglycosides, such as streptomycin that are known to increase ribosomal decoding errors across multiple codons [33, 39]. Plating of equivalent colony forming units (CFU) of strain HWS19 on low-dose streptomycin-agar led to the recovery of significantly fewer colonies compared with wild-type (Fig. S1), confirming HWS19 was hypersusceptible to streptomycin. The basis of our suppressor screen, therefore, was as follows: plating of HWS19 onto low-dose streptomycin-agar should select for strains with mutations that decrease mistranslation on this high mistranslating background (Fig. 1B, 1C). Suppressors identified from selection on low-dose streptomycin agar were then tested using our custom dual-luciferase mistranslation reporter system to verify that they had decreased error rates (Fig. 1C).

To identify mutants leading to decreased mistranslation, we plated 1×10⁷ CFU onto each of 6 agar plates containing 1 µg/mL streptomycin. The remaining survivors were transformed with dual-luciferase reporter plasmids that measured specific mistranslation errors of asparagine for aspartate [12, 17, 40] to identify low mistranslating candidates (Fig. 1C). We initially identified 4 suppressor mutants with lower mistranslation rates compared with the parent HWS19 strain, and comparable to wild-type M. smegmatis (Fig. 1D). Sequencing the strains revealed three of the four had mutations in the 16S rRNA methyltransferase gidB (Msmeg_6940) – Table S1. The remaining suppressor (C-23) had multiple mutations including in the gene, tuf, coding for the aminoacyl-tRNA chaperone EF-Tu. (Table S1). We subsequently sequenced just the gidB gene of a further 10 candidate suppressors and identified a further seven with a total of three additional independent mutations in gidB (Table S2). All but one of these further suppressors also had reduced mistranslation rates (Fig. S2), strongly supporting that mutations in gidB are able to suppress the high specific mistranslation rates in strain HWS19 that arise from mischarging tRNA.

Loss of function mutations in GidB can cause low-level streptomycin resistance: lack of methylation in 16S rRNA interferes with streptomycin binding [41, 42]. We were, therefore, concerned that our candidates from the screen may simply represent selection against streptomycin. To determine whether the observed phenotype was independent of streptomycin selection, we deleted gidB in both the HWS19 and parental (wild-type) M. smegmatis backgrounds in the complete absence of streptomycin selection. Deletion strains were complemented with a chromosomally integrated plasmid expressing gidB from its native promoter (Fig. S3). We measured specific mistranslation rates using two complementary reporter systems: the dual-luciferase system, used in the initial screen, as well as a dual-fluorescent reporter system (Fig. S4, S5) that uses flow cytometry to measure relative mistranslation rates [12]. This reporter, a GFP-mRFP fusion protein had a glutamate necessary for GFP fluorescence mutated to glutamine, abrogating green, but not red fluorescence [12]. Increased specific mistranslation of Gln→Glu would result in increased GFP fluorescence and hence increased GFP/RFP ratios. Measurement of specific mistranslation rates with both reporters verified that deletion of gidB significantly increased translational fidelity in strain HWS19, and this phenotype could be readily complemented (Fig. 2A, B, left panel). Surprisingly, deletion of gidB had no phenotype using the same reporters on a wild-type background (Fig. 2A, B, right panel).

Our results suggested that deletion of gidB increased translational fidelity only in a mutant strain (HWS19) that had basal extremely high rates of mistranslation arising from lack of quality control of mischarged asparagine and glutamine tRNAs and that deletion of gidB had no phenotype in wild-type mycobacteria grown under standard laboratory conditions. We wondered whether deletion of gidB had a phenotype in wild-type mycobacteria, but only under conditions associated with increased rates of mistranslation. We had previously shown that exposure of wild-type mycobacteria to rifampicin at 1x minimal bactericidal concentration (MBC) could select for survival and growth of wild-type mycobacteria due to two reversible and non-genetic mechanisms: increased rates of specific mistranslation [12] and a semi-heritable survival program [43], and not due to mutations in the rifampicin’s target rpoB or other mutations. We plated wild-type M. smegmatis on non-selective LB medium or low-dose (1xMBC) rifampicin-agar (Fig. 3A). As before, deletion of gidB isolated from non-selective medium had no phenotype (Fig. 3B). However, bacteria that survived and grew on rifampicin-agar had increased rates of mistranslation as measured by the dual-fluorescent reporter, in keeping with prior observations [12, 43]. These increased mistranslation rates were reverted by deletion of gidB in fully complementable manner (Fig. 3B), suggesting that deletion of gidB increased translational fidelity under growth conditions causing high rates of mistranslation, even in wild-type bacteria.

The use of a dual-fluorescent reporter and flow cytometry for measurement of relative mistranslation rates allowed us to measure the impact of gidB deletion on different subpopulations of mycobacteria in the same experiment: i.e. we could selectively gate on heterogeneous subpopulations that had relatively high or low GFP/mRFP ratios, representing high and low mistranslation rates specifically (Fig. 3C). This allowed us to ask the question as to whether deletion of gidB was sufficient, as well as necessary, for the discrimination of mischarged tRNA. We gated on bacteria with the highest (10%) and lowest (10%) green/red (i.e. relative mistranslation rates) ratios from the two experimental conditions, i.e., bacteria isolated from ‘standard’ non-selective LB agar, and from low-dose rifampicin agar, which enriched for bacteria with higher average mistranslation rates. Deletion of gidB had no observable phenotype on translational fidelity in bacteria isolated from non-selective media, even when only the cells with the highest apparent mistranslation rates were analysed (Fig. 3D). By contrast, in bacteria enriched for higher mistranslation rates (rifampicin exposure) gidB deletion could revert the mistranslation phenotype in a complementable manner (Fig. 3E). Therefore, gidB deletion appears necessary, but not sufficient for translational fidelity, and it is only within environmental contexts that enrich for high rates of mistranslation that deletion of gidB causes increased ribosomal discrimination of misacylated tRNA in otherwise wild-type mycobacteria.

For ribosomal decoding, there is consensus that there is a trade-off between speed versus accuracy [44, 45]. We first measured growth rates of both WT M. smegmatis and HWS19 strains with/without deletion of gidB to determine if the deletion impacted general fitness. Growth rates in both cases were identical between the “wild-type”, gidB deletion and complemented strains (Fig. S6). We then measured translational efficiency through induction of Nluc luciferase and measurement of Nluc activity, as before [32]. Deletion of gidB had no impact on efficiency of Nluc translation in either strain, suggesting that in axenic culture, lack of GidB does not impact rate of synthesis of this protein (Fig. S7).

The previous experiments confirm that deletion of gidB increased discrimination of physiologically mischarged Glu-tRNA^Gln and Asp-tRNA^Asn. We next wanted to ask whether this discrimination of mischarged tRNA was a general property that would allow discrimination of any mischarged tRNA. To test this idea, we leveraged the unique properties of tRNA^Ala and their cognate synthetase. To ensure the accurate charging of tRNAs, most aminoacyl tRNA synthetases depend on recognizing the anticodon of their cognate tRNAs as an identity element. For alanyl synthetase (AlaRS), the sole identity element in all clades of life is a G³•U base-pair in the tRNA acceptor stem, and AlaRS can charge any tRNA with this identity element, regardless of anti-codon identity, i.e. the G³•U base-pair is both necessary and sufficient for alanine aminoacylation by AlaRS [46–48]. Therefore, the anticodon of any alanyl-tRNA can be mutated to any triplet, and the tRNA will still be charged with aminoacyl-alanine [46] and will mediate specific translational errors. We had previously mutated the anticodon of a mycobacterial alanyl-tRNA to CCA, coding for tryptophan, which would result in specific mistranslation of alanine for tryptophan at UGG codons [17] – Fig. 4A. To measure specific mistranslation, we modified the dual-luciferase reporter system. We identified a critical alanine residue in Renilla luciferase, which, when mutated to tryptophan (A214W), resulted in >20-fold decrease in activity (Fig. S8). This reporter would thus be able to discriminate mistranslation errors greater than 5%/codon of tryptophan to alanine. We transformed the mutant alanyl-tRNA (tRNA_CCA^Ala), cloned into a tetracycline-inducible plasmid into wild-type and ΔgidB M. smegmatis, along with the specific dual-luciferase reporters. Induction of tRNA_CCA^Ala resulted in increased specific mistranslation of tryptophan for alanine, as expected. Deletion of gidB did not increase translational fidelity, even in this high mistranslating context (Fig. 4B). Next we transformed the mutant alanyl-tRNA and reporters into The high basal mistranslation (Gln→Glu, Asn→Asp, specifically) strain HWS19 and measured the rates of Trp→Ala mistranslation using the new Ren-FF dual-luciferase reporter. Deletion of gidB decreased rates of Trp→Ala mistranslation as measured by rescue of Renilla activity in the HWS19 background, suggesting that in this context, ΔgidB ribosomes could discriminate against even Trp→Ala mistranslation (Fig. 4C). Surprisingly, this phenotype is only observed in this strain with high rates of ‘physiological’ mistranslation and not in the WT strain. Taken together, these results suggested deletion of gidB was necessary but not sufficient for discrimination of mischarged tRNAs and that a further environmental context, possibly generated via excess ‘physiological’ mistranslation or some other stressor was also required.

We tested the physiological relevance of increased translational fidelity caused by gidB deletion in rifampicin antibiotic tolerance. We had previously demonstrated that increased, specific, mistranslation due to Glu-tRNA^Gln and Asp-tRNA^Asn mischarging caused increased tolerance to the antibiotic rifampicin due to mistranslation of critical residues in the drug target, RpoB [12, 17, 36]. Since antibiotic tolerance is mediated by a subpopulation of bacteria that resist killing [49–59], we hypothesised that deletion of gidB, and the subsequent increase in translational fidelity, would render the most tolerant subpopulation susceptible to rifampicin. Deletion of gidB resulted in significantly increased killing of M. smegmatis exposed to rifampicin in axenic culture in both the high mistranslating strain HWS19 and, crucially, also in wild-type mycobacteria (Fig. 5A, B). This suggests that rifampicin treatment is a sufficient condition for deletion of gidB to exert its effect.

As a first step to identify the molecular mechanism of changes in adaptive mistranslation, we wished to resolve the structure of ribosomes from HWS19 and HWS19-ΔgidB cells for analysis by cryo-electron microscopy. Sucrose gradient ultracentrifugation and ribosome subunit profiling of ribosomes from the strains did not reveal difference in the distribution of different ribosome species, suggesting that deletion of gidB did not alter ribosome biogenesis (Fig. S9). Three dimensional reconstructions of 70S fractions isolated from sucrose gradient and the models refined into those maps did not reveal significant changes in structure, likely due to the limited resolution (Fig. 6). Analysis of the resulting models does point to a hypothesis for the modulation of mistranslation in ΔgidB ribosomes as a function of GatCAB activity: an interaction between 16S G507 (the target of GidB methylation) and S12 Asn46 (Fig. 6). When GatCAB functions, minimizing mistranslation, as in WT M. smegmatis, Asn46 is reliably coded as Asn; however, in the HSW19 background and other conditions with diminished GatCAB activity, Asn46 will be coded as Asp at a much higher frequency. The interaction between G507 and the S12 loop containing Asn46 has previously been implicated in stabilizing tRNA-mRNA interactions during aminoacyl-tRNA selection in the decoding process in M. tuberculosis [42]. Phenotypic misincorporation from Asn to Asp likely alters this interaction and the conformational dynamics of this key loop. Further structural work, at higher resolution and including structures of translating ribosomes, will be needed to unravel how the interactions between the G507 methylation, tRNA identity, mRNA structure, and antibiotic binding affect the relative rates of adaptive mistranslation.

Wild-type M. smegmatis mc²-155 [73] and its derivatives were cultured in Middlebrook 7H9 Broth supplemented with 0.2% glycerol, 10% ADS (albumin-dextrose-salt) and 0.05% Tween-80 with corresponding antibiotics. Wild-type E. coli HK295 and its derivatives were cultured in LB Broth with corresponding antibiotics. E. coli DH5a and TOP10 were used for transformation of plasmids and amplification. LB-Agar was used for both M. smegmatis and E. coli in plating experiments. If not otherwise mentioned, bacteria were grown in 37 C, 220 rpm shaker or maintained in 37C incubator.

The 500 bp upstream and downstream regions of gidB were amplified from mc²-155 genomic DNA. The zeocin resistance cassette was amplified from plasmid pKM-Zeo-Lox (A kind gift from Eric J. Rubin lab). Then the zeocin resistance cassette flanking 500 bp upstream and 500 bp downstream region of gidB were assembled by overlapping extension PCR for use as an allele exchange substrate (AES), which was verified by sequencing. WT mc²-155 and HWS19 transformed with pNIT(kan)::RecET::sacB recombineering plasmid were grown to OD₆₀₀∼0.4, and expression of recET was induced with 10 μM isovaleronitrile (IVN) for 5 h, then competent cells were made by standard methods and transformed with 2 μg of the AES. The cells were recovered for 4 h and selected on LB agar plates containing 20 μg/mL zeocin. The recombinants were confirmed by PCR for verifying the zeocin cassette integration and the appropriate genomic context (Figure S3) (See verification primers sequences in Table S3). gidB deletion strains were then cured of the recombinase plasmid prior to further experiments.

For gidB complementation construction, several plasmids were constructed. The chromosome integrating mycobacterial plasmid pML1357 (Addgene #32378) was used as the backbone vector. The hygromycin resistance cassette was replaced with a kanamycin resistance cassette and streptomycin resistance cassette as pKML1357 and pSML1357 respectively. gidB from wild-type M. smegmatis with and without native promoter were amplified and cloned into integrative plasmids pKML1357/ pSML1357 to construct pKML1357-Pnative_gidB, pKML1357-Psmyc_gidB, pSML1357-Pnative-gidB. Then the plasmids were transformed into Δ gidB strains and selected onto LB agar with corresponding antibiotics for gidB complementation strains. ΔgidB::Pnative_gidB (KanR) strains were used for dual luciferase mistranslation assay when measuring N to D and Q to E mistranslation rates. Δ gidB::Psmyc_MsmgidB (KanR) were used for the dual-fluorescence mistranslation assay. Δ gidB::Pnative_gidB (StrepR) strains were used for dual-luciferase mistranslation assays when measuring W to A mistranslation as the pACET-Renilla-Firefly construct carries a kanamycin resistance marker. In addition, gidB mRNA levels in gidB deletion and complementation strains were verified by RT-qPCR (Figure S3).

For measuring mistranslation rates in M. smegmatis, the shuttle plasmids pTetG-Renilla-Firefly (WT), pTetG-Renilla-D120N-Firefly, pTetG-Renilla-E144Q-Firefly under control of a tetracycline-inducible promoter were used as previously [17]. For measurement of tryptophan to alanine mistranslation rates, the mycobacterial chromosome integrating plasmid pACET-Renilla-Firefly, where expression of the reporter is under control of an acetamide-inducible promoter was used [17]. The Renilla gene was mutated by site-directed mutagenesis to Renilla-A214W.

In mycobacteria, the assay was performed as previously [12]. Briefly, strains with pTetG-Renilla-Firely were cultured to OD600>3, and diluted to OD600∼0.2 into fresh 7H9 medium supplemented with 50μg/ml hygromycin and 100 ng /ml anhydrotetracycline for inducing dual luciferase. After 7-8 hours, cells were collected by centrifugation at 12000rpm for 3min, lysed with 40μL passive lysis buffer in 96-well white flat bottom plate for 20-30min (plate shaking at 400rpm, room temperature), then reacted with 50μL substrate Firefly reagents, 360rpm shaked 5s followed by measuring the Firefly luminescence by Fluoroskan Ascent FL luminometer with 1000ms integration time. Then 50μL Stop&Glo reagent was added followed by measuring the Renilla luminescence immediately. For strains with pACET-Renilla-Firely and pTet-tRNA_CCA^Ala [17], the strains in experimental group were diluted to OD600∼0.2 into fresh 7H9 medium supplemented with 50μg/ml hygromycin, 20% acetamide and 100ng/mL anhydrotetracycline for inducing dual luciferase and tRNA_CCA^Ala respectively. The later procedures were same as above. The corrected Renilla/Firefly representing the mistranslation level is calculated as follows: Corrected Renilla / Firefly (N to D) = Renilla(D120N)/FireflyRenilla(WT)/Firefly\begin{document}$\frac{\operatorname{Renilla}(D 120 N) / F i r e f l y}{\operatorname{Renilla}(W T) / F i r e f l y}$\end{document}, Corrected Renilla / Firefly (Q to E) = Renilla(E144Q)/FireflyRenilla(WT)/Firefly\begin{document}$\frac{\operatorname{Renilla}(E 144 Q) / F i r e f l y}{\operatorname{Renilla}(W T) / F i r e f l y}$\end{document}, Corrected Renilla / Firefly (W to A)= Renilla(A214W)/FireflyRenilla(WT)/Firefly\begin{document}$\frac{\operatorname{Renilla}(A 214 W) / F i r e f l y}{\operatorname{Renilla}(W T) / F i r e f l y}$\end{document}.

The pUVtetOR-GLR (green-linker-red) reporter was adapted as previously [12] with some modifications. Briefly, a mutated green fluorescent protein (E222Q) was fused to wild-type monomeric red fluorescent protein (mRFP) with a flexible GGSGGGGSGGGSSGG linker. A C-terminal tag (AANDENYAAAV) targeting the protein for proteolytic degradation [74] was added by PCR.

The procedure of measuring relative mistranslation rates by flow cytometry was as previously described [12] with some modifications. Strains with pUVtetOR-GLR were cultured to OD600∼0.2 and then 100ng/ml ATc was added for inducing dual-fluorescence. After 3 hours, cells were diluted into 1mL PBS to OD600∼0.05. The GFP and RFP signals of the sample were collected by flow cytometry (BD LSR Fortressa), being excited/ detected by 488nm/520nm laser line and 561nm/585nm laser line respectively. For the strains after high mistranslating selection: Strains expressing pSML1357-Aph_D214N and pUVtetOR-GLR were spread onto LB agar containing 50μg/ml Hygromycin and 2μg/ml Kanamycin as experimental group and LB agar containing only 50μg/ml Hygromycin as control group. After 5 days, colonies were scraped from the plates and cultured into 7H9 with 100ng/ml ATc for 3 hours. The following procedures were same as above. For the strains under rifampicin condition: strains with pUVtetOR-GLR were spread onto LB agar no selection plates and LB agar containing 20μg/ml rifampicin. After 5 days, colonies were then scraped from the plates and cultured into 7H9 with 100ng/ml ATc for 3 hours. The following procedures were same as above. Data were analyzed by FlowJo 10.4 for Windows 10. The gating strategy was as follows: Live bacteria were gated to eliminate debris as P1 based on SSC-A and FSC-A, then stringent gating on single cells was applied using a tandem gating strategy based on FSC-A and FSC-H as P2, subsequently using SSC-A and SSC-H as P3. Single cells with positive red fluorescence were acquired as P4 based on the red fluorescence signal. See Figure S4. Relative mistranslation rates were analysed as a histogram of GFP/ RFP ratio as P5 . The mean values of GFP/RFP ratio were calculated by FlowJo software (Flowjo 10.4 for Windows 10).

Wild-type and strain HWS19 M. smegmatis were cultured in 7H9 medium until OD600∼1. Cells were diluted to OD600∼0.3 into fresh 7H9 medium containing 20μg/ml rifampicin and cultured into 37°C, 220rpm shaker. Prior to addition of rifampicin, an aliquot was removed for calculation of cell numbers at time zero. At indicated time points, aliquots were taken, washed 3x in PBS and resuspended in PBS prior to being plated onto LB-agar. At least 3 10-fold dilutions were plated per time point. After 3-5 days, visible colonies were counted. The counts at different time points were normalized by counts at time zero for analysis.

To assess the relative translation rate, we adapted a previously described protocol [32] with minor modifications. M. smegmatis strains harboring the Nluc reporter construct was first grown in 7H9 medium to a stationary phase (ODLLL > 3). Cultures were then diluted to an ODLLL of 0.2 in fresh 7H9 medium, and anhydrotetracycline (ATc) was added to a final concentration of 50 ng/mL to induce reporter expression. After 5 minutes of induction, cells were pelleted, washed twice with fresh 7H9 medium (without ATc), and resuspended in the same medium supplemented with or without 200 μg/mL chloramphenicol (a translation inhibitor). Aliquots were collected every 10 minutes and immediately snap-frozen in liquid nitrogen. Once all time points were collected, luminescence was measured using a luminometer. All experiments were performed in triplicate using three independent biological replicates.

HWS19 and HWS19-ΔgidB strains were grown in Middlebrook 7H9 Broth supplemented with 0.2% glycerol, 10% ADS (albumin-dextrose-salt) and 0.05% Tween-80 to OD 0.8. Bacteria pellet were collected by centrifuge at 4000rpm at 4C for 10 minutes. Pellet was washed and resuspend with polysome buffer (50 mM Tris-actate pH=7.2, 12 mM MgCl2, 50mMNH₄Cl, 1mM DTT). Bacteria resuspension was lysed by MP FastPrep-24 bead beating system and the lysate was centrifuged at 12,000rpm 4C for 10 minutes. Cleared lysate was loaded on to 10% - 50% linear sucrose gradient with Beckman SW40Ti rotor at a speed of 39, 000rpm at 4C for 4 hours. Gradient were fractioned with continuous A260 measurement and the correspond 70S fraction were collected, buffer exchanged for downstream cryo-EM studies.

Samples were diluted in milliQ ddH₂O and deposited onto glow-discharged (EMS-100 Glow Discharge System, Electron Microscopy Sciences, 30 s at 15 mA and 22 mbar) copper Quantifoil (Quantifoil Micro Tools) grids, 1.2/1.3 spacing, 300-mesh, coated in 2 nm amorphous carbon. Grids were incubated for 30 s under 100% humidity and 10°C, blotted once at force 3 with Whatman #1 filter paper, and plunge-frozen in liquid ethane using a FEI Vitrobot Mark IV (Thermo Fisher). Grids were screened for ice quality on an FEI Talos Arctica (Thermo Fisher, 200 kV) or FEI Titan Krios (Thermo Fisher, 300 kV) transmission electron microscope at UCSF. WT strain non-mutant samples were shipped by dry shipper to the National Center for CryoEM Access and Training (NCCAT) at NYSBC. These were imaged on an FEI Titan Krios (Thermo Fisher, 300 kV) electron microscope with a Falcon IV camera using Leginon and without an energy filter. All other samples were imaged at UCSF on an FEI Titan Krios (Thermo Fisher, 300 kV) electron microscope with a Gatan K3 direct electron detector, using SerialEM and with an imaging filter (20 eV slit). Datasets at UCSF were collected with fringe-free imaging using a nine-shot beam-image shift approach with coma compensation. Further details are reported in Table 1. For all datasets, image stacks were binned by a factor of 2, motion corrected, and dose weighted using UCSF MotionCor2. Contrast transfer function (CTF) initial parameters were determined using CTFFIND4 in cisTEM (development version). After excluding images with poor CTF fits or poor ice quality, dose-weighted micrographs were subjected to unsupervised particle picking with a soft-edged disk template. Two-dimensional classification of the resulting particle stacks was used to select the images carried forward. Ab initio reconstructions were carried out for each dataset to produce initial references for three-dimensional reconstruction, which proceeded through auto refinement and manual refinement in two to three classes as necessary to exclude non-ribosome particles and 50S subunits. Final reconstructions were generated without sharpening from the classes yielding the 70S ribosome in the latest (highest resolution) refinement cycle prior to the emergence of overfitting artifacts in the Fourier shell correlation (FSC). Map resolutions are reported as particle FSC at 0.143.

An initial atomistic model of the 70S ribosome was generated by fitting the 5o5j and 5o60 PDB models of the small and large ribosomal subunits in Mycobacterium smegmatis, respectively, into the final map of the HWS19 unmutated dataset using UCSF ChimeraX. This model was then subjected to several iterations of rigid body fitting in UCSF ChimeraX or PyMOL, molecular dynamics model fitting in ISOLDE, model building/trimming/correction in Coot or UCSF ChimeraX, and real-space model refinement with anisotropic displacement parameter (ADP) refinement in Phenix, not necessarily in that order, until convergence. Restraints for residues not distributed with Phenix libraries were generated using Schrödinger and Phenix. The final HWS19 unmutated model was re-fitted and re-refined into the other three maps, with further model trimming as needed where map quality was not sufficient to resolve the structure. Figures were prepared using the PyMOL Molecular Graphics System (Schrödinger).

All statistical methods are described in figures legends, presented as mean with SD. The p values and statistical significance were calculated using GraphPad prism software (Prism 8 for Mac). Two-tailed unpaired Student’s t test was used to compare means between groups. ***p<0.001, **p<0.01, *p<0.05, ns p>0.05.

HWS optimised the conditions of and performed the initial screen. ZB performed most experiments with assistance from JYH and YC. YXC purified the 70S ribosomes, interpreted data and modified all figures. HJ performed additional experiments for the revised manuscript. ZB and YXC analysed the results. IYand JSF solved the cryo-EM structure and MTD and JSF revised the analysis and prepared revised structural figures. BJ conceived of and supervised the study. ZB, YXC and BJ wrote the manuscript with input from the other authors.