Genomic tness proling of Acinetobacter baumannii reveals modes of action for common biocides and mechanisms of biocide-antibiotic antagonism

Biocides, such as antiseptics and disinfectants, are used ubiquitously for hygiene in households and for life-saving infection control in hospitals. An increasing concern is that the widespread use of biocides may contribute to the emergence and spread of multidrug-resistant bacteria. We performed transposon directed insertion site sequencing (TraDIS) to identify genes and key cellular pathways of the multidrug resistant nosocomial pathogen Acinetobacter baumannii, that affect host tness during exposure to a panel of ten structurally-diverse and clinically-relevant biocides: silver nitrate, benzalkonium, cetyltrimethylammonium bromide (CTAB), chlorhexidine, triclosan, chloroxylenol, polyvidone iodine, bleach, glutaraldehyde and ethanol. Multiple genes encoding proteins localised either in the cell envelope or in the cytoplasm were shown to affect biocide susceptibility. These proteins are involved in multiple processes including fatty acid biogenesis, multidrug eux, the tricarboxylic acid cycle, cell respiration and cell division, suggesting that these biocides may have intracellular targets in addition to their known effects on the cell envelope. Based on the importance of cell respiration genes for A. baumannii tness on biocides, we proposed and conrmed that apart from triclosan, the other 9 biocides at sub-inhibitory concentration can dissipate the membrane potential and lead to A. baumannii tolerance to antibiotics that have intracellular targets. Our results support the concern that residual biocides in clinical or community environments can promote the development of antibiotic resistance in pathogenic bacteria.


Introduction
Multiple drug resistance (MDR) in bacterial pathogens is an alarming public health issue 1 . Bacteria can resist drugs by expressing e ux pumps to export drugs out of the cell, altering outer membrane permeability to reduce drug accumulation, or expressing enzymes for drug inactivation by hydrolysis or modi cation 2 . Alternatively, bacteria can gain intrinsic resistance mutations which prevent target recognition or cause target bypass. The Gram-negative opportunistic human pathogen A. baumannii is a signi cant worldwide threat for immunocompromised patients who are hospitalized in Intensive Care Unit (ICU) wards. This is due to the emergence of A. baumannii clonal lineages with high level resistance to antibiotics and tolerance to desiccation 3,4 . The World Health Organisation (WHO) recently listed carbapenem resistant A. baumannii as a "top priority pathogen" for development of new therapies.
Biocides are widely used for disinfection and cleaning in both households and hospitals, but little is known about their impact on the emergence and spread of MDR infections. In contrast to antibiotics, the use of biocides is largely unregulated. The recommended in-use biocide concentrations are normally orders of magnitude higher than the bacterial minimum bactericidal concentration (MBC). Biocides are presumed to have multiple antibacterial targets or non-speci c targets, reducing the likelihood of bacteria developing resistance to these compounds 5 . However, outbreaks of nosocomial MDR infections are not uncommon. There is evidence suggesting that biocide exposure at sublethal concentrations can select for antibiotic-resistant bacteria under laboratory conditions 6 , and clinical antibiotic resistant isolates typically have reduced biocide susceptibilities 7 . One of the most commonly known biocide and antibiotic cross-resistance mechanisms is the activity of MDR e ux pumps, such as AcrAB-TolC from Escherichia coli 8 , AdeIJK and AdeABC from Acinetobacter baumannii 9 , and NorA from Staphylococcus aureus 10 .
Biocide modes of action are generally poorly characterized, except for cell lysis via interaction with the phospholipid membrane 5 . However, some biocides are known to have mode of action that start intracellularly. For example, silver ions can interact with the thiol group of exposed cysteine residues leading to enzyme inactivation and disruption of cellular iron homeostasis 11,12 , and they also cause cell membrane proton leakage 13 . The known silver resistance determinants are primarily involved in reducing silver ion intracellular concentration, including membrane transporters for silver ion export 14 , and proteins for neutralization or reduction of silver ions to the inactive metallic form 15 . Triclosan targets an intracellular essential enzyme enoyl reductase (FabI) in fatty acid synthesis 16 . The two key determinants for high-level triclosan resistance are MDR e ux pumps and either mutations in the fabI gene or the acquisition of alternative Fab proteins not recognizable by triclosan 17 . Apart from the drug e ux pumps mentioned earlier, members from the proteobacterial antimicrobial compound e ux (PACE) family confer resistance to antibacterial surfactant compounds including chlorhexidine and benzalkonium 18 . These observations suggest the surfactant biocides are also likely to have intracellular targets, or these pumps can expel them from the membrane.
Knowledge of both biocide modes of action and tolerance mechanisms is important for improved biocide use for effective and sustainable infection control. Here we performed the rst systematic genome-wide screen to identify genes that affect susceptibilities of an MDR A. baumannii strain to ten clinically important biocides (Table S1), using transposon-directed insertion-site sequencing (TraDIS) 19 .

Results
TraDIS experimental outcomes. TraDIS assays were performed on a global clone II multidrug resistant A. baumannii strain BAL062 transposon mutant library, containing >100,000 unique Tn5 insertions (1 insertion every 38 bp) 20 , to assess the global genetic response to biocide treatments. Ten structurally distinct biocides were chosen for this study; these are all either listed as "essential medicines" by the WHO or commonly used in clinical settings and/or as household products (Table S1). The A. baumannii TraDIS library was exposed to a sub-inhibitory concentration of each biocide (¼ × the minimum inhibitory concentration (MIC) of the wild-type strain), and genomic DNA was extracted for TraDIS sequencing and analysis as previously described 21 . Gene-wise mutant abundance data were compared to untreated control samples grown under identical conditions for the same length of time to identify mutations that affect tness in the presence of biocides, while controlling for any general impacts on growth rate. Collectively, the biocide treatments revealed a range of genes harbouring a reduced mutant population (i.e. decreased insertion read counts; 3-120 genes), representing potential biocide tolerance determinants, or with an expanded mutant population (i.e. increased insertion read counts; 7-100 genes) whose inactivation is potentially bene cial under biocide-treatment (Table S1).
We examined genes previously associated with biocide resistance and found their functions re ected in the TraDIS results. For example, the MDR e ux pump-encoding operon adeABC and its activator genes (adeRS) showed signi cantly decreased insertion read counts (e.g. adeB had a 115.4-fold decrease) after treatment with benzalkonium, a known substrate of AdeABC; whereas no change was seen after treatment with glutaraldehyde, which is not known to be a substrate of AdeABC ( Figure 1A).
Genes previously associated with antibiotic resistance were identi ed, for example as lpsC -a glycosyltransferase and LOS-synthesis gene involved in polymyxin resistance in A. baumannii 22 but not previously implicated in biocide resistance -showed a 5.8-fold decrease in insertions following triclosan treatment ( Figure 1A). We also revealed potential new roles in biocide resistance for previously characterised genes, e.g., the insertions in the pyrimidine synthesis gene pyrC showed a 35.9-fold increase in AgNO 3 -treated cultures compared to the control ( Figure 1A).
As proof-of-principle phenotypic validation of this TraDIS data, we measured the MICs of the four isogenic single-gene knockout Tn26 mutants from A. baumannii AB5075 23 for the genes outlined above: adeB, lpsB, pyrC and adeR, for benzalkonium, triclosan, AgNO 3 and glutaraldehyde respectively. In each case, the biocide MIC fold change for the mutant relative to wild-type was consistent with the TraDIS results ( Figure 1B).
Outer membrane lipooligosaccharide. A. baumannii lacks a homolog of the O antigen ligase WaaL, which transfers O-antigen polysaccharide onto the outer-core of LPS, suggesting that A. baumannii LPS may comprise only lipid A and core oligosaccharide 24 . We observed that insertions in a lauroyl acyltransferase encoding gene, lpxL, increased in frequency following treatment with the cationic surfactants chlorhexidine and cetyltrimethylammonium bromide (CTAB) (21-and 2.3-fold respectively), suggesting that this gene decreased tness at sub-inhibitory concentrations of these two biocides ( Figure 2A). The predominant glycolipid molecule in the outer lea et of A. baumannii OM is hepta-acylated lipid A 25 , whereas an lpxL mutant generates a hexa-acylated lipid A 26 . This change in fatty acid pro le may impact the effectiveness of these biocides against A. baumannii.
Cell surface polysaccharides. In A. baumannii, there are two gene clusters known to be involved in capsule polysaccharide biosynthesis (K-locus) and LOS outer core oligosaccharide biosynthesis (OClocus) respectively 27 . We identi ed 9 genes in the K-locus with signi cant changes in insertion counts after treatment with one or more biocides. For instance, insertions in the gna gene, encoding a UDPglucose dehydrogenase, increased during treatment with chlorhexidine and bleach (207 and 3.2-fold, respectively), implicating capsule polysaccharide in biocide sensitivity.
Insertions in an initiating glycosyltransferase-encoding gene, pglC, had decreased read-coverage (9.9fold) when treated with AgNO 3 , while treatment with the other 9 compounds did not have an effect, suggesting this gene may mediate tolerance only for AgNO 3 (Figure 2A). pglC is located in the variable region of the K-locus, but its presence is conserved in this gene cluster across A. baumannii strains 27,28 . It is required for the construction of capsular glycan repeat-units and glycosylation of glycoproteins 28 , and the deletion mutant affects capsule production not LOS 29 . In addition to pglC, insertions in 7 other genes in the K-locus also displayed decreased frequency following AgNO 3 treatment (Figure 2A), implying that the capsular polysaccharide may play a speci c role in AgNO 3 tolerance, whereas alteration of LOS glycoforms impacts susceptibility to biocides more broadly.
Insertions in seven genes in the OC-locus had altered frequency following biocide treatment ( Figure 2A). For instance, insertions in the glycosyltransferase gene lpsC decreased 5.8-fold following exposure to benzalkonium, suggesting a speci c role for lpsC in benzalkonium tolerance. Further, insertions in another glycosyltransferase gene, gtrOC3, dropped dramatically (78.8-fold) upon chlorhexidine treatment, suggesting gtrOC3 mediates chlorhexidine tolerance ( Figure 2A). Insertions in the glycosyltransferase gene lpsB, which is outside of the two polysaccharide biosynthesis loci and is involved in LOS core biosynthesis, also showed changes following exposure to 7 biocides ( Figure 2A). Previously, it has been shown that disruption of these glycosyltransferases in A. baumannii produces truncated LOS 30,31 . Thus, genes predicted to encode glycosyltransferases, that catalyze the linking of sugars of the LOS outer oligosaccharide core, may be important in biocide tolerance.
Chaperone-usher pilus. Another potential cell surface biocide tolerance determinant identi ed in this study is a chaperone-usher (CU) pilus assembly gene cluster csuA/BABCDE (Figure 2A) 32 , insertions in which decreased after treatment with eight biocides (Figure 2A). For example, insertions in all the csu genes decreased with most biocides: 2.2-to 3.2-fold decrease for AgNO 3 , 2.8-to 3.8-fold decrease for benzalkonium, 7.2-to 8.4-fold decrease CTAB, and 5-to 7-fold decrease for triclosan ( Figure 2A). Consistent with their known function as csu operon activators, insertions in the genes of the twocomponent system BfmRS were also shown to decrease following treatments with seven same biocides ( Figure 2A). Interestingly, insertions in a transcriptional regulator (BAL062_01328) directly upstream of the csu operon also decreased in frequency following treatment with the same eight biocides and the fold change was similar to that of csu gene mutants (Figure 2A), indicating that BAL062_01328 encodes a potential local transcriptional activator of the csu operon. Together these results suggest that the mature pilus system is required for a biocide tolerance phenotype.
Membrane transport. Sixteen drug e ux systems have been characterized or identi ed in A.
baumannii 18,33 . One of the most clinically signi cant drug transporters in A. baumannii is a tripartite e ux system AdeABC 34 , comprised of a membrane fusion protein (MFP) AdeA, a resistance-nodulationdivision (RND) e ux pump AdeB, and an outer membrane factor (OMF) AdeC, respectively. This system is constitutively overexpressed in various clinical A. baumannii MDR strains, typically due to mutations in the two-component transcriptional activator genes adeRS 35 .
Tn5 insertions in adeB signi cantly decreased in frequency 115.4-, 2.2-, 85.6-, 7.9-and 2.0-fold in the mutant pools that were treated by benzalkonium, cetrimonium, chlorhexidine, triclosan and chloroxylenol, respectively ( Figure 2B), suggesting that adeB mediates resistance to these compounds. Benzalkonium and chlorhexidine have previously been characterized as AdeB substrates 9 . Surprisingly, changes in frequency of Tn5 insertions in adeA and adeC were only seen in the mutant pools treated by benzalkonium and chlorhexidine ( Figure 2B), where insertions in adeA had 3.3-and 78.8-fold decrease, and insertions in adeC had 2.4-and 11.2-fold decrease, respectively ( Figure 2B). This suggests AdeABC functions as a tripartite system in pumping out these two compounds. These fold-changes were lower than the changes of adeB caused by the same compounds, indicating that there might be alternative OMF and MFP proteins that form a tripartite RND e ux system with AdeB.
Insertions in adeR decreased 7.6-and 26.7-fold, and those in adeS decreased 7.5-and 240-fold when treated by benzalkonium and chlorhexidine respectively, whereas no change was observed in the other three conditions that also selected against insertions in adeB ( Figure 2B). These observations together suggest that benzalkonium and chlorhexidine could induce the overexpression of AdeABC through AdeRS 36 , whereas the other three compounds cannot.
The other important RND system that has been characterized in clinical A. baumannii MDR strains is AdeIJK, which has a similar range of substrates as AdeABC 9 . The TraDIS data from this study for adeIJK is similar to that for adeABC ( Figure 2B Figure 2B). The respective changes in frequency of insertions in the RND pump encoding gene adeJ and MFP encoding gene adeI were lower than seen for adeK, suggesting AdeK might not only form a tripartite drug e ux system with AdeI and AdeJ, but also with AdeAB ( Figure 2C), consistent with previous reports 38,39 . Also consistent with previous ndings that the transcriptional regulator AdeN represses the transcription of adeIJK 40 , the insertions in adeN increased 1.5-, 99.7-, 1.7-and 1.3-fold after treatment with benzalkonium, chlorhexidine, triclosan and chloroxylenol respectively ( Figure 2B).
The frequency of Tn5 insertions in another RND membrane transporter gene BAL062_00031 and MFP encoding gene BAL062_00030 reduced 7.31-and 7.41-fold respectively when exposed to triclosan, suggesting these genes play a role in triclosan resistance ( Figure 2B). There is no OMF gene located adjacent to BAL062_00030 and BAL062_00031. This system may use AdeK as its OMF partner, because adeK is the only OMF gene in which insertions reduced in frequency (27-fold) in the mutant pool exposed to triclosan ( Figure 2B).
Four other drug e ux pumps were also implicated in mediating biocide resistance. The MFS transporter AmvA is a multidrug exporter with substrates including benzalkonium, and chlorhexidine 41,42 . The frequency of insertions in amvA decreased 1.6-and 106.2-fold after exposure to benzalkonium and chlorhexidine respectively, and additionally showed a 19.4-fold decrease when treated by AgNO 3 ( Figure   2B). Furthermore, the frequency of insertions in amvR, encoding a transcriptional repressor of amvA 20 , increased 5.2-and 84.7-fold following exposure to AgNO 3 and chlorhexidine, respectively ( Figure 2B). This suggests that AmvA may also recognize silver ions as a substrate or may transport other compounds related to silver detoxi cation.
The frequency of insertions in tetR, immediately upstream of a tetracycline e ux pump gene tetA, increased 21.1-fold following exposure to chlorhexidine ( Figure 2B). TetR is a known transcriptional repressor of tetA 43 , and this result suggested that tetA overexpression reduces susceptibility to chlorhexidine. In contrast, chlorhexidine did not affect the frequency of insertions in tetA, which is likely because TetR does not recognize chlorhexidine as a ligand and is not responsive to it. These are consistent with our previous nding that E. coli expressing A. baumannii TetA had a signi cant increase in chlorhexidine resistance level 44 .
The other putative multidrug e ux pump highlighted in this study is an ABC transporter consisting of BAL062_01229, encoding a membrane permease, and BAL062_01230 encoding an ATPase. When exposed to AgNO 3 the frequency of insertions in BAL062_01229 and BAL062_01230 decreased 2.2-and 2.6-fold respectively ( Figure 2B), suggesting this ABC e ux system plays a role in AgNO 3 tolerance.
Cell Division. Various genes that are involved in peptidoglycan (PG) synthesis, cell shape determination and cell division were shown to have a decreased frequency of Tn5 insertions following treatment with silver nitrate and several other biocides ( Figure S1). Insertions in the bacterial rod shape determining genes mreB, mreC, mreD, rodA and pbp2 decreased in frequency (4.8-, 6.3-, 4.7-, 4.8-and 4.0-fold respectively) when treated by AgNO 3 , suggesting disruption of these genes resulted in greater sensitivity to AgNO 3 . In line with the phenotype of these mutants, Ag + solution has been shown to cause distortion in bacterial cell membranes and morphology 45 . Hypochlorite and ethanol also showed similar effects on the frequency of these mutants, though with lower fold changes ( Figure S1), suggesting they may also affect cell membranes and morphology.
The FtsZ-ring associated genes (zipA, zapA and rlpA) and six genes involved in peptidoglycan synthesis and hydrolysis were also found to have a role in tolerance to multiple biocides ( Figure S1). Insertions in zipA decreased in frequency when treated with eight different biocides, for instance a 6.9-fold decrease in the AgNO 3 pool, a 3.0-fold decrease in the chloroxylenol pool, and a 2.5-fold decrease in the glutaraldehyde pool. The frequency of insertions in genes in the pal-tolQ operon, which is required for OM invagination during cell division, was strongly reduced (59.3 -61.7-fold) upon chlorhexidine treatment.
The PMF was proposed to affect cell division because the cellular localization of MreB and FtsA (FtsZring associated) proteins has been shown to be PMF-dependent 46 . This may explain why TraDIS revealed that mutations in both electron transfer and cell division genes affect A. baumannii tness in the presence of AgNO 3 and other biocides.
Central Metabolism and Respiration. Tn5 insertional inactivation of several TCA cycle genes had effects on A. baumannii susceptibility to silver nitrate. The frequency of insertions in the TCA cycle genes sucC and sucD encoding succinyl-CoA synthetase β and α subunits both showed a 12.3-fold increase upon silver nitrate treatment, suggesting that absence of succinyl-CoA synthetase increased resistance to silver nitrate ( Figure S1). Similarly, previous work demonstrated that knockout strains of E. coli TCA cycle genes (ΔsucB, Δmdh) were less sensitive to silver nitrate than the wild type 12 . In contrast, insertions in two other TCA cycle genes acnA_1 encoding aconitate hydratase 1 and icd_2 encoding isocitrate dehydrogenase showed 11.7-and 12.5-fold decreases after silver nitrate treatment, respectively ( Figure S1), suggesting that these mutants had decreased tness in silver nitrate. Silver nitrate may target SucC/SucD in A. baumannii, requiring alteration of uxes through the TCA cycle, such that there is a heavier requirement for icd_2 and acnA.
Another known antibacterial effect of AgNO 3 occurs via cytoplasmic membrane proton leakage and attenuated or ceased cell respiration 13 . We measured the membrane potential change upon exposure to AgNO 3 in A. baumannii AB5075 wild type strain using 3,3'-Diethyloxacarbocyanine iodide (DiOC 2 (3)).
AgNO 3 at as low as 1/32 MIC started to cause a drop in the membrane potential of cells at exponential phase in a dose-dependent manner ( Figure 3B). The dissipation of membrane potential was re ected in decreased frequency of insertions in genes encoding proteins involved in electron/proton shuttling during respiration. For example, Tn5 insertions in cydB and cydA_1, encoding cytochrome D ubiquinol oxidase decreased by 5.5-and 7.0-fold respectively, and insertions in a ubiquinone biosynthesis gene BAL062_03562 decreased 4-fold ( Figure S1). We hypothesize that these genes might be involved in maintaining the PMF to resist silver nitrate-induced cytoplasmic membrane proton leakage and the compromised electron suppliers from the TCA cycle.
An electron transport-related operon encoding cytochrome O (cyo) ubiquinol oxidase subunits had signi cantly reduced frequency of Tn5 insertions upon chloroxylenol, benzalkonium or ethanol treatment ( Figure S1), suggesting that the electron transport chain is affected by these compounds, but in a different way to silver nitrate.
Biocide induced dissipation of membrane potential and impairment of membrane transport. The nding that cell division and cell respiration genes are required for tness in the presence of multiple biocides indicated that, like AgNO 3 , these molecules may cause dissipation of membrane potential as part of their activity. To explore this hypothesis, we rst measured the membrane potential change of A. baumannii AB5075 on exposure to AgNO 3 using 3,3'-Diethyloxacarbocyanine iodide (DiOC 2 (3)). As expected, given its known effect causing proton leakage, AgNO 3 caused a drop in membrane potential of exponential phase cells (Figure 3). This effect was apparent at as low as 1/32 MIC and was dose-dependent. We then measured the membrane potential of exponential-phase A. baumannii AB5075 following exposure to the remaining biocides at MIC. Based on the genetic tness requirements revealed by TraDIS, we predicted that all biocides would impact membrane potential, with the exception of triclosan which was the only biocide that did not have respiratory among its resistance determinants. As shown in Figure 3A all compounds except for triclosan caused a drop in membrane potential. This suggests that dissipation of membrane potential might be a direct or downstream antibacterial effect of multiple biocides.
Cell membrane transport activities are energy dependent. If the biocides in this study do induce dissipation of membrane potential, they should also be able to compromise the solute e ux e cacy of the pumps that rely on membrane potential as energy source. Acri avine is a uorescent substrate for AdeB, AdeJ and AmvA ( Figure 4A), which utilise PMF for substrate transport. Chlorhexidine and benzalkonium can induce AdeABC expression but not AdeIJK and AmvA; whereas triclosan does not affect the expression of any of these pumps. If benzalkonium and chlorhexidine induce the expression of AdeABC but have no impact on membrane potential, cells treated by either of these two biocides should have reduced acri avine accumulation; however, if the biocides dissipate the membrane potential, despite e ux pump induction, higher acri avine accumulation would be expected. To investigate the hypothesis above, we focused on how triclosan, ethanol, chlorhexidine and benzalkonium affect acri avine accumulation in A. baumannii AB5075 wild type and its amvA transposon insertion mutant (ΔamvA). As expected, at ½ MIC, triclosan treatment did not lead to any change, whereas all the other three biocides increased acri avine accumulation in the cells of both bacterial strains ( Figure 4B). This suggests that, regardless of adeB induction, the dissipation of membrane potential induced by ethanol, chlorhexidine and benzalkonium signi cantly increased overall acri avine accumulation.
Biocides' impact on antibiotic potency. The uptake of the poly-cationic aminoglycosides requires membrane potential (both the proton gradient (DH + ) and electrical potential (Dy) across the cytoplasmic membrane) and drug-sensitive ribosomal binding sites for irreversible drug uptake 47 . Our next hypothesis is that the biocides that cause membrane potential dissipation can compromise aminoglycoside uptake and thus antagonise the killing effects of the antibiotic. As resistant bacterial strains accumulate less aminoglycoside antibiotic than sensitive strains 48 , we used an aminoglycoside-sensitive strain A. baumannii ATCC17978 rather than the highly resistant AB5075 or BAL062 strain to test our hypotheses. We showed that apart from triclosan, the other biocides including benzalkonium, chlorhexidine, CTAB and ethanol at ¼ MIC increased A. baumannii ATCC17978 survival when treated with gentamicin ( Figure 5A). Although benzalkonium, chlorhexidine and ethanol can induce AdeB expression, because AdeB does not confer resistance to gentamicin 39 , the antagonism with these biocides is unlikely due to drug e ux. These observations raised the question of whether the biocides can also compromise the antibacterial effects of the other antibiotics that have different intracellular targets, such as uoroquinolones and tetracyclines, and the ones targeting cell envelopes, such as colistin and β-lactams.
Uptake of tetracyclines into the cytoplasm is PMF dependent but likely involves passive diffusion of the uncharged form 49 . Similarly, the zwitterionic form of the uoroquinolones has been proposed to passively cross the cytoplasmic membrane, but due to their different protonation behaviour from tetracyclines, uptake of some uoroquinolones has been shown to be negatively correlated with PMF 49,50 . Comparing to the combination with triclosan or solo-antibiotic treatment, benzalkonium and chlorhexidine signi cantly increased A. baumannii survival rates with amikacin (another aminoglycoside), cipro oxacin ( uoroquinolone) and tigecycline (glycylcycline) ( Figure 5B). The observed biocide antagonism with amikacin and tigecycline is consistent with our hypothesis. In the case of cipro oxacin, there are certain discrepancies in published data about the impact of PMF on the antibacterial effects of different uoroquinolones 50 . Although cipro oxacin interaction patterns with triclosan, chlorhexidine and benzalkonium are similar to the other three antibiotics tested, we cannot conclusively state that dissipation of the membrane potential is the primary cause of the antagonism with biocides.
Dissipation of the membrane potential may not antagonise the antibacterial effects of imipenem and colistin, because the primary targets of these antibiotics are located on the cell surface. As expected, triclosan, benzalkonium and chlorhexidine at ¼ MIC have no impact on imipenem killing ( Figure 5B). Synergism was observed between benzalkonium and colistin, but not for triclosan ( Figure 5B).
Chlorhexidine also seemed to marginally enhance colistin killing. The synergy between these compounds is likely because benzalkonium, chlorhexidine and colistin all targeting cell membranes. Collectively, our data suggest that the biocides that can dissipate membrane potential can promote A. baumannii tolerance to various classes of antibiotics which have intracellular targets.

Discussion
TraDIS enabled us to investigate the potential tolerance/resistance determinants and modes of action of 10 diverse biocides that are commonly used for disinfection in hospital and in personal hygiene products. These compounds have traditionally been thought to target the cell envelope, but our data suggests they have additional intracellular targets and need to penetrate through the cell envelope to exert their full antibacterial effects. Insertions in genes from multiple cellular pathways had statistically signi cant changes in frequency as a result of the biocide treatments, including the TCA cycle, electron transport chain, amino acid biosynthesis, nucleoside biosynthesis and biotin biosynthesis (SI 2). Various A. baumannii drug e ux pumps, such as AdeABC, AdeIJK and AmvA were shown to be tolerance/resistance determinants for multiple biocides, supporting the idea that these compounds may have intracellular target(s). The biocides also appeared to have diverse impacts on the cell surface. Transposon insertions in genes related to LOS and lipid A biosynthesis affected A. baumannii tness in the presence of all of these 10 compounds. In contrast to LOS, capsular polysaccharide affected only AgNO 3 susceptibility out of the ten biocides tested.
While antibiotic stewardship has been a topic of increasing public health concern, the debate on whether biocide stewardship is necessary is still continuing [51][52][53][54] . This can be largely explained by the fact that biocides' in-use concentrations are normally much higher than their MBCs. It should be emphasized that all of the biocides tested here are either WHO Essential Medicines or widely used hospital and household disinfectant products. In this study, we showed that the biocides at sub-MIC levels cause dissipation of the membrane potential which can signi cantly promote A. baumannii survival rate (tolerance) under treatment with various antibiotics that have intracellular targets, including aminoglycosides, cipro oxacin and tigecycline. This is because membrane potential plays a critical role in drug uptake. However, they do not antagonise the killing effects of drugs targeting the cell envelope, such as colistin and imipenem.
Although drug e ux pumps only confer low level resistance to antibiotics, compared to their parental strains, bacteria that lack key MDR e ux systems are much less likely to develop MDR 55,56 . Furthermore, intermittent antibiotic exposures have been shown to lead to more rapid evolution of tolerance and resistance [57][58][59] and antibiotic tolerance facilitates the evolution of resistance 60 . These works and our study together suggest that factors affecting antibiotic intracellular accumulation or drug target accessibility are important in the development of drug tolerance and AMR. Previous studies have typically used direct measurement of bacterial susceptibilities to antibiotics and biocides to investigate coselection of antibiotic resistance 61 . We propose that the principal concern with respect to biocide stewardship is not the emergence of biocide resistance or antibiotic cross-resistance, but rather the presence of residual biocides, especially chlorhexidine, benzalkonium and CTAB and others that are hard to remove from the environment, which could induce antibiotic tolerance and potentially AMR development upon the co-presence of antibiotics.