Introduction

Antibiotic treatment failure is a concerning and multifaceted problem. Major focus has been on antibiotic resistance, which occurs when bacteria acquire a genetic mutation or a genetic element that allows bacteria to grow in the presence of antibiotics^{1, 2}. However, bacteria can also develop phenotypic changes in response to stress that allows them to survive antibiotic treatment^{1, 2}. This phenomenon is known as antibiotic tolerance when the entire bacterial population responds this way, or antibiotic persistence when only a subpopulation is recalcitrant to antibiotics³. These phenotypic changes are transient and typically induce a slowed growth response that renders antibiotics less effective². Once the stress is removed, bacteria resume growth, which is a major concern for relapsing infection⁴. Growing evidence also suggests that antibiotic tolerance and persistence contribute to the development of antibiotic resistance^5–8. We currently lack strategies to target antibiotic tolerant bacteria, making it important to understand what pathways promote bacterial survival, so we can leverage this information to improve antibiotic efficacy.

Yersinia pseudotuberculosis (Yptb) is a Gram-negative bacterium in the class of Gammaproteobacteria that is a natural pathogen of humans and rodents. Yptb has been used as a model to study multiple aspects of pathogenesis of enteric pathogens^9–13 and recently has been studied in a mouse model of antibiotic persistence¹⁴. In a previous study we investigated the phenotypic response of Yptb to doxycycline treatment and found expression of the gene tusB was significantly downregulated in response to doxycycline¹⁵. TusB (2-thiouridine subunit B) is involved in a sulfur relay system that transfers a sulfur group onto the wobble position of tRNAs encoding for glutamate (Glu), glutamine (Gln), and lysine (Lys)¹⁶. TusB complexes with itself and other Tus proteins (TusC and TusD homodimers) forming a heterohexamer that mediates the transfer of sulfur from TusA to TusE^{16, 17}. TusE then transfers sulfur to MnmA, which catalyzes the final 2-thiolation reaction, adding the sulfur group onto the second carbon of the uridine (U34) in the tRNA anticodon¹⁶. This modification restricts binding of the tRNA to purines (A and G), rather than pyrimidines (C or U), thereby increasing translation fidelity and efficiency.

Translation efficiency is generally defined as the rate of protein production per mRNA molecule¹⁸. Initiation is primarily thought to be the rate limiting step of translation, but elongation can be a limiting step as well¹⁹. During elongation, the speed at which proteins are translated is impacted by the available pool of tRNAs and competition between cognate and non-cognate tRNAs²⁰. tRNA modifications that stabilize the interaction between the anticodon and its cognate codon increase translation efficiency by promoting more rapid decoding. tRNA modifications can thus impact protein translation depending on which codons are decoded by the modified tRNA, and how enriched those codons are in a given transcript^21–23. In yeast, absence of s²U has been shown to impact the metabolic state of cells²⁴ and led to proteotoxic stress due to improper protein folding during translation²⁵, but has been postulated to not have a major impact on translation efficiency, as measured by correlating transcriptional changes with ribosome footprint changes²⁴. In contrast, loss of s²U seems to impact translation in Plasmodium falciparum, by biasing the translation of certain proteins depending on their enrichment of lysine codons, impacting artemisinin resistance²⁶. In E. coli, it has been shown that s²U mutants have delayed expression of specific proteins, RpoS, Fis, and FtsZ, and impaired cell division^{27, 28}. However, it is not known whether s²U has a global impact on translation or if this tRNA modification impacts antibiotic susceptibility in bacteria.

Here, we find that deletion of tusB and loss of the s²U modification renders Yersinia tolerant to multiple ribosome-targeting antibiotics, and the RNA polymerase-targeting antibiotic, rifampicin. We hypothesized that loss of the modification slows translation at Glu, Gln, and Lys codons. In line with this, we find that ribosomes pause more frequently at the specific codons that rely on the tRNA modification for proper decoding. We find that ribosomal proteins themselves are highly enriched in these codons and downregulated at the protein level, but not the transcript level, indicating they may be translated less efficiently in the absence of s²U. Our results support the model that a reduction in ribosomal proteins is globally reducing translation and is responsible for slowing bacterial growth and inducing antibiotic tolerance.

Results

Deletion of tusB results in a complete loss of mnm⁵s²U and induces changes in other modifications

To understand what impact tusB expression has on antibiotic susceptibility, we created a tusB deletion strain (ΔtusB). First, to confirm that deletion of tusB results in loss of the s²U modification, we performed chromatography-coupled tandem quadrupole mass spectrometric (LC-MS) analysis to detect relative tRNA modification levels in ΔtusB compared to our wild-type (WT) strain (Figure 1). We also compared relative tRNA modification levels in WT +/- doxycycline (dox) treatment to see if downregulation of tusB would be sufficient to lower tRNA modification levels. We chose to look at tRNA modification levels in the early log phase of growth to match our RNA-seq timepoint where we saw downregulation of tusB under WT+dox conditions¹⁵. Additionally, early to mid-log phase is when the s²U modification should be highest in WT cells, prior to nutrient (particularly sulfur) limitation²⁹.

Figure 1.

Hierarchal clustering analysis shows that the tusB deletion strain clusters together regardless of doxycycline treatment and WT + dox samples also cluster together (Figure 1). Importantly, as expected, we found a complete loss of s²U in ΔtusB (Figure 1, blue box). We saw minimal downregulation of mnm⁵s²U (average 0.9-fold change WT dox vs. WT untreated) and cmnm⁵s²U (average 0.87-fold change WT dox vs. WT untreated) in response to doxycycline.

Interestingly we detected other changes in relative tRNA modification levels in ΔtusB compared to WT, and in response to doxycycline treatment. The sulfur modifications, s²C and ms²i6A, were downregulated in response to doxycycline (both in WT and ΔtusB) and somewhat in response to tusB deletion alone (Figure 1), which could indicate low sulfur availability²⁹. Additionally, we also detected upregulation of many modifications commonly found on ribosomal RNA (m⁵C, m¹A, m³U, m²G, and m⁶⁶A). Our RNA isolation enriches for tRNA based on size and can include smaller ribosomal RNA fragments if these are abundant.

Loss of 2-thiolation tRNA modification induces tolerance to specific ribosome and RNA-polymerase targeting antibiotics

We tested the growth and antibiotic susceptibility of ΔtusB and found that it was significantly slower growing and more tolerant to multiple antibiotics including the ribosome-targeting antibiotics, doxycycline and gentamicin, and the RNA polymerase-targeting antibiotic, rifampicin (Figure 2). We detected a slight decrease in antibiotic susceptibility to chloramphenicol, although this was not statistically significant. These phenotypes were rescued by integrating the tusB gene back into the genome (tusB rescue, Figure 2). Additionally, to confirm that phenotypes were specific to loss of the tRNA s²U modification, we complemented the deletion strain with the Bacillus subtilis pathway for 2-thiolation. B. subtilis possess an abbreviated version of the sulfur relay system where the cysteine desulfurase, YrvO, transfers sulfur groups directly to MnmA, which catalyzes the final 2-thiolation reaction³⁰. As expected, expression of yrvO-mnmA restored growth and antibiotic susceptibility of ΔtusB, although this was not statistically significant for doxycycline (Figure 2). Chloramphenicol susceptibility was similar to ΔtusB when yrvO-mnmA was expressed, indicating that loss of the modification does not greatly impact chloramphenicol susceptibility. Importantly, phenotypes were only complemented when a functional copy of yrvO was expressed and not when the active cysteine residue (C325A) of YrvO was mutated or with an empty vector control (Extended Data Figure 2)³⁰.

Figure 2.

Overall, these results indicate that loss of the s²U tRNA modification induces tolerance to the ribosome and RNA-polymerase targeting antibiotics that were tested. Interestingly, we observed no difference in susceptibility of ΔtusB to the DNA-gyrase targeting antibiotic, ciprofloxacin, (Figure 2) indicating that this tolerance phenotype may be specific to antibiotics that target translation and transcription. Since transcription and translation are generally coupled in bacteria, we hypothesized that this might indicate translation is impacted by loss of s²U.

ΔtusB has a starvation-like metabolic signature and decreased levels of ribosomal proteins

To further understand the impact of loss of s²U, we performed RNA-seq and mass spectrometry to define the transcriptome and proteome, respectively, of ΔtusB compared to WT. 259 genes were significantly downregulated, and 182 genes were significantly upregulated at the transcript level (p_adj < 0.05 and >2-fold change). At the protein level, 166 genes were significantly downregulated, and 197 genes were significantly upregulated (p-value < 0.05 and >2-fold change in protein abundance).

KEGG Pathway Enrichment analysis of differentially regulated genes from our RNA-seq dataset shows that ΔtusB upregulates multiple amino acid biosynthesis pathways (Figure 3A). Many of these are also upregulated at the protein level in our mass spectrometry dataset (Figure 3B). We hypothesized this could indicate the stringent response is being induced by loss of the tRNA modification. In vitro studies have shown the s²U modification improves tRNA synthetase binding³¹ making it plausible that loss of the modification could result in uncharged tRNAs binding to the ribosome, inducing the stringent response. However, deleting the alarmone synthetase, relA, in the tusB deletion background did not rescue the slowed growth or antibiotic tolerance phenotype (Extended Data Figure 3), indicating the ΔtusB phenotype is not linked to the stringent response.

Figure 3.

RNA-seq and proteomic analyses generally revealed that ΔtusB has a starvation-like metabolic signature. ΔtusB significantly downregulated oxidative phosphorylation (Figure 3A, B) as indicated by downregulation of several genes in the TCA cycle and multiple subunits of the electron transport chain. This type of metabolic shift is often associated with antibiotic tolerance and persistence and is consistent with the slowed growth phenotype of ΔtusB.

Surprisingly, our proteomic analysis revealed that 37 of the 54 ribosomal protein subunits were significantly downregulated in ΔtusB (Figure 3B). Most of these ribosomal subunits did not show up as hits in our RNA-seq dataset and the few that did (p<0.05) were not strongly downregulated or upregulated at the transcript level (less than 2-fold change) (Extended Data Table 1). This also indicates that the stringent response is likely not be responsible for the expression changes in ΔtusB, since the ppGpp alarmone has been shown to repress transcription of ribosomal protein genes³². We hypothesized that this difference between our RNA-seq and proteomics datasets could indicate that the ability to translate these proteins is impaired in ΔtusB.

Ribosomes pause more frequently at Glu, Gln, and Lys codons in ΔtusB

To understand what impact tusB deletion has on translation, we performed ribosome profiling to determine if loss of the s²U tRNA modification impacts ribosome speed. We adopted a ribosome profiling approach developed by Mohammad et al.^{33, 34}, allowing us to identify ribosome abundance and positioning at single codon resolution. Ribosome pause scores are calculated by taking the average ribosome density at a single codon normalized to the average ribosome density across a given gene and averaging this for all genes in the genome. As expected, the relative ribosome codon occupancy (ΔtusB pause score normalized to WT) was moderately increased at Glu, (GAG and GAA), Gln, (CAA), and Lys (AAA and AAG), indicating that ribosomes paused more frequently at these codons in ΔtusB (Figure 4). No change in ribosome pausing at CAG was expected since Yptb have one Gln tRNA that is not modified with s²U (tRNA^CUG) that is likely sufficient for decoding CAG³⁵. In contrast, Yptb seem to more heavily rely on modified tRNAs for decoding Lys (4 tRNA^UUU and only one tRNA^CUU)³⁵ and Glu (5 tRNA^UUC and no tRNA^CUC)³⁵, which explains why we see ribosome pausing at both isoacceptor codons for Lys and Glu.

Figure 4.

In addition to indicating slower translation, ribosome pausing can also be a sign that proteins are being mistranslated and/or not properly folded²⁵. Since s²U is important for stabilizing the interaction between the tRNA anticodon and its cognate codon, we briefly investigated if the absence of s²U impacts translation fidelity, with the idea that loss of s²U would increase the chances that a near cognate tRNA could bind and incorporate the wrong amino acid. We detected potential misincorporation of amino acids in a preliminary LC-MS/MS run, considering the following changes as modifications in peptides: glutamate to aspartate, glutamine to histidine, and lysine to asparagine. To understand if these misincorporations occur more frequently due to loss of s²U, we measured the relative abundance of modified to unmodified peptides in ΔtusB vs. WT (Extended Data Figure 4A, Supplemental Data 8). Interestingly we detected on average higher frequency of misincorporation at Glutamine CAA versus Glutamine CAG codons in ΔtusB compared to WT (Extended Data Figure 4B), consistent with our finding that ribosomes pause at CAA and not CAG (Figure 4). However, with a small sample size we cannot conclusively say whether misincorporation of amino acids is more frequent due to loss of s²U. Additionally, it does not seem that proteins are being widely mistranslated in ΔtusB, given that we saw no indication of proteotoxic stress in our differential expression analysis (Figure 3A, B).

Overall, our results indicate that absence of s²U impacts ribosome speed. The ribosome pausing we detected at Glu, Gln, and Lys codons (Figure 4) is comparable to what has been reported in yeast^{25, 36}, although we detected relatively higher pause scores at Gln CAA and Lys AAG codons. These values are considered a moderate increase in ribosome occupancy, indicating that the ribosome is transiently pausing at these sites rather than stalling. Although transient pausing at one site may not have a great impact on protein output, we hypothesized that if transcripts were highly enriched in these codons, this would significantly slow down the ribosome and decrease translation of a given transcript.

Switching Gln codons from CAA to CAG slightly improves translation of GFP

To test whether enrichment of codons with ribosome pausing is sufficient to slow translation and impact protein output, we created a codon-optimized GFP reporter construct where all CAA codons (6/8 total Gln codons) are switched to CAG (Figure 5A). We focused on Gln codons since we saw a clear increase in ribosome pausing at CAA, but not CAG (Figure 4). As expected, ΔtusB grew slower than WT (Figure 5B). However, ΔtusB growth was slightly improved when expressing the CAG construct (Figure 5B), consistent with others have reports that bacterial growth is impaired by expressing gfp with codons that slow the ribosome²⁰. Oddly, we saw lower GFP fluorescence from the CAG construct in the WT strain compared to native GFP, which was unexpected since CAA and CAG are used equally in protein coding sequences in Yptb and thus switching these codons was not predicted to bias translation in the WT strain (Figure 5C). However, GFP fluorescence from the CAG construct was not lower than native GFP in ΔtusB, instead it was slightly higher (Figure 5C). As a proxy for GFP translation, we quantified GFP fluorescence relative to gfp transcript levels at 4h (Figure 5D). Importantly we can see GFP fluorescence per relative gfp transcript level is trending lower in ΔtusB vs. WT (p-value = 0.0552) for native GFP, suggesting GFP translation is lower in ΔtusB. GFP fluorescence per relative gfp transcript level was slightly improved in ΔtusB with the CAG construct, suggesting that replacing CAA codons with CAG slightly improves GFP translation, although this was not statistically significant.

Figure 5.

Overall, these results indicate that translation is lower when s²U is absent and is partially improved by replacing CAA codons with CAG. We expect that GFP translation was not significantly improved by switching only Glutamine codons since the other codons, Glutamate and Lysine, likely also contribute to the lower translation of GFP in ΔtusB

Since GFP protein levels can also be impacted by protein decay and degradation, we measured GFP decay using a ssrA tag to target the GFP protein to the ClpXP proteosome^{37, 38}. Unstable GFP (gfp-ssrA) had a half-life of 57 minutes in WT. In contrast, unstable GFP (gfp-ssrA) had a half-life of 187 minutes in ΔtusB, indicating that protein degradation is impaired (Extended Data Figure 5). Additionally, stable GFP also had a longer half-life in ΔtusB (277 minutes) compared to WT (165 minutes) (Extended Data Figure 5). These findings may be explained by our RNA-seq and proteomic analysis which indicates that ΔtusB downregulates the ClpXP proteosome (clpX is downregulated 1.79-fold by RNA-seq (padj =1x10^8) and 3-fold in protein abundance (p = 0.0292); clpP is downregulated 1.69-fold by RNA-seq (padj = 1.49x10^8) and 3.77-fold in protein abundance (p=0.00508), see Supplemental Data 9). It is unclear why ΔtusB downregulates the ClpXP proteasome, but this suggests that ΔtusB is impaired in protein degradation, allowing us to conclude that the lower GFP fluorescence of ΔtusB (Figure 5C, D) is not due to increased degradation of GFP proteins.

Lys AAG codon enrichment is highly associated with subset of downregulated ribosomal proteins

To look globally at the impact of loss of the s²U tRNA modification on biasing translation, we analyzed codon usage for all protein-coding genes in the Yersinia pseudotuberculosis IP2666 genome, with the idea that enrichment of Glu, Gln, and Lys codons would be highly associated with downregulated proteins if these codons were biasing translation. Codon Z-scores, detailing relative codon over- or under-use in a gene (relative to genome averages), were calculated for each codon in each gene by subtracting the average global codon frequency from the gene codon frequency and normalizing to the standard deviation of that codon’s usage throughout the genome³⁹. As a result, codon Z-scores represent how many standard deviations above or below the global frequency a specific codon is used in a given gene. We performed principal component (PC) analysis using the codon Z-scores for all genes that were differentially regulated at the protein level (2-fold change in protein abundance, ΔtusB vs. WT, p-value < 0.05). PC scores are calculated for each gene based on their codon usage and plotted in Figure 6A, revealing a cluster of downregulated proteins along PC1. We performed principal component regression analysis using the log2fold change in protein abundance (ΔtusB vs. WT) as the dependent variable and found several codons that significantly contribute to the variance in protein expression. Codons of interest, Glu GAA (p-value=0.0345), Lys AAG (p-value <0.0001) and Lys AAA (although not significant, p-value = 0.1757) corresponded with the cluster of downregulated proteins along PC1 in the lower right quadrant of the loadings plot (Figure 6B, C). The codon Z-scores are particularly high for Lys AAG amongst the downregulated proteins in this subset. Interestingly 18 of the 21 downregulated proteins are ribosomal proteins, indicating that enrichment of Lys AAG codons may explain why these proteins are less abundant even though none were significantly downregulated at the transcript level (Figure 3B). Although we were also expecting Gln CAA codons to bias translation of proteins, principal component regression analysis indicated they did not significantly contribute to the variance.

Figure 6.

A combination of Glu, Gln and Lys enrichment likely contributes to decreased ribosomal protein levels

Finally, we also looked at the combined impact of all five codons that increased ribosome pausing, with the idea that enrichment of a combination of Glu, Gln (CAA) and Lys codons would have more impact than enrichment of a single codon, and that codons with higher ribosome pause scores would have a greater impact on the translation of a given protein. We took the sum of the codon Z-scores for Glu (GAA), Glu (GAG), Gln (CAA), Lys (AAA), and Lys (AAG) for each gene and weighted (multiplied) each codon Z-score by the average ribosome pause score for that respective codon. We refer to this calculation as the “cumulative pause predictor score”, with the idea that a higher score will predict a gene will be translated less efficiently due to increased ribosome pausing along the transcript. Consistent with this idea, we saw significantly increased cumulative pause predictor scores among proteins that were downregulated in ΔtusB compared to upregulated proteins (Figure 6D). Ribosomal proteins in particular had significantly higher cumulative pause predictor scores than all other downregulated proteins (Figure 6E), suggesting that not only enrichment of Lys AAG, but likely enrichment of a combination of all five of these codons is slowing down the translation of ribosomal proteins when s²U is absent. Importantly, we do not see this same trend in five other codons selected as a control that did not have changes in ribosome pausing (Extended Data Figure 6). Additionally, for the 54 ribosomal proteins, we see that higher cumulative pause predictor scores correlate with decreased protein abundance (Figure 6F).

Overall, these results suggest that translation of ribosomal proteins is significantly reduced when s²U is lost due to an enrichment of codons that slow down the ribosome.

Discussion

This work provides evidence for a novel mechanism by which loss of s²U induces antibiotic tolerance. It also adds to the growing evidence that modulation of tRNA modifications can widely impact expression of genes through codon biases in translation^{21, 23, 40}. In other reports, s²U has been assumed to not have a drastic impact on protein translation output when only a moderate increase in ribosome pausing is seen^18,19,25. However, here we see ribosome pausing highly correlates with genes that are significantly downregulated at the protein level, but not at the transcript level.

Although pausing at one site alone is likely not sufficient to significantly slow down translation, our results suggest that enrichment of ribosome pausing sites can lead to reduced protein levels. Ribosomal proteins seem to be the most impacted by this codon bias in translation as indicated by their high enrichment in Glu, Gln, and Lys codons and the significant reduction in ribosomal protein levels. We expect that this reduction in ribosomal proteins is globally reducing the translational capacity of the cell and is responsible for inducing tolerance to ribosome and RNA polymerase-targeting antibiotics.

The s²U tRNA modification is well conserved among bacteria¹⁶ and ribosomal protein sequences are also generally conserved across bacterial species, including prominent ESKAPE pathogens that are highly resistant to antibiotics (Extended Data Figure 7). Therefore, it is plausible that this could be an antibiotic tolerance mechanism used broadly by bacteria to better survive antibiotics, although this would depend on the codon usage and tRNA availability for each species.

Future work will be needed to understand what conditions in the host environment are sufficient to reduce s²U levels. In our study doxycycline treatment only moderately lowered s²U tRNA modification levels (Figure 1). However, given the broad downregulation of the 2-thiolation pathway in response to gentamicin (Extended Data Figure 1), we expect that gentamicin may have a stronger impact on modulating s²U levels and promoting survival.

Given that s²U levels are modulated in response to sulfur availability²⁹, it is likely that conditions within the host environment, other than antibiotic stress, also result in lowered s²U levels within bacterial pathogens. This highlights that the mechanism we have defined here may be one used broadly by bacteria to adapt to conditions where it is beneficial to slow growth, such as low nutrient conditions. Fast growing cells need to have a high number of ribosomes since protein synthesis is limited by the availability of free ribosomes ⁴³, and ribosome abundance is linearly correlated with growth rate⁴⁴. However, during slow growth, it is beneficial for bacteria to slow ribosome protein synthesis since ribosome biogenesis and translation are energetically demanding processes. Therefore, it is interesting to speculate that the impact of s²U on ribosomal protein levels we see here could be a previously unknown mechanism by which bacteria reduce translation of ribosomal proteins in response to conditions where it is beneficial to slow growth and save energy.

Surprisingly we detected some changes in other tRNA modifications (Figure 1), some of which are modifications also found on ribosomal RNA (rRNA). Particularly under starvation conditions, ribosomal RNA is more highly degraded, which could explain why we detected increased levels of modifications commonly found on rRNA in ΔtusB ^41,42. However, it is unlikely that rRNA degradation is responsible for lower ribosomal protein levels in ΔtusB. 16S rRNA levels were not different by qPCR in WT versus ΔtusB (see Supplemental Data 7). Additionally, given the reduced capability of ΔtusB to degrade GFP (Extended Data Figure 5), along with downregulation of the ClpXP protease, it is unlikely that ribosomal proteins are being readily degraded in ΔtusB. Instead, the level at which ribosomal proteins are downregulated correlates with the enrichment of codons that rely on s²U for proper decoding, indicating that these codons bias their translation in the absence of s²U.

It is important to note that we also saw metabolic changes in ΔtusB that could be contributing to the slowed growth and antibiotic tolerance phenotypes (Figure 3). However, if reduced metabolism were the main contributing factor, we would expect to broadly see antibiotic tolerance^45–47. Instead, loss of s²U seems to induce tolerance to only antibiotics that target transcription and translation (Figure 2), consistent with our model that loss of s²U is impacting the translational capacity of the cell. It is unclear what drives these metabolic changes in ΔtusB, but interestingly, others have noted similar metabolic changes in response to loss of s²U in Saccharomyces cerevisiae and have postulated that 2-thiolation acts as a signal to the cell that nutrients are low ^25,29. Loss of s²U thus results in a compensatory increase in synthesis of sulfur-containing amino acids²⁵. Additionally, carbon metabolism is altered and shifts away from oxidative phosphorylation and towards carbohydrate storage²⁵, similar to what we see in our transcriptional and proteomic analysis of ΔtusB (Figure 3).

In summary, here we identified a novel mechanism by which changes in s²U tRNA modification levels reduce translation of ribosomal proteins and induce slowed growth and antibiotic tolerance. Future work will be needed to identify conditions sufficient to lower tRNA modification levels, to further bolster the evidence that modulating s²U levels is a phenotypic response to stress that promotes bacterial survival.

Methods and materials

Bacterial strains

All experiments were done with the Yersinia pseudotuberculosis IP2666 strain¹⁵. Deletion strains are generated by amplifying 1kb each of the 5’ and 3’ regions flanking the gene of interest (homology arms). The 5’ homology arm includes the start codon and 6 base pairs of the beginning of the gene and the 3’ homology arm includes 6 base pairs of the end of the gene and the stop codon. These homology arms are fused together and cloned into the pSR47s suicide vector⁴⁸. The construct is subsequently conjugated into WT Yersinia pseudotuberculosis (for ΔtusB ΔrelA, the relA pSR47s deletion construct was conjugated into the ΔtusB Yersinia pseudotuberculosis strain). Sucrose selects for bacteria that have complete integration of the insert (two recombination events)⁴⁸. Deletion of the gene was confirmed by PCR. The tusB rescue strain was generated by amplifying the gene and 1kb upstream and downstream of the gene from WT chromosomal DNA as a single PCR product. This was inserted into pSR47s and conjugated into the ΔtusB Yersinia pseudotuberculosis strain. Rescued clones were confirmed by PCR. yrvO-mnmA complemented and control strains were generated by transforming the plasmid constructs (pDS145 yrvO-mnmA, pDS220 yrvO(C325A)-mnmA, and pARA13)³⁰ into ΔtusB Yersinia by electroporation⁴⁸.

Antibiotic susceptibility assays

To assay viability of exponential phase bacteria, overnight cultures (16h, at 26°C with rotation) were sub-cultured into 2mL LB and grown for 2h at 26°C before diluting again in 2mL LB to a starting bacterial concentration of about 10^4 CFUs/mL. Cultures were treated with antibiotics at the MIC for bacteriostatic antibiotics (1µg/mL doxycycline, 25µg/mL chloramphenicol) or MBC₉₀ concentration for bactericidal antibiotics (2µg/mL gentamicin, 2µg/mL rifampicin, or 0.1µg/mL ciprofloxacin) or left untreated. Cultures were grown for 4h at 37°C. Samples were plated for CFUs every 2h. For yrvO-mnmA experiments overnight cultures and subcultures were treated with 0.2% arabinose throughout the entire experiment to induce expression of the construct or as a control.

Small RNA extraction and processing to purify tRNA

Overnight cultures (16h, at 26°C with rotation) were diluted into 50mL LB and grown at 37°C for 2h and then approximately 10^9 CFUs were pelleted. Bacterial pellets were stored at -80°C before further processing. 1mL of Trizol was added to each cell pellet and samples were rigorously vortexed for 30 seconds, followed by room temperature incubation for 5 minutes. 200µL of chloroform was added to each sample followed by shaking by hand to mix and room temperature incubation for 3 minutes. After centrifugation at 12000xg for 15mins at 4°C, the aqueous phase was isolated and brought to a concentration of 30% ethanol to precipitate large RNAs, then applied to a spin column from the Purelink miRNA isolation kit (Thermo). After centrifugation at 12000xg for 1 minute, the flowthrough containing small RNAs was collected, brought to an ethanol concentration of 70%, and applied to a second spin column. Samples were washed twice with the kit’s wash buffer and spun dry, followed by elution in nuclease-free water. A nanodrop was used to determine RNA concentration and purity and quality was assessed by Agilent 2100 Bioanalyzer Nanochip. 1µg RNA was digested at 37°C for 6 hours in a 50µL reaction containing 5mM Tris-HCl pH 8.0, 2.5mM MgCl2, 0.1 µg/mL coformycin (adenosine deaminase inhibitor), 0.1mM deferoxamine (antioxidant), 0.1 mM BHT (antioxidant), 0.083U/µL benzonase, 0.1U/µL calf intestinal alkaline phosphatase, 0.003 U/µL phosphodiesterase I, and 50nM of 15N-dA internal standard.

LC-MS/MS of ribonucleoside modifications

Three technical replicate injections of 200ng for each sample were analyzed by chromatography-coupled triple-quadrupole mass-spectrometry (LC-MS/MS). A Waters Acuity BEH C18 column (50 x 2.1 mm inner diameter, 1.7 micron particle size) was used with an Agilent 1290 HPLC system at 25C, 0.3mL/min flow rate, in a gradient of Buffer A (0.02% formic acid in water) and Buffer B (0.02% formic acid in 70% acetonitrile) (see Supplemental Table 1). The HPLC was coupled to an Agilent 6495 Triple Quad mass spectrometer with an electrospray ionization source in positive mode with 200C dry gas temperature, 11L/min gas flow, 20psi nebulizer pressure, 300C sheath gas temperature, 12L/min sheath gas temperature, 3000V capillary voltage, and 0V nozzle voltage. Product ions were detected using multiple reaction monitoring (MRM) mode, with retention times and optimal collision energies determined by synthetic standards (see Supplemental Table 2). The sum of 260nm UV signals of the 4 canonical nucleotides, as measured by an in-line UV detector, was used as a normalization factor.

qPCR to measure transcriptional regulation of s²U modification pathway in response to antibiotics

Cultures of bacteria were grown at 37°C for 2h and treated with the MIC of antibiotics (1µg/mL doxycycline, 25µg/mL chloramphenicol, or 15µg/mL gentamicin) or left untreated. Approximately 10^7 CFUs were pelleted and then resuspended in Buffer RLT (QIAGEN) + β-mercaptoethanol. RNA was isolated using the RNeasy kit (QIAGEN). DNA was removed using the on-column DNase digestion kit (QIAGEN). RNA was reverse transcribed using the Protoscript II First Strand cDNA synthesis kit (NEB). Approximately 30ng of cDNA was used as a template for qPCR using SYBR Green PCR Master Mix according to the manufacturer’s protocol (Applied Biosystems). Relative transcript levels were measured using the DCT and 2^-ΔCT method (using 16S transcript levels as an endogenous control).

RNA-seq

Overnight cultures (16h, at 26°C with rotation) were diluted into 4mL fresh LB and grown at 37°C for 2h. Approximately 10^7 CFUs were pelleted and then resuspended in Buffer RLT (QIAGEN) + β-mercaptoethanol. RNA was isolated using the RNeasy kit (QIAGEN). DNA was removed using the on-column DNase digestion kit (QIAGEN). Purified RNA was shipped to Novogene (Novogene Corporation Inc), for library preparation, sequencing, and bioinformatic analysis. Sequencing reads were aligned to IP2666 Yersinia pseudotuberculosis genome assembly GCF_003814345.1. DESeq2 was used to determine significant differences in transcript levels between ΔtusB and WT samples. Genes were considered significantly differentially regulated if there was greater than 2-fold change in transcript levels and based on an adjusted p-value of 0.05 (p_adj, adjusted p-value based on transcript size).

Protein extraction

Briefly, overnight cultures (16h at 26°C with rotation) were diluted into 50mL fresh LB and then grown for 2 hours at 37C. Approximately 10^9 CFUs were pelleted and stored in -80C. Bacterial pellets were subsequently washed in PBS then resuspended in 4X cell pellet volume of Bacterial Protein Extraction Buffer (BPER, Thermo-Fisher). After incubation for 15 minutes at room temperature, samples were centrifuged at 15,000xg for 5 mins at 4C. Supernatants (soluble fraction) and pellets (insoluble fraction) were saved at -80C. Pellets were subsequently processed to extract the insoluble fraction by washing with 1:10 diluted BPER and resuspending in 8x cell pellet volume of Inclusion Body Solubilization Reagent (Thermo-Fisher). Samples were vortexed for 1 minute and shaken on a platform shaker for 30 minutes at room temperature before being sonicated in an ice water bath for a total of 5 minutes (30 seconds on pulse, with 10 seconds off in between pulses). Samples were then centrifuged at 13,000xg for 15min at 4C and the supernatant was saved at -80°C.

LC-MS/MS to detect relative protein expression changes in ΔtusB vs. WT

40µg of protein for each sample was processed for mass spectrometry using SP3 cleanup and on-bead trypsin/LysC digest to remove contaminants and digest samples into peptides. Peptides were subsequently labelled with tandem mass tags (TMT). Peptides were separated by HPLC (Thermo-Fisher Vanquish NEO UHPLC) using a self-packed fused silica column over a 90 minute gradient. Eluted peptides were sprayed into an Exploris 480 mass spectrometer. Survey scans (full ms) were acquired on an Orbi-trap within 375-1,500 m/z using a data dependent mode. The full scan MS was followed by MS/MS for the top 15 peptides in each cycle with dynamic exclusion of 45s. Precursor ions were individually isolated with 0.7 Da and fragmented (MS/MS) using an HCD activation collision energy of 36. Precursor and fragment ions were analyzed at a resolution of 120,000 and 30,000, respectively. Reporter ions at resolution of 45,000 (Turbo TMT). Raw mass spectral data files were searched using Proteome Discoverer (version 3.1, Thermo Scientific) and the GCF_003814345.1 Yersinia pseudotuberculosis IP2666pBI genome assembly. Search parameters were 10 ppm mass tolerance for precursor ions. Fixed modification was carbamidomethylation of cysteine, TMTpro on Nterm, TMTpro on lysine residues. Variable modification was oxidation on methionine residues (maximum of 3). A false discovery rate (FDR) of 1% was used to validate high confidence PSMs. Normalized protein abundances were averaged across three biological replicates. Protein abundance ratios are calculated by normalizing the average protein abundance of ΔtusB to the average protein abundance of WT. Pairwise comparisons were done with One-way ANOVA using a p-value of 0.05 and 2-fold change in protein abundance as cut offs for significantly differentially regulated proteins.

Detection of misincorporation of amino acids in ΔtusB vs. WT

A preliminary LC/MS-MS run was done on one biological replicate using the soluble and insoluble fractions from WT and ΔtusB samples, similar to above (with TMT labeling). Raw mass spectral data files were searched in Proteome Discoverer v3.1 (ThermoFisher Scientific) using 3 mutations as variable modifications: Glutamate (E) changed to Aspartate (D), Glutamine (Q) changed to Histidine (H), and Lysine (K) changed to Asparagine (N). Peptides were prioritized based on relative abundance in each sample and selected in order to have a good representation of all the codons we expected there to be changes at (CAA, GAA, GAG, AAA, no mutations were identified at AAG), and codons we expected to not see increased changes at in WT compared to ΔtusB (CAG). A subsequent LC/MS-MS run (without TMT labeling) was done to discover additional peptides with potential misincorporations before targeting these peptides by data-independent acquisition (DIA) to determine the relative abundance of modified and unmodified peptides in each sample. 10µg of protein for each sample was processed using SP3 buffer exchange and on-bead trypsin/LysC digest. Peptides were desalted on OASIS HLB uElution plates (Waters), according to manufacturer’s instructions. Peptides were separated by a 100-minute linear gradient. MS1 scans were acquired in the Orbitrap detector of the Exploris 480 mass spectrometer from 300-1250 m/z with the following settings: RF lens setting of 50%, 120,000 resolution at 200 m/z, normalized AGC of 101%, maximum injection time set to Auto, and Easy-IC internal lockmass calibration turned on. Precursor ions in each window were fragmented by HCD at 30% NCE. DIA product ion (MS2) scans were acquired in the Orbitrap detector using the following settings: 140−1800 m/z, 30,000 resolution at 200 m/z, normalized AGC of 1000%, and a maximum injection time of 54 ms.

Raw data files were searched with Chimerys against a customized Yersinia pseudotuberculosis UniProt FASTA database (proteome accession UP000255087, 1263 entries), with the addition of sequences of the OmpA protein (UniProt accession P38399) containing E307D, Q309H, and Q273H substitutions. Search criteria were tryptic cleavage (maximum 2 missed), peptide length 7-30, peptide charge 2-6, 20 ppm fragment ion mass tolerance, Cys carbamidomethylation as a fixed modification, and Met oxidation as a variable modification (max 3/peptide). Peptide identifications were validated by Chimerys at 1% false discovery rate (FDR) based on an auto-concatenated decoy database search.

Ribosome profiling

Ribosome profiling was done as described in Mohamed et al.³³. Overnight cultures (16h, at 26°C with rotation) were diluted into 200mL LB and grown at 37°C for 5 hours before 100mL (Approximately 10^10 CFUs) was flash frozen by spraying into liquid nitrogen. Frozen droplets of bacterial culture were stored at -80°C before 50g of frozen droplets were lysed by cryo-mill. Lysate was cleared by centrifugation at 9,000xg at 4°C for 10mins and then ribosomes were pelleted over a sucrose cushion. Approximately 12.5 AU of sample was digested with MNase to remove RNA not protected by ribosomes and then 70S monosomes were isolated by fractionation over a 10%-50% sucrose gradient. RNA (or ribosome protected fragments) was isolated by phenol chloroform extraction and then samples were run on a 15% TBE Urea gel to selectively isolate fragments within the size range of 15-45nt, which has been previously shown to be the size range of fragments protected by bacterial ribosomes³⁴. cDNA libraries were prepared, sequenced and reads were mapped to the IP2666 Yersinia pseudotuberculosis GCF_003814345.1 genome assembly. All reads that mapped to ribosomal RNA, tRNA, and non-coding RNA were excluded. Ribosome occupancy was measured at the single codon level at the A site of the ribosome. Pause scores were calculated by taking the average ribosome density at a codon in a given gene and normalizing it to the average ribosome density across that gene for all genes in the genome.

Codon bias analysis

Codon usage was calculated for each gene in the Yersinia pseudotuberculosis IP2666 genome by measuring the frequency of each codon used in each gene, otherwise known as gene codon frequency. To calculate codon Z-scores, the gene-specific codon frequency was normalized to the average global usage (all ORFs in genome) of that codon and the standard deviation of that codon. Codon Z-scores for all up and downregulated proteins (2-fold change in protein abundance and p-value <0.05), were put into a data matrix and analyzed by principal component analysis. Principal component regression analysis was done using the log2fold change in protein abundance as the dependent variable. To analyze codon usage among genes that were regulated differently at the protein level versus the transcript level, we analyzed genes that were, 2-fold upregulated at the protein level and 1-fold or lower downregulated at the transcript level, and 2-fold downregulated at the protein level and 1-fold or higher upregulated at the transcript level. All graphing and analysis was performed in GraphPad Prism.

Codon-optimized GFP reporter

The gfp DNA sequence, with all CAA codons switched for CAG, was synthesized by Genewiz. The CAG optimized gfp was then cloned into pMMB207267 downstream of the P_tac (IPTG inducible) promoter and then transformed into WT and ΔtusB Yersinia by electroporation. Overnight cultures (16h, at 26°C with rotation) were sub-cultured into 4mL LB +1mM IPTG or untreated and grown for 8 hours at 37°C. Every 2h a 200ul sample was taken to measure optical density (OD600) and fluorescence (480/520) in a black walled 96 well plate with plate reader. GFP fluorescence is normalized to OD600 to account for bacterial density. At 4h, 1mL of sample was taken to isolate RNA to quantify gfp transcript levels by qPCR as described above.

Data availability

All data generated is included in the submitted files.

Extended Data Figures

Extended Data Figure 1.

Extended Data Figure 2.

Extended Data Figure 3.

Extended Data Figure 4.

Extended Data Figure 5.

Extended Data Figure 6.

Extended Data Figure 7.

Extended Data Table 1.

Acknowledgements

We thank the Davis lab for their support and feedback on this project. We could not have done this work without the help of Bob Cole, Josh Smith, and Tatiana Boronina at the Mass spectrometry core at Johns Hopkins School of Medicine. We could also not have done the ribosome profiling without the help of Allen Buskirk, Annie Campbell, and Rachel Green at Johns Hopkins School of Medicine. We also thank Patricia Dos Santos for her advice and for generously sharing plasmids with us. We would also like to thank Bob Ernst for his help coming up with the idea for the codon-optimized reporter and Satoshi Kimura for his advice on quantifying tRNA modification levels. This work was supported by NIH (NIAID) grants AI154116 and AI175307 to K.M.D. K.L.C. was also supported by training grant AI007417-26 through NIAID.

Additional information

Author contributions

Writing – original draft: K.L.C; Writing-review and editing: K.M.D; Conceptualization: K.L.C, K.M.D, P.C.D, T.J.B; Investigation: K.L.C, A.M; Methodology: T.J.B, P.C.D; Formal Analysis: K.L.C, T.J.B, A.M; Funding Acquisition: K.M.D

Declaration of generative AI and AI-assisted technologies

The authors declare that no AI or AI-assisted technologies were used in the writing process.

Funding

National Institute of Allergy and Infectious Diseases (AI154116)

National Institute of Allergy and Infectious Diseases (AI175307)

National Institute of Allergy and Infectious Diseases (AI007417-26)

Additional files

Supplemental Data 1. tRNA modification analysis (related to Figure 1).

Supplemental Data 2. Antibiotic susceptibility assays (related to Figure 1 and Extended Figures 2 and 3).

Supplemental Data 3. KEGG Analysis (related to Figure 3).

Supplemental Data 4. Ribosome pause scores (related to Figure 4).

Supplemental Data 5. CAG optimized GFP fluorescence and qPCR analysis (related to Figure 5).

Supplemental Data 6. RNA-seq, mass spectrometry, and codon usage analysis (related to Figures 3 and 6).

Supplemental Data 7. qPCR analysis for expression of s²U pathway genes with doxycycline and gentamicin treatment (related to Extended Data Figure 1).

Supplemental Data 8. Detected peptides with amino acid misincorporations (related to Extended Data Figure 4).

Supplemental Data 9. GFP decay in WT and ΔtusB (related to Extended Data Figure 5).

Supplemental Tables 1 and 2. Buffer gradient and compound table for LC-MS/MS analysis (related to Figure 1).