Nitroxoline has a broad activity spectrum against Gram-negative bacteria

Although nitroxoline (Fig. 1a) has been used for decades against uncomplicated UTIs and has a good safety profile^1,4,5,6, minimum inhibitory concentration (MIC) distributions are available from EUCAST only for eight bacterial species and seven genera¹⁸, with a clinical breakpoint determined only for E. coli² (Fig. 1b). To investigate whether nitroxoline could be repurposed against other bacterial pathogens, we systematically profiled its susceptibility against 1000–1815 strains from 34 Gram-negative species. This set included some of the most relevant pathogens in the current antimicrobial resistance crisis, such as A. baumannii, Burkholderia cenocepacia complex and Stenotrophomonas maltophilia (Fig. 1b, c and Supplementary Fig. 1a, b).

We measured MICs in three orthogonal ways (Fig. 1b, Methods): broth microdilution (1000 isolates, Fig. 1c), disk diffusion (1815 isolates, Supplementary Fig. 1a) and agar dilution (1004 isolates, Supplementary Fig. 1b). We observed good concordance between this study and EUCAST¹⁸ for the eight overlapping species (Supplementary Fig. 1c) and across the three methods, with the best agreement between broth and agar dilutions (Pearson correlation, R = 0.94) (Supplementary Fig. 1d–f).

From these results, we confirmed that several Gram-negative species were comparably or more susceptible to nitroxoline than E. coli (MIC₅₀ = 4 µg/ml for broth and agar dilution, lower than the EUCAST breakpoint of 16 µg/ml for E. coli²). For E. coli we confirmed this value, except for one strain out of 251 clinical isolates with MIC twofold the breakpoint. Nitroxoline was active against Enterobacterales (median broth MIC: 4 µg/ml) and Moraxellaceae, which had the lowest MIC values (median broth MIC: 2 µg/ml) and comprise pathogens for which therapeutic options are currently limited, such as A. baumannii. Only P. aeruginosa exhibited a MIC₅₀ of 32 µg/ml above the clinical breakpoint (Fig. 1c). Overall, these results indicate nitroxoline as a promising antibacterial option against Enterobacterales and Acinetobacter spp.

Nitroxoline is active against intracellular Salmonella Typhi and bactericidal against A. baumannii

Salmonella spp. were among the most susceptible species to nitroxoline (average MIC in broth: 3.45 µg/ml; maximum MIC in broth: 8 µg/ml in only 1/11 tested strains) (Fig. 1c and Supplementary Fig. 1a, b). Since Salmonella can invade and persist in the host intracellularly³¹, we tested whether nitroxoline could also act on intracellular Salmonella Typhi, the leading serovar responsible for enteric fever³². Using an in vitro infection assay (gentamicin protection assay³³, Methods), we showed that nitroxoline treatment results in a strong decrease (>97% compared to solvent control) of intracellular S. Typhi in infected HeLa cells, for both clinical isolates tested (Fig. 1d).

As other metal chelators³⁴, nitroxoline is classically considered bacteriostatic³⁵. While we confirmed this in E. coli for concentrations eight times its average MIC in broth (32 µg/ml) (Supplementary Fig. 2a), we detected partial bactericidal activity (decrease of at least 3 log₁₀ colony-forming units (CFU) /ml at 24 h³⁶) against A. baumannii, for concentrations as low as 8 µg/ml, i.e. three times its average MIC of A. baumannii in broth (2.79 µg/ml) (Fig. 1e and Supplementary Fig. 2b, Methods). Cells released their cytoplasmic content and lysed as early as 4.5 h of incubation with the drug (Fig. 1f and Supplementary Movies 1, 2, Methods). To our knowledge, this is the first evidence of nitroxoline’s bactericidal activity.

Nitroxoline antagonises beta-lactams and synergises with colistin

To explore potential combinatorial regimens, we tested nitroxoline in combination with 32 antibiotics in E. coli BW25113. The drug panel included all main classes of antibiotics used against Gram-negative bacteria, but also other metal chelators and antibiotics only effective against Gram-positive bacteria, such as vancomycin (Methods, Supplementary Data 1).

We uncovered extensive antagonism with beta-lactams (bactericidal cell-wall targeting drugs), including penicillins, cephalosporins and carbapenems (Fig. 2a and Supplementary Figs. 3, 4). Like other antagonisms between bactericidal and bacteriostatic drugs (such as nitroxoline at the concentration tested here), these interactions could be based on the fact that bactericidal drugs are more effective on actively dividing cells, and slowing down division with a bacteriostatic agent can alleviate their action³⁷.

a Nitroxoline interacts with several antibiotics in E. coli. Nitroxoline combinations were tested in 8 × 8 broth microdilution checkerboards in E. coli BW25113 (Supplementary Fig. 3). Bliss interaction score distributions are shown for each combination (n = 98 scores corresponding to 7 × 7 dose-combinations in two biological replicates). The median (central line), first (lower hinge) and third quartile (upper hinge) are shown for each boxplot. Whiskers correspond to 1.5x IQR from each hinge. The numbers stand for cumulative Bliss scores for each combination (Methods). b Nitroxoline resensitizes colistin-resistant K. pneumoniae and E. coli. Growth (OD_595 nm at 10.75 h, corresponding to the beginning of stationary phase for the untreated control for each strain) was measured in the presence of serial twofold dilutions of colistin, supplemented or not with 0.75 µg/ml nitroxoline and normalised by no-drug controls. Three K. pneumoniae and two E. coli strains (dashed lines) and their isogenic colistin-resistant descendants (solid lines) were tested, including experimentally evolved and clinical isolates (framed in black, Supplementary Data 2). One K. pneumoniae clinical isolate carries the mcr-1 positive natural plasmid pKP2442 and, therefore, lacks a parental strain. Mean and standard error across four biological replicates are shown. c Nitroxoline resensitizes a colistin-resistant K. pneumoniae clinical isolate in vivo. G. mellonella larvae were infected with the indicated isolate and treated with single drugs, their combination or were left untreated (solvent-only control). The mean and standard error are shown across four independent experiments for each condition. p = 0.0255 and p = 0.0098 comparing colistin-nitroxoline with colistin and untreated, respectively (log-rank test). NX nitroxoline, COL colistin. Source data are provided as a Source Data file.

One of the most potent synergies of nitroxoline was with the OM-targeting drug colistin (Fig. 2a and Supplementary Figs. 3, 4), whose toxicity limits its therapeutic use as a last-resort agent³⁸. Nitroxoline could, therefore, be used to lower colistin concentrations required to achieve therapeutic success, preventing toxicity. To explore this possibility, we tested whether nitroxoline could not only potentiate colistin action on sensitive strains, but also resensitize colistin-resistant strains. We showed that the addition of nitroxoline at sub-MIC concentration (0.75 µg/ml) decreases the MIC of E. coli and K. pneumoniae colistin-resistant strains (clinical and experimentally generated) from two- to four-fold, even below colistin EUCAST breakpoint (2 µg/ml)² in three cases (Fig. 2b). To confirm this synergy in vivo, we infected Galleria mellonella larvae with an mcr-1 positive, colistin-resistant K. pneumoniae clinical isolate (Methods, Supplementary Data. 2). The addition of nitroxoline improved the survival of the infected larvae by twofold at 72 h post-infection compared to the monotherapy with colistin or nitroxoline alone (Fig. 2c).

Nitroxoline perturbs the OM in E. coli

In addition to colistin, nitroxoline synergised with all large-scaffold antibiotics, including drugs that are normally excluded by the OM and therefore are not active against Gram-negative bacteria, such as macrolides, rifampicin, novobiocin and vancomycin (Fig. 2a and Supplementary Fig. 3). Altogether, this suggested a direct effect of nitroxoline on the OM permeability. To obtain a broader view of nitroxoline’s direct and indirect effects, we performed two-dimensional thermal proteome profiling (2D-TPP)³⁹ on E. coli BW25113 to monitor the abundance and stability of proteins upon nitroxoline exposure. TPP is based on the principle that changes in the interactions of a protein (e.g. with a drug) can lead to changes in its thermal stability⁴⁰. We exposed bacteria (whole-cell) or lysates to multiple nitroxoline concentrations and subjected them to a gradient of temperatures, capturing nitroxoline effects on both protein abundance and stability (Methods). While changes in lysates will typically only detect direct target(s) of drugs, as the biochemical environment of the cell has been disrupted, whole-cell changes provide a snapshot of both direct and indirect effects. We could not detect any significant change in lysate samples, suggesting that nitroxoline does not directly target a protein. In whole-cell samples, we observed a decrease in the abundance/stability of OM proteins (OMPs) and members of LPS biosynthesis and trafficking (Lpt) machinery (Fig. 3a, Supplementary Fig. 5a, b and Supplementary Data 3). These effects are similar to those caused by genetic perturbations known to influence OM stability⁴¹.

a Nitroxoline decreases the abundance and stability of outer membrane proteins and Lpt machinery. Volcano plots depicting abundance (left) or stability (right) changes upon nitroxoline exposure in whole-cell 2D-TPP. Results are based on n = 5 independent experiments (four drug concentrations and a vehicle control). Effect size and statistical significance as log₂ (F-statistic) (Methods) are represented on the x- and y-axis, respectively. The F-statistic was transformed to 1 when 0 before the log₂ transformation. Proteins are colour-coded according to their Gene Ontology (GO) annotation (Supplementary Fig. 5a). b Nitroxoline effects profiled by chemical genetics on an E. coli whole-genome single-gene deletion mutant library⁴³. Effects are expressed as multiplicative changes of mutant fitness compared to the plate median (approximating wild-type). Significance was obtained from an empirical Bayes’ moderated two-sided t-statistics, Benjamini–Hochberg adjusted (two independent clones per mutant, three replicates per condition, Methods, Supplementary Data 4). Genes are colour-coded as in Fig. 3a (GO in Supplementary Fig. 5c). c Nitroxoline directly affects OM permeability. NPN fluorescence upon exposure of E. coli BW25113 to nitroxoline, positive (polymyxin B, EDTA) and negative (chloramphenicol, untreated samples) controls. Data points represent the average for each of the four biological replicates per condition. The horizontal line and error bars indicate mean and standard error. ns p > 0.05; **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001 (two-sided Welch’s t-test using the chloramphenicol control as the reference group). EDTA, p = 0.008; Polymyxin B, p = 0.00004; no-drug control, p = 0.053; nitroxoline 0.45 µg/ml, p = 0.0008; nitroxoline 0.9 µg/ml, p = 0.003; nitroxoline 1.8 µg/ml, p = 0.002. d Nitroxoline is more potent upon chemical perturbation of the OM. EOP assays with tenfold serial dilutions of E. coli BW25113 cells plated onto no-drug control plates, 0.5% SDS + 0.8 mM EDTA, 0.45 µg/ml nitroxoline, or their combination. Four biological replicates were tested for each condition. Source data are provided as a Source Data file.

We combined this data with chemical genetics, in which we systematically mapped nitroxoline effects on the fitness of deletion mutants of every non-essential gene in E. coli^42,43. We found that mutants involved in similar processes, including LPS biosynthesis and transport, were more sensitive to nitroxoline, except for mutants involved in the first reactions of heptose incorporation into LPS inner core biosynthesis (gmhA, gmhB and waaC), which were more resistant (Fig. 3b, Supplementary Fig. 5c and Supplementary Data 4). This suggests the heptosyl-Kdo₂ moiety of the lipopolysaccharide inner core as a minimum requirement for nitroxoline activity, since deletion of mutants catalysing most downstream LPS biosynthetic reactions, starting with waaF, are more sensitive to nitroxoline.

To provide direct evidence of nitroxoline’s effect on the OM, we quantified OM disruption using the hydrophobic probe 1-N-phenylnaphthylamine (NPN), which emits fluorescence upon exposure of the OM phospholipid layer⁴² (Methods). Sub-MIC concentrations of nitroxoline resulted in a significantly higher fluorescence than control samples (unexposed to any drug or to the non-OM-affecting antibiotic chloramphenicol). As positive controls, we used OM-targeting antibiotic polymyxin B and EDTA, another metal chelator that disrupts OM by sequestering LPS-stabilising divalent cations⁴³. Although lower than in positive controls, OM disruption by nitroxoline occurred even at 1/9 MIC (0.225 µg/ml, MIC = 2 µg/ml in E. coli BW25113, Fig. 3c).

To corroborate nitroxoline’s action on the OM, we tested its activity against the OM-defective E. coli strain lptD4213⁴⁴ and in OM-perturbing conditions (0.5% SDS and 0.8 or 0.4 mM EDTA)⁴⁵ using an efficiency of plating (EOP) assay (Methods). The lptD4213 mutant was more susceptible to nitroxoline than wild-type E. coli (Supplementary Fig. 5d), in agreement with LptD decreased stability (Fig. 3a and Supplementary Fig. 5b) and with its loss-of-fitness already observed in the chemical genetic data, where the lptD4213 mutant was included⁴² (Fig. 3b). Furthermore, nitroxoline synergised with the OM-perturbing conditions at concentrations tenfold lower than MIC (Fig. 3d and Supplementary Fig. 5e).

It is possible that nitroxoline acts similarly to EDTA, chelating metals necessary for the stability of the OM⁴⁶. While EDTA action is based on chelation of both Mg²⁺ and Ca²⁺, the two main cations involved in LPS stability⁴⁷, nitroxoline has been shown to preferentially complex with Mn²⁺ and Mg²⁺, with variable reports on the effect of Ca²⁺ supplementation on MIC^21,48. This might explain nitroxoline’s smaller effect than EDTA on OM integrity (Fig. 3c). However, the abundance of OMPs and Lpt machinery proteins was also altered (Fig. 3a and Supplementary Fig. 5a, b), suggesting an effect of nitroxoline on the regulation of the levels of these proteins.

Nitroxoline acts as a zinc and copper metallophore

Nitroxoline is reported to chelate Mn²⁺ and Mg²⁺ and reach the intracellular milieu²¹, but a broader and more resolved view of its effects on metal homoeostasis is missing. From the 2D-TPP (Fig. 3a and Supplementary Fig. 5a) and chemical genetic data (Fig. 3b and Supplementary Fig. 5c), we identified distinct profiles for proteins involved in the import, intracellular utilisation, and export of metals, consistent with responses to copper (Fig. 4a) and zinc (Fig. 4b) increase.

a–c Nitroxoline affects metal homoeostasis inducing copper and zinc detoxification responses, as determined by 2D-TPP. Heatmaps show the relative remaining soluble fraction compared to the vehicle control at each temperature to highlight changes in protein abundance and thermal stability profiles of the Cu(I) exporter CopA, the periplasmic copper oxidase CueO (a), the transcriptional regulators Zur and ZntR, zinc importer ZnuA and exporter ZntA (b), and the manganese importer MntH (c), are shown. d Nitroxoline increases intracellular levels of copper, zinc and manganese. Synchrotron-based nano-XRF measurements on E. coli untreated or exposed to nitroxoline (1 µg/ml), expressed as elemental areal density (ng/cm²). The mean and standard error across ≥5 cells are shown (Methods). ns p > 0.05; *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001 (two-sided Welch’s t-test). Copper, p = 0.01; iron, p = 0.73; manganese, p = 0.0006; zinc, p = 0.0005. Source data are provided as a Source Data file.

At nitroxoline concentration around MIC (2 µg/ml), we observed an increased abundance of the copper-exporting P-type ATPase CopA (twenty-fold) and the multicopper oxidase CueO (fivefold), which upon Cu(I) excess remove copper from the cytoplasm and oxidise it in the periplasm, respectively⁴⁹ (Fig. 4a and Supplementary Data 3). Accordingly, the deletion mutant of cueR, which encodes the positive regulator of copA⁵⁰, is more sensitive to nitroxoline (Fig. 3b). A similar effect has been reported for other quinolines, forming complexes with copper^51,52, which may be extruded in the periplasm by CopA as a copper-overload defence mechanism. We observed similar effects for zinc, with the stabilisation of the zinc-responsive regulators Zur and ZntR and consistent changes in Zur- and ZntR-regulated proteins, such as subunits of the zinc importer ZnuABC (repressed by Zur) and the exporter ZntA (positively regulated by ZntR)⁵³ (Fig. 4b). Accordingly, ΔzntR was more sensitive to nitroxoline, whereas we could not detect significant fitness changes in znuABC deletion mutants (except for a slight decrease for ΔznuB). This suggests that the limitation of zinc uptake is not sufficient to confer nitroxoline resistance, and that other mechanisms (e.g. zinc efflux, active drug efflux) have a stronger impact on fitness upon nitroxoline exposure (Fig. 3b and Supplementary Data 4).

Zinc and copper intoxication has been associated with the disruption of iron-sulfur (FeS) clusters^54,55,56,57 and a compensatory induction of iron-uptake proteins (derepressed by Fur), and FeS cluster biogenesis (suf genes, induced by IscR). Accordingly, upon nitroxoline exposure, we observed decreased stability of Fur and an associated increase of Fur-repressed proteins involved in enterobactin biosynthesis (EntABF), recycling (Fes), receptor (FepA) and importing system (FepBCDG)⁵⁸. The stability of IscR increased, with the associated decrease of isc operon and increase of suf operon members⁵⁹ (Fig. 3a and Supplementary Fig. 5f). The increased abundance of the manganese importer MntH (Figs. 3a, 4c) has also been reported as a consequence of copper stress⁶⁰, which is consistent with the increased sensitivity of the corresponding mutant (Fig. 3b). Since MntH is repressed by Fur, its increase is consistent with the observed Fur destabilisation (Supplementary Fig. 5f).

To confirm the impact of these effects on intracellular metal concentrations, we performed synchrotron-based nano-X-ray-fluorescence (XRF) on nitroxoline-treated and untreated E. coli, confirming a four-fold copper, twofold zinc, and tenfold manganese increase in treated cells (Fig. 4d). Overall, our data suggest pleiotropic effects of nitroxoline on metal homoeostasis, consistent with its activity as ionophore for copper, previously reported for clioquinol in cancer cells⁵², and zinc, as shown for other quinolines⁶¹.

Nitroxoline resistance is based on conserved mechanisms across species

Our results suggest that nitroxoline does not have a direct protein target, but rather exerts pleiotropic effects on OM integrity and metal homoeostasis, which might underpin the previously observed low frequency of resistance^1,6,22,28,30. To explore resistance mechanisms across different species, we evolved resistance to nitroxoline in vitro in three species: E. coli, for which nitroxoline is already used, and two Gram-negative species, K. pneumoniae and A. baumannii, for which nitroxoline could be repurposed considering its low MIC (Fig. 1c and Supplementary Fig. 1a, b). We calculated the frequency of resistance in two strains for each species. Overall, across the three species, we found 19/24 and 21/24 lineages with frequency of resistance below 10⁻¹⁰ at four times and eight times the MIC, respectively, confirming previous reports on E. coli and K. pneumoniae^1,6,22,28,30 and showing this for the first time for A. baumannii (Supplementary Data 5). This further supports the repurposing of nitroxoline on a broader range of Gram-negative bacteria.

We performed whole-genome sequencing (WGS) on 12 E. coli, 8 K. pneumoniae and 6 A. baumannii sensitive and evolved resistant strains (fold increase MIC ≥4 compared to parental-sensitive strain) (Methods, Fig. 5a and Supplementary Data 2), and performed proteomics on a subset of them (Fig. 5b, Supplementary Fig. 6a, b and Supplementary Data 6). Mutations across species primarily affected transcriptional repressors of RND-type efflux pumps: emrR (previously reported in E. coli²⁹), oqxR (K. pneumoniae), adeL and tetR/acrR (A. baumannii) (Fig. 5a).

a Whole-genome sequencing of experimentally evolved nitroxoline-resistant strains (Supplementary Data 2). Nitroxoline MIC values are indicated below each strain. Mutation effects are colour-coded. Strains on which proteomics was performed (Fig. 5b) are indicated in bold. K. pneumoniae strains whose sensitive parental strain lacks oqxR are marked with an asterisk. The in-patient evolved K. pneumoniae clinical isolate 8_R1 is indicated in italics. b Protein abundance changes in nitroxoline-resistant strains. Selected proteins annotated as efflux pumps or porins are shown and clustered according to Pearson’s correlation. Hits are marked with an asterisk (adjusted p value ≤0.05 and at least twofold abundance change (Supplementary Fig. 6a, b and Supplementary Data 6). Species are colour-coded, as in Fig. 5a. c Wild-type emrR and oqxr complementation restores nitroxoline susceptibility. The experimentally evolved E. coli strain 1_R4 with emrR D109V mutation (Fig. 5a and Supplementary Fig. 6c) and K. pneumoniae strain 4_R1 harbouring the oqxR G60-L67 duplication (Fig. 5a, d and Supplementary Fig. 6d) are shown. Nitroxoline MIC was measured by broth microdilution. Mean and standard error across four biological replicates are shown. ns p > 0.05; *p ≤ 0.05 (Wilcoxon test. For E. coli: empty plasmid vs no-plasmid, p = 0.608; empty plasmid vs wild-type efflux pump, p = 0.042; no-plasmid vs wild-type efflux pump, p = 0.042. For K. pneumoniae: empty plasmid vs no-plasmid, p = 0.217; empty plasmid vs wild-type efflux pump, p = 0.042; no-plasmid vs wild-type efflux pump, p = 0.042). d Amino acid changes resulting from oqxR mutations. The domain annotation of OqxR was obtained from its closest annotated structural homologue NsrR (Methods). e Efflux pump inhibitors resensitize nitroxoline-resistant strains. Nitroxoline MIC was measured by broth microdilution. Mean and standard error across four biological replicates are shown. Resistant strains are shown as shaded plots next to their parental-sensitive strains. For results on all strains, see Supplementary Fig. 6g. p values are only shown when significant: *p ≤ 0.05; **p ≤ 0.01 (Wilcoxon test). Source data are provided as a Source Data file.

We found emrR mutations in all 12 evolved nitroxoline-resistant E. coli strains (Fig. 5a). This is consistent with previous reports²⁹ and our chemical genetic data, where the emrR deletion mutant was more resistant, and the knockouts of its regulated pump emrAB were more sensitive to nitroxoline (Fig. 3b). To verify the clinical relevance of these mutations, we assessed them in 14 clinical isolates with reduced susceptibility (MIC ≥8 µg/ml, i.e. at least two times the MIC₅₀ measured in this study for E. coli), finding distinct mutations from experimentally evolved strains (Supplementary Fig. 6c). We also found emrR mutations in two K. pneumoniae nitroxoline-resistant strains, which, as previously shown for E. coli²⁹, had higher EmrA and TolC protein levels. An A. baumannii resistant strain, although lacking any mutation of efflux pump regulators, also exhibited a fourfold increase in EmrA levels (Fig. 5b and Supplementary Data 6).

Surprisingly, we did not detect any increase in EmrA and only a slight (<2-fold) increase of TolC in three E. coli emrR-mutated strains, which showed instead a decreased abundance of porins OmpD, OmpF and LamB (Fig. 5b). For at least two of these strains (2_R1 and 2_R4), this could depend on missense mutations of envZ, that regulates porin expression via OmpR^62,63, increased in these strains (Fig. 5b). Alternatively, porin abundance changes could be explained by mutations in the lon gene (Fig. 5a), previously associated with nitroxoline resistance²⁹ and resulting in the stabilisation of the Lon protease substrate MarA⁶⁴. which regulates the expression of several drug resistance determinants, including porins^65,66, and is also increased in these strains (Fig. 5b). Given the unexpected proteome changes in emrR-mutated E. coli, we sought to confirm the functional relevance of these mutations, complementing a strain carrying a recurring missense mutation (D109V, Supplementary Fig. 6c) with wild-type emrR, thereby restoring nitroxoline susceptibility (Fig. 5c).

The most frequent genetic alterations in K. pneumoniae-resistant strains were mutations in oqxR, the transcriptional repressor of the OqxAB efflux pump, in agreement with recent reports³⁰. We identified oqxR mutations in 5/8 experimentally evolved K. pneumoniae strains and in all 14 clinical isolates sequenced (Fig. 5d). We identified a mutational hotspot, common to clinical isolates and experimentally evolved strains: a duplication of eight amino acids (G60_67dup) resulting in a loop addition (Fig. 5d and Supplementary Fig. 6d). This mutation also emerged in a patient after four-month prophylaxis with nitroxoline (K. pneumoniae urine isolate 8_R1, Fig. 5a), confirming its relevance for in vivo evolution of resistance. Accordingly, oqxR-mutated strains coclustered in the proteomics data (Fig. 5b and Supplementary Fig. 6b) and showed an increased abundance of OqxA (BepF), OqxB (OqxB3) and TolC (Fig. 5b and Supplementary Data 6). To further demonstrate the impact of this duplication on resistance, we complemented a G60_67dup-positive strain with wild-type oqxR, restoring nitroxoline susceptibility (Fig. 5c).

In resistant A. baumannii the most common mutations affected two transcriptional regulators: adeL, repressing the expression of the efflux pump AdeFG(BepF)-OprC, and a transcriptional regulator of the acrR/tetR family (Fig. 5a and Supplementary Fig. 6e, f). We performed proteomics on a strain carrying an adeL mutation resulting in a premature stop codon, which accordingly showed an increased abundance of all components of the efflux pump AdeFG-OprC (Fig. 5b). Another resistant strain, although not carrying any mutation in efflux pump regulators, exhibited a fourfold increase in EmrA abundance, which could explain its resistance (Fig. 5b).

From the mutational spectrum and proteomic changes that we observed across species, increased drug efflux via RND pumps appeared as a conserved strategy to achieve nitroxoline resistance. To verify this hypothesis, we tested the impact of the efflux pump inhibitors (EPI) 1-(1-naphthylmethyl)-piperazine (NMP) and phenylalanine-arginine β-naphthylamide (PAβN) on nitroxoline susceptibility. We observed a decrease in nitroxoline MIC both in resistant and susceptible strains, independent of their specific mutations and generally more marked for PAβN, previously reported as an inhibitor of RND pumps in E. coli⁶⁷ and of AdeFG in A. baumanni⁶⁸ (Fig. 5e and Supplementary Fig. 6g, Methods).