Strains and plasmids are listed in Supplementary Table 1. Lysogeny broth (LB, Acumedia) was used for genetic manipulation. Mueller–Hinton broth (MH, Difco) was used for Kirby–Bauer assay, while cation-adjusted MH (MH II, Difco) was used for all the other assays. When required, IPTG was added at the indicated concentrations. All chemicals were purchased from Merck.

To obtain the constructs for IPTG-inducible expression of mcr-1, eptA, or the truncated variant eptA^ΔC–ter, the coding sequence of each gene was amplified by PCR using the genomic DNA of P. aeruginosa PAO1 (for eptA and eptA^ΔC–ter) or the plasmid pHNSHP45 (for mcr-1) as the template (Supplementary Table 1). The amplicons were digested with the restriction enzymes EcoRI and XhoI and individually cloned into the sequencing plasmid pBlueScript II KS (pBS; Stratagene), previously digested with the same enzymes. The construct pBSeptA was used as template for PCR-mediated site-specific mutagenesis using the “Q5 Site-Directed Mutagenesis” kit (New England BioLabs), according to manufacturer’s instructions, and a primer pair specifically designed to introduce an A-to-G substitution leading to the mutation T278A, thus obtaining the eptA^T278A allele (Supplementary Table 1). After verification of the cloned fragments through restriction analysis and DNA sequencing, the inserts were excised from pBS and sub-cloned, using the same enzymes, into the shuttle vector pME6032, which replicates in P. aeruginosa and allows IPTG-inducible expression of cloned genes (Heeb and Haas, 2001). Primers used for cloning, site-specific mutagenesis and sequencing are listed in Supplementary Table 2. The constructs were transferred into P. aeruginosa strains by transformation.

The expression level of eptA was determined by quantitative reverse transcription PCR (qRT-PCR). Bacteria were cultured in MH II in the presence of increasing concentrations of IPTG until mid-log phase, harvested by centrifugation and treated with RNAprotect Bacteria Reagent (Qiagen). RNA was purified using the RNeasy Mini Kit (Qiagen), treated with DNase, and re-purified with the RNeasy MinElute Cleanup Kit (Qiagen). cDNA was reverse transcribed from 0.5 μg of RNA with Prime Script RT Reagent Kit (Takara). Appropriate dilutions of the cDNA were used as template for qRT-PCR in a AriaMx Real-Time PCR System (Agilent) using TB Green Premier EX Taq master mix (Takara) and the primers listed in Supplementary Table 2. Relative gene expression with respect to the housekeeping gene rpoD was calculated using the 2^{–ΔΔ Ct} method (Livak and Schmittgen, 2001).

Bacteria were cultured in MH II until early stationary phase, harvested by centrifugation and lipid A was extracted using the ammonium hydroxide-isobutyric acid-based procedure (Liu et al., 2017) with previously described modifications (Lo Sciuto et al., 2018). Samples were analyzed by matrix-assisted laser desorption-time of flight (MALDI-TOF) in the negative-ion mode (Lo Sciuto et al., 2018).

Growth assays were performed in MH II in 96-well microtiter plates at 37°C (200 μL per well). Growth was measured as optical density at 600 nm (OD₆₀₀) in a Tecan Spark 10M microtiter plate reader.

Pairwise competition experiments were performed by inoculating each strain of the pair of interest at 5 × 10⁵ cells/mL in MH II. After 16 and 20 h at 37°C, cultures were serially diluted in saline and plated onto MH II agar and MH II agar containing 2 μg/mL colistin and 0.5 mM IPTG, to evaluate the number of cells carrying pMEeptA or pMEmcr-1 (that develop colonies both in the presence and absence of colistin) and the number of cells carrying the empty plasmid pME6032 (that develop colonies only in the absence of colistin). Preliminary experiments were performed to select the lowest colistin concentration that inhibits the growth of cells with the empty vector without affecting the plating efficiency of those carrying pMEeptA or pMEmcr-1. The competitive index was calculated as the ratio between the number of cells expressing EptA or MCR-1 and the number of cells carrying pME6032 at 16 or 24 h divided by to same ratio at time 0 (Lo Sciuto et al., 2020).

Colistin MIC was determined through the broth microdilution method in MH II. Strains were cultured at 37°C for 8 h, and then refreshed at 5 × 10⁵ cells/mL in the presence of increasing concentrations of colistin (up to 128 μg/mL). MIC was recorded after 24 h at 37°C. Each strain was tested in at least six independent experiments.

Bacterial cell suspensions in saline were normalized at 0.5 McFarland standard and swabbed onto MH agar plates. Disks containing ciprofloxacin (5 μg), imipenem (10 μg), gentamicin (10 μg), novobiocin (30 μg), erythromycin (15 μg), rifampicin (5 μg), or tobramycin (10 μg) (Becton Dickinson) were placed on the surface of the inoculated plates, and growth inhibition halo diameters were measured after 24-h incubation at 37°C.

Bacteria were cultured in MH II until mid-exponential phase, harvested by centrifugation, resuspended in saline at OD₆₀₀ of 0.5 and dispensed in microtiter plate wells in the presence of increasing concentrations (0–5%) of sodium dodecyl sulphate (SDS) (200 μL per well). Sensitivity to the SDS lytic effect was assessed by determining the turbidity (OD₆₀₀) of cell suspensions after 5-min incubation at room temperature (Lo Sciuto et al., 2014), using a Tecan Spark 10M microtiter plate reader.

Permeability assays were performed as previously described (Hancock and Farmer, 1993; Kwon et al., 2019), with few modifications. Bacteria were cultured in MH II until mid-exponential phase, harvested by centrifugation, resuspended in 5 mM HEPES (pH 7.2) at OD₆₀₀ of 2 and dispensed in microtiter plate wells in the presence of 1-N-phenylnaphthylamine (NPN) or propidium iodide (PI) at 10 μM or 20 μg/mL final concentration, respectively. Fluorescence was measured in a Tecan Spark 10M microtiter plate reader (excitation at 350 nm and emission at 420 nm for NPN; excitation at 580 nm and emission at 620 nm for PI) after 5 min at room temperature, subtracted of the background values of samples without NPN or PI, and normalized to the OD₆₀₀ of the cell suspension.

Sequence alignments were generated with Clustal Omega.¹ Residue conservation analysis was performed with the ConSurf server.² The TMHMM server was used for transmembrane helix prediction.³ Three-dimensional modeling was performed with SWISS-MODEL,⁴ using the Neisseria meningitidis EptA protein (PDB 5FGN) as the template.

Statistical analysis was performed with GraphPad InStat, using one-way analysis of variance (ANOVA) followed by Tukey–Kramer multiple comparison test.

To investigate the contribution of PEtN-modification of the lipid A moiety of LPS to colistin resistance in P. aeruginosa, and to evaluate the effect of this lipid A modification on bacterial fitness, two plasmids were generated for the IPTG-inducible expression of either the endogenous lipid A PEtN transferase EptA or the exogenous analog MCR-1. These plasmids, named pMEeptA and pMEmcr-1, were individually introduced in two P. aeruginosa reference strains, namely PAO1 and PA14, that are distantly related and are representative of the two major strain groups of the P. aeruginosa population structure (Freschi et al., 2015).

To confirm that phosphoethanolamination of lipid A occurs in the strains expressing MCR-1 or EptA, the lipid A was extracted from PAO1 and PA14 carrying the empty vector pME6032, pMEeptA, or pMEmcr-1 cultured in the presence of 0.5 mM IPTG. Lipid A was then analyzed by MALDI-TOF mass spectrometry (Figure 1). The lipid A spectra were characterized by a major peak corresponding to diphosphorylated penta-acylated lipid A (m/z 1446), that represents the main lipid A form of laboratory P. aeruginosa strains (Lo Sciuto et al., 2018, 2020). Less abundant peaks were also identified, such as tetra-acylated diphosphorylated lipid A (m/z 1276), penta-acylated monophosphorylated lipid A (m/z 1366), penta-acylated triphosphorylated lipid A (m/z 1526), and hexa-acylated lipid A (m/z 1616). Notably, some lipid A forms were found with different hydroxylation states, due to the presence/absence of a hydroxyl group in the two secondary C12 acyl chains (Lo Sciuto et al., 2019).

As expected, additional peaks were exclusively detected in the lipid A spectra of cells carrying pMEeptA or pMEmcr-1. The addition of a PEtN group results in a + 123 shift in the m/z value of lipid A peaks (Nowicki et al., 2015; Liu et al., 2017). The peaks at m/z 1489 and 1569 correspond to the phosphoethanolaminated penta-acylated mono- or diphosphorylated form, respectively, while the peak at m/z 1739 corresponds to the phosphoethanolaminated hexa-acylated diphosphorylated form. Overall, lipid A analysis confirmed that both MCR-1 and EptA expressed from the pMEmcr-1/eptA constructs are functional and that they efficiently modify lipid A with PEtN. Notably, PEtN-modified lipid A was not observed in both PAO1 and PA14 carrying the empty plasmid (Figure 1), confirming that the endogenous copy of eptA, that is present in the genome of both strains,⁵ is not or only poorly expressed in standard laboratory media (Nowicki et al., 2015; Liu et al., 2017; Lo Sciuto et al., 2018).

To investigate the effect of MCR-1 or EptA expression on colistin resistance, MIC assays were performed in the presence of different IPTG concentrations (Table 1). As expected, strains carrying the empty plasmid were susceptible to colistin, as inferred from MIC values lower than the clinical breakpoint for P. aeruginosa (2 μg/mL; EUCAST). P. aeruginosa expressing MCR-1 showed an IPTG-dose dependent increase in colistin MIC, confirming the capability of MCR-1 mediated lipid A phosphoethanolamination to confer colistin resistance in P. aeruginosa (Liu et al., 2017). Notably, colistin MIC was fourfold higher for PA14 pMEmcr-1 than for PAO1 pMEmcr-1 (Table 1). This is in line with the results of a recent study about the contribution of Ara4N-modified lipid A to colistin resistance, that was found to be significantly greater in PA14 with respect to PAO1 (Lo Sciuto et al., 2020). These results confirm previous evidence that PA14 is more responsive than PAO1 to lipid A modifications that confer colistin resistance (Gutu et al., 2015). Notably, when colistin MIC was directly compared between MCR-1 expressing strains, cultured in the presence of 0.5 mM IPTG, and recombinant strains that constitutively express the arn operon responsible for Ara4N modification of lipid A (PAO1/PA14 PrpsA:arn), in which the arn promoter was replaced with the promoter of the housekeeping gene rpsA (Lo Sciuto et al., 2020; Supplementary Table 1), it appeared that lipid A phosphoethanolamination is as efficient as aminoarabinosylation in conferring colistin resistance to P. aeruginosa (Table 1).

In contrast, colistin resistance in EptA overexpressing strains did not correlate with the IPTG concentration in the medium. Indeed, colistin MIC was higher at 0 (for PAO1) or 0.125 mM IPTG (for PA14) and reverted at levels comparable to the empty plasmid controls at 0.5 mM IPTG (Table 1). The significant increase in colistin MIC observed in strains carrying pMEeptA under non-inducing conditions prompted us to verify lipid A phosphoethanolamination and eptA expression levels at different IPTG concentrations. Mass spectrometry analysis revealed that lipid A was efficiently phosphoethanolaminated even in the absence of IPTG (Supplementary Figure 1). This is in line with the results of the qRT-PCR analysis, which showed that eptA expression was hundreds-fold higher in pMEeptA carrying strains under non-inducing conditions as compared to empty plasmid controls (Figure 2). Overall, these experiments revealed that P. aeruginosa EptA can promote colistin resistance at levels almost comparable to those conferred by the exogenous PEtN transferase MCR-1, even if proper EptA levels seem to be critical for optimal induction of resistance.

The relevant decrease in colistin MIC observed in EptA expressing cells cultured in the presence of 0.5 mM IPTG (Table 1) led us to hypothesize that EptA could exert some toxicity when expressed at very high levels (about 10,000-fold higher than the expression level of the empty plasmid controls; Figure 2), that could somehow counterbalance the positive effect of PEtN-modified lipid A on colistin resistance. To evaluate whether EptA overexpression causes any fitness costs, we compared planktonic and colony growth between EptA/MCR-1 expressing cells and the empty plasmid controls. As shown in Figure 3A and Supplementary Figure 2, MCR-1 expression did not affect growth of any strain at any IPTG concentration. In contrast, the expression of EptA caused a reduction in the growth kinetics of both PAO1 and PA14 in an IPTG-dose dependent manner, with maximum growth inhibition observed at IPTG concentrations ≥0.5 mM (Figure 3A and Supplementary Figure 2). Colony growth over time was also slightly delayed for EptA overexpressing strains as compared to the other strains (Figure 3B), corroborating the negative effect of EptA on P. aeruginosa growth kinetics. The growth defects of EptA overexpressing cells were further investigated through competition experiments, by co-culturing cells carrying pMEeptA (or pMEmcr-1 as the control) with those with the empty vector in the presence of 0.5 mM IPTG. The results are shown as competitive index, representing the relative growth of EptA (or MCR-1) overexpressing strains with respect to the empty plasmid controls (Figure 3C). In agreement with the results of mono-culture assays, cells expressing MCR-1 showed a competitive index close to 1, indicating that their growth kinetic was overall comparable to that of cells lacking mcr-1. On the other hand, both PAO1 pMEeptA and PA14 pMEeptA showed a much lower competitive index (0.1 and 0.2, respectively), implying that EptA overexpressing cells were outcompeted by cells harboring the empty plasmid.

To investigate whether the detrimental effect EptA overexpression on P. aeruginosa growth correlated with defects in the integrity and/or permeability barrier of the cell envelope, we compared the sensitivity profile to different antibiotics belonging to different classes of P. aeruginosa PAO1 and PA14 carrying the empty plasmid pME6032 or the pMEmcr-1/pMEeptA derivatives cultured in the presence of 0.5 mM IPTG, by using the Kirby–Bauer disc diffusion test. MCR-1 expressing strains did not show any difference in the resistance profile with respect to the empty plasmid controls, while EptA overexpressing strains showed a slight increase in susceptibility to some antibiotics, i.e., ciprofloxacin, gentamicin, tobramycin, and imipenem (Figure 4A). All strains remained completely insensitive to high molecular weight antibiotics, such as novobiocin, erythromycin, and rifampicin, that are poorly active against P. aeruginosa because of poor diffusion across the OM and that are generally used to highlight defects in the OM integrity (Nikaido, 2005; Lo Sciuto et al., 2014, 2018). This suggests that EptA overexpressing cells have minor defects (if any) in cell envelope homeostasis. This was confirmed by the analysis of the sensitivity profile to the lytic effect of SDS, that was overall comparable between MCR-1/EptA overexpressing cells and the empty plasmid controls (Figure 4B). To further investigate the possible effect of PEtN and/or PEtN transferases on the P. aeruginosa cell envelope, the OM and IM permeability was assessed through the fluorescent probes NPN and PI, respectively (Hancock and Farmer, 1993; Kwon et al., 2019; Figure 4C). No differences were observed in NPN and PI fluorescence between MCR-1 expressing cells and empty plasmid controls. In contrast, EptA overexpression slightly increased NPN and PI uptake in PA14 but not in PAO1, suggesting that EptA might have a strain-specific effect on OM and IM integrity. Notably, such increase in membrane permeability can hardly explain the observed growth defects of EptA expressing cells, as both mono- and co-culture assays revealed that the negative impact of EptA overexpression on growth is more severe in PAO1 than in PA14 (Figure 3).

Overall, these assays revealed that MCR-1 and lipid A phosphoethanolamination do not influence cell envelope homeostasis in P. aeruginosa, while EptA overexpression has a marginal effect on the cell envelope permeability barrier, which is, however, unrelated to the growth defects of EptA expressing cells.

The finding that EptA but not MCR-1 overexpression is deleterious to P. aeruginosa strongly suggests that the growth defects might not be related to PEtN-modified lipid A. This, however, does not rule out that the PEtN transferase activity might be involved in the noxious effects of EptA overexpression. In fact, while most PEtN transferases have high substrate specificity (Zhang et al., 2019), a PEtN transferase of Campylobacter jejuni is more promiscuous, being able to add PEtN to (1) lipid A, (2) heptose I of the LPS inner core, (3) a Thr residue of the flagellar protein FlgG, and (4) the N-terminally linked glycans of some periplasmic proteins (Cullen et al., 2013). Considering that P. aeruginosa EptA has been poorly characterized so far (Nowicki et al., 2015), we attempted to verify the relevance of PEtN transferase activity for EptA-dependent toxicity. To this aim, we generated a construct (pMEeptA^T278A) to express a catalytically inactive EptA variant (EptA^T278A), mutagenized in a conserved catalytic residue essential for N. meningitidis EptA activity (Xu et al., 2018a; Supplementary Figure 3). As expected, lipid A was not modified with PEtN in EptA^T278A overexpressing cells (Figure 5A and Supplementary Figure 4) and, accordingly, colistin MIC for these cells was identical to that for the empty plasmid controls (Table 1). Interestingly, overexpression of EptA^T278A caused growth defects comparable to those caused by wild type EptA (Figure 5B), indicating that PEtN transferase activity is not responsible for P. aeruginosa EptA deleterious effects. We therefore decided to search for possible unique feature(s) of P. aeruginosa EptA, by evaluating the evolutionary conservation among 150 representative EptA orthologs through the ConSurf server (Ashkenazy et al., 2016). This analysis highlighted a 25-aa C-terminal tail in the periplasmic domain of P. aeruginosa EptA that is absent from all other EptA orthologs (Figure 6). The IPTG-inducible expression of a truncated EptA variant lacking the last 22 amino acids (EptA^ΔC–ter) in P. aeruginosa cells, followed by mass spectrometry analysis of lipid A, revealed that the C-terminal extension is not required for lipid A phosphoethanolamination (Figure 5A). In contrast, cells expressing the EptA^ΔC–ter variant showed less severe growth defects as compared to cells expressing wild type EptA (Figure 5B), implying that the C-terminal tail is partly responsible for the growth inhibitory effect of EptA overexpression. Moreover, EptA^ΔC–ter expressing strains showed high colistin MIC even in the presence of high IPTG concentrations (Table 1), supporting the hypothesis that the inverse proportion between wild type EptA expression and colistin resistance is likely due to the negative impact of high EptA levels on P. aeruginosa growth. Although previous studies have shown that also other PEtN transferases can negatively impact bacterial physiology (Xu et al., 2018b; Yang et al., 2020), this is the first report that associates PEtN transferase toxicity to a specific and evolutionary divergent protein domain. Further studies are required to elucidate the mechanistic basis of EptA-mediated toxicity in P. aeruginosa.