The gene sequence encoding the membrane protein PmrC was cloned into the pETM13 expression vector. The recombinant vector was transformed in ClearColi BL21(DE3) - Biosearch Technologies competent cells to overpopulate the protein. The ClearColi cells, which have a genetically modified lipopolysaccharide (LPS), were grown in 2.5 L of Luria Bertani (LB) medium (Applichem) medium containing 50-μg ml⁻¹ Kanamycin at 37°C to an optical density (OD) 600 of ∼0.6–0.7. Isopropyl β-d-1-thiogalactopyranoside (IPTG) was added to a final concentration of 0.8 mM, and the culture was incubated overnight at 25°C. All purification procedures were performed at 4°C. The harvested cell pellet derived from 2.5 L of culture was resuspended in lysis buffer (200-mM NaCl, 5% (v/v), glycerol, and 50-mM 2-[4-(2-hydroxyethyl)piperazin-1-yl]ethanesulfonic acid [Hepes], pH 7.4) containing complete protease inhibitor cocktail (Roche Diagnostics, Mannheim, Germany) and was lysed using a digitally controlled high-pressure homogeniser LM20 Microfluidizer® (Microfluidics^TM). The lysate was centrifuged at 10,000 rpm for 10 min, and the supernatant, containing inside-out membrane vesicles (ISOVs), was centrifugated at 23,000 rpm for 2 h. The resultant pellet was solubilised in a buffer containing 200-mM NaCl and 50-mM Hepes (pH 7.4), and 5% (v/v) of glycerol using a Potter–Elvehjem homogenizer. The solubilised suspension was incubated with stirring for 1 h in the presence of 1% n-dodecyl-β-d-maltopyranoside (DDM). The supernatant, enriched with 1 ml of Ni-NTA resin (Qiagen) per 1 mg of protein, was incubated with stirring for 1 h. It was subsequently loaded onto an Econo-Column (Bio-Rad Laboratories) (Bio-Rad) that had been pre-washed with 10 Column volumes of a buffer containing 10 mM imidazole and 0.1% DDM. The membrane protein PmrC with the His tag at the C-terminus was eluted in the same buffer containing 250 mM imidazole. The eluted protein was dialyzed overnight in the presence of a buffer containing 200 mM NaCl, 20 mM Hepes (pH 7.4), glycerol 5, and 0.03% DDM. The protein was concentrated using a 100-kDa cutoff concentrator and injected in a Superdex S200 (Cytiva) 10/300 size exclusion column (GE Healthcare) equilibrated with the dialysis buffer. The fraction corresponding to PmrC was concentrated to 1.3 mg ml⁻¹ for further analysis.

Circular dichroism (CD) spectra were measured using a Jasco J-810 spectropolarimeter with a Peltier temperature control system (Model PTC-423-S, Jasco Europe, Cremella (LC), Italy). Far-ultraviolet (UV) measurements at a protein concentration of 0.2 mg/ml were carried out at 20°C in 20-mM phosphate buffer pH 7.4 using a 0.1-cm optical path length cell and varying the wavelength from 260 to 198 nm at 20 nm min⁻¹ scan rate. The molar ellipticity per mean residue, [Ѳ] in deg.·cm²·dmol⁻¹, was calculated from the following equation: [Ѳ] = [Ѳ]obs × MRW × (10 × l × C)⁻¹, where [Ѳ]obs is the ellipticity measured in degrees, MRW is the mean residue molecular mass, C is the protein concentration in g/L, and l is the optical path length of the cell in cm. Mean residue molecular mass is calculated from MRW = M/(N − 1), where M is the protein’s molecular mass (in Da) and N is the number of amino acid residues. This value for PmrC is 113,12. CD melting curve was registered as a function of temperature (range: 20–90°C) at their maximum Cotton effect wavelengths with a scan rate of 1°C/min.

The three-dimensional (3D) structure of the whole PmrC protein was obtained using homology modelling and the MODELLER 9.22 program¹ (Webb and Sali, 2021) starting from the crystal structure of integral membrane protein lipooligosaccharide phosphoethanolamine transferase A (EptA) from Neisseria meningitidis (PDB code 5fgn, sequence identity 36.38%) as a template (Anandan et al., 2017). In parallel, modelling was performed using the most recently released AlphaFold3.0 modelling server,² which generated five models, ranked by AlphaFold3.0 ranking score (Abramson et al., 2024).

The selection of potential PmrC binder candidates was performed by high-throughput docking of the LifeChemicals commercial molecular database on the PmrC 3D model. The protein 3D structures were prepared using the Protein Preparation Wizard tool of the Schrödinger Suite 2015–4 (Schrödinger, LLC, New York, NY, USA). The binding site occupied by DODECYL-BETA-D-MALTOSIDE (Anatrace) in the 5fgn PDB structure was used as a guide to define the PmrC pocket to target after superimposing the two proteins. The grid box was built using DODECYL-BETA-D-MALTOSIDE as the centre and as a dimensional reference. The LifeChemicals database,³ downloaded as smi files from ZINC⁴ (Irwin et al., 2012), was treated with the Schrödinger Suite 2015–4 (Schrödinger, LLC, New York, NY, USA) to generate all the corresponding 3D structures at pH 7.4 ± 1.0 using the LigPrep tool (part of Schrödinger – official website⁵, Epik LigPrep tool (part of Schrödinger – official website (See footnote 5)), and OPLS2005 LigPrep tool (part of Schrödinger – official website (See footnote 5)) as the force field. A first docking calculation of all the compounds against the protein model was performed using Glide LigPrep tool (part of Schrödinger – official website (See footnote 5)) (Friesner et al., 2006) in standard precision (SP). Based on the docking scores, 2000 hits were selected. These selected molecules were then docked in extra precision (XP) mode and visually inspected to choose finally 44 molecules The 44 molecules were then subjected to a standard protocol of induced fit docking (IFD) using Glide. Finally, 26 molecules were chosen for purchase from LifeChemicals.

The thermophoretic measurements were conducted using the Monolith NT.115 device (NanoTemper Technologies, Munich, Germany) equipped with a red detection channel. For MST analysis, recombinant PmrC was labeled with the fluorescent dye RED-NHS 2nd Generation (NanoTemper Technologies) provided by Protein Labeling Kit RED-NHS (NanoTemper Technologies) (NanoTemper Technologies). This fluorescent dye carries a reactive NHS-ester group that reacts with primary amines to form a covalent bond. Briefly, 90 μl of a 10-μM PmrC solution in labeling buffer (130-mM NaHCO₃, 50-mM NaCl, and pH 8.2) was mixed with 10 μl of 300-μM dye RED-NHS Second Generation and incubated in the dark for 30 min at room temperature. Following labeling, the protein bound to the fluorophore was separated from any unreacted excess fluorophore using a specific column provided in the NanoTemper kit. The concentration of PmrC after labeling with the fluorescent dye was determined using an ultraviolet–visible (UV–Vis) spectrophotometer, with a labeling efficiency of approximately 25%. The MST experiment was conducted in a buffer containing phosphate buffer solution (PBS) at pH 7.4 and 0.03% DDM. Approximately 10 μl of labeled PmrC at 13 nM was mixed with 10-μl of 16 serial dilutions of compound F3205-0035. The final concentration of labeled PmrC was 6.5 nM in all samples, while the concentration of F3205-0035 ranged from 3.6 mM to 0.1 μM. Samples were then loaded into 16 premium-coated capillaries provided by NanoTemper Technologies, and fluorescence was recorded for 20 s using 100% laser power and 40% MST power. The instrument temperature was maintained at 25°C throughout all measurements. After recording the MST time traces, the data were analysed. The dissociation constant (Kd) value was calculated based on the ligand concentration-dependent changes in the normalised fluorescence (F_norm) of labeled PmrC after 14 s of thermophoresis. The assay was conducted in duplicate, and the reported values were generated using MO Affinity Analysis software provided by NanoTemper Technologies.

A. baumannii Ab4451 (Durante-Mangoni et al., 2015) and Ab60537 (Stabile et al., 2023) clinical isolates belonging to a collection previously established at the Department of Molecular Medicine and Medical Biotechnology, University of Naples Federico II, were routinely grown on Luria–Bertani agar plates (Del-tek) at 37°C under aerobic conditions. Stock cultures were stored in 10% glycerol and maintained at −80°C until use. In biological assays, the strain was grown in fresh cation-adjusted Mueller–Hinton broth (CA-MHB) (Sigma-Aldrich). A stock solution of colistin sulfate salt (Sigma–Aldrich) at 50 mg/ml concentration was obtained by dissolving the powder in H₂O.

MIC values for s-Phen and colistin against planktonic bacteria were assessed using a manual microdilution method in accordance with the guidelines suggested by the European Committee for Antimicrobial Susceptibility Testing (Esposito et al., 2019). Briefly, overnight grown bacterial culture was diluted in saline to 0.5 McFarland turbidity using a BD PhoenixSpec™ nephelometer (Becton Dickinson, Franklin Lakes, NJ, USA), followed by a 1:100 further dilution in CA-MHB to a final concentration of approximately 5 × 10⁶ colony forming unit (CFU)/ml. The bacterial culture (100 μl) was then added to a 96-well microtiter plate containing 100 μl of 2-fold serial dilutions in CA-MHB of s-Phen or colistin and incubated at 37°C. Each plate contained wells with no compounds as a positive growth control and serially diluted colistin or s-Phen in CA-MHB as a negative control. In each well DMSO concentrations were nontoxic for bacterial growth. After 18 h of incubation, the absorbance of bacterial culture at 595 nm was determined using a microplate reader (Bio-Rad Laboratories S.r.l., Hercules, CA, USA). The MIC values for the tested compounds were determined as the lowest concentration that inhibited bacterial growth.

Checkboard assays were carried out to test the MIC of the combination of colistin and s-Phen (De Gregorio et al., 2020). Briefly, 100 μl of the bacterial culture at a concentration of 5 × 10⁶ CFU/ml was added to a 96-well microtiter plate, where each well contained 100 μl of colistin (0–512 μg/ml), s-Phen (0, 200, and 400 μM), or colistin plus s-Phen (at the same final concentration to the previously mentioned one). The plates were incubated at 37°C for 18 h, and MIC values were measured by recording OD₅₉₅ of each well.

A. baumannii cultures were grown in LB broth at 37°C with continuous shaking until reaching an OD at 600 nm (OD600) of 0.4. Cells were then harvested from 2 ml of culture by centrifugation, and total RNA extraction was carried out according to the previously reported method (Migliaccio et al., 2021). Subsequently, the extracted RNA underwent treatment with RNase-free DNase (Invitrogen), and quantification was performed using a NanoDrop 1,000 spectrophotometer (Thermo Scientific, Winsford, UK).

For reverse transcription polymerase chain reaction (RT-PCR) analysis, 1,000 ng of total RNA was reverse transcribed using the QuantiTect Reverse Transcription Kit (Qiagen), according to the manufacturer’s instructions. The quantitative RT-PCR was performed using the SYBR Green master mix (Thermo Fisher Scientific) as previously described (Vollaro et al., 2020) with the ABI 7500 FAST instrument (Thermo Fisher Scientific) (Applied Biosystems). The primers used for PCR are listed in Supplementary Table S4. The 16S rRNA gene was used as an internal control for relative gene expression quantification based on the mean of three independent experiments. Negative controls without reverse transcriptase were included to detect possible DNA contamination. Fold changes in gene expression were determined using the 2^−ΔΔCt method (Livak and Schmittgen, 2001).

The haemolytic activity of s-Phen was evaluated as previously described (Esposito et al., 2019). Briefly, 190 μl of a red blood cell (RBC) suspension diluted 50-fold in PBS was incubated with 10 μl of s-Phen in a 96-well plate. The negative control was PBS, and the positive control was PBS with Triton X-100 (Sigma-Aldrich) (1%). The plate was incubated at 37°C for 1 h, and then the plate was centrifuged at 500 rpm for 5 min to pellet the intact RBCs. After centrifugation, 100 μl of the supernatant from each well was transferred to a new 96-well plate. The absorbance of the supernatant was measured at 450 nm using a microplate reader. The haemolysis rate was calculated by using the following formula: 100 × (A_sample − APBS)/(A_{Triton X-100} − APBS), where A_sample is the experimental absorbance of s-Phen, APBS is the control absorbance of untreated erythrocytes, and A_{Triton X-100} is the absorbance of 1% Triton X-100-lysed cells. Statistical analyses were performed using GraphPad Prism 8 software (GraphPad, San Diego, CA, USA). All data points are expressed as the mean ± standard deviation (SD) of three independent experiments, each performed in triplicate. The statistical significance of the differences was determined using one-way analysis of variance (ANOVA), followed by Dunnett’s multiple comparison test to compare the groups.

PmrC is long been recognised as a major responsible for colistin resistance in A. baumannii (Gogry et al., 2021; Hamel et al., 2021; Nurtop et al., 2019; Seleim et al., 2022; Trebosc et al., 2019). Despite its importance, a biophysical characterisation is still unavailable, likely due to its molecular complexity. PmrC is a 533-residue membrane protein (Figure 2A). Thus, we expressed and purified it in high yields with the detergent dodecyl-β-d-maltoside (DDM; Figure 2B). Far-UV CD spectroscopy spectrum of PmrC is typical of well-structured folds with high α-helical content, with typical minima at 208 and 222 nm (Figure 2C). Thermal unfolding curves were recorded by following the CD signal at 222 nm as a function of temperature, using a 1°C/min heating rate to investigate the heat-induced changes in the protein secondary structure. The thermal unfolding curve exhibits the characteristic sigmoidal profile expected of two-state systems, with a melting temperature T_m = 60°C (Figure 2D). Consistent with the complexity of the molecule, its folding is fully irreversible (Figure 2E).

We homology modelled PmrC using MODELLER (Webb and Sali, 2021) and the structure of EptA from N. meningitis, sharing 37% sequence identity with PmrC, as a template (Anandan et al., 2017). This model was fully superposable with that computed using AlphaFold3.0 (Abramson et al., 2024), with a root mean square deviation on Cα atoms of 2.0 Å. The PmrC structure model is formed by two distinct domains with completely different features. The N-terminal domain (residues 1–194, orange) is mostly hydrophobic and contains 5 transmembrane helices (α1–α5; Figure 2E). This domain is predicted to be fully immersed in the membrane double layer. Consistently, the protein electrostatic potential map of the protein shows patches of positively charged residues at the two extremities of the N-terminal domain (Figure 2F), likely mediating charge interactions with the phosphate groups of the membrane phospholipids. The C-terminal domain of PmrC (residues 200–533) is an alpha-beta domain typical of the hydrolase-type fold (Figure 2E). Structural alignment using DALI (Holm et al., 2023) shows consistently similar domains in sulfatases and phosphatases (Supplementary Table S1). Similar to EptA from N. meningitis (Anandan et al., 2017) and MCR-1 of E. coli (Suardíaz et al., 2021), PmrC C-terminal domain contains a Zn²⁺ binding site, composed by Glu229, Thr267, Asp443, and His444 (Figure 2E).

The best characterised PetN transferase catalytic mechanism is that of MCR-1 of E. coli, which shares 37% identity with PmrC (Suardíaz et al., 2021). MCR-1 is an integral, Zn²⁺-dependent inner-membrane protein. The zinc site is tetrahedrally coordinated in MCR-1 residues Glu246, Asp465, His466, and Thr285, corresponding to Glu229, Asp443, His444, and Thr267 in PmrC, respectively (Figure 2E). In MCR-1, the reaction path for PetN transfer from the bacterial lipid membrane to the protein involves a nucleophilic attack of Thr285 (267 in PmrC) on the phosphate centre of the phospholipid, concerted with the activation of Thr285 by Glu246 and a proton transfer from His395 to the dephosphorylated lipidic leaving group (Suardíaz et al., 2021). To inactivate this reaction mechanism, a virtual screening campaign based on high throughput docking (HTPD) was conducted on the modelled structure of PmrC. Docking of the LifeChemicals (See footnote 2) commercial database was performed in Standard Precision (SP) mode using Glide (Schrödinger, LLC, New York, NY, 2015). The top 200 compounds based on docking scores were saved and redocked in extra precision (XP) mode. After visual inspection, 44 compounds were selected and subjected to Induced Fit Docking (IFD) with Glide. Visual inspection criteria were steric complementarity of compounds with the binding site, hydrogen bonds, and hydrophobic contacts (Fischer et al., 2021). Finally, after IFD, 26 compounds were chosen for purchase and subsequent testing (Supplementary Table S2).

The selected compounds were solubilised for experimental assessment of their ability to interact with PmrC. The majority of soluble molecules under conditions compatible with binding experiments were F3320-0359, F3205-0035, F6761-7637, F6523-3945, and F6231-0947 (Supplementary Table S2), which were tested for their binding to PmrC using MST. We observed that, among the tested compounds, increasing concentrations of F3205-0035 clearly affect the thermophoretic motion of PmrC (Figure 3A; Supplementary Figure 1). F3205-0035 is a 1-(3-methoxypropyl)-3-{[4-(piperidin-1-yl)phenyl]amino}pyrrolidine-2,5-dione, and is here denominated as s-Phen. Although the low solubility of s-Phen did not allow us to reach saturation, we observed a clear dose–response curve, and the data fitting provided a K_d value in the high micromolar range (Figure 3A). Docking of s-Phen in the catalytic site cleft of PmrC shows that s-Phen fills a deep pocket located at the interface between the N-and C-terminal domains (Figures 3B,C). Several hydrophobic interactions characterise the binding to residues of the N-terminal domain, including Phe74, Phe78, Leu82, Val84, Ile85, Ile86, and with Pro376 and Leu455 of the C-terminal domain (Figure 3C). In addition, hydrogen bonding interactions occur with the main chain atoms of Gln90 (O) and His456 (N; Figure 3C). Of note, s-Phen does not contact any of the residues hitherto known to contribute to PmrC catalysis, as the majority of its interactions with the protein occur with the N-terminal domain. However, it occupies the same hydrophobic cavity of DDM (Supplementary Figure 2), which mimics acyl chains of lipid A, as previously suggested for the homolog EptA from N. meningitis(Anandan et al., 2017). Therefore, s-Phen likely acts by blocking the cavity, which accommodates the hydrophobic acyl chains of lipid A (Supplementary Figure 2).

The efficacy of s-Phen in contrasting colistin resistance was tested in the two colistin-resistant clinical isolates, Ab60537 (Stabile et al., 2023) and Ab4451 (Durante-Mangoni et al., 2015). The colistin-resistant strains Ab4451 and Ab60537 present amino acid substitutions in the PmrB sensor kinase of the PmrAB TCS. It has been shown that mutations in the PmrAB lead to the overexpression of pmrC, conferring resistance to colistin (Arroyo et al., 2011; Trebosc et al., 2019). First, we performed quantitative reverse transcription-PCR (qRT-PCR) to investigate pmrABC gene expression in two colistin-susceptible/colistin-resistant pairs of A. baumannii clinical isolates (Ab60536/Ab60537 and Ab4450/Ab4451). As shown in Figure 4, the colistin-resistant isolates Ab60537 and Ab4451 had only slightly increased expression of pmrA (1.84- and 10.9-fold, respectively) and pmrB (1.77- and 2.33-fold, respectively) compared to their colistin-susceptible counterparts Ab4450 and Ab60536, whereas they showed a marked increase in pmrC expression (314.5- and 41.45-fold, respectively). This suggested that colistin resistance in the isolates Ab60537 and Ab4451 is due to PmrC overexpression. After this preliminary check, we tested the five compounds F3320-0359, F6761-7637, F6523-3945, F6231-0947, and s-Phen for antimicrobial activity. Consistent with binding studies, we observed that all compounds do not act as antimicrobials by themselves, as in all cases, MIC values were > 150 μg/ml. However, when added to colistin, s-Phen was able to drastically reduce the MIC value of colistin in both Ab60537 and Ab4451-resistant isolates (Table 1). In the presence of s-Phen at its maximum concentration of 138 μg/ml, colistin MIC decreased 16-fold (from 128 to 8 μg/ml) for Ab60537 and 64-fold (256–4 μg/ml) for Ab4451 (Table 1). We also measure the cytotoxicity of s-Phen by haemolysis assay against red blood cells to assess its safety profile. The results indicate that s-Phen is not toxic, as evidenced by haemolysis levels similar to those of the negative control (Figure 5). Indeed, s-Phen does not cause measurable haemolysis at any of the concentrations tested.