Biofilm-driven LPS remodeling and colanic acid overproduction mediate biocide adaptation and antibiotic cross-resistance in E. coli

Abstract

The identification of the determinants driving antimicrobial resistance is a prerequisite for improving the control of resistance emergence and dissemination. Disinfectant biocides, daily used in food-processing industries, have already been associated with the cross-selection of antibiotic-resistant bacterial populations. However, very few studies have addressed this issue using a biofilm model, the predominant bacterial lifestyle in food-processing environments. In this work, we examined the adaptation of Escherichia coli biofilms to four biocidal active substances over one month, and assessed their subsequent effects on antibiotic resistance. Exposure to N-(3-aminopropyl)-N-dodecylpropane-1,3-diamine (TMN) and benzalkonium chloride significantly increased the emergence of rifampicin-resistant (RifR) variants in biofilms. Genomic analyses revealed that the RifR variants selected upon TMN exposure recurrently harboured mutations in genes related to lipopolysaccharide (LPS) biosynthesis that conferred low-level rifampicin resistance in biofilm. These variants displayed altered LPS profiles, a more negative surface charge, and reduced membrane permeability. Proteomic and phenotypic analyses supported a metabolic reorientation of envelope sugar precursors, with decreased modulation of LPS synthesis and a marked induction of the colanic acid biosynthetic pathway in TMN-selected variants. This shift resulted in increased matrix production and reinforced biofilm-associated tolerance. Together, these data identify outer membrane reprogramming, linking LPS modulation with colanic acid overproduction, as a previously unknown mechanism of TMN adaptation that simultaneously promotes antibiotic cross-resistance in E. coli biofilms.

Introduction

Antimicrobial resistance (AMR) is one of the foremost challenges for public health in the coming decades. Recent years have seen a marked increase in antibiotic resistance among bacteria, including the emergence of multidrug-resistant strains that complicate therapeutic strategies¹. Although natural resistance mechanisms have existed within bacterial communities for centuries, antibiotics being often derivatives from natural substances produced by microorganisms, the selection of resistant clones has been accelerated not only by the use or misuse of antibiotics but also by other factors directly or indirectly related to human activities. The One Health concept further underscores the need for a multisectoral approach to tackle AMR, emphasising the interconnection of human, animal, and environmental health as a continuum in the propagation of resistance determinants². Increasingly, attention is being paid to co-selectants, i.e. non-antibiotic agents, especially in light of recent studies³. Among these co-selectants, chemical disinfectants—a subclass of biocides—has been receiving more and more attention, especially since the COVID crisis⁴. Disinfectants are used massively in a wide variety of sectors, including the food industry, where they safeguard consumers from pathogenic bacteria. Disinfectant biocides differ from antibiotics by their use on inanimate surfaces, but can share similar cellular targets and related modes of action. A large number of publications have hence shown that bacteria often use the same mechanisms to adapt to both biocides and antibiotics^5,6. However, the amount of knowledge available on the different biocide families differs greatly. Some of them have already been extensively studied⁵, such as quaternary ammonium compounds (QACs) or chlorhexidine, and have been recurrently found to affect antibiotic resistance. Nevertheless, the effects of the vast majority of biocidal active substances have barely been characterised, despite their frequent use. N-(3-aminopropyl)-N-dodecylpropane-1,3-diamine (TMN) for example, is a biocidal active substance increasingly used in food-processing industries, particularly in an admixture with QACs or as an alternative⁷. Therefore, we need further insights into TMN modes of action and possible effects on antimicrobial (cross-)resistance.

Moreover, while current data predominantly focus on planktonic bacteria, most of the bacterial populations in food-processing environments exist within biofilms. Biofilms are three-dimensional structures that bacteria form on surfaces through the production of an extracellular matrix composed mostly of polysaccharides, lipids, proteins and extracellular DNA⁸. The 3D structuration of biofilms generates a variety of microenvironments, resulting in broad heterogeneity of metabolic states for embedded bacterial populations. The diffusion of molecules in the structure may be reduced, leading to oxygen and nutrient gradients. While bacteria around the periphery of the biofilms have access to every nutrient they need, bacteria in the centre do not. This heterogeneity has a strong effect on bacterial metabolism, with internal subpopulations often exhibiting slow growth and persister phenotypes. Such a collective and organised lifestyle generally enhances bacterial tolerance to environmental stresses, including disinfectant biocides, through both a physical barrier effect and by inducing physiological heterogeneity and resistant phenotypes^5,9. Recently, we showed that modulation of the biofilm structure following exposure to a biguanide substance could promote the emergence of adaptive (physiological) cross-resistance to aminoglycoside¹⁰. Biofilms are therefore dynamic edifices that evolve in response to internal and external signals, potentially influencing evolution towards antibiotic resistance¹¹. These biological structures may not only enhance bacterial survival but could also facilitate the emergence of stable genetic resistance through mutational events, as already reported^5,12,13. Additionally, the extracellular matrix has been shown to limit the penetration of biocides into the structure, as already reported for quaternary ammonium compounds¹⁴. Such an effect increases the possibility for bacteria to be exposed to sub-lethal doses, leading to variability in microenvironmental conditions between cells, which is a feature that cannot be observed in planktonic populations, where biocide and bacteria are more homogeneously distributed and exposed. These factors collectively shape biofilms into unique ecosystems that could play a specific role in bacterial adaptation strategies and cross-resistance phenomenon.

Given the widespread use of biocides in food-processing environments and the limited understanding of how bacterial lifestyle influences adaptation, this work aimed to investigate the evolutionary trajectories of E. coli exposed to commonly used food-chain biocidal substances under both planktonic and biofilm conditions. Using adaptive laboratory evolution, we specifically examined the emergence of rifampicin-resistant (RifR) variants as a proxy for adaptive diversification. Rifampicin is commonly used in evolutionary experiments on E. coli to quantify variants emergence¹⁵, because mutations are usually not impacting the fitness of the variants in this species¹⁶. Our results showed that RifR variants were positively associated with the biofilm lifestyle, supporting the idea that biofilm-associated environments promote distinct adaptive pathways. Among these variants, TMN exposure was specifically associated with mutations in lipopolysaccharide biosynthesis pathways. Further characterisation revealed a molecular redirection toward colanic acid production, which correlated with enhanced extracellular matrix production and reduced TMN sensitivity in biofilms. Together, these findings highlight the importance of considering bacterial lifestyle when evaluating adaptation to biocides and demonstrate that biofilm conditions can directly shape evolutionary outcomes, with potential consequences for cross-resistance emergence in food-processing environments.

Results

TMN and benzalkonium chloride (BAC) promote biofilm-associated emergence of RifR variants

The biofilms of eight E. coli strains were repeatedly exposed over one month to four biocidal active substances including BAC, TMN, sodium hypochlorite (SHY) and polyhexamethylene biguanide (PHMB) belonging to different families, at minimal concentrations previously determined to inhibit the growth of planktonic bacteria (MIC) (Fig. 1). Depending on the biocides, different effects on RifR variant emergence were observed (Figs. 2 and S1). Interestingly, TMN exposure consistently stimulated the emergence of RifR variants in biofilms with an increase in RifR variants quantity in seven out of the eight E. coli strains compared with the H₂O control (Fig. 2A). BAC exposure was also correlated with a substantial increase of RifR variants in biofilms formed by five out of the eight strains. Conversely, exposure to PHMB and SHY did not significantly promote the emergence of resistance and even sometimes resulted in a decrease in the quantity of RifR variants isolated compared with the H₂O control (Fig. S1). It should be noted that three out of the eight strains (Ec1655, Ec223 and Ec956) were completely eradicated by the MIC of SHY after seven days of exposure, demonstrating that the biofilm’s lifestyle did not result in higher cell survival with this biocide compared with the planktonic state. This also explains the small number of RifR variants isolated for SHY. The dynamics of RifR variant emergence varied greatly depending on the E. coli strain over the one-month exposure to biocides. In the Ec1655 strain exposed to TMN, RifR variants were selected quickly, reaching the maximal concentration of countable resistant clones after one week of exposure. Differently, Ec956 initially produced many RifR variants from t0, but the effect of the biocides was reflected as a smaller decrease in variants over the exposure period compared with the H₂O control group. The effect of TMN was also more pronounced on some strains, such as Ec1655, Ec723 or Ec956, which displayed important differences in the number of RifR variants isolated compared with the H₂O control group throughout the experiment. In contrast, other strains like Ec775, Ec478 or Ec223 only showed differences at certain time points. Interestingly, the general effect of TMN and BAC on RifR variant emergence highlighted in biofilms was not observable when analysing supernatant populations, suggesting a specific adaptive response of biofilms to biocides (Fig. 2B). In three out of the eight strains (Ec223, Ec009, Ec478), neither BAC nor TMN exposure had any significant effect on RifR variant counts in supernatants compared with the H₂O control. In Ec1655, at some timepoints, variants abundance was even reduced compared to the control group. Only Ec956 showed a lasting increase in RifR variants in supernatants over the first two weeks of exposure to TMN compared with H₂O.

Fig. 1. Description of the four weeks experiment cycle to monitor strain evolution.

Biofilms were pre-formed in microplates, during 72 h at room temperature without biocide (in green). Then the experiment started and was composed of four one-week cycles. On day 1 (D1) of the cycle, a part of the supernatants was sampled (5 µL) and plated on a rifampicin-supplemented plate and the rest was discarded. 40 µL of fresh medium was added, and one quarter of the bottom of the well was scratched to resuspend a section of the biofilm, from which 5 µL were also plated on a rifampicin-supplemented plate. Then, wells were filled back to 200 µL with MIC concentrations of biocide. From D2 to D7, supernatants were carefully removed to not detach the biofilm and replaced with 200 µL of fresh biocide-full medium. 2 biological replicates were performed with 8 independent lineages (separated wells in the microplate) per condition.

Fig. 2. Effects of TMN and BAC on the quantity of RifR variants in E. coli biofilms.

Evolution curves representing the number of CFUs/mL of rifampicin-resistant variants enumerated on TSA plates supplemented with 20 mg/L of rifampicin and collected from biofilms (A) or supernatants (B) exposed to H₂O (dashed lines), BAC (in green) or TMN (in blue). Error bars represent the standard error of the mean. Statistical analysis was performed using a Mann–Whitney test for the biocide condition and the H₂O condition, separately, for each week. P-values are represented as follows: * = [0.05–0.01]; ** =]0.01–0.001]; *** = <0.001.

TMN selects Rifᴿ variants harbouring recurrent mutations in LPS-biosynthesis genes

Genomes of RifR variants isolated in biofilms were sequenced and compared with genomes of their corresponding parental strains to identify mutations underlying the adaptation and resistance phenotypes. Detailed information about the genetic modifications identified are presented in Table S1. Figure 3 displays the genetic targets carrying point mutations in RifR variants isolated from H₂O, TMN or BAC conditions, classified by their level of resistance to rifampicin (in mg/L) and their corresponding Gene Ontology (GO) biological process categories. The most prevalently mutated gene was rpoB, encoding the DNA-dependent RNA polymerase β-subunit, known as the primary target of rifampicin¹⁷. However, the proportions of mutations in rpoB varied with each biocide. Among RifR variants selected after exposure to TMN, only 56% exhibited a mutation in rpoB, while this percentage reached 67% with H₂O. Although BAC has already been associated with rpoB mutations¹⁸, it did not show a correlation in this study, with 65% of BAC-selected variants having rpoB mutations. Other pleiotropic regulators, like hfq, dksA or rpoS were also recurrently mutated. The general stress response was impacted, with mutations in rpoS¹⁹ and in hfq, a chaperone protein known to have a post-transcriptional effect on rpoS²⁰. These mutations were not, however, clearly associated with exposure to a specific biocide, and some were also co-occurring with rpoB mutations. Three variants, exclusively from the BAC condition, were mutated in genes related to the stringent response (dksA²¹; gppA²²), a stress response activated under nutrient depletion, which is consistent with the conditions within biofilms. Other mutated genes were linked to a general stress in the biofilm, as barA, hslU and slyD, involved in diverse responses. Mutations were also recurrently observed in the rbs operon, involved in D-ribose transport and catabolism, and more specifically in rbsR, the repressor of the ribose operon²³. This genetic target seemed particularly associated with the BAC condition, as four of the five rbsR variants emerged under BAC exposure and only one in the control condition (H₂O). The link with rifampicin was however not clear, as four of the five rbsR variants also carried a mutation in rpoB.

Fig. 3. TMN selects for LPS-mediated variants with low-level rifampicin resistance.

Table representing the median level of rifampicin resistance in mg/L for every variant with its associated mutated genetic targets. Columns represent the most represented GO biological process in which the gene is implicated. The grey-coloured genes are mutated in strains isolated in the H₂O condition, the green ones are mutated in strains isolated in the BAC condition and the blue ones are mutated in strains isolated in the TMN condition.

Interestingly, the second-most prevalent gene category was related to LPS biosynthesis, with six mutated variants. Moreover, these variants were associated with the TMN condition, as four out of the six variants originated from this condition. Mutations were mostly related with the production of the saccharidic part of the LPS. Variants mutated in the waaU gene (Ec723_TMN_2; Ec723_H2O_1; Ec956_BAC_1), encoding a putative heptosyltransferase²⁴, were the only ones that were also isolated in H₂O and BAC conditions. Mutations in the rfbB gene, encoding a dTDP-glucose 4,6-dehydratase 1 involved in rhamnose biosynthesis²⁵ were observed in two strains isolated after exposure to TMN (Ec223_TMN_5 and Ec775_TMN_5). Finally, wbbK is identified as a putative glycosyltransferase²⁶. Interestingly, two other genes, that were un-annotated in the Gene Ontology database, yfdI and htrL, were isolated under TMN exposure as well, and could also be closely related to LPS biosynthesis. The yfdI gene (Ec1655_TMN_2), encodes a glucosyltransferase, which mediates O-Antigen modulation when introduced in Shigella flexnerii²⁷, and htrL (Ec478_TMN_3) encodes a putative glycosyltransferase²⁸, referred on Uniprot as involved in LPS synthesis. Variants with mutations in LPS biosynthesis genes mostly carried single mutations. Only one of them (Ec723_TMN_2) had an additional mutation, in barA. Most importantly, these mutations were never associated with mutation in rpoB gene, suggesting a direct effect of the LPS-biosynthesis gene mutations in RIF resistance. Finally, mutations in LPS biosynthesis genes were mainly related to low levels of rifampicin resistance (20–60 mg/L), in contrast to the stronger resistances observed in rpoB-mutated strains (40- > 200 mg/L) as shown in Fig. 3.

Large genomic modifications were also observed in three of the six LPS-biosynthesis variants selected under TMN conditions (Ec223_TMN_5, Ec775_TMN_5 and Ec723_TMN_2), often resulting from the genomic plasticity associated with mobile genetic elements (MGEs) (Table S1). In the Ec223_TMN_5 and Ec723_TMN_2 variants, these modifications corresponded to phage excisions. The analysis of Ec775_TMN_5 revealed that the MGEs in this strain were mostly related to transposon excisions. MGEs were excised in many strains: 41% of TMN variants, 29% of BAC variants and 33% of H₂O variants. Some strains, like Ec723 and Ec956, displayed high genome plasticity, with more than 10 identified MGEs. Nevertheless, no correlation was observed between MGE excision and the conditions under which the variants were isolated.

LPS-associated mutations alter cell surface charge and reduce outer-membrane permeability

The specific association between TMN and LPS biosynthesis genes in RifR variants prompted us to further characterise the six corresponding variants (yfdI and htrL were included in the LPS-involved genes). First, an SDS-PAGE analysis of the variants’ LPS was performed to verify if the LPS structure was impacted by the mutations (Fig. 4A). Most of the variants exhibited pronounced perturbations in LPS structure, with clear differences in their band profiles compared with their corresponding parental strains (red boxes). Five out of the six variants showed modifications indicated by the disappearance of bands between 10 kDa and 15 kDa. These variants also had modified profiles in the heavier bands, between 20 kDa and 35 kDa, with either reduced intensity or the disappearance of some bands. Conversely, one variant, Ec1655_TMN_2, did not show a marked difference in its LPS profile compared with its parental strain. It should be noted that the Ec1655 parental profile is already lighter than the other parental strains because the strain is known to have a transposon insertion in one LPS biosynthesis gene, leading to an O-antigen production defect, which explains the lack of bands above 15 kDa²⁹. Because of the lack of differences between Ec1655_TMN_2 and its parental strain, we decided to exclude this strain from the rest of the analyses, and to focus on variants with phenotypical differences.

Fig. 4. LPS biosynthesis variants have a lower membrane potential, which reduces global cell permeability.

SDS-PAGE representing the LPS profile of the parental strains and variants with silver nitrate staining A) (red boxes show the differences between parental strains and variants); Membrane potential analysis of the LPS biosynthesis variants and their parental strains (B); Crystal Violet uptake assay of the LPS biosynthesis variants and their parental strains to assess their permeability (C). Statistical analysis was performed with a Mann–Whitney test comparing variants with their parental strains. P-values are represented as follows: * = [0.05–0.01]; ** =[0.01–0.001]; *** = <0.001.

Given the negative charge carried by LPS molecules²⁸, we investigated whether the mutations impacted the membrane potential. A zeta potential analysis was performed on the variants and their parental cells to compare global membrane charge (Fig. 4B). Interestingly, four out of the five variants were significantly different from their parental strain, with a more negative potential than their parents. The most notable difference was observed for the Ec223_TMN_1 variant, with a reduction from −19.9 mV (SD ± 0.98) in the parental strain to −24 mV (SD ± 1.67) in the variant. Conversely, the Ec223_TMN_5 variant was the only strain that showed LPS structural modifications on the SDS-PAGE without also showing a modification in the membrane charge.

The negative charge of the LPS is known to help maintain cell structure through interactions with divalent cations of the LPS molecules²⁹. Hence, a Crystal Violet permeability assay was performed with the five LPS biosynthesis variants (Fig. 4C) to evaluate the impact of LPS modifications on cellular membrane permeability. This assay measures the absorbance of Crystal Violet in supernatants, which means that a higher OD represents a lower internalisation and permeability of the cells. Substantial differences were observed in every strain compared with their parental strains, showing reduced cell membrane permeability. Particularly strong differences were observed for both variants of the Ec223 strain and the Ec478_TMN_3 variant.

TMN-selected LPS variants exhibit enhanced biofilm tolerance to the biocide

As the LPS variants were associated with the TMN condition, we decided to evaluate if their sensitivity to this biocide was also modified. We analysed the killing kinetics of TMN on the variants, in planktonic (Fig. S2) and biofilm (Fig. 5) conditions. Interestingly, the variants had no significant differences in mortality with their parental strains in planktonic condition. However, inactivation kinetics were different in biofilms. Except Ec775_TMN_5, all the variants showed significatively reduced mortality at at least one timepoint. The biggest differences were observed for the two variants of the Ec223 strain, which showed reduced mortality in biofilms during the whole kinetic. For other variants, particularly Ec478_TMN_3, differences seemed to occur mostly at the beginning of the kinetic, even though some tendencies were also observed at the end of the experiment for Ec723_TMN_2. Interestingly, most variants already showed a higher number of cells before TMN exposure, which could confer them an advantage upon the beginning of the biocide stress, particularly, as observed, at earlier timepoints.

Fig. 5. LPS biosynthesis variants have a reduced sensitivity to TMN in biofilms.

Killing kinetics of TMN (400 mg/L corresponding to 60x the MIC) against the parental strains and their variants in the biofilm condition. Error bars represent the standard error of the mean. Statistical analysis was performed with a Mann–Whitney test comparing variants with their parental strains. P-values are represented as follows: * = [0.05–0.01]; ** =]0.01–0.001]; *** = <0.001.

TMN triggers envelope reprogramming toward colanic acid-rich matrix production

The production of matrix components, in particular exopolysaccharides and amyloid fibres, was quantitatively evaluated in parental strains and variants through the Congo red assay (Fig. 6A). Coherently with the biofilm killing kinetics we observed an increase in polysaccharide production by the variants in comparison to their parental strains.

Fig. 6. LPS biosynthesis variants have an increased production of biofilm matrix, related to a redirection of their metabolism from LPS production to colanic acid production.

Congo red fluorescence analysis in parental strains and variants biofilms (A). GO categories that were significatively enriched among proteins with a log2FC ± 1 in response to TMN (compared to H₂O condition) in parental strain and variant biofilms (B). Heatmaps showing every significatively modified (<0.05) abundances of proteins involved in LPS biosynthesis exclusively (C), in colanic acid exclusively (D) and in both pathways (E), in parental strain and variant biofilms in response to TMN. Statistical analysis of Fig. 6A was performed with a Mann–Whitney test comparing variants with their parental strains. P-values are represented as follows: * = [0.05–0.01]; ** =]0.01–0.001]; *** = <0.001.

In order to precise the link between biofilm and exopolysaccharide production, bacterial proteomic profiles were compared between a biocide-free condition and a TMN-supplemented condition, in parental strains and variants biofilms. An enrichment analysis of every protein with an abundance modification of log2FC ± 1, based on GO biological processes, was performed to characterise the responses to TMN in parental strains and in variants (Fig. 6B). Heatmaps representing these abundances modifications are also shown in Table S2 for the parental strains and Table S3 for the variants. Some pathways were activated in both parental strains and variants, while others were specific to either parents or variants. First, the “Acetyl-CoA metabolic process”, the “Cellular response to sulfur starvation” and the “L-threonine catabolic process to propionate” biological processes were significantly modified in presence of TMN in both the parental strains and the variants. Then, TMN exposure led to a significant enrichment of the “Alkanesulfonate transmembrane transport” (FDR = 0.041) and “Phenylacetate catabolic process” (FDR = 0.029) pathways in the parental strains. Differently, three pathways were significantly enriched in variants, including “Citrate metabolic process” (FDR = 0.039), “L-lysine catabolic process” (FDR = 0.039) and particularly “Colanic acid biosynthetic process” (FDR = 0.0096).

The production of colanic acid, known as an E. coli polysaccharide composing capsular and biofilm matrix exopolysaccharide, is often closely linked to LPS production³⁰. Thereby, we analysed more precisely the response to TMN in parental strains and variants in both of these pathways (Fig. 6C–E). LPS biosynthesis proteins abundances were remarkably less modified in response to TMN in variants compared to their parental strains, with a global abundance modification of 32 proteins in parental strains, while only 16 modifications were observed in variants. Some proteins had strong differences between parental strains and variants. The Lpt cluster, involved in LPS transport³¹, was highly modified in parental strains, but almost never in variants. Other proteins, like Ais, putatively related to core oligosaccharide synthesis, ArnB, involved in lipid A phosphate modification³², and RffG, a paralog of RfbB involved in LPS and Enterobacterial Common Antigen production³³, were also recurrently modified in response to TMN, in parental strains, while being rarely modified or not at all in variants. In contrast with LPS production, an increase in colanic acid production was observed in the variants, with 10 protein abundance modifications in variants in response to TMN, while only 3 modifications were observed in parental strains. Particularly, the Fcl protein, participating in the fucose residue synthesis of colanic acid³⁴, was modified in response to TMN, in every variant. Other proteins, like Wza and Wzc³⁵, were also specifically modified in variants.

Further investigation was performed on other biofilm components producing proteins to see if the variants response to TMN was impacted in other pathways (Tables S2 and S3). Interestingly, the CsgBCD proteins, implicated in curli production³⁶, were downregulated in Ec223 and Ec723, but without differences between parental strains and variants. In contrast, the abundance of the YcfJ protein, involved in biofilm maturation³⁷, was strongly increased, in response to TMN, in every strain and every variant.

Discussion

Antimicrobial resistance constitutes one of the major health challenges of the coming decades. One possible risk factor contributing to antibiotic resistance is the use of biocides, which may select for antibiotic cross-resistant variants through their broad and often untargeted effects^5,6. In this study, four biocidal active substances (BAC, PHMB, TMN and SHY), representative of molecules used in food chain disinfection processes, were used to continuously expose E. coli biofilms for one month. Previous studies using this experimental framework investigated how exposure to biocides influences the emergence of antibiotic resistance, including gentamicin and ciprofloxacin. In particular, we previously observed an association between exposure to PHMB and the emergence of gentamicin-resistant variants¹⁰. Here, we monitored their effects on rifampicin resistance. This antibiotic is commonly used to study the evolutionary processes of bacterial populations^15,38, and is a useful tool for quantifying mutation rates through the rifampicin resistance assay³⁹. Rifampicin is also considered as a critically important antibiotic for human health⁴⁰. Although it is primarily used against Mycobacterium tuberculosis, rifampicin can also be active against E. coli, including in the treatment of travellers disease⁴¹.

Independently of biocidal exposure, recurrent mutations in the RifR variants were detected, in the rpoB gene, which was expected as we selected them on rifampicin-supplemented plates and this gene codes for the RNA Polymerase β subunit, the primary target of rifampicin in E. coli and other genera like Mycobacterium^17,42. This observation aligns with the central role of rpoB mutations in the emergence of rifampicin resistance among E. coli as extensively described in previous works⁴³. Mutations in rpoB conferred here the highest levels of rifampicin resistance, with MIC values ranging from 40 to 200 mg/L (Fig. 3), but these mutations can also carry fitness costs^44,45.

While PHMB and SHY presence did not induce RifR variant emergence compared to H₂O, BAC and TMN exposure consistently promoted the selection of RifR variants across several E. coli strains. Both biocides were associated with mutational events in specific pathways. BAC has been associated to mutations in rbsR and stringent response genes (dksA, gppA). As RbsR regulates D-ribose metabolism and biofilm inhibition, its loss-of-function may enhance D-ribose degradation and promote biofilm formation⁴⁶, while also affecting purine metabolism by shifting from de novo synthesis to the salvage pathway⁴⁷. In parallel, the stringent response, activated under nutrient limitation, downregulates cellular activity including purine synthesis⁴⁸, suggesting a potential functional connection between these pathways under BAC exposure. However, the link between these mutations and rifampicin resistance remains unclear, as they are frequently found in association with rpoB mutations, making it difficult to distinguish their individual effects from those attributable to rpoB mutations. Interestingly, TMN exposure was recurrently associated with mutations in LPS biosynthesis pathways in RifR variants, suggesting a pivotal role of LPS modulation in TMN adaptation and cross-resistance to rifampicin. LPS biosynthesis variants were selected in five of the eight tested strains when exposed to TMN, demonstrating the recurrence of this phenomenon despite the various genetic backgrounds of the bacterial strains. These LPS biosynthesis variants exhibited low-level rifampicin resistance (from 20 to 60 mg/L) and did not carry additional mutations in rpoB (Fig. 3). Several studies have reported that low-resistance variants emerged preferentially in biofilms (where the bacteria is exposed to reduced concentrations of antimicrobials), while high-resistance variants emerged preferentially in planktonic cultures^49,50. Biofilm-associated protection thus favours the fixation of low-cost resistance mutations⁴⁹. Here, in a majority of the variants, the effect of mutations on the LPS structure and associated phenotype was clear, with modifications in the LPS SDS-PAGE profile, the membrane charge and the permeability. Only one variant, Ec1655_TMN_2, did not reveal any changes in its structural and membrane properties. This variant is mutated in yfdI, a gene which was previously associated with O-Antigen synthesis²⁷. However, Ec1655 does not possess any O-Antigen, because of a transposon insertion in the wbbL gene, essential in O-Antigen production²⁹. The lack of O-Antigen was verified per SDS-PAGE (Fig. 4) and explains why Ec1655_TMN_2 was not phenotypically modified like the other variants. In contrast, other variants displayed characteristic LPS alterations and more negative surface charge values. Alteration of the surface charge is a known mechanism of adaptation to cationic antimicrobials, which interact with the LPS negative charge and interfere with the equilibrium built with divalent cations present in the external medium⁵¹. It was however not the case in this study, as survival in presence of TMN was not modified in planktonic conditions. However, the O-Antigen is also known to have effects on uncharged molecules, like rifampicin. O-Antigen regulates membrane fluidity through indirect effects on the anionic part⁵² and changes in the O-antigen chain length have been associated with decreased cell permeability⁵³. Murata et al.⁵⁴ demonstrated that modifications in LPS lipid A synthesis affect bacterial resistance to uncharged molecules, and that LPS was essential for the robustness of the membrane and survival against rifampicin. This observation supports the recognised role of LPS in limiting permeability to hydrophobic compounds, like rifampicin⁵⁵. Consistently, additional analyses of susceptibility of RifR variants to 15 antibiotics from various classes (Fig. S3) revealed a systematic two-fold increase of azithromycin MIC, a hydrophobic compound. However, the magnitude of the effect remains modest relative to the sensibility of the broth microdilution method used for MIC determination. Killing kinetics showed that the variants had increased survival in biofilms under TMN exposure compared to their parental strains, with some displaying slower death rates and higher basal biofilm production. Consistently, independent Congo Red analysis revealed overproduction of matrix components in most LPS biosynthesis variants, supporting enhanced biofilm matrix formation. Proteomic analysis confirmed distinct responses to TMN between parental and variant strains, with increased abundance of LPS biosynthesis proteins in parental strains and increased colanic acid biosynthesis proteins in variants. Deletion of LPS biosynthesis genes has previously been linked to colanic acid overproduction, a key component of the E. coli extracellular matrix, likely due to increased availability of sugar precursors^30,56,57. Colanic acid is also negatively charged and helps to maintain proton-motive force and membrane potential⁵⁸. This could be another explanation for the variations observed on individual cells. This negative charge could also help variants survival in biofilms, through the interaction and sequestration of the cationic biocide by the anionic exopolysaccharide. Interestingly, exposure to some positively charged antimicrobials already showed some colanic acid induction in E. coli⁵⁹.

Other differences were observed in E. coli matrix components production in response to TMN, with a reduced abundance of the CsgBCD proteins and an induction of YcfJ, but without differences between parental strains and variants (Tables S2 and S3). This could demonstrate that TMN stimulates and accelerates biofilm formation, the unexposed biofilms being still in the initial adhesion step, in which Csg proteins are involved, while the exposed biofilms are already maturing. It could also suggest a metabolic redirection, favouring the production of matrix components which helps to survive the biocide stress, colanic acid being rather involved in biofilm architecture and maturation, essential in case of antimicrobial stress^60,61.

TMN triggered distinct pathway modifications in parental strains and variants (Fig. 6B, Tables S2 and S3), including adaptive metabolic shifts linked to biofilm lifestyle, such as citrate and phenylacetate metabolism, which may both reflect nutrient adaptation and promote biofilm formation^62,63. TMN response was also strongly associated with alterations in polyamine-related pathways, including lysine catabolism and cadaverin production⁶⁴, putrescine degradation, and ornithine metabolism, which may enhance stress resistance, stabilise cell components, and counteract TMN activity through membrane reinforcement^65,66. These findings support the hypothesis that variant evolution in the presence of biocides is driven by biofilm tropism.

Conclusion

This work reports an association between exposure to biocidal active substances commonly used in the food industry and antibiotic resistance selection in E. coli biofilms. This work revealed the mechanism of bacterial adaptation to TMN for the first time, providing increased knowledge of how bacteria evolve in the presence of this biocide at both transient (proteomic) and long-term (genomic analysis) scales. Interestingly, mutations occurring in LPS biosynthesis genes selected during TMN exposure resulted in benefits for RifR variants exposed to antimicrobials. Phenotypical analysis revealed a novel mechanism of resistance involved in metabolic reorientation of the sugar precursors from LPS biosynthesis to exopolysaccharide production. These findings support the diversity of adaptive strategies among bacteria and their interplay, which together ultimately lead to the evolution and survival of bacterial populations under stress. It revealed specific factors of adaptation linked to evolution in biofilms which are of prime interest, regarding bacterial lifestyle in food-processing industries. Such insights are of prime interest for understanding how bacteria adapt to biocides and provide potential markers for cross-resistance monitoring in the food chain.

Methods

Bacterial strains

The E. coli strains used in this study are listed in Table 1. Strain Ec1655 corresponds to the reference strain MG1655, Ec223 is a strain from the INRAE CIRM-BP collection of bacterial pathogens for humans and animals originally isolated from chickens, and the other six strains were from a collection belonging to the French National Reference Laboratory for Antimicrobial Resistance (NRL-AR) hosted in ANSES Fougères, and composed of E. coli strains isolated from the pork and poultry industries. Strains were conserved at −80 °C in cryoprotective solutions (Mast Group, Bootle, UK). Ec223 and the six additional strains were selected from a prior screening based on their antibiotic resistance profiles and their biofilm-forming capacity in our experimental system.

Table 1.

E. coli strains used with their origin

Name in the study	Origin (isolation date)
Ec1655	E. coli K-12 substr. MG1655 reference strain
Ec223	CIRM-BP Collection (1977)
Ec775	NRL-AR collection, Chicken (2014)
Ec009	NRL-AR collection, Chicken (2014)
Ec956	NRL-AR collection, Pork (2015)
Ec709	NRL-AR collection, Pork (2017)
Ec723	NRL-AR collection, Pork (2017)
Ec478	NRL-AR collection, Chicken (2018)

Experimental evolution with biocides and isolation of antibiotic-resistant variants

The exposure to biocides was performed as follows (Fig. 1). Strains were grown overnight in 1/10 TSB medium. On the following day, strains were diluted in fresh 1/10 TSB to reach an OD_600nm of 0.2 ( + /− 0.02). For each strain, five columns of eight wells were filled with 190 µL of 1/10 TSB and 10 µL of the subcultures were then added to reach a final OD_600nm of 0.01. Plates were incubated at 20 °C for 1 h. Supernatants were then removed and replaced with fresh 1/10 TSB medium to keep only the fixed cells. Plates were incubated for 72 h at 20 °C.

After 72 h, the first cycle began. On the first cycle day, 5 µL of supernatants were first pipetted onto 20 mg/L rifampicin-supplemented plates, to evaluate RifR variants emerging in supernatants. Supernatants were then removed and 40 µL of fresh 1/10 TSB was added. A quarter of the biofilm was scratched and resuspended and 5 µL of the biofilm suspension was pipetted onto the rifampicin plates, which were then incubated for 48 h at 37 °C. Finally, fresh media with previously identified planktonic MIC concentrations of biocides was added. Each column corresponded to one condition, ultrapure water being used as a control and the biocides being used separately for the other four conditions. Then, each day for 4 days, the medium was renewed with fresh 1/10 TSB medium supplemented with H₂O or previously evaluated MIC concentrations of biocides (BAC = 6.25 mg/L; PHMB = 1.25 mg/L; TMN = 6.67 mg/L; SHY = 100 mg/L). Finally, on the last day, media were renewed for the following 72 h, until the new cycle began. The cycle was repeated four times over the month. A first one-month experiment was performed with H₂O and all four biocides, then the experiment was repeated with only the H₂O control group and the BAC and TMN groups, the results of which are presented in Fig. 2. The PHMB and SHY results are presented in Fig. S1 with the results of H₂O that were obtained only in the first experiment.

Confirmation of variants and phenotypic resistance stability

After colonies development on rifampicin-supplemented plate, stability of resistance phenotypes was assessed by subculturing RifR variants on TSA plates without rifampicin every day for four days. On the 5th day, variants were streaked again over 20 mg/L rifampicin-supplemented TSA plates to observe whether the variant was still able to grow on antibiotic-supplemented medium after being spread over non-selective media.

Levels of rifampicin resistance were characterised using TSA plates supplemented with various rifampicin concentrations. Each variant and its parent were first grown on a fresh TSA + 5% sheep blood plate and incubated overnight at 37 °C. On the following day, variants were resuspended in TSB 1/10 to reach a final concentration of 0.2 OD_600nm. Next, 5 µL of each culture was pipetted onto TSA rifampicin-supplemented plates at different concentrations (20 mg/L; 40 mg/L; 60 mg/L; 80 mg/L; 100 mg/L; 150 mg/L; 200 mg/L) and on one antibiotic-free plate. Plates were incubated overnight at 37 °C. Experiment was performed two times if the concentration was identical between the two experiments and, if not, a third experiment was realised and the median was displayed.

Enterobacterial repetitive intergenic consensus – PCR (ERIC-PCR) was performed as previously described¹¹ in order to check that variants were indeed derivatives of associated parental strains. DNA was extracted using the InstaGene kit (Bio-rad, Marnes-la-Coquette, France) and amplified using a LightCycler® 480 thermocycler (Roche Diagnostics, Meylan, France) with primers ERIC1-R (ATGTAAGCTCCTGGGGATTCAC) and ERIC2 (AAGTAAGTGACTGGGGTGAGCG) and GoTaq Flexi polymerase (Promega, Charbonnières-les-bains, France) as follows: 95 °C for 2 min for initial melting; 30 cycles at 95 °C for 1 min, 54 °C for 1 min, 72 °C for 4 min; final extension at 72 °C for 8 min followed by incubation at 4 °C. PCR products were then checked on 1% agarose gel and migrated over 90 min at 110 V before being revealed using a GelRED stain (Biotium, Brumath, France).

Genomic analysis of RifR variants

The whole genome of parental and variant strains was sequenced. First, DNA was extracted using the Macherey-Nagel Nucleospin Tissue kit (Düren, Germany) according to the manufacturer’s instructions. DNA concentrations were then checked with BiospecNano quantification. Illumina sequencing was performed on the strains at the Institut du Cerveau et de la Moelle épinière (Paris, France) using the Novaseq 6000 SP. Data from the raw reads were analysed and filtered with the fastp tool (v0.23.4)⁶⁷ using --correction, --cut_front and --cut_right options with a threshold of 25. Adapters were automatically detected and removed from the reads with --detect_adapter_for_pe option. Unicycler (v0.5.0)⁶⁸ was then used to perform de novo assembly on the genome of parental strains, keeping only contigs longer than 200 bp. These newly reconstituted genomes were annotated using Bakta (v1.8.2)⁶⁹ with the full database and E. coli as --genus and --species options. Afterwards, variant calling was performed with the variant strains exposed to a biocide and rifampicin using the annotated genomes of parental strains. Snippy (v4.6.0) was used to search for point mutations. The workflow described previously is available on the following GitHub repository: https://github.com/AB2R/WGS_pipeline.git. Mutated genes were analysed using the EcoCyc database⁷⁰. The genomes of the variants and their parental strains were compared and the variants’ genomes were assembled using Unicycler (v0.5.0)⁶⁸. They were then compared against the genomes of the parental strains using BLAST⁷¹. Proksee⁷² was used to generate the genome ring figures. A search for genes linked to mobile genetic elements was carried out on the genome of the parental strains using the mobileOG⁷³ database. GO analysis was performed on Uniprot (https://www.uniprot.org/).

To confirm the presence of the mutation, we conducted complementary targeted Sanger sequencing on all parental strains and derived variants (Eurofins Genomics, Nantes, France).

SDS-PAGE characterisation of LPS through silver nitrate staining

LPSs were extracted using the Sigma-Aldrich lipopolysaccharide (LPS) isolation kit (Saint-Louis, USA), according to the manufacturer’s instructions. A SDS-PAGE analysis was performed as in Chalabaev et al.⁷⁴. LPS extracts were reduced with β-mercaptoethanol for 5 min at 95 °C and migrated by SDS-PAGE using mini-Protean tris-tricine precast gels (16.5% acrylamide, BIO-RAD) and a mini Protean 3 migration system (BIO-RAD). A Tris/Tricine/SDS buffer (16610744, BIO-RAD) and a Tricine sample buffer 10X (1610739, BIO-RAD) were used. Fixation was performed in a solution of EtOH 30% (Sigma-Aldrich) and acetic acid 10% (UN2789, VWR) for 1 h. The gel was then washed for 3 × 5 min in ultrapure water (MilliQ, Millipore) before incubation in 0.7% metaperiodate (98%, 013798.22, Thermo Scientific) for 10 min. Another washing step of 3 × 5 min in ultrapure water was performed before incubation in a 0.02% sodium thiosulfate solution (99%, 217263, Sigma-Aldrich) for 1 min. Yet another washing step of 2 × 1 min in ultrapure water was performed, followed by incubation in 25 mM AgNO3 (>99%, 209139, Sigma-Aldrich) for 10 min. The final washing step using ultrapure water was for 15 s. The gel was revealed in 50 mL ultrapure water supplemented with 1.75 g of potassium carbonate (>99%, 209619, Sigma-Aldrich), 15 µL of formaldehyde (37 wt in water, 252549, Sigma-Aldrich) and 6.25 µL of thiosulfate 10% and incubated until the bands were revealed. Finally, the gel was incubated for 30 min in a stop solution composed of 4% Tris base (TG791, Sigma-Aldrich) and 2% acetic acid.

Determination of cell membrane surface charge

Bacteria were grown overnight on TSA plates at 37 °C. On the following day, bacteria were resuspended in 5 mL of TSB 1/10 and grown overnight at 37 °C. On the third day, cultures were centrifuged at 2000 g/10 min/4 °C. Two washing steps were performed with 5 mL of 15 mM NaCl and 2000 g/10 min/4 °C centrifugations. Then the OD_600nm was adjusted to 0.02 in 15 mM NaCl. Zeta potential was measured using a Zetasizer Nano ZS90 (Malvern, UK). Next, 900 µL of the cell suspension was added to folded capillary Zeta cells (DTS 1070, Malvern Panalytical). The following settings were used: equilibration time 50 s, auto mode, 3 measurement replicates, 30-s pause between replicates. Between each sample, the cuvette was rinsed three times with ultrapure water. At the end of the experiment, the cuvette was rinsed with 70% ethanol. There were five biological replicates for this experiment.

Assessment of membrane permeability modifications

Outer membrane permeability was evaluated using the Crystal Violet assay⁷⁵. First, cells were grown on TSA plates overnight at 37 °C. On the following day, cells were resuspended in 5 mL of 1/10 TSB medium and incubated overnight at 37 °C. On the 3rd day, bacterial cultures were first normalised by their OD_600nm so that parental strains and variants would be at the same concentrations. The bacterial suspension was centrifuged at 4500 g/5 min/4 °C. Cells were then washed twice with 15 mM NaCl and centrifuged at 4500 g/5 min/4 °C. After the second centrifugation, cells were resuspended in 15 mM NaCl supplemented with 10 µg/mL of Crystal Violet. Suspensions were then incubated for 10 min at 37 °C followed by a centrifugation step at 13,400 g/15 min/4 °C. Finally, the OD_590nm was read. A solution of 15 mM NaCl was used as a blank. The OD_590nm represents the amount of Crystal Violet that was not internalised by the cells, so a higher OD_590nm represents a strain which is less permeable than one with a lower OD_590nm. There were eight biological replicates for this experiment.

TMN killing kinetics measurements

Biofilms of the parental strains and the variants were prepared as described in part 2 of the Methods section. After 72 h incubation at 20 °C, the supernatants were removed. In one well, supernatant was replaced with 100 µL biocide-free PBS and 100 µL of polyvalent neutraliser were added. Then, for each timepoint, one well was discarded from its supernatant and replaced with 100 µL of PBS supplemented with 400 mg/L of TMN (60x the MIC). At 30 s, 1 min, 2 min, 5 min, 10 min, 20 min and 30 min, 100 µL of polyvalent neutraliser were added into the corresponding well. Biofilms were scratched and resuspended in the medium. 20 µL of the suspension were transferred in 180 µL of PBS. Serial dilutions were performed until 10⁻⁵. 6 spots of 10 µL were plated for each dilution and timepoint on TSA plates. Plates were incubated at 37 °C during 24 h and counted on the following day. This experiment was repeated 6 times, and each replicate corresponds to the mean value of the 6 spots.

Planktonic mortality kinetics were evaluated in a similar way. Overnight cultures were prepared in 5 mL TSB 1/10 medium and incubated at 37 °C. On the following day, cells were harvested through a 4000 g/10 min/8 °C centrifugation. Supernatants were removed and cells were resuspended with 10 mL PBS. Cultures were equilibrated in 10 mL of PBS to a 0.1 OD_600nm concentration. 500 µL were sampled and mixed with 500 µL of polyvalent neutraliser. TMN was added to the suspension at 133 mg/L (20x the MIC) and the suspension was vortexed thoroughly. At the 30 s, 1 min, 2 min, 5 min, 10 min, 20 min and 30 min timepoints, cultures were vortexed and 500 µL were sampled and mixed with 500 µL of polyvalent neutraliser. 20 µL of these cultures were transferred in 180 µL of PBS. The following steps (serial dilutions, agar spots and counting) were performed as for the biofilms in the previous paragraph. This experiment was repeated 3 times, and each replicate corresponds to the mean value of the 6 spots.

Biofilm matrix quantification

Biofilms of the variants and their parents were prepared as described in part 2 of the Methods section and grown for 72 h, without biocide. Then, the medium was replaced with a fresh TSB 1/10 medium. Plates were incubated for an additional 24 h. Finally, 50 µL of Congo Red at a concentration of 40 µg/mL were added to each biofilm. Fluorescence was quantified using a FLUOstar OPTIMA (BMG Labtech, Champigny-sur-Marne, France) with an excitation wavelength of 520 nm and an emission wavelength of 570 nm. There were 10 biological replicates with three technical replicates each time for this experiment.

Sample preparation for proteomic analysis

Biofilms of the parental strains and variants were prepared as described in part 2 of the Methods section. After the 72 h initial formation, medium was replaced either with fresh TSB 1/10 or with TSB 1/10 supplemented with MIC concentrations of TMN (6.67 mg/L). Plates were incubated at 20 °C during 24 h. Medium was then replaced with 100 µL fresh TSB 1/10 medium. Biofilms were scratched, resuspended and pooled in one Eppendorf. Two centrifugations of 8000 g/5 min/4 °C and resuspensions of the pellets in 1 mL of PBS were performed. A third centrifugation was then performed and cells were resuspended in 100 µL of lysis buffer (Tris 1 M, Sodium dodecyl sulfate 2%, cOmplete Mini EDTA-free Protease Inhibitor Cocktail (Roche, Switzerland) (1 Tablet in 10 mL)). Solution was then transferred in a new Eppendorf containing 0.1 mm diameter Zirconium beads at a 1:3 ratio (beads : volume). A mechanical lysis was performed with three 1 min cycles at a frequency of 25 Hz on a MM200 bead beater (Retsch, Germany). Cells were cooled in ice during 30 s between each agitation step. Finally, a 13000 g/15 min/18 °C centrifugation was performed and supernatants were stored at −80 °C.

A dosing step was performed with the Pierce^TM BCA Protein Assay Kit (Thermo Scientific, USA). Briefly, a standard range was prepared with bovine serum albumin concentrations ranging from 25 to 2000 µg/mL. The BCA reagent was added to both samples and controls, which were incubated for 30 min at 37 °C. Absorbance was measured at 570 nm using a FLUOstar Omega (BMG Labtech, Germany). The standard curve was generated and samples concentrations were quantified accordingly.

Fifteen micrograms of proteins were used for subsequent steps. Proteins were first reduced and alkylated using a solution containing 5% SDS, 20 mM Tris (2-carboxyethyl) phosphine (Thermo Scientific, USA), 180 mM Ammonium bicarbonate (ABC) (Thermo Scientific, USA), 50 mM iodoacetamide (Sigma Aldrich, Germany) and water for a final adjusted volume of 50 µL. Samples were then incubated in the dark, 15 min at room temperature. 2.5% orthophosphoric acid was added, followed by 300 µL of binding buffer (90% Methanol and 100 mM ABC). The samples were transferred to an S-Trap^TM column (Protifi, Farmingdale, NY, USA) and centrifuged at 4000 g/30 s/18 °C. Six additional centrifugations were performed, each following the addition of 150 µL of binding buffer. Trypsin (MS grade, Fisher scientific, USA), resuspended in 25 mM ABC at 0.2 µg/µL, was added to achieve a final trypsin to protein ratio of 1:50. The columns were incubated overnight at 37 °C. After digestion, the columns were washed with 40 µL of 50 mM ABC, followed by 40 µL of 0.2% formic acid with centrifugation at 4000 g/30 s/18 °C after each step. Peptides were eluted with 40 µL of 50% acetonitrile and the samples were evaporated using a Savant^TM SpeedVac^TM SPD120 (Thermo Scientific, USA). Samples were resuspended in 20 µL of 0.1% trifluoroacetic acid (TFA) and centrifuged at 10,000 g/10 min/18 °C. Pierce C18 Tips (Thermo Scientific, USA), pre-equilibrated with 100% acetonitrile and 0.1% TFA, were used for sample cleanup. Samples were loaded onto the tips, washed with 0.1% TFA, and eluted in 10 µL of 0.1% formic acid and 50% acetonitrile. The eluted samples were then evaporated a second time and finally resuspended in 20 µL of 0.1% formic acid, transferred into vials and stored at −20 °C.

Proteomic data acquisitions

Peptide mixtures (200 ng) were analyzed in biological quadruplicates using a Tims-ToF Pro2 mass spectrometer equipped with a Captive spray ion source with a 20 µm emitter and coupled to a nanoElute 2 UHPLC system (all From Bruker Daltonik GmbH, Bremen, Germany). Chromatographic separations were performed using a single-column separation setup with a reprosil-pur C18 column (PepSep Ten Series; 75 µm inner diameter, 100 mm length, 1.9 µm particle size, 120 Å pore size). The flow rate was set to 200 nL min⁻¹ using a gradient from 2% mobile phase A (water, 2% acetonitrile, 0.1% formic acid) to 27% mobile phase B (100% acetonitrile, 0.1% formic acid) over 10 min. This was followed by column cleaning (95% mobile phase B) and regeneration for 5 min at a flow rate of 250 nL min⁻¹, resulting in a total run time of 15 min. The mass spectrometer operated in the DIA-PASEF mode⁷⁶ with an estimated cycle time of 0.74 s. The mass range was set to 354.1–1200 m/z, and the mobility range was 0.68–1.33 Vs cm⁻² (1/K0). A total of 29 DIA-PASEF mobility windows were defined, each with a mass width of 25 Da and a single 100 ms TIMS scan. The collision energy was ramped linearly from 20 eV at 1/K0 = 0.6 Vs/cm² to 59 eV at 1/K0 = 1.6 Vs/cm².

Data analysis and statistics for proteomics

DIA data were analysed using DIA-NN (version 1.9.1) in library-free search mode, employing an in silico fasta database derived from the Uniprot E. coli K-12 fasta database (extracted the February 12th, 2025, 4357 sequences). Precursor length and charge ranges were set to 7-30 amino acids and 1-4, respectively. Precursor and fragment ion m/z ranges were defined as m/z 300–1800 and m/z 200–1800, respectively. Trypsin/P was specified as the protease, allowing for one missed cleavage. N-terminal methionine excision and carbamidomethylation of cysteine were set as fixed modifications. Label-free quantification (LFQ) was performed using match between-runs (MBR) with a precursor False Discovery Rate (FDR) cutoff of 1.0% and heuristic protein inference enabled. The quantification strategy was fixed to Quant UMS (High precision). All other parameters were let at default settings. Statistical analyses were performed using Perseus software (version 1.6.15). Proteins were set if more than 50% abundances values were valid. Missing values were imputed from a normal distribution with a width of 0.3 and a downshift parameter of 1.8. Overall, protein abundance variations were considered significant if the calculated ANOVA p-values were below 0.05. For pairwise comparisons between experimental conditions, a Student’s t test was applied. For each test, the permutation-based FDR threshold was set to 0.05, with 250 randomisations.

A GO biological process enrichment analysis was performed using ShinyGO⁷⁷. The complete list of detected proteins was used as a background. GO Categories with less than two modified proteins were automatically excluded. P-values were calculated using the hypergeometric test, and FDRs were computed via the Benjamini-Hochberg method. GO categories with FDR values > 0.05 were automatically excluded.

Uniprot database⁷⁸ was used to compare the proteins and their associated GOs. Proteins with a GO term containing either “lipopolysaccharide” ; “lipid A” ; “O-Antigen” were considered as implicated in LPS production. Proteins with a GO term containing “colanic acid” were considered as implicated in colanic acid production.

Statistical analysis

Statistics were generated using Graphpad Prism. Throughout the current study, a two-tailed Mann–Whitney test was used to compare two groups. It was first used to compare the different biocides with the H₂O control group for each week, in both biofilms and supernatants. It was then used to compare the variants with their parental strain. Each timepoint was also compared during the killing kinetics between the variant and its parental strain. The P values are represented in the study by the symbols: * = [0.05–0.01]; ** =]0.01–0.001]; *** = <0.001.

Author contributions

R.C.: Conceptualisation, Formal analysis, Methodology, Investigation, Validation, Visualisation, Writing—original draft, Writing—review & editing. P.L.: Data curation, Formal analysis, Investigation, Methodology, Software, Writing—original draft. T.L.: Methodology, Investigation, Data Curation. L.P.: Formal Analysis, Investigation, Visualisation. M.B.: Formal analysis, Investigation, Visualisation. P.H.: Investigation. J.D.: Investigation. A.H.: Methodology. C.S.: Project administration, Resources. R.B.: Conceptualisation, Funding acquisition, Supervision, Writing—original draft, Writing—review & editing. A.B.: Conceptualisation, Funding acquisition, Project administration, Resources, Supervision, Visualisation, Writing -original draft, Writing—Review & editing.

Data availability

All sequencing data were deposit in NCBI under the BioProject accession number PRJNA947807. Details of accession numbers for assembled genomes of parental strains and sequence reads of variant strains were provided in Table S4. The complete proteomic dataset is available in the PRIDE partner repository⁷⁹ under the identification number: PXD071663 (Username: reviewer_pxd071663@ebi.ac.uk, Password: udSvAbpcszp2). Full raw data set is provided as a supplementary file.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s44259-026-00224-5.