An Evolutionarily Conserved Mechanism for Intrinsic and Transferable Polymyxin Resistance

ABSTRACT Polymyxins, a family of cationic antimicrobial cyclic peptides, act as a last line of defense against severe infections by Gram-negative pathogens with carbapenem resistance. In addition to the intrinsic resistance to polymyxin E (colistin) conferred by Neisseria eptA, the plasmid-borne mobilized colistin resistance gene mcr-1 has been disseminated globally since the first discovery in Southern China, in late 2015. However, the molecular mechanisms for both intrinsic and transferable resistance to colistin remain largely unknown. Here, we aim to address this gap in the knowledge of these proteins. Structural and functional analyses of EptA and MCR-1 and -2 have defined a conserved 12-residue cavity that is required for the entry of the lipid substrate, phosphatidylethanolamine (PE). The in vitro and in vivo data together have allowed us to visualize the similarities in catalytic activity shared by EptA and MCR-1 and -2. The expression of either EptA or MCR-1 or -2 is shown to remodel the surface of enteric bacteria (e.g., Escherichia coli, Salmonella enterica, Klebsiella pneumoniae, etc.), rendering them resistant to colistin. The parallels in the PE substrate-binding cavities among EptA, MCR-1, and MCR-2 provide a comprehensive understanding of both intrinsic and transferable colistin resistance. Domain swapping between EptA and MCR-1 and -2 reveals that the two domains (transmembrane [TM] region and phosphoethanolamine [PEA] transferase) are not functionally exchangeable. Taken together, the results represent a common mechanism for intrinsic and transferable PEA resistance to polymyxin, a last-resort antibiotic against multidrug-resistant pathogens.

the Klebsiella pneumoniae carbapenemase (KPC), has called into question the use of carbapenem, a ␤-lactam antibiotic, as an effective treatment option against severe MDR infections (5,6). Consequently, the cationic antimicrobial peptide polymyxin (regardless of its partial renal toxicity) was reintroduced as a last-resort option for the clinical treatment of patients infected by the carbapenem-resistant Gram-negative bacterium (7). However, it seems that that bacteria have developed either an acquired resistance to colistin or a chromosomally encoded "intrinsic" resistance ( Fig. 1) (8). Intrinsic resistance is naturally occurring and is due to the functional expression of certain chromosomal genes (such as eptA of Neisseria meningitidis). Acquired resistance to colistin is evidently related to mutations in two-component regulatory systems involved in the lipopolysaccharide synthesis pathway (e.g., PhoP/PhoQ in K. pneumoniae [9] and Pseudomonas aeruginosa [10] and PmrA/PmrB in Salmonella enterica [11][12][13], Acinetobacter baumannii [14], and P. aeruginosa [15]). Transferable resistance to colistin is conferred/acquired through plasmid-borne genes (i.e., mcr-1 and mcr-2) (16,17). Therefore, new strategies are urgently required to reverse/bypass colistin resistance in Gram-negative pathogens (18).
Colistin primarily interacts with the lipopolysaccharide (LPS) layer of the Gramnegative bacterial outer membrane (19)(20)(21). A structural alteration of LPS species that reduces the overall negative charge of the bacterial surface and interferes with the electrostatic interaction between colistin and LPS (22) renders the bacteria resistant to colistin ( Fig. 1A) (17). Of all the covalent modifications of LPS that occur, the addition of phosphoethanolamine (PEA [also called pEtN]) to lipid A is the most prevalent (12,23). EptA in Neisseria, a member of the PEA transferase family, is a well-known player in catalyzing the transfer of PEA to the 1-phosphate or 4=-phosphate [1(4=)-phosphate] position of lipid A glucosamine (GlcN) moieties (22,24). However, unlike the chromosomal eptA (22), mcr-1, a newly identified gene encoding a PEA transferase, is harbored on diversified plasmids that disseminate into a variety of Enterobacteriaceae species and confers a transferable resistance to colistin (25,26). In fact, mcr-1 has spread to over 40 countries or regions, covering 5 of the 7 continents (27,28), and is the most-frequently detected genetic element of mobilized colistin resistance, especially on swine farms with massive use of colistin. Very recently, MCR-2 was determined to be a new variant of MCR-1, with around 80% identity and exclusively encoded on an IncX4-type plasmid (16,17). This seems to be a rare form in that it has only been detected in Belgium (29,30). In addition, three more mcr-like genes have been elucidated (28), namely, mcr-3 (31)(32)(33)(34)(35), mcr-4 (36), and mcr-5 (37). The MCR-like enzymes encoded by these genes exhibit relatively low levels of similarity to MCR-1 and -2 (28). Together with other research groups (38)(39)(40)(41)(42), we have defined a zinc-requiring catalytic motif in MCR-1 that is a prerequisite for colistin resistance (43,44). In addition, we have also deciphered the partially genetic/biochemical mechanism of MCR-2 for colistin resistance (17). However, we are not aware of how MCR-2 interacts with the phosphatidylethanolamine (PE) substrate, due to a lack of structural/chemical evidence.
In this study, we report mechanistic insights into the structure and function of the two PEA transferases EptA and MCR-2. In addition to the overall architecture, we also illustrate a parallel between their substrate entry cavities. More importantly, we propose that a possible "ping-pong" mechanism is shared among EptA, MCR-1, and MCR-2 ( Fig. 1B and C). This mechanism features the PEA transfer via the formation of an intermediate product, enzyme-bound PEA. Our finding reveals that parallels are present between intrinsic and transferable resistance to colistin (Fig. 1D to F), a last-resort treatment option for severe infections by carbapenem-resistant superbugs.

RESULTS
Discovery of a cavity for PE substrate entry in EptA. The Neisseria EptA (LptA [lipid A PEA transferase]), which belongs to the YhjW/YjdB/YijP superfamily, is a previously known intrinsic determinant of colistin resistance. The X-ray structure of full-length EptA revealed a previously uncharacterized helical membrane domain (see Fig. S1 and S2 in the supplemental material) (24), suggesting residues that might be (Continued on next page) A Common Mechanism for Polymyxin Resistance ® involved in PE substrate binding (45). However, experimental evidence for this is still missing. We therefore applied molecular docking to reanalyze the complex structure of EptA with a PE substrate analogue, the detergent n-dodecyl ␤-D-maltoside (DDM) ( Fig. 2A and B). As a result, we propose a cavity ( Fig. 2B and D), 12 residues of which are essential for PE substrate entry/binding ( Fig. S2A to D). Among them, 5 amino acids (E240, T280, H378, D452, and H453) are involved in the interaction with zinc (Fig. S1, S2B, and S2F) and the remaining 7 residues (N106, T110, E114, S325, K328, H383, and D465) are implicated in recognition of the physiological substrate, PE ( Fig. S2D and G). To elucidate the function of these residues that define the cavity, we generated 12 point mutants of EptA and demonstrated their effectiveness using colistin resistance assays. Even though all EptA point mutants are expressed well in Escherichia coli, as verified by Western blotting (Fig. S2E), none of the five mutants with point mutations of EptA with an alanine substitution in a zinc-binding site can support any significant growth in the presence of colistin (Fig. S2F). As for the point mutants with mutations of the 7 PE substrate-interacting sites, 3 are partially inactive in the trials of colistin resistance on Luria-Bertani agar (LBA) plates, namely, those with the N106A (bearing a change of N to A at position 106) (4 g/ml), S325A (2 g/ml), and T110A (1 g/ml) mutations (Fig. S2G), while the remaining four mutants are fully nonfunctional, i.e., their mutations confer no additional resistance to colistin, in agreement with the results for the negative control (0.5 g/ml) (Fig. S2G). Similar scenarios were also observed in the colistin MIC measurements (Fig. S2H). The aforementioned in vivo data thus confirm the importance of cavity-forming residues in the ability of EptA to provide resistance to colistin.
As described by Anandan et al. (45), we also assessed the role of this cavity in the enzymatic activity of EptA in vitro, and found it to be similar to that of MCR-2 (see Fig. 4). As anticipated, the results are in agreement with those of our colistin trials (Fig. S2) and show that (i) each of the 5 alanine mutants of EptA with defects in zinc binding have no detectable activity and (ii) 3 of 7 EptA derivatives with mutations in the PE-recognizing cavity (N106A, T110A, and S325A) possess partial activity, whereas the other 4 (E114A, K328A, H383A, and D465A) are inactive (data not shown). Subsequently, we subjected the lipid A of E. coli cells expressing eptA mutants to matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF MS) (  Table 1). Second, 3 of 12 EptA mutants (N106A, T110A, and S325A) still retain partial activity in transferring the PEA moiety to lipid A, as evidenced by an additional appearance of PPEA-1(4=)-lipid A (m/z, 1,920.077 to 1,920.195) ( Table 1). Third, the remaining 9 mutants of EptA are nonfunctional in catalyzing the formation of PPEA-1(4=)-lipid A, since the MS peak of lipid A alone is consistently at an m/z of 1,796.719 to 1,797.069 (Table 1). In addition to the three mutants (E114A, E240A, and T280A) with partial activity (Table 1), six more derivatives of EptA (K328A, H378A, H383A, D452A, H453A, and H456A mutants) are involved in the enzyme activity (Table 1). These results constitute a functional definition of a cavity in EptA for PE lipid substrate entry.
Similarity among PE-binding cavities of EptA, MCR-1, and MCR-2. The phylogeny of the MCR-like proteins indicates that EptA might be evolutionarily distinct from MCR-1 and -2 ( Fig. 3), which generally agrees with their differing levels of colistin resistance ( Fig. 4; Fig. S2). However, it seems that EptA and MCR-1 have evolved substrate-binding cavities with similar conformations that require identical sets of cavity-forming residues (12 conserved amino acids in total) ( Fig. S1 and S2). Based on the assumption that MCR-2 possesses a similar cavity as well, we analyzed its overall structure through superposition of the modeled structure of full-length MCR-2 on that of EptA (Fig. S3). Again, as observed with EptA (45), the overall architecture of MCR-2 contains an N-terminal transmembrane (TM) domain and a periplasm-facing catalytic domain at the C terminus (Fig. S3A). The TM domain, spanning the inner membrane, includes six ␣-helices (Fig. S3A), and the catalytic domain has a hydrolase fold (Fig. S3B) that consists of 10 ␣-helices and 7 ␤-sheets (Fig. S3A). The two domains are connected by four short periplasmic loops (PH2, PH2=, PH3, and PH4), a bridge helix (BH), and a long, coiled loop ( Fig. S3A and B). The presence of a long, coiled loop between the BH and ␤-sheet 1 (S1) facilitates the flexible movement of the catalytic domain to capture PE substrate molecules. Therefore, structural similarities are observed among EptA (45), MCR-1, and MCR-2 ( Fig. 2; Fig. S3).
Since DDM, a nonionic detergent ( Fig. 2A), is only an analogue of the physiological substrate of EptA, PE (Fig. 2B), and is used simply for convenience in performing crystallography, we utilized molecular docking to analyze the true PE substrate by replacing the detergent in EptA-DDM (PDB code 5FGN) (Fig. 2C). This revealed a cavity in EptA for PE substrate entry/binding (Fig. 2D). The energy-scoring component of DOCK 6.7 is based on the implementation of force field scoring. The best score for PE and EptA from docking is Ϫ95.11 kcal/mol (the energy of the van der Waals interaction between PE and EptA is Ϫ85.25 kcal/mol, and the energy of electrostatic interaction The mass data for MCR-1 are adapted from the recent description by our research group (44).
Xu et al.  (Fig. 4D). However, functional/experimental evidence validating this is lacking. Thus, we applied site-directed mutagenesis to obtain 12 point mutants of MCR-2 ( Fig. 4E) and tested their roles both in vivo and in vitro ( Fig. 4F to I). Prior to functional assays, Western blot analysis demonstrated that all 12 point mutants of MCR-2 are expressed equally as well as the wild-type in E. coli (Fig. 4E). As determined from the results of LBA plate assays, each of the MCR-2 mutants (with 1 of 5 putative Zn 2ϩ -binding sites mutated) lost the ability to provide colistin resistance (Fig. 4F). Among the 7 point mutants with defects in PE lipid-recognizing sites, 3 MCR-2 mutants retained partial colistin resistance (i.e., those with the N106A [4 g/ml], S328A [2 g/ ml], and T110A [1 g/ml] mutations) (Fig. 4G), while the remaining 4 mutants were fully inactive (Fig. 4G). This is generally consistent with colistin MIC measurements (Fig. 4H).
As described for EptA and MCR-1 ( Fig. 5 and 6), we also purified all 12 mutant MCR-2 proteins to homogeneity and evaluated the role of the cavity-forming residues in the catalytic activity of MCR-2 using an in vitro enzymatic reaction with the fluorescently Level of colistin resistance (μg/ml)  4I and 5A to C) (45). Thin-layer chromatography (TLC) assays suggested that (i) 3 mutants of MCR-2 (N106A, T110A, and S328A) retain partial enzymatic activities similar to that of the wild-type MCR-2 in that an intermediate product, NBD-glycerol, is produced (Fig. 4I), while (ii) the other 9 MCR-2 point mutants have no detectable activity, as evidenced by NBD-glycerol being absent and only the substrate, NBD-glycerol-3-PEA, being observed, almost identical to the results for the negative controls (Fig. 4I). The differences in the enzymatic activities of MCR-2 mutants were further demonstrated through their ability to modify lipid A anchored on the bacterial outer membrane, as determined using MALDI-TOF MS ( Due to the distinct phylogenetic placement of EptA compared to those of MCR-1 and -2 (Fig. 3), we examined the relationship between the two protein domains (transmembrane region and PEA transferase domain) of these enzymes and their relationship with the evolution of protein function in MCR-like enzymes (Fig. 6). Domain swapping was applied in engineering an array of hybrid colistin resistance genes (Fig. 6A). In total, six separate hybrid proteins were constructed, denoted as follows: (i) TM-MCR-1, a modified MCR-1 carrying the TM region of EptA ("TM" in hybrid protein designations); (ii) TM1-EptA, a derivative of EptA with TM replaced by the TM region of MCR-1 ("TM1" in hybrid protein designations); (iii) TM-MCR-2, a modified version of MCR-2 with TM of EptA; (iv) TM2-EptA, a mosaic version of EptA in which TM is replaced with the TM region of MCR-2 ("TM2" in hybrid protein designations); (v) TM1-MCR-2, a derivative of MCR-2 with TM1 of MCR-1; and (vi) TM2-MCR-1, a modified version of MCR-1 where TM2 of MCR-2 is present (Fig. 6A). Subsequently, they were functionally evaluated via colistin MIC measurements. In agreement with our recent observations made with MCR-2 (17), only the two mosaic versions (TM1-MCR-2 and TM2-MCR-1) allow the recipient E. coli cells to exhibit appreciable resistance levels, with a MIC of 4.0 g/ml (Fig. 6B). Intriguingly, all four derivatives of MCR carrying the TM region of EptA seem to be inactive (colistin MIC of 0.25 g/ml) and have results indistinguishable from those of the negative controls (Fig. 6B). Western blot analysis demonstrates that all the mosaic proteins are expressed equally as well as the native versions, despite the apparent inconsistency in migration rates in SDS-PAGE (Fig. 6C). Subsequently, we obtained the  Table S1.
A Common Mechanism for Polymyxin Resistance ® actual masses of TM-MCR-1 (60.52 kDa) and TM1-EptA (65.85 kDa) via MS and confirmed their sequence identities with~60% coverage. More importantly, circular dichroism (CD) analyses confirmed that all the hybrid proteins exhibit the typical CD spectra of an ␣-helix structure (Fig. S4D to G), almost identical to those of EptA (Fig. S4A), MCR-1 (Fig. S4B), and MCR-2 (Fig. S4C). This validates our recent biophysical description of MCR-1 and its derivatives (44). In general consistency with crystallographic observations made by different research groups (38)(39)(40)(41)43), our results from inductively coupled plasma mass spectrometry (ICP-MS) detected the presence of zinc in the full-length MCR-1 and -2 (EptA) proteins, as well as the six mosaic derivatives (Fig. 6A;  Fig. S5). These results ruled out the possibility of improper expression/proteolytic degradation of mosaic proteins (Fig. 6C). The inconsistent apparent masses might be due to differences in charge during the separation by SDS-PAGE. This might imply that EptA and MCR-1 and -2 have incompatible TM region and PEA transferase modules ( Fig. 6A; Fig. S1), which is at least in part (if not completely) consistent with the scenario that they are evolutionarily distant (Fig. 3).
As described by Anandan et al. (45), we further assayed the enzymatic activities of EptA/MCR and derivatives in vitro (Fig. 4I, 5, and 6D). It is thought that EptA-/MCRoriginated enzymes can catalyze the cleavage of the PEA moiety from the alternative substrate NBD-glycerol-3-PEA ( Fig. 5A and B), giving a fluorescently tagged product, NBD-glycerol ( Fig. 5A to C). As expected, TLC experiments showed the following: (i) the NBD-glycerol product band consistently appears upon the addition of EptA/MCR-1/ MCR-2 protein in DDM micelles, whose migration rate is faster than that of the substrate NBD-glycerol-3-PEA (Fig. 6D); (ii) the two hybrid versions of MCR-1 and MCR-2 (TM1-MCR-2 and TM2-MCR-1) retain hydrolytic activity on NBD-glycerol-3-PEA (Fig. 6D); and (iii) the four mosaic forms of MCR with EptA modules inserted lose the enzymatic ability to transfer the PEA moiety (Fig. 6D). Of note, both the NBD-glycerol-3-PEA substrate (Fig. 5B) and its resultant NBD-glycerol product (Fig. 5C) were verified by liquid chromatography-mass spectrometry (LC-MS). Driven by the fact that an intermediate of PEA-threonine 280 (T280)-EptA is present in the soluble EptA of Neisseria meningitidis (24), we speculated that PEA might be released as an adduct of PEA-T285-MCR-1 (equivalent to PEA-T283-MCR-2) ( Fig. 1B and 5A). Subsequently, MALDI-TOF MS analyses of lipid A species were performed to trace the in vivo transfer of the PEA moiety from its physiological donor phosphatidylethanolamine (PE) to the acceptor LPS-lipid A ( Fig. 1C and 7). In addition to the single lipid A peak seen in negative controls (  (Fig. 7G to J). These MS results match the scenarios seen with TLC enzymatic trials in vitro (Fig. 6D) and assays for colistin resistance (Fig. 6B). In summary, these data suggest that protein evolution has rendered a functional differentiation among EptA, MCR-1, and MCR-2 ( Fig. 3 to 7).

DISCUSSION
Since MCR-1-mediated colistin resistance is spreading globally and is becoming a serious risk for human health, colistin will be formally banned as an animal growth booster in China. It is now being reconsidered for human clinical applications as well. In general, colistin resistance is physically dependent on bacterial surface remodeling, which interferes with (and even impairs) static-electron interaction between the cationic antibiotic peptide and its primary target, the negatively charged lipid A moiety of LPS species. Chemical modifications of the lipid A moiety anchored to the outer leaflet of the outer membrane of Gram-negative bacteria decrease the net negative charge of the surface and confer colistin resistance (Fig. 1).
Although MCR-2 was earlier suggested to confer an appreciably stronger resistance to colistin than MCR-1 in E. coli, our data here show that three colistin resistance proteins (EptA, MCR-1, and MCR-2) share a similar PE lipid substrate-recognizing cavity ( Fig. 2 and 4; Fig. S2). This parallel nature of the cavities of these three protein allows them to exploit an identical mechanism for enzymatic catalysis (Fig. 1). In fact, in addition to EptA, which according to Anandan et al. (45) utilizes a possible ping-pong mechanism of catalysis, our biochemical data confirm that MCR-2 (and/or MCR-1) might also do so. This is, probably, a common machinery among the family of PEA lipid A transferases, since they consistently share the same cavity, consisting of 12 essential residues, for the entry of the PE lipid substrate molecule. In addition to the structural  insights gained from the cavity of MCR-1 (Fig. 2), we utilized a structure-guided functional mapping approach to finely dissect the physiological roles of 12 core residues of the PE binding cavity in both EptA and MCR-2 ( Fig. 4; Fig. S2 and S3). The domains (TM region and PEA transferase) of the closely related MCR-1 and MCR-2 are functionally exchangeable (Fig. 6 and 7), while those of EptA and MCR-1 (and/or MCR-2) are not (Fig. 6 and 7), implying structural variations among PE-occupied cavities. Considering the lower level of similarity between EptA and MCR-1 (and/or MCR-2) (Fig. S1), we believe that EptA might represent an ancestral variant of MCR-1 (MCR-2). Indeed, this hypothesis is also supported by the evolutionary enhancement of the colistin resistance of MCR-1/MCR-2 (MIC of 4 g/ml) (Fig. 4 and 6B) compared to that of EptA (MIC of 2 g/ml) (Fig. 6B; Fig. S2).
In addition to the chromosomally encoded colistin resistance protein EptA, two plasmid-borne PEA transferases (MCR-1 [46] and MCR-2 [16,17], with 81% amino acid identity) that can confer colistin resistance to recipient bacterial strains have already been identified. Thus, it seems very plausible that the rapid and global dissemination of colistin resistance might proceed not in the way of clonal expansion (intrinsic resistance), but through plasmid-based conjugative transfer (transferable resistance) (27). In particular, several genetic variants of mcr-1 with point mutations have been detected, including mcr-1.2 (47), mcr-1.3 (48), mcr-1.5 (49), and mcr-1.6 (50). This highlights the possibility that mcr-1 is under antibiotic selection pressure. Since MCR-3 (558 amino acids) (31), another recently discovered PEA transferase, exhibits closer similarity to EptA (53.1% amino acid similarity and 36.4% identity) than it does to MCR1 and MCR 2 (44.1% and 45.2% similarity and 32.2% and 31.3% identity, respectively), we tend to believe that MCR-3 and other MCR proteins, like MCR-4 and MCR-5, might be EptA-like enzymes rather than MCR-1 and -2-like proteins (Fig. 3). This speculation is validated by the comparable levels of colistin resistance obtained in our trials (8 g/ml on LBA plates for both EptA [ Fig. S2F and G] and MCR-3 [not shown]). Moreover, Kieffer and coworkers also reported an MCR-like gene from Moraxella osloensis (AXE82_07515) encoding a 548-amino-acid product with 73.6% and 75.2% similarity (61.4% and 63.1% identity) to MCR-1 and MCR-2, respectively (51). The product of this MCR-like gene, AXE82_07515, only possesses 44.3% similarity and 29.1% identity to MCR-3, which might indicate a growing confusion in the nomenclature used for MCR-like genes. It is probable that mcr-3 and AXE82_07515 both might be ancestors/intermediates for mcr-1 and -2 genes, though this hypothesis needs further experimental evidence. Also, the convergent/divergent nature of the evolution of this branch of MCR-like enzymes (MCR-3, -4, and -5) from MCR-1 and -2 is unexplored. Unexpectedly, another member of the PEA transferase family, encoded by locus Z1140 (539-amino-acid product) in E. coli O157:H7 strain EDL 933 (accession number AE005271), did not play any detectable role in our trials of enzymatic activity and colistin resistance (not shown), although it displayed 54.3%, 46.8%, 47.3%, and 64.3% similarity (33.6%, 34.7%, 52.5%, and 36.2% identity) to EptA, MCR-1, MCR-2, and MCR-3, respectively ( Fig. 3; Fig. S1). The protein encoded by Z1140 might be a relic of the developing or degenerative family of PEA lipid A transferases.

MATERIALS AND METHODS
Strains, plasmids, and growth conditions. All E. coli strains and plasmids used here are listed in Table S1 in the supplemental material (52). The two strains DH5␣ and BL21(DE3)pLysS were used for DNA manipulation and protein expression, respectively. E. coli MG1655 was used a recipient for eptA, mcr-1 and -2, and their mutants. To create hybrid genes of eptA and mcr-1 and -2, overlapping PCR was performed as described by Sun et al. (17) with appropriate sets of specific primers (Table S2). The Mut Express II fast mutagenesis kit version 2 (Vazyme Biotech Co., Ltd.) with proper primers (Table S2) was used to generate an array of eptA and/or mcr-2 mutants, in which pBAD24::eptA and/or pBAD24::mcr-2 acted as a template. The isopropyl-␤-D-thiogalactopyranoside (IPTG)-inducible pET21a expression vector and arabinose-inducible plasmid pBAD24 (Table S1) were utilized for protein expression and functional assays, respectively, for EptA and/or MCR-1 and -2 or its derivatives. All the recombinant plasmids were verified by direct DNA sequencing. Both liquid Luria-Bertani (LB) broth and solid LB agar (LBA) plates were used to maintain E. coli cultures, and appropriate antibiotics, such as kanamycin and colistin, were supplemented when necessary.
Measurement of colistin resistance. The MIC of colistin was determined using a liquid broth dilution test with cation-adjusted Mueller-Hinton broth (CAMHB) as recommended by EUCAST (17). First, an overnight culture from a single colony was diluted 100-fold in fresh CAMHB medium and collected for determination of the MIC when its optical density at 600 nm (OD 600 ) reached 0.5. Subsequently, the cultures were diluted to an OD 600 of 0.05 in CAMHB medium supplemented with various levels of colistin (0, 0.25, 0.5, 1.0, 2.0, 4.0, 8.0, and 16.0 g/ml) and incubated with shaking at 200 rpm at 37°C for 16 h. The colistin MIC was recorded using optical density measurements. A concentration of 0.2% arabinose was added into CAMHB medium to induce the expression of pBAD24-borne eptA and/or mcr-1 and -2 and their derivatives in E. coli.
To judge the formation of colonies, these LBA plates were incubated at 37°C overnight and supplemented with 0.2% arabinose to promote protein expression.
Expression and purification of membrane proteins. One-liter amounts of mid-log-phase cultures of bacteria encoding MCR-1 and -2, EptA, and their variants (OD 600 of 0.8 to 1.0) were induced with 0.5 mM IPTG overnight in a shaker (220 rpm) at 18°C. Bacterial cells were centrifuged (5,000 rpm for 25 min) at 4°C prior to washing with 1ϫ phosphate-buffered saline (PBS) (17). The cell pellets were resuspended in buffer A (20 mM Tris-HCl [pH 8.0], 100 mM NaCl, 5 mM DNase I, 1 mM phenylmethylsulfonyl fluoride [PMSF], 2 mM MgCl 2 ) to 20% (wt/vol) and then lysed by passage through a French press (JN-Mini, China) at 500 lb/in 2 once and 1,300 lb/in 2 twice. Following a round of centrifugation (16,800 rpm for 1 h at 4°C), the resultant supernatant was subjected to further centrifugation (38,000 rpm for 1 h at 4°C). The pellet contained the membrane protein of interest (17,43).
LC-QTOF MS. The identities of the four chimeric proteins of MCR-1 and -2 (TM-MCR-1, TM1-EptA, TM-MCR-2, and TM2-EptA) were examined using a Waters quadrupole time of flight (QTOF) API-US mass spectrometer (53,54). A band obtained from SDS-PAGE separation of purified protein was digested with trypsin (G-Biosciences St. Louis, MO), and the resultant peptides were analyzed with a Waters Atlantis C 18 column (0.03-mm particle size, 0.075 by 150 mm). Finally, the acquired data were subjected to further analyses through the Waters ProteinLynx Global Server 2.2.5, Mascot (Matrix Sciences), and BLAST against the NCBI nr database.

ICP-MS.
As recently performed by us (44), with minor changes, inductively coupled plasma mass spectrometry (ICP-MS) was adopted to detect the presence of zinc ions in MCR-1, MCR-2, EptA, and their derivatives. The protein samples (~0.2 mg/ml) were analyzed using a NexION 300 ICP-MS instrument (PerkinElmer, USA) switched to collision-cell mode. The mass-to-charge ratio (m/z) was quantified using the kinetic energy discrimination (KED) mode with helium as the carrier gas (57).
LC-MS. As in our recently conducted experiments (44), the identities of both the substrate of NBD-glycerol-3-PEA and the product of NBD-glycerol (from the MCR-1-catalyzed half reaction) were determined by using a liquid chromatography-mass spectrometry (LC-MS) system (Agilent Technologies 6460 triple-quadrupole LC-MS) (58). The electrospray ionization (ESI) source was connected with the mass spectrometer, and the neutral loss ion (m/z, 141) mode was set to scan the positive ion. Using a Zorbax SB C 18 analytic chromatography column (2.1 by 50 mm, 3.5 m), the samples were eluted with a solution of methanol-0.1% methanoic acid (95:5) at 0.3 ml/min. Extraction, purification, and identification of LPS-lipid A. LPS-lipid A was extracted routinely according to previously established protocols (24,59,60). First, bacterial cultures grown on LBA plates with an appropriate level of colistin were stripped and washed with washing buffer containing 30 mM Tris-HCl. Lysozyme was added to the bacterial cells (suspended in a buffer of 30 mM Tris-HCl and 20% sucrose) to disrupt the cell wall. The lysates were dissolved in 3 mM EDTA and subjected to further sonication. Next, the crude LPS was precipitated following the removal of cell debris by 1 h of centrifugation at 16,000 rpm. The crude LPS was resuspended in 30 mM Tris-HCl and 0.2% SDS and subjected to successive treatments of DNase I (25 g/ml), RNase A (100 g/ml) at 37°C for 1 h and proteinase K at 37°C for 1 h. The reaction was stopped by incubation at 100°C for 1 h. The purified LPS-lipid A was acquired after the removal of SDS contaminants by two rounds of washing with acidified ethanol and 95% ethanol. The purified lipid A species in LUG loading buffer [250 mM Tris-HCl (pH 6.8), 10% (vol/vol) SDS, 1% bromophenol blue, 10% (vol/vol) glycerol, 5% (vol/vol) 2-mercaptoethanol] were visualized with SDS-PAGE (10%) coupled with silver staining, and analyzed by MALDI-TOF MS (ultrafle-Xtreme; Bruker) in negative ion mode with the linear detector (24,59,60).
Bioinformatics, structural modeling, and molecular docking. Sequence alignment of PEA lipid A transferases was conducted with Clustal Omega (https://www.ebi.ac.uk/Tools/msa/clustalo/) and processed via the program ESPript 3.0 (http://espript.ibcp.fr/ESPript/cgi-bin/ESPript.cgi) (61). The transmembrane regions of EptA and MCR-1 and -2 were predicted using TMHMM server version 2.0 (http://www .cbs.dtu.dk/services/TMHMM/). The structure of MCR-2 was modeled using Swiss-Model software (62), in which N. meningitis EptA (PDB code 5FGN) acted as the structural template (45). MCR-2 (MCR-1) exhibits 35.4% (35.6%) identity to EptA, and its modeled structure possesses a coverage score of 96% (amino acids 10 to 538) compared with that of EptA. The global model quality estimation (GMQE) score is 0.7, and the QMEAN value (which provides a global and local absolute quality estimate for the modeled structure [63]) is Ϫ4.02, implying a reliable qualified structural prediction. The ready-to-dock chemical structure of PE (ZINC identification number [ID] ZINC32837871) and head group of PE (ZINC ID ZINC02798545) were sampled from the ZINC database (64). UCSF DOCK 6 software (version 6.7) was applied to predict binding patterns of the PE molecule versus EptA and the head group of PE to MCR-1 and MCR-2 (65). LigPlotϩ was used to illustrate the diagrams for possible ligand-protein interaction (66).
Phylogenetic analyses. A collection of MCR-like proteins were aligned using Clustal Omega (https:// www.ebi.ac.uk/Tools/msa/clustalo/) (67). The phylogenetic tree was generated and visualized using TreeView (68). In addition to EptA, the MCR-like proteins included the following: MCR-1 and its 8 variants  (46). The protein encoded by locus Z1140 (accession number AAG55285), a putative PEA transferase of E. coli O157:H7 strain EDL 933, was used as an internal reference in the phylogenetic tree.