An anti-virulence drug targeting the evolvability protein Mfd protects against infections with antimicrobial resistant ESKAPE pathogens

Abstract

The increasing incidence of antibiotic resistance and the decline in the discovery of novel antibiotics have resulted in a global health crisis, particularly, for the treatment of infections caused by Gram-negative bacteria, for which therapeutic dead-ends are alarming. Here, we identify and characterize a molecule, NM102, that displays antimicrobial activity exclusively in the context of infection. NM102 inhibits the activity of the non-essential Mutation Frequency Decline (Mfd) protein by competing with ATP binding to its active site. Inhibition of Mfd by NM102 sensitizes pathogenic bacteria to the host immune response and blocks infections caused by the clinically-relevant bacteria Klebsiella pneumoniae and Pseudomonas aeruginosa, without inducing host toxicity. Finally, NM102 inhibits the mutation and evolvability function of Mfd, thus reducing the bacterial capacity to develop antimicrobial resistance. These data provide a potential roadmap for the development of drugs to combat antimicrobial resistance.

Introduction

The fight against pathogenic bacteria is becoming one of the greatest challenges for our societies, particularly with the spread of multi- and extensively-drug resistant bacteria, which are seriously jeopardizing the use of traditional antibiotics¹. Estimates suggest that at least 700,000 people die annually from drug-resistant infections, and without intervention, this number could rise to 10 million by 2050, becoming the major cause of death worldwide². In particular, infections caused by the ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter spp.) are increasingly associated with therapeutic dead-ends. As such, they are at the top of the World Health Organization’s priority list of bacteria for which novel therapeutics are urgently needed. The identification and characterization of antibiotics with new molecular scaffolds, innovative targets and/or novel mechanisms of action, associated with a low potential for resistance induction, are urgently needed and particularly challenging for Gram-negative pathogens. Indeed, since 1990, only six new classes of antibiotics have been approved³. Recent efforts have begun to revitalize antibiotic research, with compounds being explored either through drug repurposing or through molecular mechanisms that are often redundant with traditional antibiotics. This is the case for finafloxacin, a fluoroquinolone antibiotic recently approved to treat ear infections caused by P. aeruginosa^4,5, darobactin, which specifically targets Gram-negative bacteria⁶, teixobactin, which is only effective against Gram-positive bacteria⁷, and the newly developed irresistin-16, which exhibits bactericidal activity against both Gram-negative and Gram-positive bacteria without inducing resistance, at least thus far⁸. These findings pave the way for the discovery of new molecules, a crucial step in providing effective antibiotics that act on novel targets, thereby avoiding cross-resistances.

In this context, this study focuses on a new bacterial target, the Mutation Frequency Decline (Mfd) protein, and its inhibition by innovative small molecules. Mfd is a non-essential transcription-repair coupling factor, ubiquitous in bacteria and absent in eukaryotes⁹. Mfd recognizes RNA polymerase (RNAP) stalled at noncoding lesions and utilizes ATP to translocate along DNA, likely forcing RNAP forward and ultimately dissociating it from the bulky DNA template¹⁰. Mfd then stimulates DNA repair by recruiting components of the Nucleotide Excision Repair system^11,12,13,14. The structure of Mfd has been solved, revealing the presence of an ATPase motor module, which is highly conserved among bacteria and is functionally necessary^12,15. Stress-induced mutagenesis can help pathogens generate drug-resistant cells during antibiotic therapy. Mfd has been associated with the development of antibiotic resistance in Campylobacter jejuni^16,17, and Helicobacter pylori Mfd-deficient cells are sensitized to antibiotics^16,17. Furthermore, Mfd has been suggested to be an evolvability factor that promotes hypermutation in bacteria, thereby accelerating the evolution of antimicrobial resistance (AMR)¹⁸. In line with that, a recent study identified, through an in vivo screen, a molecule called ARM-1 for Antimicrobial Resistance Molecule that targets Mfd to reduce mutagenesis and significantly delays antibiotic resistance development across highly divergent bacterial pathogens¹⁹. Taken together, these data emphasize that blocking Mfd could hinder molecular evolution, thereby inhibiting the development of resistance in many bacterial pathogens.

Furthermore, Mfd is critical for virulence in the Gram-positive Bacillus cereus and in the Gram-negative Shigella flexneri species and it confers resistance to nitric oxide (NO) stress^20,21. The production of reactive nitrogen species is a key step in the immune response following infection²². NO induces lesions in bacterial DNA, thereby limiting bacterial growth within hosts. Mfd is involved in DNA repair following DNA damage and is thus required for bacterial resistance to the host response. Consequently, in both species, a mutant lacking Mfd is severely impaired in its virulence capacity in cell and animal models. As Mfd is widely conserved in the bacterial kingdom, these data highlight a mechanism—possibly ubiquitous in bacteria—that overcomes the host immune response and tackles its mutagenic properties.

The function of Mfd in the bacterial bypass of host immunity, its role as an evolvability factor during antimicrobial resistance, its ubiquity in the prokaryotic world as well as its absence in higher eukaryotes, all support the selection of this protein as an innovative bacterial target for anti-infective therapies. Molecules that inhibit Mfd activity would neither directly inhibit nor kill bacteria, but instead curb bacterial evolution while impeding the bacteria’s ability to resist the host immune response. Thus, Mfd inhibitory compounds should boost the immune system’s response against pathogenic bacteria while acting exclusively at the inflammation site, thereby preventing collateral damage.

Herein, we report a targeted, structure-based, high-throughput in silico screening, using the druggable ATP binding site of Mfd, to identify molecules capable of acting as competitors to the cognate ligand ATP, thereby inhibiting Mfd activity. One molecule, NM102, was identified as a potent molecule that specifically targets the ATP-binding site of Mfd. In silico molecular analysis and in vivo assays revealed that NM102 displays antimicrobial activity, specifically in the context of infection, even against clinically relevant Gram-negative bacterial pathogens such as resistant K. pneumoniae, P. aeruginosa, and pathogenic E. coli. In animal models, NM102 was able to block infections with these pathogens without inducing host toxicity nor damaging the microbiota. Finally, NM102 also blocked the function of Mfd as an evolvability factor, thereby reducing the bacteria’s ability to develop antimicrobial resistance.

Overall, our findings identify and characterize a promising antimicrobial candidate that could be used either alone, as a booster for the host immune system, or in combination with other antibiotics by inhibiting Mfd’s evolvability activity.

Results

Identification of promising hits targeting Mfd in silico

To identify antimicrobials with a new mechanism of action, we performed a rational, targeted, and structure-based high-throughput in silico screening of small molecules to challenge their efficiency to accommodate to the ATP binding site of the bacterial protein Mfd. 3D modeling of E. coli Mfd in an active conformation was designed (Supp Fig. 1). Using this template, a large library of 4.8 million compounds was screened in silico, to challenge their capacity to bind the ATPase site of Mfd. This virtual screening led to the identification of 95 molecules, selected for subsequent experimental validation (Supp Fig. 1).

Inhibition of Mfd activity in vitro

The 95 molecules were challenged to inhibit the ATPase function of E. coli Mfd in vitro (Fig. 1A). The inhibition rates of Mfd activity ranged from 10% to 85%. The molecule with the highest inhibition rate (85%), named NM102, was selected for further analysis. Dose-response assays on ATPase activity (Fig. 1B) and Lineweaver-Burk plot (Fig. 1C) both revealed a competitive mode of inhibition of NM102 to ATP with an IC₅₀ of 29 ± 0.1 µM and a K_i of 27 ± 1.9 µM.

Fig. 1: Identification of NM102 as an ATP competitor for Mfd activity.

A In vitro high-throughput screening for inhibitors of Mfd-C ATPase activity. The indicated compounds were tested at a final concentration of 100 mg/mL. The data were normalized to those of the DMSO control. Mfd-C ATPase activity was measured at 0.35 µM at an ATP concentration of 1 mM. The results are the average of two independent experiments done in duplicate with standard deviation. B Michaelis Menten plot for inhibition activity of NM102 (0–100 µM) on E. coli Mfd-C (0.35 µM) ATPase activity with ATP (0–0.3 mM). The results are the average of at least three independent experiments with standard deviation. The IC₅₀ was computed using GraphPad Prism7.05. C Lineweaver-Burk plot showing NM102 inhibition of Mfd-C ATPase activity through a competitive mode of action. The results are the average of at least three independent experiments with standard deviation. Source data are provided as a source data file. D Chemical structure of the NM102 compound.

NM102 is a small molecule with a chemical scaffold relatively close to ATP as it is composed of an indole-like ring close to the adenosine ring followed by a ribose-like ring and eventually polar sulfur groups that could mimic phosphate moieties (Fig. 1D).

Specificity of NM102 binding to Mfd

To measure the specific binding of NM102 to Mfd, Isothermal Titration Calorimetry (ITC) experiments were conducted (Fig. 2A). ITC is based on the determination of the heat involved during the binding of a ligand to a receptor, measuring power units as a function of time. NM102 showed affinity towards Mfd, and the isotherm for the titration of the ligand against the molecule indicated a significant exothermic behavior. The obtained stoichiometry value (n = 1.1) indicates a 1:1 binding interaction between the Mfd protein and NM102 with a K_d of 83 ± 9 µM. To assess the specific and competitive binding of NM102 in the ATP pocket of Mfd, we measured the interaction of ATP with Mfd in the absence and in the presence of NM102 (Fig. 2B). In the absence of NM102, ATP binds to Mfd with a K_d of 145 ± 9 µM. By contrast, in the presence of NM102, ATP binds to Mfd with a K_d of 430 ± 50 µM. Thus, NM102 binds to Mfd with a better affinity than ATP and successfully inhibits Mfd-ATP interaction which decreased by ca 70% (A = Ka/Kapp = 0.34). These data demonstrate that NM102 binds specifically to the ATP binding site of Mfd and inhibits Mfd-ATP binding in a competitive manner.

Fig. 2: NM102 affinity to Mfd.

A ITC measurement of binding of NM102 to Mfd. The heat rate in the presence of NM102 was compared with and without NM102 and plotted as a function of time (higher panel), and the normalized fit was measured with an increased mole ratio (lower panel). B ITC thermodynamic parameters associated with the interaction of Mfd with ATP in the absence or presence of NM102. K_d, dissociation constant K_d, K_a, association constant, A, ratio of K_a in the presence vs absence of inhibitor NM102.

To assess the specificity of the NM102 inhibitor towards the ATPase activity of Mfd, we evaluated its capacity to decrease the ATPase activity of various eukaryotic proteins, including functionally related and unrelated ATPase proteins. The activity towards four proteins was tested: ERCC3, ERCC6, XPD, and yUpf1 (Fig. 3A). NM102 sharply inhibited Mfd ATPase activity, while it had no effect on all other ATPase protein activity, strongly suggesting that NM102 is selective for Mfd activity. In addition, the activity of NM102 was also tested towards the bacterial helicase RecG since the library screen that identified NM102 was performed on the Mfd sequence of E. coli morphed with respect to the active ADP binding site of RecG (Supp Fig. 1). Consistently, a slight activity of NM102 was observed, nevertheless significantly lower than for Mfd, further emphasizing NM102 specificity for Mfd. (Fig. 3B).

Fig. 3: NM102 specifically binds the ATP binding site of Mfd.

A NM102 inhibition of bacterial Mfd and eukaryotic proteins (ERCC3, ERCC6, XPD, yUpf1) ATPase activity. ATPase activity of the proteins, in the absence and presence of 100 µM of NM102, was assessed by using BIOMOL® Green reagent microtiter-plate assay. Data were normalized to that of the DMSO control. The graph shows the mean of at least two independent experiments with standard deviation. B Michaelis Menten plot for inhibition activity of NM102 (0 to 100 µM) on E. coli RecG (0.35 µM) ATPase activity with ATP (0 to 0.3 mM). The results are the average of two independent experiments done in duplicate with standard deviation. C Binding energy measured in silico in kcal/mol between ATP or NM102 and Mfd (left) anf Upf1 (right). The bindings of ATP and NM102 are shown in cyan, and orange, respectively. The graph shows the distribution of binding energies obtained for twenty poses of the substrate/ligand couple by AutoDock. Source data are provided as a source data file.

NM102 binds to E. coli Mfd in silico with a better affinity as compared to ATP

Additionally, we investigated the specificity of NM102 through computational docking studies of the ATP pocket of the E. coli Mfd to measure the binding affinity (Fig. 3C). NM102 showed a marked enhanced affinity for Mfd as compared to its cognate ATP ligand, with binding energies of −9.8 kcal.mol⁻¹ vs -6.5 kcal.mol⁻¹, respectively, consistently with ITC data. The docking (Fig. 4A) highlighted that Mfd accommodates NM102 at its ATP binding site, within the same binding pocket that encompasses the motif I with residues from Asp629 to Glu636, that also includes the catalytic Lys634. Markedly, NM102 engages in π-stacking interactions with Phe597 and Phe599, a polar interaction with Thr635, and charge interactions with Glu636 plus Arg905 (Fig. 4A panel right). Additionally, we were able to define a sphere of interaction through twenty-one positions involved in the binding of both ATP and NM102 (Suppl. Table 1). NM102 is a slightly longer molecule as compared to ATP, especially in its extended conformation. Thus, its accommodation extends to interact with amino-acid residues that are remote from the catalytic center, namely Gln664, His665, Asn668, and D876 of motif V. This feature could explain the increased affinity for NM102 compared to ATP.

Fig. 4: NM102 docking in the ATP binding site of Mfd.

A Upper part: strip of the eight domains with the following color code blue D1a-D2-D1b, orange D3, pink D4, yellow D5, green D6, and red D7. Lower part: Mfd rendered surfaces of conformations L₀ inactive (pdb id 2eyq) and L₁ active (pdb id 6 × 26), and their corresponding view at 90°. The color code of the surface corresponds to the strip of domains shown above. Right part, top insert: Close-up of Mfd from E. coli with the docking of ATP (in stick with carbon colored in cyan) and down insert: similar view with NM102 (in stick with carbon colored in deepteal). The residues involved in the sphere of binding of ligands are shown as sticks. The color code of the domains is respected with residues of D5 and D6 in yellow and green, respectively, and residues of motifs I and II in red and blue, respectively. B Close-up of the docking of ATP (cyan sticks) and NM102 (orange stick) in Mfd of E. coli harboring the site-directed mutations F597A, F599A, K634A, and E730Q. For the sake of comparison, position of ATP in the non-mutated Mfd is shown as black lines.

In parallel to Mfd, bindings of ATP and NM102 were simulated in Mfd that was mutated in silico at its catalytic center (Fig. 4B). Substitutions were performed and rotamers were selected for Phe597, Phe599, Lys634A and Glu730Q. Lys634A and Glu730Q since they are known to define the active site⁹. Phe597 and Phe599 were selected because they could be crucial for the recognition and pose of the indole-like ring of the ligand. Accordingly, they were mutated into alanine residues to abolish the apolar contribution of their aromatic ring, while preserving local and global folding of the protein. Importantly, the docking showed that mutated Mfd completely changed the accommodated poses of both ATP and NM102, and that their alternative positions were inconsistent with a competitive occupancy of the ATP site by ATP. These data further highlight the interaction of NM102 at the ATP binding site of Mfd.

In silico, NM102 binds to the Mfd active site of the ESKAPE bacteria

Mfd is widely observed in the bacterial kingdom¹⁰. To assess whether NM102 could more generally bind and inhibit the function of Mfd from a wide range of bacteria implicated in AMR, multiple sequence alignment and homology modeling of Mfd were performed for one strain of each species of the ESKAPE bacteria (Supp Fig. 2). The alignment revealed that Mfds are distinctly clustered into Gram-positive and Gram-negative bacteria, nevertheless they overall share more than 45% of sequence similarity along 1200 amino acid residues of the ESKAPE bacteria. Markedly, this average on Mfd full-length sequence does not reflect the trend observed for each functional module. Indeed, modules composed of domains D1a-D3, D4, D5-D6, and D7 show 47%, 58%, 68% and 44% of sequence similarity, respectively (Supp Fig. 3). Particularly, the ATPase motor module, composed of D5 and D6 domains, is by far the most conserved. It shares 68% and 36% sequence similarity and identity within the ESKAPE strains, respectively. Furthermore, the ATP binding site, which locates at the interface between D5 and D6, showed fairly high to strict conservation of the catalytic motifs, the so-called motifs I to VI (Supp Fig. 2).

To profile the affinity of NM102 towards Mfds from the ESKAPE bacteria and identify the probable conserved features of binding, NM102 was docked into each 3D homology model of Mfd and the binding energy was measured (Suppl. Fig. 4). Molecular binding in all the modeled Mfds revealed that both ATP and NM102 maintain similar interactions at the ATP binding site, with residues conserved in all species, particularly the catalytic Lys634 and Glu730 that are strictly conserved, and Phe597 and Phe599 that are replaced by tyrosine residues in S. aureus, A. baumannii, and E. faecium (Mfd E. coli numbering). Their mode of binding can be clustered into two rather isoenergetic groups. One cluster gathers NM102 pose similar in Mfd E. coli, S. aureus, A. baumannii, K. pneumoniae, and P. aeruginosa, while the other gathers an NM102 pose similar in Mfd E. faecium and E. cloacae. Also, 21 residues were identified as identical between species and always positioned within the 5 Å coordination sphere (Suppl. Table 1). Particularly, P1 Phe597 and P2 Phe599 could be responsible for the π stacking pose of the adenine-like moiety while polar P12 Thr635, and charged P11 Glu636 and P21 R905 could interact closely with the sulfonamide and phosphate-like groups of NM102. Overall, ATP binding ranged from −5.5 to −8 kcal mol⁻¹, whilst NM102 binding ranged from -9.0 to −10.3 kcal.mol⁻¹, suggesting that NM102 binds to all ESKAPE Mfds with a significantly better affinity as compared to the physiological ATP.

Taken together, our results strongly suggest that NM102 could compete with ATP at the ATP binding site in all Mfd proteins of the ESKAPE group bacteria, hence, specifically inhibiting Mfd ATPase activity. As the challenging urgency is to identify molecules efficient against Gram-negative pathogenic bacteria of the ESKAPE group, the rest of the study is focused on clinically relevant K. pneumoniae, P. aeruginosa and E. coli strains.

NM102 has antimicrobial activity in the context of nitric oxide stress

Mfd is a non-essential protein in unstressed conditions. However, we have previously shown that Mfd is essential to resist NO stress during infection^20,21. Our innovative antimicrobial strategy consists in identifying inhibitors that have no antibiotic activity under unstressed conditions, but are only active under stressed conditions, where Mfd is required. NM102 was thus first tested for its antimicrobial activity in the absence of stress. Under these conditions, no antimicrobial activity on K. pneumoniae was observed (Fig. 5A). Conversely, under NO stress conditions, NM102 inhibited clinical K. pneumoniae survival (IC₅₀ of 78.8 µM) (Fig. 5B). These results highlight that NM102 has an antimicrobial activity against K. pneumoniae under NO stress conditions.

Fig. 5: Antimicrobial effect of NM102 on Gram-negative bacteria during NO stress.

Strains were grown to exponential phase in LB medium. The bacteria solution was prepared in RPMI medium and dispatched in 96-wells plate. Bacteria were exposed for 4 h at 37 °C to 50 µL of increasing concentrations of NM102 alone A K. pneumoniae DOU or with NOC-5 as a NO donor, B K. pneumoniae DOU, C E. coli ATCC 25922, D P. aeruginosa CIP27853). The bacteria survival rate was calculated by normalizing bacteria load against control without NM102. The results reported are mean ± SD of four independent experiments each in triplicates, P values are calculated against the condition without NM102, using One Way ANOVA (****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05, the exact P values are provided as source data). IC₅₀ was computed using Graph Pad 7.05. Source data are provided as a source data file.

When further investigated in the presence of other important bacterial pathogens, NM102 inhibited bacterial survival of clinical E. coli (IC₅₀ of 61.2 µM, Fig. 5C) and, although to a lesser extent, of P. aeruginosa (IC₅₀ of 94.2 µM, Fig. 5D) under NO stress conditions. These data show that NM102 acts against Gram-negative pathogenic bacteria under conditions that mimic immune stress.

NM102 is effective in vivo

To assess the antimicrobial activity of NM102 as an antimicrobial in vivo, Bombyx eri larvae were injected with P. aeruginosa CIP27853 and treated with NM102 (Fig. 6A). The insects were then crushed and Colony-Forming Units (CFU) were counted by plating. Treatment with NM102 resulted in a drastic 2.5 log reduction in bacterial load, showing that NM102 is effective against P. aeruginosa in vivo.

Fig. 6: Antimicrobial effect of NM102 in vivo in insects and mice.

A P. aeruginosa CIP27853 wt and ∆mfd strains were grown to late exponential phase in LB medium at 37 °C and injected into forth instar silkworm larvae without or with NM102 (2.78 µg/larvae). Boosts of NM102 (6.97 µg/larvae) were administrated at 4 h and 8 h post-infection. At 24 h post-infection larvae were crushed and the content of each tube was serially diluted on LB plates for CFU numeration. Each graph represents the values obtained for n insects with the mean value indicated. Each dot represents one larva (Pa wt n = 10, Pa wt +NM102 n = 10, Pa ∆mfd n = 15, Pa 3/18/25mfd + NM102 n = 15). P values are calculated against the condition wild type-strain without NM102, using Mann-Whitney test (p < 0.0001). For mice experiments, K. pneumoniae ATCC 700603 wt and ∆mfd (B), K. pneumoniae ALE (C), P. aeruginosa CIP27853 (D) were grown to late exponential phase and diluted in PBS. Intranasal (i.n.) bacterial administration was performed through slow instillation of 20 µL of bacterial suspension (1 × 10⁷ CFU). 200 µL of NM102 nanoformulation (NF-NM102 at 1.5 mg/kg or 6 mg/kg) or empty nanoformulation (control) were administrated in mice via the intraperitoneal (i.p.) route (B). Alternatively, 20 µl of a mixture containing bacterial suspension (1×10⁷ CFU) and NM102 nanoformulation (1.5 or 6 mg/kg) or empty nanoformulation was administered via the i.n. route (C, D). MPN (10 mg/kg) was inoculated via the i.p. route. Mice were sacrificed by cervical dislocation after 24 h and the log CFU in the lung was calculated per gram of organ. Each graph represents the bar plot obtained for n mice with mean value indicated: (B) CTR n = 9, NM102 1.5 mg/kg n = 4, NM102 6 mg/kg n = 9, MPN 10 mg/kg n = 6, ∆mfd n = 6, ∆mfd + NM102 n = 10; C CTR n = 8, NM102 1.5 mg/kg n = 8, NM102 6 mg/kg n = 8, MPN 10 mg/kg n = 8; (D) CTR n = 7, NM102 6 mg/kg n = 7, MPN 10 mg/kg n = 8, NM102 1.5 mg/kg+ MPN n = 8, NM102 6 mg/kg + MPN n = 7. P values are calculated using a two-tailed Mann-Whitney test (***p < 0.0005; **p < 0.001; *p < 0.005). The exact P values, minima, maxima, center, bound of box, and whisker are all provided as source data. Source data are provided as a source data file.

To confirm that NM102 targets Mfd in vivo, the effect of NM102 was investigated on the survival of the P. aeruginosa mutant ∆mfd. First, the amount of CFU of the ∆mfd mutant was lower compared to the CFU recovered following insect infection with the wild-type strain, showing that Mfd is a critical virulence factor for P. aeruginosa. Second, NM102 did not decrease the bacterial load of the ∆mfd mutant, confirming that NM102 is specifically active on Mfd in vivo. It is interesting to note that NM102 decreased bacterial load to the same level that could be reached with the ∆mfd mutant, highlighting that NM102 had a maximum of activity, completely inhibiting the Mfd activity, similarly to a complete deletion of the target.

NM102 nanoformulation is a potent antimicrobial agent active in vivo against clinically-relevant pathogens

To further investigate the potential of NM102 in vivo, a mouse lung model of infection with K. pneumoniae and P. aeruginosa was chosen²³. Due to its poor solubility in most common solvents (Suppl. Table 2), the above studies were performed using a solution of NM102 in DMSO. However, DMSO is known to be toxic to mice at high concentrations²⁴. Thus, an alternative to the DMSO vehicle was developed to further assess the therapeutic effect of NM102 and to offer a more viable medication. Herein, a DMSO-free nanoformulation was prepared via a sonication assisted-nanoprecipitation method in the presence of poly(d-L lactide-co-glycolide) (PLGA) (Suppl. Fig. 5A-D). The resulting NM102 nanoformulation (drug content = 0.6 mg/mL) exhibited an average diameter of 180 nm, with narrow particle size distribution (polydispersity index, PDI = 0.2) and a neutral surface charge (ζ potential = −2 ± 1 mV). An “empty” nanoformulation, used as a control, was prepared according to the same protocol but without the drug (average diameter = 136 nm ± 7, PDI = 0.08 ± 0.04, ζ potential = −4 ± 1 mV). The NM102 nanoformulation showed a good colloidal stability for up to 2 weeks at both 4 °C and room temperature (Suppl. Fig. 5E, F).

Mice were infected intranasally with K. pneumoniae and treated via the intraperitoneal route with the NM102 nanoformulation. After 24 h, the CFU in the lung was assessed (Fig. 6B). The NM102 nanoformulation significantly reduced the bacterial burden in lungs compared to control group treated with empty nanoformulation in a dose-dependent manner (by 4 times with 1.5 mg/kg of NM102 and by 15 times with 6 mg/kg of NM102, corresponding to 75% and 94 % reduction of the bacterial load, respectively), showing that NM102 nanoformulation acts as an efficient antimicrobial against K. pneumoniae infection in vivo. The efficacy of the NM102 molecule was compared under the same conditions to that of meropenem (MPN), a beta-lactam antibiotic that is commonly used in the treatment of various infections including K. pneumoniae and P. aeruginosa²³. NM102 nanoformulation showed similar efficacy to MPN in reducing the K. pneumoniae CFU in the lung of the infected mice at 6 and 10 mg/kg, respectively (Fig. 6B), resulting in 94% and 93% reduction of the bacterial load, respectively (Suppl. Fig. 6).

To confirm that NM102 targets Mfd in vivo, mice were also infected with the K. pneumoniae ∆mfd mutant. As expected, the ∆mfd mutant was less virulent than the wild type strain. MPN inhibited the bacterial burden of the ∆mfd mutant (Suppl. Fig. 6). By sharp contrast, the NM102 nanoformulation showed no effect on the bacterial burden of the ∆mfd mutant, further highlighting the inhibitor specificity toward the Mfd target in vivo (Fig. 6B, Suppl. Fig. 6). NM102 decreased the CFU recovered of the wild type strain to the level of CFU obtained with a ∆mfd mutant, and NM102 had no impact on the ∆mfd mutant, which further highlights that NM102 acts on Mfd in vivo.

The efficacy of NM102 was further investigated on the clinical K. pneumoniae ALE strain, which is resistant to several antibiotics including MPN (Fig. 6C, Suppl. Table 3). The NM102 nanoformulation at 6 mg/kg drastically decreased by 23 times the bacterial load in the lung (75% reduction of the bacterial load). Strikingly, MPN at 10 mg/kg had no significant effect on the resistant K. pneumoniae ALE, highlighting the potential of the NM102 molecule on resistant Gram-negative bacteria in vivo.

For P. aeruginosa, the NM102 nanoformulation at 6 mg/kg decreased by 14 times the mean bacterial load (although not statistically significant), and MPN at 10 mg/ml decreased the bacterial load by 26 times (Fig. 6D). The combination of the two treatments was drastically efficient against P. aeruginosa in an NM102-dose dependent manner compared to single treatments (decrease by 127 times for NM102 1.5 mg/kg + MPN 10 mg/kg, and by 730 times for NM102 1.5 mg/kg + MPN 10 mg/kg). These data suggest the possible use of NM102 nanoformulation as an antimicrobial drug during combination therapy.

Overall, NM102 proved to be an antimicrobial molecule active against clinically-relevant pathogens, including multidrug-resistant bacteria that may be associated with difficult-to-treat infections and sometimes therapeutic failure.

NM102 is not toxic for the host

In the development of anti-infectives, an important feature that needs to be assessed is their impact on the host. When tested for toxicity towards human cells (Fig. 7A), NM102 and the NM102 nanoformulation did not induce cytotoxicity either to HeLa or to Vero cells at the highest doses used in insect and mouse experiments, respectively. MPN showed no toxicity to HeLa cells, but induced cytotoxicity in Vero cells at 10 and 100 µM.

Fig. 7: NM102 is not toxic to the host.

A HeLa and Vero cells were cultured in DMEM at 37 °C and 5% CO₂. Cells were treated for 1 h with MPN at 10 and 100 µM, NM102 at 10 and 300 µM, NM102 nanoformulation at 1.29 and 5.16 mM or their respective control 10% DMSO or empty nanoformulation. Cytotoxicity was assessed using CellTiter96®AQ_ueous. Values for treated cells were normalized to the untreated control. The results reported are mean ± SD of three independent experiments each in triplicates. B Mice were i.p. treated with NM102 (6 mg/kg) or control. Changes in the mice body weight were assessed for 7 days. Source data are provided as a source data file.

Treatment with NM102 had neither impact on insect or mouse survival nor on the mice weight (Fig. 7B). Importantly, PLGA, which was used for the nanoformulation of NM102, is an FDA-approved copolymer for which neither systemic nor organ toxicity has been reported in animal models^25,26.

NM102 may partially protect the host microbiome

As NM102 is active on bacteria only in the context of a stress-like inflammation, we hypothesized that following pulmonary infection, the inflammation will be localized at the site of infection and that NM102 treatment should not affect the distant gut microbiota species, in contrast to treatment with conventional antibiotics.

To assess the innocuity of NM102 towards the gut microbiome in vivo, mice were first treated with NM102 without bacterial infection, and feces were recovered after treatment. The state of the microbiome was assessed by quantifying the overall quantity of bacteria in the feces by qPCR (Supp Fig. 7). NM102 did not impact the overall quantity of bacteria in the feces, after 1 and 7 days of treatment. By contrast, streptomycin decreased sharply the overall quantity of bacteria in the feces after 1 day.

We further assessed the composition of the gut microbiome following nasal infection by K. pneumoniae and NM102 nanoformulation treatment. We did not detect changes in gut microbiome diversity between untreated and NM102-treated mice. The family-level microbiome profile, following NM102 treatment, largely resembled that of untreated mice (Fig. 8A). The Shannon indexes of empty nanoformulation and MN102 nanoformulation-treated mice were similar to untreated mice (Fig. 8B). Similarly, we did not detect significant differences in beta-diversity between treatments using weighted and unweighted UniFrac distances (Fig. 8C). These results indicate that the nasal NM102 treatment does not induce changes in the richness or the structure of the gut microbiome.

Fig. 8: NM102 does not impact the gut microbiome.

Gut microbiome diversity is unaffected by NM102. Mice were infected with K. pneumoniae and left untreated or treated with NM102 nanoformulation (NF-NM102) or the empty nanoformulation (NF) as control. After 24 h, feces were collected and the microbiome diversity was analyzed by 16S rRNA gene sequencing. Relative abundance of bacterial families (A), Shannon index (B), and weighted and unweighted UniFrac distances (C) were determined. The bar colors represent bacterial families with mean abundance >1%. Each dot (B, C) and each column (A) represents the value obtained for one mouse. Error bars in Fig. 6B represent mean ± SEM. The sample size in Fig. 6 A, B, C is n = 7 mice per group. The statistical test in Fig. 6B is permutational anova (PERMANOVA). The statistical test in Fig. 6C is a two-tailed Wilcoxon test with Benjamini-Hochberg correction for multiple comparisons. Source data are provided as a source data file.

NM102 is an anti-evolvability drug to combat AMR

Mfd has been shown to play a critical role in promoting bacterial resistance to commonly used antibiotics and has been proposed as a target for the development of anti-evolvability drugs¹⁸. We thus assessed whether NM102 could also inhibit Mfd evolvability activity. First, the mutation rate of E. coli wild type, ∆mfd mutant, and ∆mfd complemented strains was assessed following exposure to rifampicin that commonly induces a high frequency of resistance in vitro (Fig. 9A). The mutation frequency of the ∆mfd mutant was significantly reduced compared to that of the wild type strain, confirming the role of Mfd in the induction of mutations. Second, the strains were treated with NM102 resulting in a significant reduction in the mutation frequency of the wild-type strain to a level similar to that of the ∆mfd strain. NM102 had no impact on the mutation frequency of the mutant strain, further highlighting its specificity to the Mfd target. To assess for potential polar effects of the mfd mutation, we complemented the E. coli ∆mfd strain with the entire mfd gene. An intermediate value of mutation frequency was obtained for the complemented strain, indicating that the complementation is likely partial, which is not surprising due to the use of an inducible promoter. More importantly, treatment with NM102 led to a statistically significant decrease in the mutation frequency, consistently with the phenotype of the wild-type strain, showing that the complementation restored mfd function.

Fig. 9: NM102 inhibits Mfd evolvability activity.

A The mutation rate of E. coli wt, Δmfd, and Ω∆mfd/mfd strains were measured following exposure to NM102 (100 µM) or the DMSO control, using the frequency of spontaneous accumulation of resistant mutants to rifampicin as reference measurement. Data in this figure correspond to 12 independent cultures. Error bars show SEM. Lea Coulson analysis was used on webSalvador to determine significance using an efficiency parameter of ε = 0,5 (^∗p < 0.05; ^∗∗p < 0.01; ^****p < 0.001, the exact P values are provided as source data). Resistance evolution of E. coli to rifampicin (B) and streptomycin (C) was measured following exposure to NM102 (100 µM) or the DMSO control. Line plots show the median antibiotic concentration at each sampled time point for three independent experiments. Source data are provided as a source data file.

Then, the impact of NM102 on the evolution rate was evaluated following treatment of E. coli with several antibiotics. NM102 significantly decreased the evolution rate of E. coli following treatment with rifampicin (Fig. 9B) or streptomycin (Fig. 9C) by eight and two folds, respectively. These data strongly suggest that NM102 targets the evolvability activity of Mfd and could thus be used as a combinatory drug to decrease antimicrobial resistance.

Discussion

To address the need for novel bacterial drug targets and the design and synthesis of unique anti-infectives with new mechanisms of action, especially towards Gram-negative bacteria, we have identified and characterized an innovative bacterial target, the Mutation Frequency Decline protein, Mfd, and promoted its inhibition by a newly discovered therapeutic molecule. Mfd is a nonessential transcription repair coupling factor that is ubiquitous and conserved in bacteria but absent in eukaryotes. In this study, we identified a molecule, NM102, as an inhibitor of Mfd activity. We showed its dual capacity, first to display antimicrobial activity in animals against K. pneumoniae and P. aeruginosa, two major nosocomial Gram-negative bacteria of the ESKAPE group, and second to inhibit Mfd evolvability function, thus reducing the frequency of antibiotic resistance appearance. As such, NM102 is active on its own to prevent bacterial infections. Furthermore, it is also predicted to strongly reinforce the efficacy of existing antibiotics and prevent the emergence of resistance, an aspect rather unique and essential in the field of drug development and antibiotic therapy.

Since Mfd is not an essential protein, standard minimum inhibitory concentration (MIC) assays could not be used to identify inhibitory molecules. Therefore, to identify the NM102 molecule, we developed a targeted approach based on the 3D structure of the Mfd’s active site. This rational approach allowed screening of a large library of molecules to identify potential inhibitory molecules. Hence, we propose a strategy that can be applied to other non-essential bacterial targets, providing a rapid and efficient pathway to screen the potential of small molecule without the need to develop laborious and specific in vitro activity tests. This structure-based method allowed us to predict the druggability of the target based on the 3D structural descriptors (i.e., polarity, hydrophobicity, volume) of ligand-bound cavities in the Mfd ATP-binding site²⁷. The main advantage of such method is its high interpretability regarding pocket properties. Additionally, once a potentially druggable pocket has been identified, it can be screened using various in silico tools to propose potential ligands for experimental validation and further optimization.

We chose to target the ATP-binding domain of Mfd to identify inhibitory molecules because: (i) ATP plays a central role in Mfd activity; (ii) the 3D-structure of the ATP-binding site structure in Mfd is known; (iii) ATP-binding sites, despite their large structural diversity, are known to be druggable, as evidenced by inhibitors of many targets (e.g., protein kinases, HSP-90a)²⁸; and (iv) although a large proportion of proteins have ATP-binding sites, these sites exhibit significant structural diversity, resulting in distinct specificity and selectivity for ligands²⁸.

We report that NM102 is a potent ATP-competitive inhibitor in silico and in vitro. In silico, its molecular mode of interaction explains how NM102 is competitive and docks into the ATP pocket with greater affinity than ATP. Notably, its length could reinforce the interaction between D5 and D6 domains, which is expected for ATPase activity, through direct and closer interaction with the Q664-H665 stretch and an ancillary salt bridge between the D876 and R905 residues. Consistently, NM102 competed with ATP in vitro. Moreover, such interaction could trap Mfd in its active ATPase conformation and prevent it from engaging in the subsequent DNA translocation step. Indeed, Mfd undergoes a functional cycle to couple ATPase to DNA translocation and then switch to UvrA recruitment. NM102 interacts strongly through polar interaction with the catalytic residue E730, whose mutation into glutamine is defective in ATP hydrolysis. Furthermore, E730 is sandwiched between R733 and R953, and R953A has been reported to result in the loss of DNA translocation⁹. Any modification of interaction within the E730 nexus, which connects both ATPase and DNA translocation activities, could impact the coordination of the two activities, which must be exquisitely tuned. Notably, the pseudo-adenine part of NM102 conserves π-stacking with F599, similar to that seen in the adenine moiety of ATP.

Furthermore, we have mutated the four residues F597A, F599A, K634A, and E730Q that are possibly critical in all species for the binding of NM102 in the ATP binding pocket of Mfd. Indeed, K634A and E730Q were previously described as important for ATP binding⁹ and F597A and F599A are evidenced in this work as crucial for the recognition of the indole-like ring of NM102. Neither ATP nor NM102 could correctly be positioned into the mutated Mfd ATP pocket, further demonstrating the specific binding of NM102 in the ATP binding pocket of Mfd. These data provide mechanistic and molecular insights into NM102 binding with respect to its target and specificity.

Regarding specificity, NM102 neither binds nor inhibits the ATPase activity of various eukaryotic ATPase proteins, including closely functionally related ATPase proteins involved in transcription-coupled nucleotide excision repair with ATP-dependent DNA helicase activity. These data further suggest that the ATP pockets of ATPase proteins exhibit structural diversity, allowing for distinct ligand specificity. The absence of toxicity of NM102 towards bacteria (in the absence of exogenous stress) and eukaryotic cells, further suggests that NM102 does not affect the activity of general ATPase proteins.

Mfd activity enables bacteria to overcome the host’s defense responses²⁰. Host defense against bacteria is predominantly mediated by cellular immune mechanisms. Various cells, such as macrophages, neutrophils, and epithelial cells produce toxic species, including NO, during most infections. NO can severely damage biological molecules, such as proteins and nucleic acids, thereby inducing DNA damage and strand breaks²⁹. NO is cytotoxic and mutagenic for various pathogens and host cells^30,31,32, and it plays an important role during infections by limiting microbial proliferation²². Bacteria express sensor proteins that can detect NO and, accordingly, activate the expression of enzymes that detoxify NO before it reaches lethal levels^33,34. Additionally, we previously demonstrated a link between NO-induced bacterial DNA damage and DNA repair by Mfd²¹. Mfd is ubiquitous in bacteria, and almost all bacteria induce an NO response from the host, so the Mfd-targeted inhibitor is expected to have a broad spectrum of activity. We have consistently shown that NM102 docks in silico to the Mfd of all ESKAPE pathogens and has antibacterial activity against Gram-negative bacteria under NO stress. As a proof of concept for efficacy against Gram-positive bacteria, we also confirmed that NM102 has antimicrobial activity against B. cereus under NO stress, the same B. cereus in which we initially identified the function of Mfd during immune stress (Suppl. Fig. 8).

Mfd activity is not limited to NO stress. Indeed, Mfd has been shown to protect against oxidative stress in Bacillus subtilis³⁵. This may explain why NM102 was efficient in vitro during ATPase assay, with a relatively low IC50 value, and also effective in vivo, while exhibiting higher IC50 values during in vitro NO stress, which may not fully represent the conditions encountered by bacteria during in vivo infections. The ability to evolve is critical for bacterial survival, particularly in the context of pathogenesis, where evading host immunity is essential and requires constant adaptation. Therefore, Mfd may be involved in bacterial virulence, likely by preventing DNA damage induced by the host through its overall immune response and/or stress conditions encountered in vivo. This question is difficult to address technically in vivo as bacteria encounter many different stresses during infection. However, we can speculate that blocking Mfd activity with the inhibitory molecule hinders the bacteria’s ability to cope with the host response (not only NO stress), allowing the immune system to effectively combat the invading pathogens. Another hypothesis could be the role of Mfd during the modulation of transcription. Indeed, it has been recently suggested that Mfd acts as a transcription factor able to modulate the activity of RNAP and gene expression. These activities may also explain the observed effect of Mfd in stressed bacteria^9,16,36. Besides, in C. difficile, Mfd deficient cells overproduce toxins involved in virulence³⁷. Thus, inhibition of Mfd may not be effective in the treatment of some pathogens.

We then focused our study on the impact of NM102 on the mutagenic property of Mfd to assess its role in antibiotic resistance. Indeed, Mfd has been shown to be a general evolvability factor promoting bacterial mutagenesis during infection⁹. Mfd is reportedly required for developing high levels of drug resistance upon initial exposure to sub-inhibitory concentrations of antibiotics, accelerating AMR development¹⁸. Additionally, stress-induced mutagenesis can help pathogens generate drug-resistant cells during antibiotic therapy³⁸. Here we show that NM102 inhibits Mfd’s function as a mutation and evolvability factor during antibiotic stress. This decreases the ability of bacteria to develop resistance to classical antibiotics. As such, NM102 can target the evolvability property of Mfd. By decreasing the resistance pressure, these new drugs may have a longer efficacy, which in turn will help limit the spread of antibiotic resistance. Interestingly, a small molecule, ARM1, has very recently been identified as an inhibitor of Mfd-dependent mutagenesis, further demonstrating the potential of fighting AMR by targeting this protein involved in the acceleration of evolution leading to increased AMR^19,39. The mechanisms leading to Mfd inhibition are different between NM102 and ARM1. Indeed, in contrast with NM102, ARM1 did not bind to the ATP binding site of Mfd and had no impact on Mfd ATPase activity¹⁹.

Combination therapies have played and will continue to play an important role in clinical treatment. We have shown that NM102 is effective as antimicrobial in vivo, both alone and in combination with meropenem, and that NM102 treatment can kill bacteria that are resistant to current antibiotics. Thus, a combination treatment, including an Mfd inhibitor coupled with a classical antibiotic, could be a promising strategy with increased antimicrobial efficacy and a reduced likelihood of resistance development at the onset of treatment. As recently proposed, targeting the evolutionary capacity of bacteria could also have broad implications beyond AMR development, from reducing cancer evolution to limiting pathogenic diversity in the context of host immunity¹⁸.

Drug loaded into nanocarriers has been widely proposed for the delivery of poorly soluble drugs, not only overcoming the drug’s physicochemical limitation but also ensuring enhanced accumulation at the site of action and sustained drug release^40,41,42,43. The design of an NM102-loaded nanoformulation helped overcome the issues related to low solubility of the drug molecule, allowing its administration in the best possible clinically acceptable solvent (i.e., water). Herein, we provide proof of the efficacy of the aqueous nanoformulation of NM102, paving the way for developing a novel antibiotic formulation suitable for widespread use.

Classical antibiotics target essential bacterial pathways and, as such, are highly efficient but not specific, thus also affecting commensal members of the microbiome⁴⁴. The gut microbiome is essential for human and animal health, as it is involved in various health-associated processes, including digestion, metabolism, and immune development, and it also protects against invading pathogens. Since Mfd is not an essential protein, its inhibition by NM102 should not impact the distant host microbiome in the absence of inflammatory stress. To test this hypothesis, we assessed the gut microbiome composition after NM102 treatment in healthy mice and mice infected intranasally with K. pneumoniae. NM102 had no detectable effect on the diversity of the gut microbiome, which can be collaterally damaged with classical, non-selective antibiotics. This is a unique feature that may help develop microbiome-friendly molecules that not only treat infectious diseases but also promote the overall health of patients. However, in enteric infections, local gut inflammation may make Mfd essential for the survival of bacterial commensals as well. Therefore, in those cases NM102-like compounds may also negatively impact the gut microbiome, causing collateral damage similar to that of current antibiotics, but with the advantage still of being effective in combating resistance.

Besides, since Mfd is a non-essential protein, targeting it should decrease the resistance pressure by NM102. NM102 by itself does not kill bacteria and therefore, resistance is not observed in the absence of any stress. In vivo, during inflammation, Mfd is essential and NM102 treatment hinders bacterial survival. This is likely to cause bacterial resistance eventually. Nonetheless, we can speculate that this resistance will be reduced in comparison to traditional antibiotics, which apply continuous pressure to both pathogenic and microbiota bacteria over the course of treatment. More research will be required on this topic.

Overall, new drugs targeting Mfd, a non-essential virulence and evolvability factor, could be used alone or as adjunctive therapy during the treatment of infections to improve the potency of current antibiotics and reduce the emergence of resistance. We are confident that these innovative molecules could provide a promising therapeutic option and expand the arsenal of drugs available to combat AMR and potentially other diseases.

Methods

Ethics statement

All animals were handled in strict accordance with good animal practice as defined by the local animal welfare bodies (Unité IERP, INRAE Jouy en Josas, France, Agreement No. 78120), and all experiments were approved by the COMETHEA ethics committee and the French Ministry of Higher Education and Research (APAFIS#10124-2017040413027917 v12). All animal experiments were performed in accordance with European Directive 2010/63/EU. All efforts were made to minimize animal suffering and adhere to the 3Rs principle (Reduce, Refine, Reuse).

E. coli Mfd modeling

As we aimed to explore the capacity of Mfd to bind ligands that could be competitive with ATP, a prerequisite was to model Mfd in a conformation conductive ligand binding, meaning that the ATP-binding site must adopt an open and accessible conformation. At the time of computation, all available crystal structures of E. coli Mfd were in an inactive form due to the so-called walker A motif closing over the active site, thus precluding any binding (pdb id 2EYQ)¹². The D2 domain of RecG helicase from Thermotoga maritima (pdb id 1GM5) is a structural homolog of Mfd, solved in an active conformation complexed with ADP⁴⁵. Thus, the nucleotide-bound active form of E. coli Mfd (K578-P780, UniProt identifier P30958) was obtained by morphing the inactive form of E. coli Mfd into the active form of RecG using the default settings of the Yale Morph2 server (http://morph2.molmovdb.org/submit.html).

High-throughput in silico screening

The Bioinfo-DB database (http://bioinfo-pharma.u-strasbg.fr/bioinfo) is an in-house developed database of 4.8 million commercially available compounds as powder (1–50 mg) filtered according to internal rules to contain only drug-like compounds. The database was first filtered to retain compounds grossly resembling nucleotides and fulfilling the following properties: (i) at least one hydrogen-bond donor, (ii) at least two hydrogen-bond acceptors, (iii) at least one aromatic ring, (iv) predicted aqueous solubility higher than 50 µM (predicted with PipelinePilot v.9.5, Dassault Systèmes, Paris), (v) topological polar surface area lower than 120 Ų. The filtered set of 1.2 million compounds was then converted into three-dimensional (3D) structures using Corina v3.40 (Molecular Networks, Erlangen, Germany). Up to four stereoisomers were created for compounds with undefined stereocenters. Stereocenters that were explicitly defined were left unchanged. Altogether, 3D atomic coordinates were defined for 1,874,034 compounds, constituting the docking set.

The docking set was anchored to the above-described nucleotide-binding site of E.coli Mfd (Phe599, Thr602, Gln605, Gly631, Phe632, Gly633, Lys634, Thr635) with the Surflex-Dock v.3066 program⁴⁶. A prototyping molecular dynamics object or “protomol” was first generated from the list of binding site residues. Compounds were then docked using default settings (excepted for the -pgeom option) of the docking engine, retaining the best ten poses according to the native Surflex-Dock scoring function.

Potential hits were selected according to the following two strategies (Suppl. Fig. 1):

Strategy A

The 178 best-scoring poses (Surflex-Dock score > 10) from the docking set were retained. Then, compound redundancy was removed (the highest score was retained if more than two poses originated from the same compounds) to yield 91 compounds. A chemical diversity selection by maximum common substructures was performed using the LibMCS algorithm (ChemAxon Ltd., Budapest, Hungary) to retain a first set (SET1) of 23 chemically diverse compounds.

Strategy B

All poses from the docking set were submitted to an interaction-based filter⁴⁷ to select 206 non-redundant compounds that absolutely fulfilled the following interactions: hydrogen bond to Gln605.OE1 atom, hydrogen bond to Gln605.NE2 atom, hydrogen-bond to Gly631.N atom, π–π aromatic stacking to Phe599. If multiple poses from the same compound fulfilled the interaction similarity filter, the top scored posed (best Surflex-Dock score) was kept. A chemical diversity selection by maximum common substructures was performed using the LibMCS algorithm (ChemAxon Ltd., Budapest, Hungary) to retain a second set (SET2) of 74 chemically diverse compounds.

Previously defined SET1 and SET2 were merged, yielding a final selection of 95 unique hits (two hits were common to both sets) that were further purchased in 5 mg quantities.

Homology modeling of Mfd from the ESKAPE species

Mfd sequences from the ESKAPE pathogens were retrieved from NCBI. They were aligned using the multiple sequence alignment MAFFT algorithm with default parameters⁴⁸. The alignment was visualized using Jalview 2⁴⁹.

For homology modeling, the recent structure of E. coli Mfd in the so-called L1 state and in complex with the RNA polymerase was chosen as template because L1 state-Mfd adopts an active and opened conformation, that renders the ATP pocket accessible. The L1 state was solved at 4.6 Å resolution, using cryo-electronic microscopy; its pdb id is 6X26⁵⁰. Based on this template, one hundred models were computed for each Mfd of the ESKAPE bacteria, in their wild-type form, using the homology building software Modeller version 9.18⁵¹. Each model was ranked and selected upon the lowest DOPE and function scores of Modeller. Eventually, the stereochemistry of selected model was checked using Procheck⁵⁰.

Modeling of mutated Mfd

To obtain Mfd with the mutations F597A, F599A, K634A, and E730Q, the best model of non-mutated Mfd was iteratively substituted four times, using the Pymol command mutagenesis. For the mutation of glutamate into a glutamine residue, the rotamer of the side-chain was selected to give a conformation compatible within the protein and avoid steric clashes.

NM102 docking

ATP and NM102 were individually docked into the active site of each modeled Mfd, in L1-state conformation, with respect to the homology modeling described above. AutoDock Tools 4.2 (ADT4) was used with a grid box conserved in size and position, centered onto the active site of the ATPase, with the genetic algorithm of Lamarck and the default parameters for 20 runs. ATP and NM102 coordinates were obtained from the ZINC database (https://zinc.docking.org/substances/home/) in mol2 format, converted into ADT4’s PDBQT format with their dihedrals set free to rotate.

For each run of each complex, an affinity score was calculated. The pose of the ligand, with the best interaction energy measured in kcal.mol⁻¹, was selected because it is the most probable one. Finally, holo models were visually inspected using PyMOL (PyMOL 2.0.7. Schrödinger, LLC), and protein-ligand interactions were characterized using the Protein–Ligand Interaction Profiler (PLIP) program⁵². The same protocol was used for Mfd mutated at F597A, F599A, K634A and E730Q.

Bacteria

K. pneumoniae DOU CTXM-15, ALE CTXM-15+, P. aeruginosa CIP27853, E. coli ATCC25922 were provided by Dr. Thierry Naas (CHU de Bicêtre, Le Kremlin-Bicêtre, France). The MIC of antibiotics was determined on those strains. Briefly, a 0.5 McFarland standard bacterial solution was prepared in physiological saline and then diluted 1/1000 in Mueller Hinton broth. Then, 50 µL of bacterial solution was added in 96 well plates containing 100 µL of various concentrations of antibiotics (Penicillin, Ampicillin, Ceftriaxone, Meropenem, Cefotaxin, Oxacillin, Streptomycin, and Ciprofloxacin). The plates were incubated overnight at 37 °C. Bacterial growth was measured in each well to determine the MIC (Suppl. Table 3).

E. coli K12 wild-type and ∆mfd mutants were previously described⁵³. The complemented E. coli K-12 Δmfd Ωmfd mutant strain was obtained by inserting into the E. coli K12 Δmfd mutant strain the complementation plasmid pET21a containing the entire mfd gene (synthesized by Biobasic, genetic environment Suppl. Fig. 9). Briefly, the receptor strain was suspended in 100 µL of a buffer containing LB, 10% PEG4000, 5% DMSO, MgCl₂ 0.01 M, MgSO₄ 0.01 M and mixed the complemented plasmid. A heat shock was performed and complemented strains were selected after regeneration on LB agar plates supplemented with 100 µg/mL of ampicillin. The corresponding mutant was named E. coli Δmfd Ωmfd.

All Strains were stored at −80 °C as 20% (v/v) glycerol stocks.

Construction of ∆mfd mutants

The P. aeruginosa CIP27853 ∆mfd mutant strain was constructed by double-crossover region deletion. Briefly, using the available sequencing information of P. aeruginosa CIP27853 strain, approximately 1-kb regions upstream (region 3360879 to 3361826) and downstream (region 3363535 to 3364310) of the mfd gene were synthesized by Genecust (Boynes, France). The two fragments were cloned and juxtaposed into the suicide vector pEXTK by Genecust. The resulting plasmid, pPa001, was transformed into electrocompetent E. coli SM10 λpir and conjugated into P. aeruginosa CIP27853 strain as previously described (Hmelo et al., 2015). Briefly, donor and recipient strains were mixe 1:2 on a sterile 0.05 μm-pore-size-membrane (VMWP02500 Milipore) placed on LB agar plates supplemented with irgasan 25 µg/ml and gentamicin 75 µg/ml and incubated for 18 h at 37 °C. After mating, bacteria were resuspended from filter in LB with IPTG 1 mM, incubated for 3 h at 37 °C, and plated on LB agar plates with Azithromycine (AZT) 100 µg/ml and IPTG 1 mM for 18 h at 37 °C. Transconjugants were selected for loss of gentamicin resistance by picking individual colonies on LB agar and LB agar supplemented with 75 μg/mL gentamicin at 37 °C. Deletion of the mfd gene by double recombination was verified by PCR using primers upstream (GAACACCACCAGTTCCACCTG) and downstream (GGTAGAGCTGGCCAATCAGC) of the cloned region and by sequencing. The resulting mutant was named P. aeruginosa ∆mfd.

The K. pneumoniae ATCC 700603 ∆mfd mutant strain was constructed by double-crossover region deletion. Briefly, using the available sequencing information of K. pneumoniae ATCC700603 strain, ~1-kb region upstream (region 2056347 to 2057137) and downstream (region 2060189 to 2060810) of the mfd gene were synthesized by Genecust (Boynes, France). The two fragments were cloned on either side of the chloramphenicol resistance cassette (CatR) from the pDK3 plasmid (Addgene) into the pUC57 vector (Addgene) by Genecust. The synthesized DNA fragments and the CatR cassette were digested with the appropriate enzymes and ligated to produce a “mfd-upstream”-“CatR”-“mfd-downstream” KpnI-EcoRI fragment, which was then inserted between the KpnI and EcoRI sites of the pDG704 vector (Addgene). The resulting plasmid, pKp002, was transformed into electrocompetent E. coli SM10 λpir and then conjugated into K. pneumoniae ATCC700603. Briefly, the donor and recipient strains were mixed 2:1 on a sterile 0.05 μm-pore-size-membrane filter (VMWP02500Milipore) placed on LB agar plates for 4 h at 37 °C. After mating, bacteria were resuspended from the filter in 1.5 mL of LB medium. Transconjugants were selected on LB agar supplemented with 25 μg/mL of chloramphenicol at 37 °C. Deletion of the mfd gene by double recombination was verified by PCR using primers located upstream (CCCCAGGATGTTGTTTTGCA) and downstream (CCGTTATCCTGCACCACATG) of the cloned region and by sequencing. The resulting mutant was named K. pneumoniae ∆mfd.

Chemicals

The 95 molecules were sourced from commercial vendors: Enamine, Chembridge, Chemspace, AbamaChem, Asinex, Chemdiv, Interbioscreen, BCH Research and Specs. Each compound was dissolved in DMSO (Sigma-Aldrich) and stored at room temperature until further use. Roswell Park Memorial Institute (RPMI) 1640 GlutaMAXTM medium was purchased from Gibco. The Nitric Oxide (NO) donor, 3-[2-hydroxy-1-(1-methylethyl)-2-nitrosohydrazino]-1-propanamine NOC-5, was purchased from Calbiochem, Sigma-Aldrich. NOC-5 was dissolved in NaOH 0.01 N to a final concentration of 200 mM. NOC-5 has a half-life time of NO release of 25 min at 37 °C. Under these conditions, a stable NO-amine complex can spontaneously release two NO equivalents²². Meropenem (MPN) was purchased from Sigma-Aldrich and dissolved in water to a stock concentration of 25 mM. Poly(lactide-co-glycolide) (PLGA) (Resomer® RG502 H, acid terminated, M_W = 7–17 kDa), poly(vinyl alcohol) (PVA) (87−89% hydrolyzed, M_W = 30−70 kDa), trehalose and all other reagents and solvents were supplied by Sigma-Aldrich (France).

Protein purification

E. coli MG1655 His6-tagged MfdC (residues 451-1148 Uniprot accession code P30958) was synthesized and purified by BioBasic Inc. (Markham, ON, Canada) with a purity >90%. Briefly, the expression plasmid pET21a containing mfdC was transformed into E. coli BL21(DE3), and expression of His6-tagged MfdC was induced with 0.5 mM IPTG at 37°C for 4 h. Cells were lysed, and protein was purified using a Ni-IDA column. Elution fractions were dialyzed against dialysis buffer (2 mM DTT, 50 mM Tris, 300 mM NaCl, pH 8.0) overnight at 4 °C.

For ITC experiment the full-length Mfd protein was synthesized by BioBasic Inc. (Markham, ON, Canada) and purified following the same protocol with a purity >90%.

The XPD/ ERCC2 protein is involved in transcription-coupled nucleotide excision repair and is an integral member of the basal transcription factor BTF2/TFIIH complex. The protein has ATP-dependent DNA helicase activity and belongs to the RAD3/XPD subfamily of helicases⁵⁴. The XPB/ ERCC3 protein is an ATP-dependent DNA helicase that functions in nucleotide excision repair. It is a subunit of basal transcription factor 2 (TFIIH) and, therefore, also functions in class II transcription helicases⁵⁵. The CSB/ERCC6 protein is a DNA-binding protein involved in transcription-coupled excision repair. It has an ATPase activity, interacts with several transcription and excision repair proteins, and may promote complex formation at DNA repair sites⁵⁶. The three purified proteins were purchased from Abbexa Ltd.

Saccharomyces cerevisiae protein yUpf1 HD (residues 221–851 Uniprot accession code P30771, which corresponds to the helicase domain) was kindly provided by Dr Hervé Le Hir (Institut de biologie de l’Ecole normale supérieure (IBENS), Paris, France) and described in ref. ⁵⁷.

Inhibition of ATPase activity in vitro

Mfd-C enzyme activity was evaluated by measuring the quantity of inorganic phosphate (PO4i) released using BIOMOL® Green reagent microtiter-plate assay (Enzo Life Sciences). For the initial drug screening, Mfd-C (0.35 µM) was incubated with DMSO (2% (v/v)) or with the 95 compounds (100 mg/mL) and 1 mM ATP for 10 min at 37 °C. To further evaluate NM102 activity, Mfd-C (0.35 µM) was incubated with NM102 at the indicated concentration (0 to 100 µM). ATPase reaction was measured in Tris buffer 0,05 M at pH 8 in the presence of ATP (0 µM to 0.3 mM) for 30 min at 37 °C. XPD (2 nM), ERCC3 (35 nM), ERCC6 (2 nM), and yUpf1 HD (127 nM) ATPase activities were measured in reaction buffer (20 mM MES pH 6.0, 100 mM potassium acetate, 1 mM DTT, 0.1 mM EDTA, 1 mM magnesium acetate, 1 mM zinc sulfate, and 5% (v/v) glycerol). Then, the proteins were incubated with 2% DMSO or 100 µM NM102. The ATPase reaction was quantified after 30 min of incubation at 30 °C. For all proteins, 50 µL of each reaction medium was transferred into clear, flat-bottom 96-well plates, and the reaction was stopped by adding 100 µL of BIOMOL® Green reagent. Absorbance at 620 nm was measured in a microplate reader (Tecan). Absorbance values were then converted to nmols of released PO4i based on a PO4i standard curve prepared as recommended by the supplier. The potency of each compound was calculated relative to the DMSO control.

The Z′ factor of the ATPase test was calculated using the formula Z′ = 1 − (3σc + +3σc−)/(|μc + −μc − |) where (σc+) and (σc−) are the standard deviations of the high and low reference controls, respectively, and |μc + –μc − | is the absolute difference between the means of the two control.

The IC₅₀ was measured using a plot of Mfd-C ATPase activity versus ATP concentration (0.002–0.3 mM) in the presence of various concentrations of NM102. Data analysis and IC₅₀ calculations were performed using GraphPad Prism 7.05 (San Diego, CA). Lineweaver-Burk plots were generated using SigmaPlot Software.

Isothermal titration calorimetry (ITC)

The affinity constant and thermodynamic parameters of the interaction between NM102 and Mfd were determined by ITC using NanoITC (TA Instruments)⁵⁸. The Mfd protein was dissolved in 0.1 M Tris + 0.2 M NaCl with 4% DMSO at pH = 8 (buffer A) and used in the assay at 20 µM. NM102 and ATP were dissolved in the same buffer and used at 200 µM. For competitive assay, Mfd was incubated with NM102 and Mfd protein interaction with ATP was evaluated. The generation of heat involved during the binding of the ligand to the receptor was measured at 25 °C. Each titration involved 20 injections at 300 s intervals of 2.5 μL aliquots of the compound NM102 solution (200 μM) into the sample cell (volume 200 μL) containing the Mfd solution (20 μM). Solutions were degassed and stabilized at 298 K. The titration cell was continuously stirred at 300 rev/min. The heats of dilution of the ligand in buffer A were subtracted from the titration data.

Fitting was performed using the independent model of the Nano Analyze software (ITC Run) to determine the association constant (K_a), the dissociation constant (K_d), the stoichiometry (n), and the enthalpy change (ΔH). For competitive assays, the ratio A = Ka/Kapp indicates the reduction of the ATP binding in the absence or presence of NM102 inhibitor. Determinations were performed by triplicate.

Nitric oxide (NO) cell-free assay

Bacteria were grown in LB medium at 37 °C with shaking at 200 rpm until the late exponential phase. Bacterial suspensions were prepared in RPMI medium at concentrations between 5 × 10⁴ and 1 × 10⁵ CFU/mL, and 150 μL aliquots were dispensed into a 96-well plate. Bacteria were exposed to NO by adding 50 μL of NOC-5 at final concentrations ranging from 0 to 1000 μM. The bacterial survival rate was quantified by plating on LB agar after 4 h at 37 °C and normalizing the bacterial load in NOC-5-treated samples to that of untreated control samples (0 μM of NOC-5). NO dose–response curves were generated by plotting log10-transformed NOC-5 concentrations and fitting the data to a non-linear model with a variable Hill slope using GraphPad Prism 7.05. The NOC-5 concentrations required to induce 10 and 20% of mortality were calculated using R and the drc package (Christian Ritz and Jens C. Strebig).

For efficacy assays, bacterial solutions were prepared as described above. Bacteria were exposed to 50 µL of NOC5 at a concentration inducing ~80–90% survival compared to the untreated condition (K. pneumoniae DOU 14 µM, P. aeruginosa 5 µM, E. coli ATCC 42 µM), in the absence (0 µM) or presence (5–80 µM) of NM102. Bacterial survival rate was quantified by plating on LB agar after 4 h at 37 °C and normalizing the bacterial load in NM102-treated samples to that of the control samples. The toxicity of NM102 in the absence of NO was assessed by exposing the bacterial suspension to NM102 at concentrations ranging from 0 to 80 µM of NM102 as described above but without NOC-5. The NM102 concentration required to induce 50% of mortality (IC₅₀) was calculated using R and the drc package.

Nanoformulation of NM102 and characterization

NM102 was formulated in the presence of poly(d-L lactide-co-glycolide) (PLGA) according to a previously described method with some modifications⁵⁹. It consists of an ultrasonic emulsification process, followed by dialysis to remove the organic solvent. First, 120 mg of PLGA and 28 mg of NM102 were dissolved in 4 mL of DMSO and then the organic phase was injected through a 23 G-needle into 50 mL of a 0.5% w/v aqueous solution of PVA under sonication in ice (5 min, amplitude 20%, cycle of sonication: 10 s ON and 10 s OFF) using a tip sonicator (Vibra-cellTM–75041, Fisher Bioblock Scientific, Belgium). The organic phase was then removed by dialysis (24 h, room temperature) using a 100 kDa MW cut-off cellulose ester membrane (Thermo Fisher Scientific, France). Empty nanoformulation was prepared according to the same protocol but in the absence of drug in the organic phase. Nanoformulations were stored at 4 °C. Intensity-weighted mean diameters (D_z) and particle size distributions were measured at 25 °C by dynamic light scattering using a Nano ZS instrument (173° scattering angle, Malvern, France) after 1/20 dilution in Milli-Q water (Millipore, France). The surface charge of the nanoformulations was determined by measuring the ζ-potential at 25 °C using the same instrument after 1/10 NP dilution in 1 mM NaCl. All measurements were performed in triplicate. The colloidal stability of the nanoformulations was assessed by measuring the evolution of the mean diameter and the particle size distribution for up to 2 weeks at 4 °C and room temperature.

For storage, 20-mL flat-bottom glass vials were filled with 2 mL of nanoformulation supplemented with 5% w/v of trehalose as a cryoprotectant. Samples were frozen at −20 °C for 12 h and then freeze-dried (24 h) using an Alpha 2–4 LD plus freeze-drier (Martin Christ, Germany) (condenser temperature <70 °C, vacuum <0.1 mbar). Freeze-dried samples were capped, weighted, and stored at 4 °C. Samples were reconstituted immediately before use by the addition of Milli-Q water, followed by gentle homogenization by pipetting and vortexing.

The drug content was determined by direct quantification of NM102 in freeze-dried NM102 nanoformulation samples (without cryoprotectant). Briefly, one sample was dissolved in DMSO (1 mL) by vortexing. The amount of NM102 was then quantified by UV spectroscopy at 282 nm (PerkinElmer UV/VIS spectrophotometer, Germany). The drug concentration was calculated using an NM102 calibration curve in DMSO (correlation coefficient >0.999), as follows: drug content (mg/mL) = weight of drug/volume of nanoformulation dispersion.

Insect assay

Fourth instar silkworm larvae Bombyx eri were purchased from « L’office pour les insectes et leur environnement » (OPIE), Guyancourt, France. Larvae were reared with Ligustrum vulgare or Ligustrum japonicum until they reached a weight of [0,6 –0,9] g. Prior to assays, Bombyx eri larvae were kept under starvation overnight and randomly distributed in groups of 30. P. aeruginosa wt and ∆mfd strains were grown in LB medium at 37 °C under agitation at 180 rpm until late exponential growth phase. Bacterial cultures were then diluted in Phosphate Saline Buffer (PBS). The last dilution to obtain the desired concentration was done in PBS with 10% DMSO with or without NM102. Bacteria were plated right before infection to determine the quantity injected. Larvae were injected with 20 µL of bacterial suspension (3 × 10³ CFU/injection in 10% DMSO) containing NM102 (2.78 µg/larvae) or 10% DMSO in the hemolymph via the last pro-leg. Larvae were placed at 27 °C. At 4 h and 8 h post first injection, larvae were injected with 50 µL of: (i) PBS, 10% DMSO or (ii) PBS, 10% DMSO with NM102 (6.97 µg/larvae) via the last pro-leg on the other side of the larvae. All injections were performed using a 1 mL hypodermic syringe with a 0.5 × 25 mm needle and an automated syringe pump (KD Scientific KDS 100). 24 h post infection, insects were placed in 2 mL sterile tubes containing 500 µL of PBS and 1.4 mm ceramic lysing beads (MP Biomedicals, Solon, OH, USA) and crushed using a Fast Prep instrument (MP Biomedicals, Solon, OH, USA). The bacterial content in each tube was assessed by serial dilutions on LB plates and CFU numeration. The log CFU was calculated per gram of larvae.

Mice infections

For all mice experiments, nine-week-old, specific-pathogen-free, female, strain C57BL, substrain 6J mice were purchased from Janvier Labs (Le Genest-Saint-Isle, France). They were housed with a maximum of five animals per cage and had ad libidum access to food and water. Mice were adapted to the environment of the facility (IERP, INRAE, Jouy-en-Josas, France) one week before the experiments. The experiments were done with 12 h dark/light cycle at ambient temperature and controlled humidity.

Prior to infection, mice were mildly anaesthetized by intraperitoneal (i.p.) injection of ketamine (79 mg/kg) and xylazine (9 mg/kg). Bacterial strains were grown to late exponential phase and diluted in PBS. Intranasal (i.n.) bacterial administration was performed through slow instillation of 20 µL of bacterial suspension (1 × 10⁷ CFU) with a micropipette (10 µL in each naris). Inoculated suspensions were plated on LB agar for CFU determination prior to injection. Empty and NM102-containing formulations were administrated in mice either: (i) via the i.p. route with 200 µL of suspension containing 1.5 or 6 mg/kg (1.29 and 5.16 mM, respectively) of drug or (ii) via the i.n. route with 20 µl of a mixture containing bacterial suspension (1 × 10⁷ CFU) and NM102 nanoformulation (1.5 or 6 mg/kg) or empty nanoformulation. MPN (10 mg/kg) was inoculated via the i.p. route. For combination therapy studies, NM102 nanoformulation and MPN were inoculated via the i.n. and the i.p. route, respectively.

Mice were sacrificed by cervical dislocation after 24 h and the left lung was placed in 2 mL sterile tubes containing 600 μL of PBS and 1.4 mm ceramic lysing beads (MP Biomedicals, Solon, OH, USA). Organs were homogenized using a FastPrep instrument (MP Biomedicals, Solon, OH, USA), and aliquots from each tube were serially diluted on LB plates for CFU enumeration. The log CFU was calculated per gram of organ.

For the toxicity study, mice were treated (i.p., 200 L/injection) with NM102 (6 mg/kg) or DMSO as control, and mice weight was monitored for 7 days.

Cytotoxicity

HeLa (human cervical epithelial cell line, ATCC CCL-2) and Vero (monkey kidney epithelial cell line, ATCC CCL-81)) cells were cultured in Dulbecco’s Modified Eagles Medium (DMEM), supplemented with 10% heat-inactivated fetal bovine serum and 1 U/mL penicillin, and 1 µg/mL streptomycin at 37 °C and 5% CO₂. Once reaching a confluency of 70%, cells were washed with Dulbeccos Phosphate Buffered Saline (DPBS) and detached by treatment with 0.05% Trypsin-EDTA. Total cell number was determined by using a Neubauer Cell chamber and cells were seeded at a concentration of 4 × 10⁴ cells/well in a flat-bottom Tissue Culture 96-well plate. Cells were then incubated at 37 °C and 5% CO₂ overnight before starting the cytotoxicity assay. Two hours before the assay, medium in each well was replaced by an antibiotic-free DMEM medium. Cells were washed and treated with the respective chemical (MPN at 10 and 100 µM; NM102 at 10 and 300 µM; NM102 nanoformulation at 1.29 and 5.16 mM) or their respective control (10% DMSO or empty nanoformulation) in a total volume of 100 µL in Hanks’ Balanced Salt Solution (HBSS) medium (ThermoFischer). After 1 h of incubation, 20 µL of CellTiter96®AQ_ueous was added to each well and cells were incubated again for 1 h. Absorbance was measured at 490 nm after 5 s of shaking to determine cytotoxicity. The background of CellTiter96®AQ_ueous in HBSS and the background of each chemical were subtracted of the respective values. Treated cells were compared and normalized to the untreated control.

Analysis of microbiome diversity

Mice were treated (i.p., 200 µL/injection) with DMSO or NM102 (6 mg/kg). Feces were collected (day 1 and day 7) and the quantity of bacteria was assessed by qPCR as previously described⁶⁰. In addition, microbiota diversity was assessed by 16S rRNA sequencing. Mice were treated (i.p., 200 µL/injection) with DMSO or NM102 (6 mg/kg) after i.n. infection with K. pneumoniae. Feces of untreated and treated mice were collected and stored at −80 °C. Microbial DNA was extracted using the Dneasy PowerSoil Pro Kit following the provider’s instructions (Qiagen, Hilden, Germany). DNA samples were sent to the NGS Competence Center Tübingen (Tübingen, Germany) for quantification, library construction, and sequencing. DNA concentration was quantified using a Qubit fluorometer (Thermo Fischer, Waltham, MA). The V4 hypervariable region of the 16S rRNA gene was amplified using primers F515 (5′-GTGCCAGCMGCCGCGGTAA-3′) and R806 (5′-GGACTACHVGGGTWTCTAAT-3′)⁶¹ and sequenced with the Illumina MiSeq sequencing platform with v2 chemistry.

To process 16S rRNA amplicon sequences, we used the dada2 v.1.21.0 package of R (v.4.2.0)⁶² following its standard operating procedure as available at https://benjjneb.github.io/dada2/bigdata.html. Briefly, after inspecting the quality profiles of the raw sequences, we trimmed and filtered the paired-end reads using the following parameters: trimLeft: 23, 24; truncLen: 240, 200; maxEE: 2, 2; truncQ: 11. The filtered forward and reverse reads were dereplicated separately and used for inference of the amplicon sequence variants (ASVs) with default parameters, after which they were merged in a per-sample basis. We next filtered the merged reads to retain only those with a length between 250 and 256 bp and carried out chimera removal.

We then performed the taxonomic assignment on the final set of ASVs using a curated dada2-formated database based on the genome taxonomy database (GTDB) release R06-RS202⁶³. A maximum-likelihood tree of the ASVs was obtained using the R packages DECIPHER v.2.24.0 and phangorn v.2.8.1⁶⁴ after performing a multiple sequence alignment with each of the representative sequences. We used the PERFect v.1.10 package of R⁶⁵ to identify spurious ASVs, defined as those that made insignificant contributions to the total covariance of the microbiome dataset.

For alpha- and beta-diversity analyses, samples were rarefied to 21300 reads/sample to account for differences in library size. To compare alpha-diversity between samples, we used the Shannon index as implemented in the vegan v.2.6-2 package of R. Global differences in Shannon index between treatments were assessed using the Kruskal-Wallis test and pairwise contrasts between treatments were performed using the Wilcoxon signed-rank test. We used the ASV-level abundance profile to obtain matrices of the weighted and unweighted UniFrac distances with the GuniFrac v.1.6 package of R⁶⁶. We assessed differences in beta-diversity estimates using the adonis function (analysis of variance using distance matrices) of the permutational multivariate analysis of variance (PERMANOVA) on the distance matrices with 9999 permutations, as implemented in the vegan v.2.6-2 package of R. For pairwise PERMANOVA tests, we used the pairwiseAdonis v.0.4 package of R (available at https://github.com/pmartinezarbizu/pairwiseAdonis). P values were adjusted for multiple comparisons using the Benjamini-Hochberg method.

Mutation rate

E. coli Keio wt, Δmfd and complemented ∆mfd/mfd strains were grown onto LB agar O/N at 37 °C. Cultures in LB broth were started from a single colony at 37 °C at 200 rpm until an OD _{600 nm} at around 0.005. Each culture was then divided in two with one culture exposed to 100 µM NM102 and the other to 2% final DMSO as control. Bacteria were incubated until the culture reached OD _600 nm of 1.2, which corresponds to the early stationary phase. Bacteria were plated onto LB agar plates with 50 µg/mL rifampicin and incubated for 24 h at 37 °C. The number of colonies was counted and used to determine the mutation frequency. The total number of bacteria was determined following serial dilution and plating onto LB agar plate. The test was performed 12 times with independent cultures. Lea-Coulson comparisons of rate analysis were performed on Web Salvador to determine significance between conditions, using an efficiency parameter of ε = 0,5 (^∗p < 0.05; ^∗∗p < 0.01; ^****p < 0,001).

Evolvability assays

A serial passage broth microdilution protocol was performed based on previously described methods¹⁸ to identify the effect of NM102 on the evolution of antibiotic resistance. Briefly, the culture of E. coli Keio WT started from a single colony and was diluted to OD _600 nm = 0.005. The bacterial suspension was subsequently transferred into a 96-well microplate (180 μL/well) and incubated with NM102, DMSO 2%, or left untreated. Treated and untreated bacteria were grown in LB with a gradient of concentrations of the indicated antibiotic to select for resistance for 18–24 h at 37 °C under stationary conditions. The OD _{600 nm} was measured and MICs were determined. Serial passage was performed using cells growing (defined by an OD _{600 nm} at a minimum of 50% of the untreated condition) in the highest concentration of antibiotic. Antimicrobial concentrations were adjusted during the process to compensate for the rise in MIC values. This process was repeated for four or five passages. Results were expressed as the median MIC of the three replicates.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.