Effect of fusA mutations on bacterial growth, antibiotic resistance, and proteome composition

To explore how fusA mutations confer resistance to AGAs, we constructed E. coli MG1655 strains harboring three frequently reported laboratory-evolved EF-G variants (F593L, A608E, and P610L) (Fig. 1a), which belong to a prominent cluster of resistance mutations in EF-G domain IV (Supplementary Data 1), along with an isogenic wild-type control generated by the same cloning strategy but retaining the wt sequence (see "Methods", Supplementary Fig. 3a-d). Resistance mutations in the same region of EF-G have been identified in ESKAPE pathogens, suggesting a shared resistance mechanism against AGAs across different bacterial species. EF-G variants F593L and A608E caused only a slight reduction in the growth rate, whereas P610L had a stronger effect (Fig. 1b). All three mutations conferred resistance to a range of AGAs, including Apr, Gen, KanA, KanB, neomycin (Neo), ribostamycin (Rib), and sisomicin (Sis) (Fig. 1c). Despite belonging to different structural classes of AGAs (Supplementary Fig. 1a), these AGAs share a common mode of action: they slow down translocation and induce misreading and error cluster formation. The P610L mutant has a stronger negative impact on cell growth (Fig. 1b), yet it conferred the same level of resistance as the F593L and A608E mutants. This suggests that the reduced growth rate alone cannot fully explain the observed resistance, thus challenging models that attribute resistance to just a slower translation rate, which would provide the cell with more time to pump out AGAs (see Supplementary Fig. 2 for details of putative resistance mechanisms; models 1 and 2). Furthermore, the mutations did not confer significant resistance to neamine (Nea), which slows translocation and induces misreading at high concentrations, but does not induce error clusters. The mutant strains also remained sensitive to streptomycin (Str), which induces errors and error clusters but does not strongly inhibit translocation. These observations show that the mutant strains are not generally tolerant towards translational misreading and proteotoxic stress. Sensitivity to spectinomycin (Spc), which inhibits translocation without inducing misreading, and kasugamycin (Ksg), which does not bind to the ribosomal A site, remained unchanged, further supporting the notion that fusA mutant strains do not exhibit broad resistance to all AGAs. Together, these observations suggest that fusA mutations specifically counteract the effects of AGAs that both disrupt translocation and induce misreading with error cluster formation.

Furthermore, the fusA mutations did not confer resistance to other bactericidal antibiotics such as carbenicillin (Car), a cell wall synthesis inhibitor, or norfloxacin (Nor), a gyrase inhibitor. Since these antibiotics -- as well as some AGAs -- have been reported to kill bacteria by escalating metabolic stress, the lack of resistance suggests that fusA mutations do not mitigate this common metabolic death pathway. Instead, their resistance appears to be specific to a subset of AGAs, likely through a more targeted mechanism affecting translation.

To determine whether fusA mutations might confer resistance to AGAs indirectly by inducing expression of resistance proteins or redirecting existing pathways (Supplementary Fig. 2, model 2), we analyzed the proteomes of the three mutant strains compared to the parental MG1655 and wt strains. The fusA mutations did not alter the expression patterns of proteins directly associated with AGA entry (e.g., porins, sugar and peptide transporters), AGA removal (e.g., efflux pumps), AGA-induced stress adaptation (e.g., ribosome silencing factors, global stress modulators), AGA response regulation (e.g., stress-associated transcription factors), and AGA-induced proteostasis maintenance (e.g., proteases and chaperones) (Supplementary Fig. 4a). Overall, the proteomic changes induced by the mutations were minimal for F593L and A608E, and moderate for P610L, with at most 63 out of 2527 proteins showing more than a 2-fold change (Fig. 1d), and correlated with growth rate differences (Fig. 1b and Supplementary Fig. 4b-d) but not with resistance patterns (Fig. 1c and Supplementary Fig. 4e). Thus, these adaptations are likely due to changes in translation rates (Fig. 1d and Supplementary Fig. 4b-f), but do not account for the observed AGA resistance, further challenging the validity of models suggesting an indirect effect of fusA mutations on adaptation (Supplementary Fig. 2, model 2).

Notably, the most upregulated proteins, such as enzymes of the Leu and Ile synthesis pathway (Fig. 1d), are regulated by translation attenuation mechanisms normally activated during starvation. Under starvation, translation slows down due to low aminoacyl-tRNA levels, promoting the formation of alternative mRNA secondary structures that enhance full-length transcription of Leu and Ile biosynthetic operons, resulting in increased expression levels of enzymes involved in Leu and Ile synthesis. The same regulatory pathway is activated in fusA mutant strains, presumably due to slower translation, stimulating the production of these anabolic enzymes under non-starvation conditions. A similar misregulation seems to occur in the pyrimidine synthesis pathway in fusA mutant strains. Under nutrient-rich conditions, high UTP levels speed up transcription of uridine-rich regulatory sequences, resulting in the loss of transcription-translation coupling and downregulation of key enzymes of pyrimidine biosynthesis, PyrB and PyrI. The observed down regulation of PyrB and PyrI in fusA mutants (Fig. 1d) is consistent with previous reports on the effects of ribosome variants, which showed that slower translation leads to a transcription-translation decoupling and a misregulation of the expression of the pyrBI operon. While fusA mutations activate attenuation pathways, suggesting alterations in local translation speed, they have little effect on cell growth, implying that EF-G variants still promote efficient mRNA translocation.

To test how fusA mutations affect translation, we performed in-vitro experiments with EF-G variants F593L, A608E, and P610L, as well as additional variants within the prominent mutation cluster in domain IV (F593C, F605I, A608V, and P610Q). We also tested five additional mutations (G117C, R371L, T674A, A678V, and Y680C) located throughout different domains of EF-G (Fig. 1a and Supplementary Data 1). Notably, the last three mutations (T674A, A678V, and Y680C) belong to a cluster of mutations that are frequently found in P. aeruginosa (residues 668-680). We analyzed the in-vitro translocation activity of the purified EF-G variants in the absence of AGAs using a time-resolved assay that monitors the kinetics of individual translocation events. The EF-G variants exhibited a moderate reduction in translocation rates, ranging from 1.3- to 4-fold (Fig. 2a). For the F593L, A608E, and P610L variants, the results align with the observed changes in growth rates and the proteome changes (Supplementary Fig. 5a). Similarly, a translation assay monitoring the time course of in-vitro synthesis of the model protein SlyD revealed that EF-G variants have only minor effects on translation speed in the absence of AGA (Fig. 2b and Supplementary Fig. 5b).

In the presence of low, subsaturating Apr concentrations, translation with wt EF-G was slower, as expected. Notably, the EF-G resistance variants did not alleviate the inhibitory effect of Apr; instead, translation was even slower than with wt EF-G. This suggests that EF-G resistance variants neither displace the drug from the ribosome nor enable efficient translation on AGA-bound ribosomes, ruling out the 'drug displacement' and 'gain-of-function' models proposing that variant EF-G accelerates translocation in the presence of the drug (Supplementary Fig. 2, models 3,4). With increasing Apr concentration, more ribosomes bound Apr, leading to a dramatic decrease in the yield of full-length SlyD (Fig. 2c and Supplementary Fig. 5c). While translation with wt EF-G continued to produce SlyD even at high Apr concentrations, the EF-G resistance variants substantially reduced SlyD formation at concentrations above 10 µM, indicating that these variants specifically impair translation on the AGA-bound ribosomes. To better assess the underlying kinetic effect on translocation on AGA-bound ribosomes, we measured the rates of individual translocation events at high Apr and KanA concentrations (250 µM), at which nearly all ribosomes carry AGAs (Fig. 2d). As expected, Apr and KanA substantially reduced translocation rates for wt EF-G, without completely blocking it. Notably, none of the twelve EF-G variants alleviated the translocation inhibition (Fig. 2d), effectively ruling out resistance mechanisms in which EF-G variants compensate for the inhibitory AGA effect (Supplementary Fig. 2, model 3) or displace the drug allosterically (Supplementary Fig. 2, model 4). Instead, for most EF-G variants, especially the frequently observed variants in EF-G domain IV including F593L, A608E, and P610L, translocation rates substantially decreased in the presence of AGAs, supporting the idea that these mutants are selectively hindered in promoting translocation on AGA-bound ribosomes (Supplementary Fig. 2, model 5).

Next, we used time-resolved cryo-EM to visualize the mechanism of translation inhibition and confirm that the AGA is not displaced. To capture various states of translocation, we slowed the reaction by lowering the temperature and by adding polyamines. Our cryo-EM analysis of EF-G P610L-promoted translocation reveals that the EF-G variant and Apr bind simultaneously to the ribosome (Fig. 3a-c, and Supplementary Fig. 6a-f and Supplementary Table 1). Superimposing the Apr-bound ribosome-EF-G P610L complex with the previously published Apr-bound ribosome EF-G wt complex shows no significant structural changes in the decoding center or drug-binding site (Fig. 3c). These findings further support the notion that EF-G variants do not displace the drug from the ribosome (Supplementary Fig. 2, model 3).

With wt EF-G, cryo-EM structures and particle populations reveal that Apr does not interfere with the early steps of translocation, including tRNA hybrid state stabilization upon EF-G binding, GTP hydrolysis, inorganic phosphate (Pi) release, up to the movement of tRNAs into chimeric states on the small ribosomal subunit (SSU) (Fig. 3d,e). However, Apr significantly slows down the final transition of the tRNAs on the SSU to the post-translocation state. With EF-G P610L, we could not detect any particles in the chimeric state, despite extensive sorting of the cryo-EM data (see "Methods"). Furthermore, we could also not find any post-Pi-release state that would correspond to a tRNA hybrid state with EF-G-GDP bound, as we found previously for wt EF-G. Accordingly, with EF-G P610L the early stages of translocation up to GTP hydrolysis are unaffected by Apr but the later steps appear to be inhibited, particularly Pi release and the tRNA movement into chimeric states. The fact that EF-G P610L and Apr can bind simultaneously, and that the mutation impairs late translocation steps, is inconsistent with a 'gain-of-function' model (Supplementary Fig. 2, model 4). Together with the kinetic analysis (Fig. 2), these results demonstrate that EF-G variants are impaired in facilitating translation on AGA-bound ribosomes. Furthermore, this also suggests that EF-G resistance variants can promote efficient translation only when intracellular AGA concentrations are low and most ribosomes are drug-free, whereas at higher AGA concentrations, translation on all ribosomes would cease, abolishing the resistance phenotype.

Next, we asked whether selective silencing of AGA-bound ribosomes by EF-G variants affects proteome integrity. To measure translation errors, we treated E. coli cultures (wt, F593L, A608E, and P610L) with Apr (Fig. 4a) and quantified amino acid misincorporation using MS. After Apr addition, the cultures continued to grow at similar rates initially, likely due to the lag phase in self-promoted AGA uptake (Fig. 4a). After 60 min, wt cells stopped growing, while mutant cells remained unaffected by the Apr exposure, consistent with their resistance phenotypes (Fig. 1c). In wt cells, the Apr treatment led to a burst of misreading within 60 min (Fig. 4b and Supplementary Fig. 7a-c). After that, error frequencies did not increase further, likely because proteostasis collapsed and cell growth ceased. In contrast, error levels in mutant strains increased gradually and remained lower than in wt cells, suggesting that fusA mutations establish resistance early in the initial phase of AGA uptake. As the accumulation of aberrant membrane proteins is crucial for AGA uptake, we analyzed error frequencies specifically in membrane proteins (see "Methods"). Similar to their effect on cytosolic proteins, fusA mutations helped to maintain the integrity of the membrane proteome, including diverse proteins involved in protein export, quality control, respiration, metabolic transport, and cell structure maintenance (Fig. 4c and Supplementary Fig. 7d).

One type of errors that crucially depends on the translation velocity is error clusters: Faster translation in wt strain should allow AGA-bound ribosomes to decode multiple codons before the drug dissociates, thereby increasing the probability of long error clusters. In contrast, selective silencing of AGA-bound ribosomes in fusA mutant strains increases the chance that AGA will dissociate from the ribosome before several elongation cycles are completed, thereby suppressing error cluster formation. To assess whether fusA mutations decelerate translation strongly enough for AGA to dissociate before multiple elongation cycles are completed, we measured error cluster formation in wt and mutant strains in vivo using targeted MS (see "Methods"). In wt cells, Apr treatment led to a rapid increase in single translation errors and error clusters and to growth arrest (Fig. 5a, b and Supplementary Fig. 8a). In contrast, in the P610L strain, the level of single errors increased gradually, reaching levels comparable to the wt only at very high Apr concentrations (128-256 µM), reflecting the attenuated AGA uptake at lower concentrations. Across all concentrations and incubation times, error clusters were dramatically reduced, with some falling below the mass spectrometer's detection limit.

Error cluster formation depends linearly on the occurrence of the initial error (Fig. 5c), with the slope reflecting the probability that an AGA-bound ribosome, after making the first error, will proceed to make a subsequent error. For example, for the wt strain, the probability of misreading the next codon D48E after the initial F47L misincorporation (E) is ~0.2 (e.g., 20%), which is a very high error rate (Fig. 5c, upper panel). Shallower slopes indicate that P610L systematically prevents error cluster formation, e.g., for the same error cluster, E is ~0.1 -- about 2-fold less than in the wt strain. However, the reduction is not uniform for all error clusters (compare examples of F47L-D51E and H67Q-D71E in Fig. 5c). Notably, the codon distance between the consecutive misreading events matters: for the clusters where the misreading events are separated by 3 codons, the probability to make a second error is decreased about 30-fold, from E ~ 0.08 to ~0.003. A systematic analysis for P610L showed that clusters with a longer codon distance decrease dramatically compared to the wt (Fig. 5d and Supplementary Fig. 8a, b), while misreading events involving single amino acid substitutions are less affected (Supplementary Fig. 8b). Similar length-dependent reductions in error cluster formation were also observed for F593L and A608E mutant strains (Fig. 5e), indicating that translocation inhibition of AGA-bound ribosomes and the corresponding reduction in error clusters are hallmarks of fusA mediated resistance.

To evaluate whether silencing of AGA-bound ribosomes is associated with resistance, we correlated the effect of different AGAs on P610L-mediated error cluster suppression with their impact on resistance (Fig. 5f). For Apr, KanA, KanB, Sis, and Rib, error clusters were strongly reduced in P610L compared to wt, correlating with a significant increase in resistance. In contrast, Neo and Gen showed a smaller reduction in error cluster formation and only a slight increase in resistance, while Str and dihydrostreptomycin (Dhs) showed almost no reduction and little resistance. Overall, the strong correlation between error cluster reduction and resistance, together with our biochemical, kinetic, and structural data, support the conclusion that EF-G resistance variants selectively silence corrupted ribosomes, thereby reducing the proteotoxic stress (Supplementary Fig. 2, model 5).

Because the proposed resistance mechanism is most effective at low intracellular AGA concentrations, we investigated how proteostasis and membrane integrity are affected in fusA mutant strains at extracellular AGA concentrations that are lethal for wt bacteria. Light microscopy revealed significant protein aggregation at the poles of Apr-treated wt cells (Fig. 6a), which is consistent with previous microscopic studies of AGA-treated cells. These aggregates form due to AGA-induced misreading, which leads to protein misfolding and the activation of the unfolded protein response. The unfolded protein response then triggers the expression of the small chaperones IbpA and IbpB, which guide unfolded proteins into aggregates. In contrast, fusA mutant cells appear unaffected by Apr treatment (Fig. 6a and Supplementary Fig. 9). Supporting our microscopic analysis and error burden evaluation (Figs. 4b, c and 5a), the quantification of these chaperones, along with their transcription factor RpoH, revealed that the unfolded protein response is strongly induced in Apr-treated wt cells but remains unaffected in resistance mutant strains (Fig. 6b), supporting the notion that fusA mutations confer resistance by maintaining proteostasis.

We then investigated whether fusA mutants affect AGA uptake. First, we assessed the membrane integrity of Apr-treated wt and P610L cells by staining them with propidium iodide, a membrane-impermeable dye. While wt cells took up the dye, indicating membrane damage, P610L cells remained unstained, suggesting that fusA mutations help preserve membrane integrity after AGA treatment (Fig. 6c). Second, to investigate the specific effect on AGA uptake, we treated wt and mutant strains with Apr and then used gentamicin-Texas Red (GTTR) as a reporter to visualize AGA uptake (Fig. 6d, e). Wt cells showed a strong increase in GTTR fluorescence after AGA treatment, indicating substantial antibiotic accumulation. In contrast, fusA mutant cells showed little or no increase in GTTR fluorescence during AGA treatment, suggesting low AGA uptake (Fig. 6e). These findings suggest that fusA mutations, by silencing AGA-bound ribosomes, contribute to resistance by limiting AGA uptake.

To test whether reduced antibiotic uptake explains the resistance phenotype of fusA mutants, we measured MIC values in the absence or presence of AgNO, a compound that permeabilizes bacterial membranes and accelerates AGA uptake (Fig. 6f). While the fusA mutant strains appeared to maintain a low-level residual AGA resistance after AgNO treatment, the differences to the wt strain were not statistically significant (one-way ANOVA with Šidák correction). The large (25-50-fold) effect of AgNO strongly supports the notion that while selective silencing of corrupted ribosomes is the primary microscopic effect of fusA mutations, the preservation of the membrane integrity is the dominant macroscopic outcome that ultimately enables the cell to survive at otherwise lethal AGA concentrations.