Non-lethal oxidative stress boosts bacterial survival and evolvability under lethal exposure

When oxygen began to accumulate by photosynthesis and aerobic respiration evolved, reactive oxygen species such as H2O2 became a challenge. Later on, multicellular organisms began to employ reactive oxygen species as potent weapons against microbes. Although bacterial defences against lethal and sub-lethal oxidative stress have been studied in model bacteria, the role of fluctuating concentrations remains unexplored. It is known that sub-lethal exposure of E. coli to hydrogen peroxide results in enhanced survival upon further exposure. Here we investigate the priming response to H2O2 close to physiological concentrations. The basis and the duration of the response (memory) were also determined by time-lapse quantitative proteomics. A low level of peroxide induces several scavenging enzymes that are quite stable, protecting cells from future exposure. We then asked if the phenotypic resistance against H2O2 alters resistance evolution against oxygen stress. Experimental evolution for H2O2 resistance revealed faster evolution and higher levels of resistance in primed cells. Several mutations were found to be associated with resistance in evolved populations affecting different loci but counterintuitively, none of them was implicated in scavenging systems. Our results have important implications for host colonisation and infections where microbes often encounter reactive oxygen species in gradients. Author summary Throughout evolutionary time bacteria were exposed to reactive oxygen species (ROS) and evolved the ability to scavenge toxic oxygen radicals. Multicellular organisms evolved also the ability to produce such oxygen species against pathogens. Recent studies also suggest that ROS such as hydrogen peroxide play an important role during host gut colonisation by its microbiota. Traditionally, experiments with different antimicrobials are carried out using fixed concentrations while in nature, including intra-host environment, microbes experience this type of stress in steps or gradients. Here we show that bacteria treated with sub-lethal concentrations of H2O2 (priming) survive far better than non-treated cells when they later encounter a stronger concentration. We also found that ‘priming’ response has a protective role from lethal mutagenesis. This protection is provided by long-lived proteins that, upon priming, remain at a high level for several generations as determined by time-lapse LC-mass spectrometry. Bacteria that were primed evolved H2O2 resistance faster and to a higher level. Moreover, mutations that increase resistance to H2O2, as determined by whole-genome sequencing (WGS), do not occur in known scavenger encoding genes but in loci coding for other functions, at least in E. coli.

Experimental evolution for H 2 O 2 resistance revealed faster evolution and higher levels of resistance in primed cells. Several mutations were found to be associated with resistance in evolved populations affecting different loci but counterintuitively, none of them was implicated in scavenging systems. Our results have important implications for host colonisation and infections where microbes often encounter reactive oxygen species in gradients.

Author summary
Throughout evolutionary time bacteria were exposed to reactive oxygen species (ROS) and evolved the ability to scavenge toxic oxygen radicals.
Multicellular organisms evolved also the ability to produce such oxygen species against pathogens. Recent studies also suggest that ROS such as hydrogen peroxide play an important role during host gut colonisation by its microbiota.
Traditionally, experiments with different antimicrobials are carried out using fixed concentrations while in nature, including intra-host environment, microbes experience this type of stress in steps or gradients. Here we show that bacteria treated with sub-lethal concentrations of H 2 O 2 (priming) survive far better than non-treated cells when they later encounter a stronger concentration. We also found that 'priming' response has a protective role from lethal mutagenesis. This protection is provided by long-lived proteins that, upon priming, remain at a high level for several generations as determined by time-lapse LC-mass spectrometry. Bacteria that were primed evolved H 2 O 2 resistance faster and to a higher level. Moreover, mutations that increase resistance to H 2 O 2 , as determined by whole-genome sequencing (WGS), do not occur in known scavenger encoding genes but in loci coding for other functions, at least in E. coli.

Introduction
The ability to solicit a stress response when encountering repeated stress relies on 'remembering' a similar event from the past (memory), a common trait to many biological entities [1]. During the course of an infection or the colonisation of a host, bacteria encounter increasing and repeated stress imposed by the host immune system [2,3]. Obata et al., for example, recently demonstrated that low levels of hydrogen peroxide (H 2 O 2 ) in the gut of Drosophila melanogaster shape the composition of the gut microbiota resulting in differential survival of the flies [3]. Here, we report how bacteria respond to low levels of reactive oxygen species (ROS) exposure and how this impacts bacterial fitness. We then investigate if the phenotypic response to a sub-lethal dose of H 2 O 2 facilitates resistance evolution, hence providing a test of the phenotype-first hypothesis, which proposes that environmentally initiated phenotypic change precedes or even facilitates evolutionary adaptation [4].
In Escherichia coli, the defences against oxidative stress depend on transcriptional regulators such as OxyR or SoxR that detect different changes in the redox balance. They also induce the production of detoxifying enzymes, DNA repair and protection systems and other proteins [5]. Oxidation of OxyR by H 2 O 2 leads to the formation of an intramolecular disulfide bond between cysteine residues 199 and 208. OxyR is deactivated by enzymatic reduction with glutaredoxin I (Grx) or thioredoxin (Trx). OxyR positively regulates catalases and peroxidases. Because the Grx/GorA system is also transcriptionally regulated by OxyR, the whole response is self-regulated by switching on or off the active form of OxyR [6].
The OxyR-mediated oxidative stress response results in scavenging of H 2 O 2 and mitigates the toxicity of this by-product of aerobic metabolism. It includes the induction of Suf proteins that form a complex to supply apo-enzymes with ironsulfur clusters. Suf system replaces the normal iron-sulfur cluster (Isc), required for critical biochemical pathways such as respiration, which is disrupted by H 2 O 2 -sensitive systems. The repair of those clusters by Suf is necessary to prevent the failure of the TCA cycle. The iron-sulfur clusters are enzymes that employ ferrous iron as a co-factor which can increase the risk of fuelling a Fenton reaction. Active OxyR also induces Dps, a ferritin-class protein, that strongly suppresses the amount of DNA damage by sequestering the unincorporated iron [9].
The spontaneous reaction of hydrogen peroxide with free ferrous iron (Fe 2+ ) at physiological pH, oxidising iron to Fe 3+ and generating hydroxyl radicals and water is named the Fenton reaction. Hydroxyl is a strong non-selective radical that damages many cellular components, particularly DNA [10,11]. H 2 O 2 impedes the function of the Fur regulatory protein and can directly damage many cell components but is less toxic than other reactive oxygen species such as hydroxyl. These radicals are responsible for DNA damage, indirectly promoted by H 2 O 2 as a consequence of the Fenton reaction [12].
A classic paper by Imlay et al. describes that the pre-treatment of E. coli with a low dose (60 µM) of H 2 O 2 can increase the survival upon subsequent exposure to an otherwise lethal dose (30 mM) [13]. However, neither the duration of such priming responses nor the molecular mechanisms of its maintenance, i.e. the memory, have been studied. Our study has two main aims. First, we investigated the main factors in the H 2 O 2 priming response and for how long the response is sustained. Second, we used this system to test the hypothesis that inducible phenotypes accelerate adaptive evolution [4]. Therefore, we experimentally evolved Escherichia coli under a growing level of H 2 O 2 with and without priming. For H 2 O 2 -resistant populations evolved after priming and nonpriming regimes, genome re-sequencing analyses were performed to identify mutations. We focus on growing bacteria (exponential phase), as this better represents an infection or colonisation of host surfaces including the gut [2,3].
Moreover, this allows us to focus on H 2 O 2 priming response in proliferating cells that differ from the stationary phase. RpoS, the master regulator of the general stress response, controls a pronounced phenotypic H 2 O 2 tolerance during stationary phase [14,15].

Priming by H 2 O 2 results in higher bacterial survival
We found that, in our conditions, the minimal inhibitory concentration for H 2 O 2 is 1 mM. This concentration was subsequently used as a reference for 30minute killing curves that show a clear dose-effect in survival rate (Fig 1a, ranging from 50 µM to 1 mM, Fig 1a, Fig 1b). We also determined that the priming response contributes to a more efficient removal of H 2 O 2 by quantifying it in the supernatant of the cultures (S2 Table). After  growth rate, or doubling time is affected. Therefore, differential survival can be only attributed to cell response but not to cell growth arrest. The cost can be explained because even very low concentrations of H 2 O 2 damage the ironsulphur clusters compromising respiration. However in low-density bacteria in rich medium, E. coli can grow very fast by fermentation [17]. This wasteful strategy allows quick generation of ATP on the expenses of the medium and can explains the small difference in carrying capacity and the area under the curve of the growth curves (Fig S1, S1 Table). In our experiments, this cost has no consequences because bacteria are maintained in low density and in exponential growth. This scenario should be similar while starting an infection or colonisation of a host's gut.
These results are in agreement with data previously reported by Imlay et al. [13], showing survival protection even to doses as high as 30 mM, a concentration close to H 2 O 2 usage as a disinfectant. Our concentrations are also in the range of some in vivo situations. For example, tailfin transection on zebrafish larvae induces a rapid increase in H 2 O 2 levels ranging from 100-200 μM in the wound margins [2]. In some cases, more than 100 μM of H 2 O 2 have been reported in human and animal eye vitreous humour and aqueous humour [18].

Duration of the priming response
As the priming response protects the cells effectively from an otherwise lethal exposure to H 2 O 2 , an important question is for how long this response remains effective. To address this we pre-treated E. coli again with 0.1 mM H 2 O 2 for 30 minutes, but we applied the higher dose (1 mM, trigger of the priming response) at different time-points (30, 60, 90, 120 and 150 minutes after the 30 min priming period, also keeping cell density constant). We observed a significant decay of the priming response from 120 minutes after H 2 O 2 pretreatment removal (Fig 2), approximately four divisions, suggesting that the priming effect is also trans-generational. After 150 minutes, the survival rate of primed populations no longer differed from naïve populations. Another study has shown long-term memory based on an epigenetic switch that controls a bimodal virulence alternation also in the scale of hours in E. coli [19].
How is the state of priming maintained for up to five generations? Bacteria can store information about recent stress via stable transcripts or proteins [19,20].
We investigated the memory at the protein level and initially studied the impact of priming and trigger concentrations (0.1 and 1 mM) on the proteome of E. coli by quantitative LC-mass spectrometry. We detected many of the known enzymes that are induced by H 2 O 2 just 5 minutes after the addition of H 2 O 2 (Fig 3, and S3 Table and S4 Table). For both concentrations used, we detected and quantified many proteins belonging to OxyR regulon.
Furthermore, many other genes, such as ahpC/F, xthA and suf operons showed a weak difference or no response at all. The samples in this experiment were taken after only 5 minutes of treatment. The rationality behind this design is that the availability of viable cells after 1 mM H 2 O 2 treatment would be too low. To explore the temporal dynamics of the proteome after a 30-minute stimulus with H 2 O 2 (0.1 mM), we followed the changes over almost 3 hours (five time points: 30, 60, 90, 120 and 150 minutes). The decline in anti-H 2 O 2 protein levels correlates well with the decrease of the response (Fig 4). Proteins such as KatG, AhpF or RecA declined slowly after removal of the H 2 O 2 , consistent with a sustained production with a minor contribution of dilution due to cell division and slow degradation rate. These results indicate that many of these proteins are stable and show a significantly higher abundance than in the control even at 150 minutes after H 2 O 2 treatment. Other proteins such as GrxA, YaaA and XthA, SufA, SufS, AcrA-AcrB completely declined at this point indicating that these proteins may be subject to proteolysis and have shorter half-lives (Fig 4) but also that the stressful situation is alleviated. The overall results of this proteomic experiment shows that the memory of the priming response in E. coli is mediated by the scavenging proteins such as KatG and AhpCF. The primary amino acid sequence is informative about the in vivo halflife a protein. A prediction of the half-life for some of the proteins, giving as that are responsible for the memory additionally supports our findings by the proteomic approach (S5 Table).
We visualised the global impact of H 2 O 2 on bacterial physiology using a network analysis based on protein-protein and function [21]. The proteome response to the priming concentration (0.1 mM H 2 O 2 during 30 minutes), resulted in a high degree of connectivity of protein-protein interactions and functional relation of both, up-and down-regulated proteins (Fig 5). If we compare our network with a large-scale protein-protein interaction network of E. coli [22], we find a wide perturbation including the most important nodes. This analysis also points to proteome-wide readjustment to cope with H 2 O 2 stress and shows the profound impact of oxidative stress across the entire proteome in a dose that does not change the growth rate in a rich medium. One possibility is that although such H 2 O 2 concentration (0.1 mM) can damage iron clusters, bacteria in early exponential phase obtain most of the energy by fermentation as shown for E.
coli previously [17]. This means that at low cell density in rich medium, the consumption of the resources does not drastically change the medium properties and it is better for bacteria to use a costly strategy by providing fast energy via fermentation that supports faster growth compared to aerobic respiration.
The memory could also be based on long-life transcripts. Therefore, we sequenced the full transcriptome after exposure to H 2 O 2 . RNAseq captured both sRNA and mRNA and we sampled just before the decline of the response (120 minutes after removal the stimulus). The transcript with the greatest induction was OxyS, a small regulatory RNA (sRNA) induced by active OxyR (S6 Table   and S7 Table). OxyS regulates several genes, and although several targets have been identified, its function is not fully understood [19]. We did not detect other transcripts under OxyR regulation.
OxyS is a potential candidate to explain the duration of priming because it is relatively stable with a half-life between 18 to 20 minutes during the exponential phase [23][24][25]. To explore this possibility, we used an oxySdeficient mutant to test its susceptibility to H 2 O 2 over 30 minutes. We did not find any significant differences in susceptibility (S2 Fig) consistent with prior reports [24,26,27]. Although OxyS did not provide us with a mechanism to explain the duration of the memory, its stability could play some role in 232 233 234 235 alleviating DNA damage as recently suggested [28]. However, such protection does not seem to have a significant impact on cell survival.
Based on the comparison of proteomic and transcriptomic data we speculate that the capacity of the cells 'to remember' the stimulus is mainly based on the stability of scavenging proteins as documented in the proteomic dataset. These scavenging proteins remain present at higher concentrations than in the control samples as long as 120 minutes after removal of peroxide. This indicates that these enzymes are not degraded but probably that their production continues after the stimulus with a minor impact of proteolysis and dilution during cell divisions which could explain the pattern that we observe in the decline of the response. It is important to consider that around 120 minutes, there were still significant levels of KatG and AhpF (compared to t=0) without providing additional protection. Thus, the memory requires the contributions of many other genes whose relative expression decreased to low levels after two hours. It is also possible that specific enzymatic activity of both proteins gets lost, since the proteomic approach is based on protein identification by sequence, not by activity.

Priming response is compromised by disrupting important H 2 O 2scavenging genes
The expression of important genes for H 2 O 2 stress survival was also confirmed by qPCR 30 minutes after the addition of the chemical (0.1 mM). We found a significant up-regulation of selected genes such as katG, ahpF/ahpC, dps, mntH and sufA (S3 Fig and S8 Table). As previously described in the literature, oxyR and fur do not change in expression level when cells are treated with H 2 O 2, since their activation relies on switching between active or inactive forms of proteins that depend on the intracellular level of H 2 O 2 or iron respectively [6,7].
Many of these transcripts showed a strong induction at the RNA level (S8 Table). After 30 minutes, the level of induction of all genes showed a stronger response at RNA level than at proteins level. In a normal situation, we would expect that a single molecule of RNA is translated into several proteins. This possibly indicates that under oxidative stress, the translation is inefficient and it is compensated with a high level of transcription probably due to damage of many cell components, including ribosomes as previously described [29].
To understand the priming response to H 2 O 2 at the molecular level, we constructed a set of mutants for the coding gene of key proteins pointed by our proteomic dataset. In proliferating E. coli, OxyR is the major regulator controlling cellular response to H 2 O 2 [30]. We explored the involvement of OxyR in the priming response since many of the differentially expressed proteins were transcribed in an OxyR-dependent fashion.
We found that by disrupting OxyR, there was a dramatic change in sensitivity with full loss of viability after 30 minutes (S2 Fig). The priming response is completely abolished (Fig 6), indicating that the enhanced survival, due to preexposure to H 2 O 2 , depends on the regulator OxyR. This regulator is a major transcription factor that protects E. coli against H 2 O 2 during the exponential phase [30,31]. The active form of OxyR positively regulates dozens of genes.
OxyR mutants accumulate ROS at much higher levels than the wild-type strain during growth even in the absence of H 2 O 2 which also accounts for its high sensitivity [30].
Next, from the most highly expressed proteins and informed by published work [6,30], we selected a set of genes which mutants were used to determine their contribution to survival ( Fig S2) and priming ( additional priming response experiment mainly those genes with increased susceptibility (Fig 6 and S2 Fig). The removal of KatG indicates that catalase importantly contributes to the priming response, but it does not fully explain the protection observed for the WT strain (compare to Fig 1B). In the case of AhpF, we also observe differences in priming response with the naïve state but not as pronounced as in the case of KatG-deficient strain. The double mutant defective in KatG and AhpF showed a dramatic decrease in the priming response but still significantly different from naïve cells. Recently, a report showed that bacteria lacking the AhpF/C system suffer a severe post-stress recovery. The absence of the AhpF/C system also contributes to the lethality of the mutants probably by the inability of the cell to cope with a low level of ROS after severe oxidative stress [34]. Another protein that illustrates an important influence on priming response is RecA, with the mutant showing a decreased priming response to H 2 O 2 . The disruption of OxyS gene did not suppress the priming response or decreased the survival rate compared to the wild-type control. Overall, our data indicate that the priming by a low dose of H 2 O 2 is multifactorial, with several OxyR-controlled proteins such as KatG, AhpCF or other factors related to DNA as RecA contributing to the priming response.

Priming response enhances the survival of evolving populations
To find out whether the priming response described above has an influence on the rate of evolution of resistance to H 2 O 2 , we used an experimental evolution approach. We evolved bacterial cultures with a treatment protocol described in detail in M&M section. Parallel populations were evolved where one group was periodically exposed to a sub-lethal concentration of H 2 O 2 , an activating signal that should protect the populations in comparison with naïve ones when later exposed to a higher dose. We continued daily intermittent exposures to H 2 O 2 doubling both, priming and triggering concentrations until extinction occurred.
We consider that a population is extinct when it shows no sign of growth during the next passage. This was also confirmed by periodical contamination checks after each passage. We observed that the extinction rate was faster for the naïve populations in comparison to primed populations (Log-rank test, p< 0.05, These results show that priming increases the evolvability of pre-treated populations.
Further analysis of the resistant populations by whole-genome sequencing revealed that all populations harboured different sets of mutations.
Surprisingly, we did not find any modification in the enzymatic scavenger systems, such as catalase or any other proteins related to peroxide protection.
There are hundreds of single nucleotide polymorphisms (SNPs) and other types of mutations that were unique to each population for both regimes (Fig 8).
These mutations probably represent many of the neutral or non-lethal changes that populations accumulated during the exposure to H 2 O 2 . Here, H 2 O 2 is not only a selective agent, but it also speeds up evolution by increasing mutagenesis. At the moment, we cannot be certain about the contribution of all mutations to H 2 O 2 resistance and they will be subject to detailed studies in the future. However, we studied two cases of the most frequent mutations in more details as proof of principle.
One first case is that of very frequent inactivating mutations in fimE. The (pCA24N-fimE, GFP minus) [36] reverted the attaching ability (Fig 9) of the mutant but also restored the original resistance to 1 mM H 2 O 2 . Fimbriae expression per se constitutes a signal transduction mechanism that affects several unrelated genes via the thiol-disulfide status of OxyR [37]. Fimbriae formation is accompanied by massive disulfide bridge formation [37] that could also contribute to titration of the exogenous H 2 O 2 limiting the intracellular damage.
Following the inversion of the phase switch to 'on state' of fimbriae production by environmental signals, this element can remain phase-locked in the 'on orientation' due to integration of insertion sequence elements at various positions the fimE gene [38]. Interestingly, fim operon expression allows E. coli to attach to abiotic surfaces, host tissues and to survive better inside macrophages protecting against the presence of extracellular antibacterials [39,40]. Reactive oxygen species (ROS) are critical components of the antimicrobial repertoire of macrophages to kill bacteria [41].
A second case that we studied is the intergenic mutations between insB1 and flhD (insB1→flhD), which were particularly highly frequent in our evolution experiment. FlhD forms an operon with FlhC and both form the master transcriptional factor that regulates transcription of several flagellar and nonflagellar operons by binding to their promoter region [42]. It is known that in E. coli MG1655, some insertions sequences such as insB1 can increase the motility of E. coli [43]. Our hypothesis here was that mutations in the intergenic region between insB1 and flhD contribute to abolishing the positive effect of insB1 insertions on motility, which in turn increases resistance to hydrogen peroxide. A second possibility is that the decrease of motility itself decreases the basal level of hydrogen peroxide due to lower metabolic demand. Flagellar motility enables bacteria to escape from detrimental conditions and to reach more favourable environments [43]. However, flagella impose an important energetic burden for bacterial metabolism by the many proteins involved in the machinery and the energy spending. For instance, one interesting study showed improved tolerance to oxidative stress in Pseudomonas putida as reflected by an increased NADPH/NADP(+) ratio, concluding that flagellar motility represents the archetypal tradeoff involved in acquiring environmental advantages at the cost of a considerable metabolic burden [44]. In our condition of oxidative stress, flagellate phenotype makes the cells more susceptible to hydrogen peroxide. These results raise an interesting question in regards to the motile vs non-motile strategy in bacteria: does flagellar activity bring diminishing returns by creating susceptibility to oxidative stress? This could have an impact on bacterial lifestyle and evolution of motility.
We can speculate about the role of some other mutations. For example, a set of changes are located in genes coding for iron-binding proteins or related with iron transport such as iceT, feoA, yaaX/yaaA or rsxC. The control of intracellular iron is crucial to decrease the adverse effects of Fenton chemistry [7]. There were also mutations that were common to both types of population, evolved under priming and non-priming conditions (Fig 8). The most frequent mutations were yodB (a cytochrome b561 homologue), intergenic mutations between insA and uspC (universal stress protein C). Another frequent mutation was in the gene yagH, belonging to the CP4-6 prophage. CP4-6 is a cryptic prophage in E. coli that could play a role in bacterial survival under adverse environmental conditions [45].

Priming alters the mutational spectrum and H 2 O 2 -induced mutagenesis in evolving populations
To assess if the evolved populations have a similar mutational spectrum, we analysed the total pool of mutations segregated by the treatments. We used the Monte Carlo hypergeometric test implemented by iMARS [46] to assess the overall differences between each mutational spectrum. Both groups, evolved under priming and non-priming conditions, differed from each other significantly (p=0.00021). ROS induces a particular type of mutations, with a characteristic signature in the DNA. The guanine base in genomic DNA is highly susceptible to oxidative stress due to its low oxidation potential. Therefore, G·C→T·A and G·C→C·G transversion mutations frequently occur under oxidative conditions [47,48]. Thus, we investigated the proportion of C→A and C→G substitutions between the two types of evolving regimes, but we did not find significant differences (p=0.056 and p=0.11 respectively, two-tailed Fisher's exact-test, Fig 11).  (Fig 1).
We also showed previously that different inoculum sizes in similar conditions to our current experiment do not influence mutagenesis [49]. We propose that one of the most important consequences of the priming response to H 2 O 2 is a drastic decrease in lethal mutagenesis. Although H 2 O 2 damages most of the cellular components [7], DNA damage is likely the major contributor to lethality.
In principle, an increase in mutation rate increases the evolvability of asexual populations [50][51][52]. How is it possible that naïve populations show lower evolvability compared to primed populations despite a higher mutation rate? A possible explanation is that evolvability can be influenced by the population size and the mutation supply. Even with an increased mutagenesis, if population size drastically decreases, the final number of mutants can be smaller when survival is improved. We also found that the mutational spectra of our evolving populations are different. Mutational spectra are a qualitative property of mutation rate that could enhance or hinder the access to beneficial mutations [53][54][55]. It is possible that the observed changes in mutational spectra between primed and naïve evolving populations could play a role in   Table) were generated in E. coli K-12 strain MG1655 following a modified methodology described elsewhere [56]. Briefly, transformants carrying the red recombinase helper plasmid, pKD46, were grown in 5-ml SOB medium with ampicillin (100 µg/ml) and L-arabinose at 30°C to an OD 600 of 0.5 and then made electrocompetent. PCR products with homology regions were generated using specific primers (S10 Table) to amplify the region of interest from using the corresponding mutants of the Keio collection [57]. The PRC-generated fragments were purified (MinElute PCR Purification Kit, Qiagen).
Competent cells in 50 µl aliquots were electroporated with 100 ng of PCR product. Cells were added immediately to 0.9 ml of SOC, incubated 1 h at 37°C, and then 100 µl aliquots spread onto LB agar with kanamycin (30 µg/ml).
The mutants were verified by PCR and the antibiotic resistance cassette was removed using the plasmid pCP20. The correct inactivation of genes was verified by PCR. To construct double mutants, single mutants obtained in MG1655 were transduced using the P1vir phage procedure as previously described [58]. used for in-solution protein digestion as described previously [61]. Briefly, proteins, demobilised in denaturation buffer, were reduced by the addition of 1 µl of 10 mM DTT dissolved in 50 mM ammonium bicarbonate (ABC) and incubated for 30 minutes, followed by 20-minute alkylation reaction with 1 µl of 55 mM iodoacetamide. As first digestion step, Lysyl endopeptidase (LysC, Wako, Japan) resuspended in 50 mM ABC was added to each tube in a ration of 1 µg per 50 µg of total proteins and incubated for 3 hours. After pre-digestion with LysC, protein samples were diluted four times with 50 mM ABC and subjected to overnight trypsin digestion using 1 µg/rection of sequencing grade modified trypsin (Promega, USA), also diluted before use in 50 mM ABC. All in-solution protein digestion steps were performed at room temperature. After the addition of iodoacetamide, the samples were protected from the light until the digestion was stopped by acidification adding 5% acetonitrile and 0.3% trifluoroacetic acid (final concentrations). The samples were micro-purified and concentrated using the Stage-tip protocol described elsewhere [61], and the eluates were vacuum dried. Re-dissolved samples were loaded on a ReprosilPur C18 reverse phase column and peptides were analysed using a nano-HPLC Dionex Ultimate 3000 system (Thermo Scientific, Germany) coupled to an Orbitrap Velos mass spectrometer (Thermo Scientific, Germany). MS and MS/MS data from each LC/MS run were analysed with MaxQuant software [62].

Identification of proteins was performed using the MaxQuant implemented
Andromeda peptide search engine and statistic analysis was carried out using the software Perseus [63].

Prediction of in vivo protein half-life and stability index. The half-life
estimation is a prediction of the time that takes for half of the amount of protein in a cell to disappear after its synthesis. It relies on the "N-end rule" (for a review see [64][65][66][67]). The instability index provides an estimate of the stability of a protein in a test tube. Based on experimental data [68], making possible to compute an instability index using the primary amino acid sequence. For these predictions, we used the online software ProtParam [69]. When available, Nend sequence was corrected to the real in vivo sequence due to methionine excision [70]. For primer design, Escherichia coli strain K-12 MG1655 complete genome (accession U00096) sequence was downloaded from NCBI database (www.ncbi.nlm.nih.gov) and used as reference. Target sequence accession number, and primer sequences for each assay can be found in S11 Table. Primers were designed using Primer Express software (Applied Biosystems, Germany) and optimised for annealing temperature of 60°C. Each primer pair and amplicon were checked for secondary structure formation using Oligo Tool Research. Sequence data are available from the NCBI SRA under BioProject accession PRJNA485867. The haploid variant calling pipeline snippy [74] was used to identify mutations in the selection lines. Snippy uses bwa [75] to align reads to the reference genome and identifies variants in the resulting alignments using FreeBayes [76]. All variants were independently verified using a second computational pipeline, breseq [77].

Determination of H 2 O 2 -induced mutagenesis.
This procedure was carried out following similar protocols to previous studies with some modifications [78,79]. Five independent cultures (5 ml each one) of E. coli MG1655 were grown in fresh LB medium to an OD 600 of ~0.2. Then, each culture was diluted killing curves a Welch's test was used. Growth curve analysis was carried out using the Growthcurver package for R. For survival analysis we used a log-rank test. Some specific analyses are mentioned elsewhere in the manuscript. All tests were performed with software R [81] except mutational spectrum analysis that was implemented using the ad hoc software iMars [46].  Asterisks indicate significant differences between each time-point pair (Welch's test, one asterisk for p<0.05 and two asterisks for p<0.01).       Additionally, the strain transformed with the plasmid pCA24N-fimE also recovers the original resistance to H 2 O 2 , that turns from 4 to 1 mM. In both panels control cells (transformed with the cloning vector pCA24N) can be also observed.    when compared to non-treated ones (control). The growth curve parameters were estimated with the Growthcurver R package [60]. Only carrying capacity and the areas under the curve have shown significant differences with a small fond-change.