Genome-wide identification of fitness-genes in aminoglycoside-resistant Escherichia coli during antibiotic stress

Resistance against aminoglycosides is widespread in bacteria. This study aimed to identify genes that are important for growth of E. coli during aminoglycoside exposure, since such genes may be targeted to re-sensitize resistant E. coli to treatment. We constructed three transposon mutant libraries each containing > 230.000 mutants in E. coli MG1655 strains harboring streptomycin (aph(3″)-Ib/aph(6)-Id), gentamicin (aac(3)-IV), or neomycin (aph(3″)-Ia) resistance gene(s). Transposon Directed Insertion-site Sequencing (TraDIS), a combination of transposon mutagenesis and high-throughput sequencing, identified 56 genes which were deemed important for growth during streptomycin, 39 during gentamicin and 32 during neomycin exposure. Most of these fitness-genes were membrane-located (n = 55) and involved in either cell division, ATP-synthesis or stress response in the streptomycin and gentamicin exposed libraries, and enterobacterial common antigen biosynthesis or magnesium sensing/transport in the neomycin exposed library. For validation, eight selected fitness-genes/gene-clusters were deleted (minCDE, hflCK, clsA and cpxR associated with streptomycin and gentamicin resistance, and phoPQ, wecA, lpp and pal associated with neomycin resistance), and all mutants were shown to be growth attenuated upon exposure to the corresponding antibiotics. In summary, we identified genes that are advantageous in aminoglycoside-resistant E. coli during antibiotic stress. In addition, we increased the understanding of how aminoglycoside-resistant E. coli respond to antibiotic exposure.


TraDIS-based sequencing and bioinformatics analysis of libraries
Bioinformatics analysis confirmed that all three Tn-libraries were highly saturated, containing 233.994 (STREP), 383.779 (GEN) and 293.609 (NEO) unique chromosomal Tn-insertions distributed across the genome (Supplementary Fig. S1 online), corresponding to (on average) more than one Tn5 insertion every 20 nucleotides.TraDIS-based sequencing and bioinformatics studies revealed that 82.37% to 95.28% of the reads of the input and output libraries mapped accurately to the E. coli MG1655 U00096.3 reference genome (Supplementary Table S2 online).Moreover, according to the TraDIS-bioinformatics analysis, 371 genes were classified as 'essential' for growth on LB agar plates across the three aminoglycoside-resistant libraries (Supplementary Table S3 online).

Identification of fitness-genes and their functional classification
TraDIS-bioinformatics analysis comparing the output libraries generated without antibiotics (control) and the output libraries obtained in the presence of ½ MIC of the corresponding aminoglycoside revealed a list of 106 fitness-genes (Fig. 1).Fifty-six genes were advantageous for growth in the presence of STREP (Table 1), while 39 (Table 2) and 32 (Table 3) genes were classified as fitness-genes for growth in the presence of GEN and NEO.Eighteen genes were identified in more than one Tn-library, and three fitness-genes (polA, pcnB and atpE encoding for the DNA polymerase I, the poly(A) polymerase I and a subunit of the ATP synthase, respectively) were in common between all three output libraries.Of the 56 genes identified as fitness-genes in the STREP library, 13 and 2 genes were shared with the GEN and NEO Tn-library, respectively.A total of 38, 23 and 27 fitness-genes were unique to the STREP, GEN and NEO Tn-libraries (Fig. 1a).The 106 fitness-genes were categorized according to their function using associated Gene Ontology (GO) terms and KEGG pathways.Many fitness-genes were involved in stress-response (n = 15), cell cycle/division (n = 13), ATP metabolism (n = 9), enterobacterial common antigen (ECA) biosynthesis (n = 8), aminoacyl-tRNA biosynthesis (n = 7) and protein export (n = 5) while the function of 14 fitness-genes remained unknown (Fig. 1b, Supplementary Table S4 online).

Functional analysis of fitness-genes in the STREP Tn-library
Genes that were classified as advantageous for growth during STREP stress are listed in Table 1.STRING analysis was conducted to acquire a more profound insight into the cellular response of STREP-resistant E. coli to STREP exposure (Supplementary Fig. S2 online).Of the 56 identified fitness-genes, 10 genes were associated with cell division (ftsLNX, minCDE, envC, dedD, zapB, yciB), 7 were involved in ATP synthesis (atpABCDEFH) and 5 Figure 1.Fitness-genes after growth of the Tn-libraries generated in aminoglycoside-resistant E. coli MG1655 in the presence of STREP, GEN or NEO.(a) Venn-diagram of genes that were classified as fitness-genes based on a significant depletion of Tn5 insertions (log 2 FC < − 2, q-value < 0.01) was created in Excel.(b) Functional classification of aminoglycoside fitness-genes according to their associated GO terms and KEGG pathways were depicted using GraphPad Prism 9 (GraphPad Software, San Diego, USA).
Table 1.List of top 30 fitness-genes in MG1655_pACYC_sul2_strAB (STREP R ) during growth with 700 m/L STREP.*Information on the gene function was obtained from the Bio::TraDIS pipeline and completed with information from EcoCyc 25 .Fitness-gene identified in two (**) or three (***) libraries.www.nature.com/scientificreports/genes were linked to bacterial stress response (ortT, hflCK, yccA, cpxR).These results correspond to enrichment in the following GO terms: proton-transporting ATP synthase complex, cytokinetic process/division septum assembly, peptidoglycan binding and the KEGG pathway oxidative phosphorylation (Supplementary Table S5 online).Furthermore, ruvA and ruvC, which are involved in the formation and resolution of holiday junctions (four-stranded DNA structures linking two separated DNA helices during homologous recombination) 20 were classified as fitness-genes.Other fitness-genes identified were the regulator encoding genes slyA, cpxR and rspR and 7 genes encoding for proteins with unknown function.Interestingly, the majority (n = 33) of these fitnessgenes were localized to the bacterial cell membrane.

Functional analysis of fitness-genes in the GEN Tn-library
The 39 fitness-genes identified in the GEN Tn-library are shown in Table 2, and the results from the STRING analysis are depicted in Supplementary Fig. S3.Notably, 12 genes were associated with heat/stress response (yccA, pspA, hflC, hflK, sspA, rpoH, ppiD, cpxR, hns, hslU, hslV, tig).The enriched GO terms for these fitness-genes were response to heat, regulation of proteolysis and proton-transporting ATP synthase complex, similar to the results obtained for the STREP Tn-library and there were no enriched KEGG pathways (Supplementary Table S6 online).Additional fitness-genes were the three gene regulators: fabR, cpxR, and hns as well as clsA and pgpA, which are both involved in cardiolipin synthesis and 10 genes of unknown function.A total of 25 fitness-genes were associated with membrane localization.

Functional analysis of fitness-genes in the NEO Tn-library
Thirty-two genes were predicted to be fitness-genes during NEO exposure (Table 3), and STRING analysis (Supplementary Fig. S4 online) demonstrated that especially the wec gene-cluster that is responsible for ECA biosynthesis was advantageous during NEO stress as seven genes from this cluster (wecABCDEFG) were classified as fitness-genes.The importance of these genes was also highlighted by the enrichment in the GO term

Validation of the role of selected fitness-genes from the STREP and GEN Tn-libraries
To validate the TraDIS-based predictions, four fitness-genes/gene-clusters, associated with different functions, from the STREP and GEN Tn-libraries were selected and subjected to mutagenesis.The genes minCDE are involved in cell division 21 , hflCK and cpxR are associated with bacterial stress response 22,23 and clsA encodes for cardiolipin synthase 24 .Deletion of these genes/gene-clusters did not impair growth of the aminoglycoside-resistant strains in the absence of antibiotic (Supplementary Fig. S5 online).However, in the presence of 700 mg/L STREP or 50 mg/L GEN, the deletion of minCDE, hflCK, cpxR and clsA led to a slower growth of the mutants than the WT strain (Fig. 2).MIC testing revealed a two-fold decrease of the MIC of STREP and GEN for the ΔminCDE and ΔhflCK compared to their aminoglycoside-resistant parent strains (Supplementary Table S8 online).The clsA mutation resulted in twofold reduction of the MIC to GEN, while for the cpxR mutants a two and fourfold reduction of the MIC to STREP and GEN, respectively, was observed (Supplementary Table S8 online).
To validate the growth dynamics of STREP-and GEN-resistant deletion mutants in direct competition with their aminoglycoside-resistant parent strains, we conducted a competition assay in which each deletion mutant (ΔminCDE, ΔhflCK, ΔcpxR, and ΔclsA) was cultivated alongside its corresponding parent strain.After 24 h of coculturing, the relative abundance of mutants and aminoglycoside-resistant WT in each sample was determined.The results revealed that under control conditions (no antimicrobials), the mutant richness ranged from 40 to 52% in the co-culturing samples (ideally 50%).However, in samples exposed to 700 mg/L STREP or 50 mg/L Table 3. List of top 30 fitness-genes in MG1655_pACYC_aph(3′)-Ia(NEO R ) during growth with 3000 mg/L NEO.*Information on the gene function was obtained from the Bio::TraDIS pipeline and completed with information from EcoCyc 25 .Fitness-gene identified in two (**) or three (***) libraries.www.nature.com/scientificreports/GEN, the mutant richness substantially decreased to between < 1 and 10% (Fig. 3a,b), illustrating that these genes were important for competition in the presence of antimicrobials.

Validation of the role of selected fitness-genes in the NEO Tn-library
Deletion mutants of four selected fitness-genes/gene-clusters in the NEO Tn-library were constructed.The selected fitness-genes are involved in ECA biosynthesis (wec gene-cluster), magnesium sensing/transport (phoPQ) or encode for lipoproteins (lpp and pal).Seven members of the wec gene-cluster (wecABCDEFG) were classified as advantageous during NEO exposure, and we selected wecA (or rfe), a gene which initiates the synthesis of ECA 26 for mutagenesis.PhoPQ is a two-component system that senses cellular magnesium levels and activates genes responsible for magnesium uptake 27 .Lpp and Pal are both lipoproteins, which are important for outer membrane integrity 28 .
In the absence of NEO, the deletion of phoPQ did not affect growth in the NEO-resistant strain while deletion of wecA, lpp and pal caused a slightly attenuated growth phenotype compared to their NEO-resistant parent strain (Supplementary Fig. S6 online).When exposing the NEO-resistant parent strain and the derived deletion mutants to a sublethal concentration of 1000 mg/L NEO, their growth phenotype exhibited considerable variability as evidenced by the standard deviation shown in Fig. 4.However, the growth of the ΔwecA, ΔphoPQ, Δlpp mutants was more affected by the NEO exposure, than the parent strain resulting in an extended lag phase, especially in the Δlpp mutant, while the Δpal mutant appeared to growth slightly better than the parent strain during NEO exposure.Furthermore, MIC testing uncovered a twofold, twofold and fourfold decrease of the MIC of NEO for the ΔwecA, Δpal and Δlpp NEO-resistant strains compared to the parent strain, respectively, while the MIC of NEO against the phoPQ mutant was not affected (Supplementary Table S8 online).
When we investigated the growth dynamics of the NEO-resistant deletion mutants in direct competition with their NEO-resistant parent strain MG1655_pACYC_aph(3′)-Ia(NEO R ), the mutant richness ranged from 99 to 11% in the co-culture samples of the ΔphoPQ, ΔwecA, Δlpp mutants after co-culture with their parental strain for 24 h under control conditions.In contrast, the mutant richness decreased considerably to less than 7% in the samples exposed to 1000 mg/L NEO (Fig. 3).An exception was the Δpal mutant, where there was no decrease of mutant richness in the co-culturing sample upon 1000 mg/L NEO stress (Fig. 3c).www.nature.com/scientificreports/

Homology of fitness-genes to human proteins
To ensure that the proteins encoded by the identified fitness-genes did not show significant homology to proteins in humans, we compared the protein sequence of MinC, MinD, MinE, HflC, HflK, ClsA, CpxR, PhoP, PhoQ, WecA, Lpp and Pal against the human proteome using the NCBI-pBlast tool.This in silico investigation showed lack of homology.However, homology was identified to other E. coli strains as well as various members of the Enterobacteriaceae family such as Salmonella, Klebsiella, Shigella and Citrobacter (Supplementary

Discussion
Aminoglycosides, which directly target the bacterial ribosome, have been used for decades in human and veterinary medicine, leading to widespread resistance towards this class of antibiotics.Little is known about cellular changes that occur in aminoglycoside-resistant bacteria during treatment, even though this could be the key to target these bacteria in the future.Additionally, it has been debated if the bacterial ribosome might not be the only target of aminoglycosides and it has been disputed if their bactericidal action could be linked to the formation of reactive oxygen species contributing to the bacterial cell death 29,30 .
TraDIS has been established as a powerful tool to identify sets of fitness-genes that are advantageous during antibiotic exposure, including exposure to colistin 19 and CTX 16 in antibiotic-resistant bacteria, as well as fosfomycin 31 , trimethoprim and sulphonamide 32 in antibiotic-sensitive bacteria.Using TraDIS, we identified 106 conditionally essential genes (STREP (n = 56), GEN (n = 39) and NEO (n = 32)) which could represent targets for helper-drugs to re-sensitize resistant E. coli to aminoglycoside treatment.
Three fitness-genes were common in the three libraries: polA, which encodes for DNA polymerase I 33 , pcnB that is involved in the polyadenylation of mRNAs 34 and atpE which encodes for a subunit of the ATP synthase 35 .ATP synthase genes were previously reported to be advantageous during aminoglycoside stress 36,37 and we did not characterize these further.A possible explanation for the Poly-A polymerase PcnB and the DNA polymerase PolA appearing on the list of aminoglycoside fitness-genes could be that both genes are involved in plasmid replication/plasmid copy number maintenance 38,39 , and thus may have affected resistance by affecting the copy number of the plasmids containing the resistance genes.However, further studies are needed to clarify this.Besides these genes, two genes were shared between the NEO and STREP Tn-libraries (tatC and lpp) and no additional genes were shared between the NEO and GEN Tn-libraries.The STREP and GEN Tn-libraries on the other hand, shared 16 fitness-genes, showing that the STREP and GEN Tn-libraries were more alike compared to the NEO Tn-library.Interestingly, most (n = 55) of the fitness-genes identified were localized to the bacterial cell membrane.This suggests that the aminoglycoside effectiveness relates to the cell membrane.This could either be the ability to penetrate the bacterial cell membrane, or it could be membrane stability related to the killing mechanism of the antimicrobial which has previously been proposed (although contested) 30 .Anyway, this suggests that the membrane in general is a promising helper-drug target.
We identified 371 genes that were classified as 'essential' for the growth on LB agar plates in the generated aminoglycoside-resistant E. coli MG1655 strains.Genes categorized as 'essential' in TraDIS are either essential for survival or their interruption is associated with a severe fitness-cost under the given conditions 40 .This number of 'essential' genes is in a similar range as previously published counts of 'essential' genes in E. coli that were determined by TraDIS analysis 16,40 , which provides strong assurance in the quality of our Tn-libraries enabling us to obtain meaningful conclusions.
To validate our TraDIS results, we constructed deletion mutants of eight selected fitness-genes.The criteria for fitness-gene selection were: (i) present in at least two antibiotic Tn-libraries or (ii) specific to NEO stress, (iii) on the list of enriched GO terms, (iv) not previously associated with aminoglycoside-resistance and (v) preferably, in a cluster where several genes belonging to the same operon/complex/cluster were classified as fitness-genes.
MIC testing and growth analysis showed that deletion of cpxR increased the efficiency of STREP and GEN, while ΔclsA only presented a higher sensitivity towards GEN.Deletion of lpp and pal increased the efficiency of NEO in the corresponding aminoglycoside-resistant strains.These results confirmed that cpxR, clsA, lpp and pal were indeed fitness-genes during exposure to one or more aminoglycoside(s).cls mutants in E. coli are sensitive to novobiocin, an antibiotic that inhibits the DNA gyrase 41 and this gene may be added to the list of genes which are important for several antibiotic stresses.minCDE and hflCK were specifically detected as fitness-genes during STREP and GEN exposure.The minCDE gene-cluster is involved in cell division by restricting Z-ring formation to the mid-cell 21,42 .Analyzing the fitnessgenes from the STREP Tn-library revealed that 10 genes were associated with cell division.This complex cellular process is executed by the 'divisome' , an assembly of multiple essential and non-essential proteins that span the cell envelope 43 .In our study, the most important genes involved in cell division were classified either as essential for growth or as fitness-genes for growth in the presence of the antimicrobials, highlighting the role of cell division genes especially during STREP stress.Essential genes, by definition, are excluded from the list of fitnessgenes, which potentially affected our STRING analysis of aminoglycoside exposed cell systems, as automated analysis may have overlooked connections between fitness-genes if they are linked via an essential gene.In our study, phenotypic analysis including MIC testing and growth analysis confirmed that minCDE played an important role during STREP and GEN exposure.In a previous TraDIS-based study, Tn-insertions into minCD have been associated with increased susceptibility to trimethoprim and sulfamethoxazole in an antibiotic-sensitive E. coli BW25113 32 , highlighting the importance of minCDE during antibiotic stress.In the GEN and STREP Tn-libraries, 12 and 5 genes were associated with cellular stress and heat response, respectively; an example was the HflCK-complex, which modulates the activity of the essential protease FtsH 23 .Interestingly, the divisome protein FtsZ has been shown to be an in vitro substrate of the protease FtsH possibly linking ftsH and hflCK to cell division 44 .Deletion of the hflCK cluster confirmed that the cluster was indeed important for growth in the STREP and GEN-resistant background in the presence of the corresponding aminoglycosides.In previous studies, disruption of hflK (and hflC 30 ) has been associated with increased sensitivity to GEN in GEN-susceptible E. coli MG1655 30 , MG1655/pTF2 16 and E. coli BW25113 45 allowing us to confide in the reliability of our data.
The wec-cluster (wecABCDEFG), which is involved in ECA biosynthesis 26 , and phoPQ, a two-component system associated with magnesium sensing and transport 27 were unique to the NEO Tn-library.As they were not identified in any other antibiotic-resistant library mentioned above, they seemed to be specific for NEO exposure.Understanding the role of these fitness-genes might increase our insight in what happens in the bacterial cell during NEO stress, and it might explain why the fitness-genes during NEO stress differ from the fitness-genes associated with the other two aminoglycosides.We decided to delete wecA, as this gene has been described as essential for the ECA biosynthesis 46,47 .The wecA deletion mutant presented a twofold decrease in MIC compared to the NEO-resistant parent strain, while no difference in MIC was observed for the phoPQ mutant.Growth analysis of the two mutants did not show an impaired growth compared to the parental strain at 1000 mg/L NEO.Here, we used 1000 mg/L NEO instead of 3000 mg/L (the original ½ MIC of the NEO-resistant strain) for growth analysis because neither the NEO-resistant mutants nor their parent strain grew sufficiently at 3000 mg/L.One potential reason for not confirming phoP and phoQ as fitness-genes during NEO exposure could be that phoPQ might only be advantageous during higher concentrations of NEO.
During validation using deletion mutants, we observed reproducible, yet not very pronounced differences (only two to fourfold change in MIC, for example) between deletion mutants and corresponding parental strains.Fitness in bacteria has been quantified by measuring the (maximum) growth rate in monoculture, MIC testing and by competition assays 48 .The latter assay is regarded as closest to the meaning of fitness in the evolutionary sense 48 , and this also resembled the situation during the TraDIS procedure, where a pool of more than 200.000 mutants was cultivated together 49 .It may also be considered similar to natural conditions, where in many situations a vast variety of different microorganisms are present at the same time and compete for growth 50 .We thus cultivated selected deletion mutants together with their aminoglycoside-resistant parental strain.After growth without any antibiotic selection, mutant and parental strains were present in almost equal amounts (40-52%) in the STREP and GEN-resistant deletion mutants.However, the wecA, lpp and pal mutants, selected from the NEO Vol:.( 1234567890 www.nature.com/scientificreports/library, presented a slightly altered growth phenotype compared to their parent strain in the absence of NEO, which was evident in the competition assay, resulting in a mutant richness ranging from 99% (ΔwecA) to 11% and 4% (Δlpp and Δpal).However, irrespective of this, when we added sublethal concentrations of aminoglycosides, the majority of mutants were reduced in richness compared to the parental strain, showing that during sublethal antibiotic exposure, they were outcompeted.The only exception to this was Δpal, but this mutant showed a very low mutant richness without the antimicrobial, and the applied prediction method might not be exact in such circumstances.Based on comparison, we suggest that testing the deletion mutants in monoculture is not an ideal validation method for TraDIS results.We compared the fitness-genes from our aminoglycoside Tn-libraries with fitness-genes identified in other antibiotic-resistant Tn-libraries to gain an insight into which fitness-genes were antibiotic specific or in general advantageous during antibiotic stress (Supplementary Table S10 online).Among 35 fitness-genes in a multidrugresistant K. pneumoniae challenged with colistin 19 , six were shared with our list of aminoglycoside fitness-genes, and among 57 genes fitness-genes during CTX exposure in ESBL E. coli 16 six were predicted as fitness-genes in the current study.Four genes namely lpp, pal, cpxR and envC were identified in all three studies.In addition to this, cpxR has been associated with an increased sensitivity to carbapenem antibiotics in a carbapenem resistant K. pneumoniae 51 .lpp deletion mutants have shown an increased susceptibility to vancomycin in E. coli 52 while pal deletion has not previously been associated with antibiotic stress.These findings are noteworthy, since the antibiotics colistin, CTX + carbapenems, vancomycin and aminoglycosides belong to different classes of antibiotics.Understanding the role of these "general" antibiotic fitness-genes during antibiotic exposure in more detail would be beneficial and could result in the discovery of potential helper-drug targets for several antibiotics.Furthermore, comparing the fitness-genes identified in our GEN Tn-library (n = 39) to the fitness-genes described in a GEN-sensitive library (n = 18) 16 , revealed that 11 fitness-genes were shared between the output lists of genes while 33 and 12 fitness-genes were uniquely identified, respectively.These results indicate that some fitness-genes are aminoglycoside concentration dependent, while others appear to be beneficial at multiple drug concentrations.
Limitations of the TraDIS methodology have been discussed and reviewed in previous studies 17,40,53,54 and these limitations also apply to the current study.Briefly, the Tn5 transposon utilized has AT insertion-site bias, the method tends to overpredict essential as well as fitness-genes, it has orientation-dependent bias, and it may overlook essential regions in non-essential genes.Finally, TraDIS may not be well suited for predictions related to very short genes and sRNA 17,40,53 .Furthermore, a limitation of the current TraDIS study was that all experiments were conducted in the laboratory adapted strain E. coli MG1655 in LB media.However, the fitness-genes that we identified are relatively conserved among Enterobacteriaceae and other bacterial genera (Supplementary Table S9 online), and the fact that some of these fitness-genes have already been described as advantageous during antibiotic exposure in either other E. coli strains 32,45,52 or in other species 19,37 supports the results.Nevertheless, further research in other bacteria including clinical isolates in in vivo settings is needed to confirm that our findings are not strain-or media-specific.
In conclusion, we present a list of 106 conditionally essential genes (fitness-genes) that become specifically advantageous during STREP, GEN and NEO exposure in aminoglycoside-resistant E. coli strains.These fitnessgenes are mainly membrane located and involved in bacterial stress response, cell cycle/division, ATP metabolism and ECA biosynthesis, highlighting the role of these biological processes under aminoglycoside antibiotic stress conditions.With the current study, we improve our understanding of the cellular changes that occur in resistant E. coli during aminoglycoside exposure.Additionally, these fitness-genes present potential candidates for novel helper-drug targets that could be utilized to re-sensitize E. coli to aminoglycoside antibiotics.

Bacterial strains
Bacterial strains and plasmids used in this work are presented in supplementary Table S1.The wild type strain employed was E. coli K12 MG1655 55 .Strains were cultivated in Luria Broth (LB) Lennox under vigorous agitation or on LB agar plates overnight at 37 °C, except for the strains containing temperature sensitive pKD46 or pCP20 plasmids, which were cultivated at 29 °C.The media was supplemented with antibiotics (Sigma-Aldrich, Copenhagen, Denmark) including 50 mg/L NEO, 50 mg/L STREP, 20 mg/L GEN or 50 mg/L chloramphenicol (CHL) where appropriate.

Preparation of competent cells and transformation
Electrocompetent cells of E. coli MG1655 were prepared as previously described with slight modifications 54 .In short, E. coli MG1655 was cultivated in LB medium overnight, afterwards diluted 1:100 in LB media and grown to an OD 600 of 0.5-0.7.The bacteria cells were centrifuged and first washed with 1× and 0.5× volume of cold Milli-Q water and then twice in 0.05× volume of 10% ice-cold glycerol followed by resuspension in 0.002× volume of 10% ice-cold glycerol.Aliquots of electrocompetent cells were stored at − 80 °C until use.
For transformation, 50 μL of electrocompetent cells were mixed with 1 μL plasmid and electroporated using an Eporator ® electroporator (Eppendorf, Hamburg, Germany) set to 2.5 kV in a 2 mm cuvette essentially as previously described 54 .After electroporation, the mixture was recovered in 900 μL pre-warmed Super Optimal broth with Catabolites repression (SOC) medium and incubated at 37 °C for 1 h.Afterwards, 10 and 100 μL of the transformed cells were plated on LB agar plates containing the corresponding antibiotic and incubated at 37 °C overnight.Strains containing temperature sensitive plasmids (e.g.pKD46 or pCP20) were incubated at 29 °C during all incubation steps.On the following day, eight to ten antibiotic-resistant colonies were tested for the presence of the correct plasmid using colony PCR with specific primers (Supplementary Table S11  The pACYC184 backbone was amplified by PCR using pACYC184 (GenBank X06403) as template.The aph(3′)-Ia(NEO R ) gene, the sul2_strAB(STREP R ) genes and the aac(3)-IV(GEN R ) gene were amplified by PCR from the E. coli strains 113026-3 (NEO R ) and 113790-2 (GEN R , STREP R ) 12 .All PCR products were generated using Q5 DNA Polymerase (New England BioLabs, Ipswich, USA) with overhang primers (listed in supplementary Table S11).
The PCR products containing the plasmid backbone and the selected resistance gene were mixed according to the HiFi cloning protocol (New England BioLabs, Ipswich, USA) and transformed into electrocompetent E. coli DH5α as described above.The plasmids were extracted using GeneJET Plasmid Miniprep (Thermo Fisher Scientific, Roskilde, Denmark) and validated by Sanger sequencing (Macrogen Europe).Next, the plasmids were transformed into E. coli MG1655 via electroporation as described above resulting in the three aminoglycosideresistant strains: MG1655 pACYC184_ΔTet R (CHL R )_aph(3′)-Ia(NEO R ) (plasmid in supplementary Fig. S7 online), MG1655 pACYC184_ΔTet R (CHL R )_sul2_strAB(STREP R ) (plasmid in supplementary Fig. S8 online) and MG1655 pACYC184_ΔTet R (CHL R )_aac(3)-IV(GEN R ) (plasmid in supplementary Fig. S9 online).These E. coli strains are referred to as STREP-, GEN-and NEO-resistant strains across the manuscript.

Antimicrobial susceptibility testing
The minimum inhibitory concentration (MIC) of STREP, NEO and GEN against the strains under study was assessed using broth micro-dilution method according to the CLSI guidelines 56 .

Bacterial growth curves
Bacterial growth experiments were conducted in three to four biological replicates using a Bioscreen C (Thermo Labsystems).Analogous to procedures for MIC analysis, a 0.5 McFarland suspension of each strain was prepared and 100 µL of the bacterial suspension was added to 10 mL LB media.Next, 100 µl of LB with bacteria were mixed with 100 µL of LB media with or without the corresponding antibiotic in a 100-well honeycomb plate.The final concentration of the antibiotics was 700 mg/L STREP, 50 mg/L GEN and 1000 mg/L NEO.The first and last row of the honeycomb plate were used as blank containing only media and four technical replicates of the same treatment were prepared.Bacterial growth was measured every 20 min under continuous, medium shaking for 24 h at 37 °C.The growth curves and calculation of standard error were generated with GraphPad Prism 9 (GraphPad Software, San Diego, USA).

TraDIS input library construction and validation
The transposome mix containing 4 µL of EZ-Tn5 < DHFR-1 > transposon (a Tn5 transposon harboring a trimethoprim (TRI) resistance gene encoding for a dihydrofolate reductase (DHFR)), 1 µL of transposase (both Epicentre, Nordic Biolabs, Täby, Sweden) and 2 µL of 100% glycerol was prepared and mixed, incubated for 30 to 60 min at room temperature and afterwards stored at -20 °C until use following manufacturer's instructions.Fifty µL of electrocompetent cells of each aminoglycoside-resistant strain were mixed with 1 µL of transposome mix.Electroporation was performed in a 0.1 mm cuvette using an Eporator ® electroporator (Eppendorf, Hamburg, Germany) set to 1.8 kV, and the cells were recovered as described above.After incubation, 100 µL cells were spread on nine large LB agar plates (15 cm) containing 20 mg/L TRI.Following overnight incubation, the TRI-resistant colonies from each plate were collected using a sterile Drigalski spatula and resuspended in 2 mL LB + 20% glycerol (Fig. 5).With this method, three high-density transposon mutant libraries were constructed in the three aminoglycoside-resistant E. coli MG1655 strains (referred to as NEO, STREP and GEN Tn-libraries).Sixteen randomly selected colonies of each aminoglycoside-resistant Tn-library were tested for random Tn5-DFRH-1 insertion by colony 'random amplification of transposon ends' (RATE) PCR as previously described 57 .
For each input library, colonies from all nine plates were counted, pooled together in one 50 mL tube, and stored at -80 °C.The same procedure was repeated, and Tn-mutants were pooled into the same tube until at least 200.000 colonies were collected resulting in three Tn-input libraries (one per aminoglycoside-resistant strain).Colony forming units (CFU) of each input library suspension were counted, and aliquots containing around 1 × 10 10 CFU in 1 mL were prepared and stored at − 80 °C (Fig. 5).

TraDIS output library construction, DNA extraction and sequencing
To create the output Tn-libraries, 100 µL of the 1 mL aliquots from each aminoglycoside-resistant input Tn-library were used for inoculation of 9.9 mL LB media supplemented with and without ½ MIC of the corresponding aminoglycoside (700 mg/L STREP, 50 mg/L GEN and 3000 mg/L NEO) (Fig. 5).The concentration of ½ MIC was selected to allow direct comparison to the previous study on CTX treatment 16 .The bacterial suspensions were incubated at 37 °C under vigorous agitation and harvested when the cultures reached an OD 600 of 4 (Fig. 5).The experiments were performed in duplicates resulting in the generation of 12 output Tn-libraries.The following steps were performed essentially as previously described 16 .Briefly, for DNA isolation, 1 mL of each output and 0.5 mL of each input library were centrifuged at 12,000 rpm for 2 min, the pellet was washed using DPBS and DNA was extracted using the Genelute™ Bacterial Genomic DNA kit (Sigma-Aldrich, Copenhagen, Denmark) according to manufacturer's instructions with elution in 100 µL water.DNA quality was assessed by NanoDrop and only DNA samples with a ratio of OD 260/280 of 1.8-2.0 and OD 260/320 of 2.0-2.2 were used for further sequencing steps (Supplementary Table S12 online).DNA concentration was measured using Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, Roskilde, Denmark) (Supplementary Table S12 online) and 4-5 μg of DNA per library were fragmented into approximately 300 bp fragments using Covaris M220 (Covaris LLC, Woburn, USA).DNA samples were prepared for TraDIS sequencing according to protocols previously described 16,18 using Tnspecific primers for PCR enrichment (Supplementary Table S13  www.nature.com/scientificreports/pooled and sequenced using a MiSeq machine with MiSeq reagent kit V2 (50 cycles) (Illumina) as previously described 18,54 .

Analysis of TraDIS data
The DNA sequences containing the Tn were analyzed using the Bio::TraDIS pipeline 18 (https:// github.com/ sanger-patho gens/ Bio-Tradis) with some modifications as described previously 16 .In short, the raw reads obtained from MiSeq sequencing were processed using the fq2bam.plscript to relocate the tag from the read name and to convert the files into SAM format.Using samtools, these SAM files were afterwards converted into BAM files, which were utilized for analysis with the check_tradis_tags and add_tradis_tags scripts.The resulting output BAM files were converted to FASTQ format with samtools.In the following step, the bacteria_tradis mapping pipeline was used to first filter the reads for Tn-tags (10 base pairs that match the 3′ end of the Tn) and subsequently to remove these tags.With SMALT, a short read mapping tool (https:// www.sanger.ac.uk/ tool/ smalt-0/), the trimmed reads were mapped against the MG1655 U00096.3 reference genome (https:// www.ncbi.nlm.nih.gov/ nucco re/545778205) and the corresponding plasmid sequence.From the resulting files, the exact insertion site of each Tn across the genome was predicted, the read counts and accurate number of unique insertion sites (UISs) per gene were determined.For visualization of the UISs on the gene level or as part of a circular genome representation, Artemis (version 18.2.0)or Artemis DNAPlotter (version 18.2.0)were utilized 58,59 .Further analysis steps were performed with the R scripts tradis_essentiality.R for gene essentiality using a combined file containing raw reads from aminoglycoside input libraries (Supplementary Table S3 online) and tradis_comparisons.R for conditionally essential genes (Supplementary Table S14 online), the scripts were included in the Bio::TraDIS pipeline.For the gene essentiality, the insertion site index (IID) (i.e. the number of Tn-insertions per gene divided by the gene length) was calculated and statistical analysis of the gamma distribution of the IID classified each gene as For each aminoglycoside-resistant E. coli MG1655 strain, a Tn-mutant library (input library) was created and cultured (100 µL) in LB (9.9 mL) without antibiotic (to generate the output library: control) or supplemented with a sublethal concentration of the corresponding aminoglycoside (to generate the output library: with antibiotic) at 37 °C.When an OD 600 of 4 was reached, the cells were harvested, genomic DNA of the input and the two output libraries (in replicates) was extracted, prepared for sequencing and sequenced as described below.The figure was created in PowerPoint using BioRender.com.

Figure 2 .
Figure 2. Growth curves of the aminoglycoside-resistant ΔminCDE, ΔhflCK, ΔcpxR and ΔclsA in the presence of 700 mg/L STREP (a) and 50 mg/L GEN (b).(a) The growth of MG1655_pACYC_strAB (STREP R ) (dark blue) in LB is presented compared to the growth of the four STREP-resistant deletion mutants minCDE, hflCK, cpxR and clsA (red) in the presence of 700 mg/L STREP.(b) The growth of MG1655_aac(3)-IV(GEN R ) in LB supplemented with 50 mg/L GEN is visualized (dark blue) and compared to the growth of the four GENresistant deletion mutants minCDE, hflCK, cpxR and clsA (red).The mean values from four biological replicates (with four technical replicates each) and the corresponding ± standard deviation were calculated and illustrated with GraphPad Prism 9 (GraphPad Software, San Diego, USA).

Figure 3 .
Figure 3. Competition between deletion mutants and parental strains with and without STREP (700 mg/L), GEN (50 mg/L) or NEO (1000 mg/L).The relative richness of deletion mutants after a 24-h co-culture experiment with STREP (a), GEN (b) or NEO (c) resistant parental strains are shown.Two sets of conditions are depicted: control conditions (dark turquoise/dark blue/dark red) and exposure with 700 mg/L STREP, 50 mg/L GEN or 1000 mg/L NEO (light turquoise/light brown/light red).Each mixture sample initially included the same concentration of KO-mutants and parental strains.The relative richness was calculated by sample sequencing and an in-house in silico analysis pipeline (detailed in "Method").The figure was depicted with GraphPad Prism 9 (GraphPad Software, San Diego, USA).

Figure 4 .
Figure 4. Growth curves of ΔphoPQ, ΔwecA, Δlpp and Δpal in the presence of 1000 mg/L of NEO.The growth of MG1655_pACYC_aph(3′)-Ia(NEO R ) (dark blue) compared to the growth of the four NEO-resistant deletion mutants phoPQ, wecA, lpp and pal (red) in LB supplemented with 1000 mg/L NEO is shown.Mean values from three biological replicates (with four technical replicates each) and the corresponding ± standard deviations were calculated and illustrated with GraphPad Prism 9 (GraphPad Software, San Diego, USA).

Figure 5 .
Figure 5. Workflow of TraDIS library preparation and sequencing in an aminoglycoside-resistant E. coli MG1655.For each aminoglycoside-resistant E. coli MG1655 strain, a Tn-mutant library (input library) was created and cultured (100 µL) in LB (9.9 mL) without antibiotic (to generate the output library: control) or supplemented with a sublethal concentration of the corresponding aminoglycoside (to generate the output library: with antibiotic) at 37 °C.When an OD 600 of 4 was reached, the cells were harvested, genomic DNA of the input and the two output libraries (in replicates) was extracted, prepared for sequencing and sequenced as described below.The figure was created in PowerPoint using BioRender.com.

Construction of aminoglycoside resistance plasmids with pACYC184 backbone and obtaining of mutant strains
online).