Mechanistic insights into transferable polymyxin resistance among gut bacteria

Polymyxins such as colistin are antibiotics used as a final line of defense in the management of infections with multidrug-resistant Gram-negative bacteria. Although natural resistance to polymyxins is rare, the discovery of a mobilized colistin resistance gene (mcr-1) in gut bacteria has raised significant concern. As an intramembrane enzyme, MCR-1 catalyzes the transfer of phosphoethanolamine (PEA) to the 1 (or 4′)-phosphate group of the lipid A moiety of lipopolysaccharide, thereby conferring colistin resistance. However, the structural and biochemical mechanisms used by this integral membrane enzyme remain poorly understood. Here, we report the modeled structure of the full-length MCR-1 membrane protein. Together with molecular docking, our structural and functional dissection of the complex of MCR-1 with its phosphatidylethanolamine (PE) substrate suggested the presence of a 12 residue–containing cavity for substrate entry, which is critical for both enzymatic activity and its resultant phenotypic resistance to colistin. More importantly, two periplasm-facing helices (PH2 and PH2′) of the trans-membrane domain were essential for MCR-1 activity. MALDI-TOF MS and thin-layer chromatography assays provide both in vivo and in vitro evidence that MCR-1 catalyzes the transfer of PEA from the PE donor substrate to its recipient substrate lipid A. Also, the chemical modification of lipid A species was detected in clinical species of bacteria carrying mcr-1. Our results provide mechanistic insights into transferable MCR-1 polymyxin resistance, raising the prospect of rational design of small molecules that reverse bacterial polymyxin resistance, as a last-resort clinical option to combat pathogens with carbapenem resistance.

Antibiotic resistance is becoming a global public health priority (1)(2)(3). Bacterial pathogens with multidrug resistance (MDR) 3 are considered to be a leading challenge to global public health in that they result in over 700,000 deaths each year (1,3). Colistin (referred to polymyxin E), a member of cationic antimicrobial peptides (CAMP), is used as a "last-resort" defense against serious infections caused by MDR-producing Gram-negative pathogens (5). However, it seems that its use as a final-line clinical option might be potentially disrupted by the emergence of transferable colistin resistance determinant MCR-1 (6,7). Structural alterations of lipid A species anchored on bacterial lipopolysaccharide (LPS) are implicated in the resistance to polymyxins. So far, at least three distinct molecular mechanisms have been identified for the chemical modification of lipid A: first, phosphoethanolamine (PEA) is attached to the 1 (or 4Ј)-phosphate position of the lipid A glucosamine (GlcN) moieties (8,9); second, the 1 (or 4Ј)-phosphate position of the lipid A GlcN moiety is modified with amino-arabinose (8,10); third, glycine modification occurs at the 3Ј-linked secondary acyl chain of lipid A (11). This indicates a significant diversity in the machineries responsible for colistin resistance.
Very recently, a new mobilized colistin resistance gene (mcr-1) was discovered from the gut microbiota of human beings and animals in southern China (12). Given the efficient transfer of plasmid-borne mcr-1 via transposon-like genetic elements (13,14), it is not surprising that mcr-1 has already disseminated into 40 countries covering 5 of the 7 continents (6). More than 10 species of Enterobacteriaceae been detected to carry the mcr-1 gene, including Escherichia coli (15) and Klebsiella pneumoniae (16). Also, an unexpectedly rich diversity/complexity has been found in the growth niches (and/or host reservoirs) of mcr-1-positive bacteria, like pigs (17,18), meats (19,20), human beings (12,(21)(22)(23), etc. In the last year alone, no fewer than eight types of plasmids have been found (17,21), some of which are even hybrid versions, like IncX3-X4 (24) and IncI2/IncFIB2 (25). Worryingly, the mcr-1 gene can also coexist with both ESBL (26,27) and NDM-1 (28) (or its variants NDM-5 (29) and NDM-9 (30)). This poses a serious threat to the use of carbapenem and colistin, two final lines of refuge antibiotics against Gram-negative pathogens with pan-drug resistance, suggesting that the current situation of public health is almost on the cusp of a "post-antibiotic era." MCR-1, a plasmid-encoded polypeptide (541 aa), is annotated as a member of the family of PEA transferases (12,31). These PEA transferases belong to the "YhjW/YjdB/YijP" alkaline phosphatase superfamily, which are generally engaged in the decoration of lipid A anchored on bacterial LPS (LPS-lipid A) with PEA (9,32,33). As a well-studied PEA transferase, the Neisseria meningitidis eptA protein product (of note: it was originally designated as LptA by Cox et al. (33) and was very recently renamed as EptA (EC 2.7.4.30) by Anandan et al. (34) to alleviate the confused nomenclature of LptA in that it is also used for the lipopolysaccharide transport system in E. coli (35)) catalyzes the transfer of PEA from phosphatidylethanolamine (PE) to the 1 (or 4Ј)-phosphate position of LPS-lipid A GlcN moieties (33). Consequently, this chemical modification of LPS-lipid A by NmEptA allows N. meningitidis to have an intrinsic colistin resistance because it weakens the negatively charged phosphate group on the bacterial surface interacting with colistin. As an integral membrane protein, MCR-1 features a core hydrolase fold-containing catalytic motif at the C terminus anchored by an N-terminal transmembrane (TM) domain with five helices (12,31). Genetic deletion of the TM domain completely impairs the function of MCR-1, suggesting that its presence ensures correct localization of this enzyme into the periplasmic face of the cytoplasmic membrane (31,36). When compared with the chromosomally encoded EptA (544 aa), the complete plasmid-borne MCR-1 (541 aa) has a 34.6% identity, whereas the TM region of MCR-1 (1-180 aa) is only 28.6% identical to its counterpart in EptA (1-178 aa). It is consistent with their situation of being in distinct subclades of phylogeny (Fig. S1A). Intriguingly, the expression of MCR-1 in E. coli can confer an appreciably robust growth relative to EptA (Fig. S2). Unlike the scenarios seen with MCR-1 and MCR-2 ( Fig. S3), domain-swapping assays revealed that the two TM domains (and/or catalytic motifs) from MCR-1 and MCR-2 are not functionally exchangeable (Fig. S4), implying the presence of a certain evolutionary distance (Fig. S1) and the possibility of different structural/biochemical mechanisms for MCR-1. Along with the other four research groups (37-40), we described the architecture of the catalytic domain of MCR-1 and functionally defined a 6 residue-requiring motif involved in zinc binding (and/or catalytic activity) (31,36). Moreover, a couple of MCR-1 variants with point mutations were recently discovered (e.g. MCR-1.2 (Q3L) (41), MCR-1.3 (I38V) (42), MCR-1.6 (R536H) (43), etc.), suggesting the existence of diversified mcr-1 populations under unknown selection pressures.
Our current understanding of MCR-1 polymyxin resistance remains fragmented in that only the crystal structure of the soluble catalytic domain is available (37)(38)(39)(40). In this paper, we attempt to close this knowledge gap. We report the modeled structure of the full-length MCR-1 integral membrane protein through structural modeling with the N. meningitidis EptA (PDB entry 5FGN) as a template (34). Also, we integrated molecular docking and genetic manipulation to present the functional definition of a cavity for the entry/binding of the PE lipid substrate. The in vitro enzymatic assays demonstrate that MCR-1 can remove the PEA moiety from an alternative substrate 1-acyl-2-{12-[(7-nitro-2-1,3-benzoxadiazol-4yl) amino] dodecanoyl}-sn-glycerol-3-phosphoethanolamine (NBD-glycerol-3-PEA), and the MALDI-TOF mass spectrometry shows in vivo evidence that MCR-1 transfers the PEA moiety to its recipient, LPS-lipid A (34). Collectively, these results give us a more complete picture of the structure and mechanism for MCR-1 polymyxin resistance.

Biochemical insights into MCR-1 catalysis
To determine its enzymatic activity in vitro (Fig. 1A), the recombinant form of MCR-1 integral membrane protein was overexpressed in an E. coli expression system and purified by nickel-affinity chromatography in the presence of 1% detergent dodecyl-␤-D-maltoside (DDM). Following gel filtration, the protein of interest was visualized by SDS-PAGE (Fig. 1B) and verified by MALDI-TOF MS (Fig. S5). An enzymatic system, similar to that described by Anandan et al. (34) for EptA activity, was established (Fig. 1, A and C). The PEA donor substrate is the fluorescent label NBD-glycerol-3-PEA. In principle, MCR-1 removes the PEA moiety from the substrate NBD-glycerol-3-PEA and gives a fluorescent product NBD-glycerol. Thin-layer chromatography (TLC) was used to separate the mixture in the above enzymatic reaction and the product NBDglycerol was visualized using blue light (455-485 nm) established by Anandan et al. (34). As anticipated, the TLC shows that the NBD-glycerol product band consistently appears upon the addition of MCR-1 protein in DDM micelles, whose migration rate is faster than that of the substrate NBD-glycerol-3-PEA (Fig. 1C). Moreover, chemical identities of both NBD-glycerol-3-PEA substrate (Fig. S6A) and the resultant NBD-glycerol product were confirmed with LC-electrospray ionization MS ( Fig. 1D and Fig. S6B). It clearly confirmed that the full-length MCR-1 enzymatically removes the PEA moiety from the lipid substrate. In contrast, the soluble truncated version of MCR-1 (lacking the TM region) failed to catalyze PEA hydrolysis, consistent with observations in colistin resistance assays.
To further detect the in vivo transfer of PEA moiety from glycerol-3-PEA to the acceptor substrate LPS-lipid A (Fig. 1E), LPS-lipid A was prepared and purified from E. coli strains with/ without the expression of mcr-1 (Fig. 1F). Unlike the negative control, MG1655, which has a single lipid A peak (m/z ϭ 1797.356), MALDI-TOF mass spectrometry reveals two unique peaks in the mcr-1-harboring E. coli strain: a bis-phosphorylated hexa-acylated lipid A (m/z ϭ 1797.416) and PPEA-1 (or 4Ј)-lipid A (m/z ϭ 1920.501), a modified form with an additional PEA (m/z ϭ 123) (Fig. 1G). This in vitro evidence when taken with the in vivo data demonstrates that MCR-1 catalysis proceeds by PEA transfer from the donor PE lipid substrate to

Mechanism for transferable polymyxin resistance
Mechanism for transferable polymyxin resistance the receiver Kdo 2 -lipid A, giving the two final products PPEA-Kdo 2 -lipid A and diacylglycerol (Fig. 1H). Given that observation of the Thr 280 -PEA enzyme adduct in EptA (34,44) is quite similar to scenarios in alkaline phosphatase-type phosphate transferase (45), we thus speculate that a PEA-enzyme intermediate is probably released from the PE lipid molecule in the first half-reaction of MCR-1 catalysis (Fig. S1B). In the second halfreaction, PEA is transferred from a MCR-1-bound PEA adduct to the 1 (or 4Ј)-phosphate position of Kdo 2 -lipid A GlcN moieties, generating PPEA-Kdo 2 -lipid A (Fig. S1C). It seems likely that MCR-1 might adopt a possible "ping-pong" reaction mechanism for enzymatic catalysis, similar to the proposal for EptA (34).

Structure-guided functional dissection of MCR-1
Overall structure of MCR-1 in full length was modeled using the structure of N. meningitidis EptA (Protein Data Bank entry 5FGN) as a template (34). The architecture consists of two discretely folded domains: an N-terminal TM region and a periplasm-facing catalytic domain (PEA transferase) at the C terminus (Fig. 2, A and B). These two domains are linked by an extended periplasmic loop and bridging helix (Fig. 2, A and B). The TM domain-spanning bacterial inner membrane includes six ␣-helices (designated as TMH1, TMH1Ј, TMH2, TMH3, TMH4, and TMH5; Fig. 2A), oriented approximately in parallel to one another (Fig. 2B). Similar to NmEptA, the longest helix, TMH5, is the only one of six TM helices completely spanning the membrane, whereas all of the other five helices (TMH1, TMH1Ј, TMH2, TMH3, and TMH4) are buried in the membrane in that the length of helices seems to be less than the average width of the membrane bilayer (30 Å) (Fig. 2B). It is possible that the clustered positively charged residues on THM5 (e.g. lysine) might be involved in cross-talking with the negatively charged headgroup of phospholipid on the surface of the inner membrane. Interestingly, two short periplasm-facing helices (PH2 and PH2Ј) are positioned on a long loop that connects TMH3 with TMH4 (Fig. 2, A and B), suggesting that they might be partially embedded in the membrane. Also, the linker between TMH4 and TMH5 is a short loop ( Fig. 2A). Two more periplasmic helices that are amphipathic in nature (PH3 and PH4; Fig. 2B) are localized on a coiled loop between TMH5 and a bridge helix ( Fig. 2A). Similar to the folding mode of NmEptA (34), the soluble catalytic domain also exhibits a hydrolase-like configuration comprising 10 ␣-helices (referred to as H1, H2, . . . H10) and 7 ␤-sheets (designated as S1, S2 . . . S7) (Fig. 2,  A and B). Also, the presence of a long-coiled loop adjacent to the catalytic motif probably ensures better capture/binding of the lipid substrate via its flexible rotation/movement centering on the TM region fixed on the inner membrane.
To test whether the two domains of MCR-1 can be functionally replaced with their counterparts in EptA, we employed domain swapping to generate two hybrid versions of MCR-1/ EptA (referred to as TM1-EptA, a modified EptA whose TM region is replaced with TM1 of MCR-1, and TM-MCR-1, a chimeric version of MCR-1 in which the TM of EptA is present) (Fig. 3A). In agreement with our observation on LBA plates (31), the minimum inhibitory concentration (MIC) of colistin due to EptA (2 g/ml) is less than that of MCR-1 (4 g/ml). However, the two hybrid versions of MCR-1/EptA consistently exhibited a 0.25 g/ml colistin MIC (identical to that of the colistin-susceptible strain MG1655, which served as a negative control) (Fig. 3B). This is in contrast to the results of domain swapping between MCR-1 and MCR-2 ( Fig. S3). It is possible that no expression or misfolding of the hybrid proteins (TM1-EptA and TM-MCR-1) occurs in E. coli. Indeed, not only did we employ Western blotting to validate appreciable expression of all of the four membrane proteins (Fig. 3C), but we also utilized MS to verify them ( Fig. 3C and Fig. S5 (A-D)). Similar to observations with the two parent versions of EptA (Fig. 4A) and MCR-1 (Fig. 4B), the two chimeric proteins of TM1-EptA (Fig.  4C) and TM-MCR-1 (Fig. 4D) exhibit typical CD spectra with the hallmark of being rich in ␣-helix. These results ruled out the aforementioned two possibilities. As expected, MCR-1 and EptA could hydrolyze the substrate NBD-glycerol-3-PEA, whereas the two chimeric versions (TM-MCR-1 and TM1-EptA) could not (Fig. 3D). This is opposite to the scenarios seen with domain swapping between MCR-1 and MCR-2 ( Fig. S3) (36), validating that the domains between MCR-1 and EptA are not functionally exchangeable (Fig. S4).

Discovery of a cavity of MCR-1 for PE substrate binding
In general agreement with the description of the truncated MCR-1 by different research groups using X-ray crystallography (31,(37)(38)(39)(40), the results of inductively coupled plasma mass spectrometry (ICP-MS) proved that zinc is occupied inside the full-length MCR-1 protein as well as its mosaic derivatives TM- A, scheme for the MCR-1-catalyzed cleavage reaction of the alternative substrate NBD-glycerol-3-PEA into NBD-glycerol and phosphoethanolamine. The fluorescent labeled substrate NBD-glycerol-PEA was used for subsequent detection of its resulting products separated with TLC assays. NBD is highlighted in magenta, whereas PEA is indicated in red. B, 12% SDS-PAGE profile of the purified membrane protein MCR-1. The purified protein with expected mass (ϳ60 kDa) is indicated with an arrow. C, TLC-based detection of the enzymatic activity of the MCR-1 protein in hydrolyzation of its alternative lipid substrate NBD-glycerol-3-PEA into NBD-glycerol. D, LC/MS identification of the NBD-glycerol-3-PEA lipid substrate and the resultant product NBD-glycerol. E, illustration for chemical modification of lipid A by the MCR-1 with an addition of PEA, giving the PEA-lipid A product. The 4Ј-position of the modification indicated is only a suggestion. F, silver-staining analyses of the LPS-lipid A species from E. coli with (and/or without) the expression of MCR-1. G, MALDI-TOF MS determination of the structural alteration of the LPS-lipid A species of E. coli carrying mcr-1. The peak of the bis-phosphorylated hexa-acylated lipid A appears at the position of mass (m/z ϭ 1797.416), whereas the modified form (PPEA-4Ј-lipid A) with an addition of PEA (m/z ϭ 123) exhibits a mass of 1920.501. The E. coli MG1655 carrying a plasmid-borne mcr-1 was grown on LBA plates (with 8 g/ml colistin) overnight. Then the colonies were stripped and subjected to the preparation of bacterial LPS as routinely conducted. The purified LPS and lipid A were separated with 15% SDS-PAGE and visualized using silver staining. Finally, MALDI-TOF mass spectrometry was applied to determine the alteration in the structure of the lipid A samples. H, scheme for the inner-membrane location of the MCR-1 enzyme, two substrates (PE and Kdo 2 -lipid A), and the resulting two products (PPEA-Kdo 2 -lipid A and diacylglycerol (DG)). Shown is a ribbon structure of the MCR-1 in which the catalytic domain in the periplasm is shown in green, and the trans-membrane domain is in magenta. Yellow sphere, Zn 2ϩ . M, protein marker; Ori, origin. In agreement with our former observations with both MCR-1 and MCR-2 (36), our MS result supports the addition of only one PEA to lipid A moieties. However, this might be at either the 1 or 4Ј position, and the position indicated in the illustration is merely a suggestion.

Mechanism for transferable polymyxin resistance
MCR-1 and TM1-EptA (Fig. 4E). Earlier studies suggested that Zn 2ϩ might be surrounded with five conserved catalytic residues. They correspond to Glu 246 , Thr 285 , His 390 , Asp 465 , and His 466 , respectively ( Fig. 4A and Fig. S1). Among them, Thr 285 acts as a putative nucleophile site for MCR-1 activity (Fig. 5A). It is noteworthy that our recent site-directed mutagenesis The catalytic domain is indicated in cyan, the transmembrane region is shown in magenta, the two important helices (termed as PH2 plus PH2Ј) are highlighted in orange, and PE substrate is illustrated in blue. The cavity region indicated with a white arrow is formed by the PH2 and PH2Ј helices plus the TM domain. The rectangle with yellow background denotes the layer of inner membrane. The structure of the MCR-1 was modeled using the N. meningitis EptA as a structural template (PDB entry 5FGN) (34), and the ribbon representation was generated using PyMOL. PH, periplasm-facing helices; H, ␣-helices; ␤, ␤-sheet; N, N terminus; C, C terminus. WB, Western blotting; Ori, origin.

Mechanism for transferable polymyxin resistance
assays demonstrate critical roles of these active sites in MCR-1 catalytic function and its resultant colistin resistance (31). However, we are not aware how the two lipid substrates (PE and lipid A) enter into (and/or bind to) MCR-1 enzyme.
The re-analyses of complex structure of DDM-EptA illustrates a cavity for DDM binding. Structural superposition of MCR-1 onto EptA allowed us to observe a similar DDM-binding cavity in MCR-1. Considering that DDM detergent is structurally similar to the real PE lipid substrate of EptA (Fig. S7A), it is reasonable to define a potential cavity for PE substrate entry (Fig. S7, B and C) by employing the approach of molecular docking (and/or molecular replacement). As expected, molecular docking of the PE substrate to NmEptA revealed six known active sites (Asn 106 , Thr 110 , Glu 114 , Thr 280 , His 383 , and His 465 ), which is consistent with the scenarios seen in the X-ray crystal structure of NmEptA (44) (Fig. S8). Among them, four residues (Thr 110 , Glu 114 , Thr 280 , and His 383 ) were also detected in trials of molecular docking of NmEptA to the PE headgroup alone (Fig. S9). This partially validates the feasibility of molecular docking in this situation. Because the full-length PE with flexible acyl chains was hard to dock into MCR-1, the PE molecule with acyl chains removed (and only the headgroup retained) was subjected to molecular docking, which also in turn gave four possible crucial residues (Gln 111 , Glu 116 , Thr 285 , and His 478 ) (Fig. S10), most of which have been verified in our earlier experiments (31). On the basis of the complex structure of the lipid substrate PE-EptA protein modeled by molecular docking, we replaced EptA with MCR-1 in the trials of structural superposition, which gave a modeled complex structure of MCR-1-PE (Fig. 2C). Although the complex structure might not be the best one with minimum energy, it does show clearly a potential cavity for PE substrate binding/entry into MCR-1 enzyme (Fig. 2D).
Fine analyses of the MCR-1-PE complex structure allows us to better define this cavity (Fig. 5A). This cavity has the following three elements: (i) TM region; (ii) PH2 plus PH2Ј; and (iii) parts of the catalytic domain (Figs. 2 (C and D) and 4A). In addition to the five zinc-interacting sites (Glu 246 , Thr 285 , His 390 , Asp 465 , and His 466 ) (Fig. 5A), we discovered seven additional residues (Asn 108 , Thr 112 , Glu 116 , Ser 330 , Lys 333 , His 395 , and His 478 ) from this cavity, which may be implicated in potential interaction with the headgroup of PE lipid substrate (Fig.  5A). It seems likely that the two periplasm-facing helices (PH2 and PH2Ј) contributed significantly to the recognition and occupation of the cavity by PE lipid substrate in that (i) each of the two PH helices has an essential residue (Glu 116 in PH2 and Thr 112 in PH2Ј) and (ii) the interspace between PH2 and PH2Ј harbors a critical site, Asn 108 . However, the structure-guided discovery of the cavity requires experimental evidence.

Functional evidence for PE lipid substrate-binding sites
To test our interpretation of this cavity, we systemically performed site-directed mutagenesis analyses. In total, 12 MCR-1 derivatives having one point mutation each were assayed in this study and were categorized into two groups: (i) five mutants with zinc-binding/catalytic site inactivated (E246A, T285A, H390A, D465A, and H466A) (31,36) and (ii) seven mutants with functional impairment of the PE lipid substrate-binding site (N108A, T112A, E116A, S330A, K333A, H395A, and H478A). The expression of the aforementioned MCR-1 mutants was verified with Western blotting (Fig. 6A). Subsequently, all of the 12 MCR-1 derivatives were purified to homogeneity ( Fig. 6B) and subjected to the in vitro enzymatic assays (Fig. 6C). Among them, only three point mutants (N108A, T112A, and S330A) were detected to retain the partial activity of hydrolyzing of NBD-glycerol-3-PEA into NBD-glycerol (Fig.  6C). To further determine levels of colistin resistance conferred by different MCR-1 mutants, two different approaches were applied: one is visualization of bacterial growth on LBA solid plates with varied colistin level (Fig. S11), and the other is measurement of colistin MIC in liquid broth dilution tests (Fig. 5). In the LBA plate assay, bacterial growth of negative control (MG1655 with/without the pBAD24 vector) is visualized only under 0.5 g/ml colistin, whereas the expression of mcr-1 in wildtype confers growth of the recipient strain MG1655 at up to 16 g/ml colistin (Fig. 5B). Among the seven point mutants of MCR-1 with deficiency in substrate-binding sites, four MCR-1 mutants (E116A, K333A, H395A, and H478A) are not successful at supporting the growth of MG1655 under the selective pressure of Ͼ0.5 g/ml colistin. Intriguingly, the remaining three MCR-1 versions with a point mutation retained partial  (36), migration rates of these membrane proteins (MCR-1/EptA protein and its mosaic versions) vary greatly, due to the different charges in SDS-PAGE. D, TLC-based detection of enzymatic activity for the MCR-1/EptA protein and its mosaic versions. A representative result from over three independent assays is given.

Mechanism for transferable polymyxin resistance
activity on the colistin LBA plates (i.e. 8 g/ml for N108A, 4 g/ml for S330A, and 1 g/ml for T112A). In general agreement with the scenarios we very recently observed (31,36), none of the five MCR-1 mutants lacking a full set of catalytic sites could allow the recipient strain MG1655 to grow with Ͼ0.5 g/ml colistin (Fig. 5B). Of note, the discrepancy in the maximal level of colistin tolerance on the LBA plates is in part due to different brands and batches of agar used here.
Using the liquid broth dilution tests as recommended by EUCAST with cation-adjusted Mueller-Hinton broth (CAMHB) (31,38), we further quantified the MIC of colistin in the E. coli MG1655 strain expressing the wildtype mcr-1 (and/or its mutants). In this case, the colistin MIC of the strain MG1655 with/without vector alone (negative control) is around 0.25 g/ml, whereas for the positive control strain MG1655 expressing the wildtype MCR-1, it is about 4 g/ml (Fig. 5C). Similar to MG1655, the negative control, 9 of 12 MCR-1 point mutants (four substrate binding-deficient mutants (E116A, K333A, H395A, and H478A) and five catalytically inactivated mutants (E246A, T285A, H390A, D465A, and H466A)) cannot confer a significant increment in colistin MIC (Fig. 5C). The MIC of colistin for the remaining three MCR-1 mutants (N108A, T112A, and S330A) was 2.0, 0.5, and 2.0 g/ml, respectively (Fig. 5C). Before MALDI-TOF MS detection of altered structure of lipid A pools, we carried out Western blotting to examine expression of the aforementioned mcr-1 and its 12 point mutants. As expected, all of the MCR-1 and its derivatives were well expressed in E. coli (Fig. 6A). Moreover, all of the versions of integral membrane enzyme MCR-1 (the wildtype and 12 mutants) were successfully extracted from . TM1-EptA, a derivative of EptA whose TM region is replaced with its counterpart in MCR-1 (B); TM-MCR-1, a derivative of MCR-1 whose TM region is replaced with its counterpart in EptA (A). A representative result from three independent tests is given.

Mechanism for transferable polymyxin resistance
bacterial membrane fractions and purified to homogeneity (Fig.  6B). It validated normal expression of MCR-1 and its derivatives. In contrast to the mcr-1-lacking strain MG1655 (Fig. 7, A  and B), the modified lipid A, PPEA-1 (or 4Ј)-lipid A can be detected in the mcr-1-positive E. coli (Fig. 7C). In general agreement with the low level of colistin resistance observed with the three mutant versions of MCR-1 (N108A, T112A, and S330A; Fig. 5), we detected partial enzymatic activity in the chemical modification of lipid A (Fig. 7, D, E, and I). No significant peak for PPEA-lipid A can be detected in the E. coli carrying any one of the other nine mutants of MCR-1 (Fig. 7, F-H and J-O). Evidently, the MS-based illustration of lipid A structures is completely consistent with observations in enzymatic assays in vitro (Fig. 6) and colistin susceptibility trials (Fig. 5), highlighting the varied importance of the above residues in MCR-1 activity (Figs. 5-7). Collectively, these results represent concrete in vivo evidence that a 12 residue-containing cavity is critical for the enzymatic hydrolysis of PE by MCR-1.

A role of MCR-1 in colistin resistance of gut microbiota
To assess contribution of mcr-1 to colistin resistance in clinical pathogens, we selected three representative clinical species (E. coli, Salmonella enterica, and K. pneumoniae) from gut microbiota. First, a PCR screen allowed us to determine the presence of mcr-1 (Fig. 8A). As expected, expression of MCR-1 conferred an appreciable level of colistin resistance to clinical strains ( Fig. 8B and Fig. S12). The colistin MIC of the mcr-1bearing E. coli E15017 (4.0 g/ml) was found to be 16-fold that of the mcr-1-negative E. coli MG1655 (0.25 g/ml) (Fig. 8B). In contrast to MG1655, the mcr-1-deficient S. enterica strain S14018 has a relatively higher basal level of intrinsic resistance to colistin (2.0 g/ml) (Fig. 8B). Relative to this, the mcr-1harboring S. enterica S10 exhibited a 4-fold increment in colistin resistance (up to 8.0 g/ml), suggesting an augmentation of colistin resistance by MCR-1 in Salmonella (Fig. 8B). The strain Kp253 of K. pneumoniae that carries mcr-1 also gave 8.0 g/ml colistin MIC (identical to that of the mcr-1-harboring S. enterica (8.0 g/ml)), which is 8-fold higher than that of the mcr-1-lacking K. pneumoniae RK14011 (1.0 g/ml) (Fig. 8B).
To unravel the metabolic basis for differences in colistin MIC of the aforementioned mcr-1-positive clinical species, we purified LPS-lipid A and verified it by silver staining (Fig. 8C), followed by MALDI-TOF mass spectrometry to detect structural alterations within LPS-lipid A (Figs. 8, C-I). In general agreement with our recent observations in the engineered E. coli strain with pBAD-borne expression of MCR-1 (31, 36) as well as the Liu et al. report with clinical isolates (46), one more unique Figure 5. Structure-guided functional dissection of the PE substrate-recognizable cavity. A, fine structural illustration of the cavity for the entry of PE substrate. The PE molecule is shown with blue sticks, and the zinc ion is presented as a pink sphere. Molecular docking and structural analyses allow us to anticipate that 12 residues might be critical for roles of this cavity in MCR-1 catalytic mechanism. Five residues (Glu 246 , Thr 285 , His 390 , Asp 465 , and His 466 , in green) are essential for the zinc ion, and the remaining seven amino acids (Asn 108 , Thr 112 , Glu 116 , Ser 330 , Lys 333 , His 395 , and His 478 , in red) are implicated into crosstalk with the head of PE substrate molecule. It seems likely that the two periplasm-facing helices (highlighted in orange) are critical for MCR-1 activity in that this short region has three crucial residues (namely Asn 108 , Thr 112 , and Glu 116 ) with the involvement of its binding to PE substrate. B, site-directed mutagenesis assays for the role of the 12 critical residue-containing cavity in the MCR-1 colistin resistance. Structure-guided site-directed mutagenesis was routinely conducted as recommended by the manufacturer. All of the strains tested here are listed in Table S1. The experiment of colistin susceptibility was conducted using the LBA plates containing with colistin in a series of dilutions. The value of resistance to colistin was extremely consistent in our five independent experiments of solid plate dilution; thus, a representative result is given. Vec, an empty pBAD24 vector. C, MIC comparison of colistin in the engineered E. coli strains with expression of the wildtype mcr-1 and its point mutants. Colistin MIC trials were determined using the microbroth dilution method, and the breakpoints were set as recommended by the European Committee on Antimicrobial Susceptibility Testing (EUCAST 2015, version 5.0) (24). All of the experiments of MIC determination were conducted more than three times, and the results were very consistent. As a result, colistin MIC is presented as recommended by EUCAST.

Mechanism for transferable polymyxin resistance
MS peak appeared at m/z 1920.546 in the clinical strain E15017, mcr-1-carrying E. coli (Fig. 8E), which corresponds to either PPEA-1-lipid A or PPEA-4Ј-lipid A, the lipid A (m/z 1796.565-1797.216; Fig. 8, D and E) with chemical modification of a PEA (m/s ϭ 123 units). The MS spectrum illustrates that (i) the lipid A peak is present at the m/z position of 1796.970 in the mcr-1-lacking S. enterica serovar Typhimurium strain ATCC 14028s (Fig. 8F) and (ii) two specific peaks consistently appear (one is m/z 1797.374 for lipid A, and the other is m/z 1919.199 for PPEA-1 (or 4Ј)-lipid A) in the mcr-1harboring S. enterica strain S10 (Fig. 8G). Of note, two peak forms (m/z ϭ 1824.049 and/or 1840.117) of the lipid A consistently occurred in K. pneumoniae, which is due to the variable length of acyl chains (Fig. 8, H and I). In the mcr-1-positive strain Kp253 of K. pneumoniae, PPEA-1 (or 4Ј)-lipid A was shown at the position of mass (m/z ϭ 1947.624, i.e. 1824.049 ϩ 123) (Fig. 8I). Collectively, MCR-1 modifies the chemical structure of LPS-lipid A with the addition of PEA at either the 1 or 4Ј position, which consequently leads to colistin resistance.

Discussion
The metabolic mechanism by which Gram-negative bacteria developed resistance to the CAMP-type antibiotic colistin mainly relies on the reduction of net negative charge of bacte-rial outer membrane by structural alteration of the LPS-lipid A moiety. MCR-1 is a newly identified member of the PEA transferase family having the ability to catalyze the addition of PEA to the phosphate group of the LPS-lipid A GlcN moiety (31). However, there still remains an uncertainty about whether the 1-or 4Ј-phosphate is modified. Given the fact that colistin is generally regarded as a last-resort antibiotic against bacterial pathogens with multidrug resistance, global dissemination of transferable MCR-1 colistin resistance poses a great challenge to public health. Although its genomic epidemiology has been studied extensively, biochemical/enzymatic catalysis mechanism of MCR-1 remains lacking. The data shown here report a modeled structure of the full-length MCR-1 integral membrane protein in complex with its real lipid substrate PE molecule (Fig.  2), defining a 12 residue-requiring cavity for PE substrate entry/binding (Fig. 5), and provide in vitro and in vivo evidence for enzymatic activity of MCR-1 in transfer of PEA moiety (Fig.  1). Of note, during our revision, Yin and co-workers (47) showed structures of catalytic domain of MCR-1 complexed with its substrate analogue ethanolamine and mapped several residues that partially overlap with findings in our study. However, the functional interpretation by Yin et al. (47) lacks rigorous quality control, such as the identity of NBD-glycerol product from the in vitro enzymatic reaction of MCR-1 and the in vivo role of MCR-1 in lipid A modification. Our data significantly complement this missing metabolic link.
In contrast with another lipid A enzyme modifier, ArnT, which has an N-terminal 11 TM helices and functions in the periplasm, MCR-1 enzyme possesses only five TM helices/domains at the N terminus and is active in the bacterial periplasm. Molecular docking of MCR-1 with its real PE lipid substrate (Fig. 2) offers a framework to better understand the possible interplay between MCR-1 and PE ligand. As of yet, the only information available is the structure of the closely related EptA in complex with the detergent DDM (34). In addition to the five Zn 2ϩ -binding residues that we recently elucidated to be critical for catalytic activity of MCR-1 (Fig. 5), we have now mapped seven more residues that are implicated in the formation of a substrate-binding cavity for a PE lipid molecule. More interestingly, we demonstrate the direct interaction of two periplasmic helices (PH2 and PH2Ј) on the MCR-1 transmembrane domain with the PE substrate. This verifies speculations made in this regard by Anandan et al. (34) based on the structure of NmEptA. Further structural dissection of MCR-1 revealed that all of the mutations found in the three recently discovered variants of MCR-1 (namely MCR-1.2 (Q3L) (41), MCR-1.3 (I38V) (42), and MCR-1.6 (R536H) (43)) are neither catalytic sites nor PE substrate-interacting sites. That is why they all are active. Of note, the enzymatic reaction exploited by MCR-1 is almost identical to that engaged in the chromosomally encoded homologous enzyme NgEptA (Neisseria gonorrhoea EptA). However, domain-swapping experiments suggest that MCR-1 is not functionally equivalent to NgEptA, implying the presence of functional differentiation among the family of PEA transferases. This seems to be in complete contrast to our recent findings that domains of MCR-1 and MCR-2 are functionally exchangeable (36).  A and B) was generated with two SDS-polyacrylamide gels because of limited wells per gel. C, TLC-based comparison of the in vitro enzymatic activities of the wildtype MCR-1 and its 12 point mutants. The TLC graph here was generated through a combination of three individual silica plates. All the experiments were conducted more than three times, and a representative result is given.

Mechanism for transferable polymyxin resistance
Taken together, these findings might represent a full mechanistic glimpse of transferable MCR-1 colistin resistance, providing a structural and functional basis for the rational design of small molecule compounds targeted at reversing resistance to colistin, a final line of defense antibiotic against lethal infections with MDR superbugs.

Bacterial strains, plasmids, and growth conditions
The bacterial strains used here include E. coli, S. enterica, and K. pneumoniae, respectively (Table S1) (12). PCR assays were conducted to screen for the presence of the mcr-1 gene in clin- Here, the PEA mass is 123 units, and the modification occurring at the 4Ј-position is only a suggestion. A representative result from more than three independent experiments was given.

Mechanism for transferable polymyxin resistance
ical strains like E15017, using specific primers (Table S2). The two strains of E. coli (DH5␣ and BL21 (DE3)) separately acted as the gene cloning host and the protein expression host, respectively (Table S1). The E. coli MG1655, a colistin-susceptible strain, functioned as a recipient strain of the mcr-1 gene and/or its mutants (Table S1). All bacterial cultures were maintained in Luria-Bertani (LB) broth. Solid LB agar plates supple-mented with appropriate antibiotics were used to either screen the mcr-1-containing clones or determine the level of bacterial colistin resistance.

Expression and purification of MCR-1 membrane protein
To express the MCR-1 integral membrane protein, the strain FYJ915 (BL21 with pET21a::mcr-1, a new construct that results

Mechanism for transferable polymyxin resistance
in appreciably more protein than that of the former construct pET28a::mcr-1 in our trials (21,31)), was used. Overnight cultures were inoculated at a 1:100 ratio into 2 liters of LB medium with 100 g/ml ampicillin and incubated at 37°C with shaking (220 rpm). At an A 600 of 1.0, 0.5 mM isopropyl ␤-D-1-thiogalactopyranoside was added to induce the expression of mcr-1. Cells were grown overnight and harvested by centrifugation (5,000 rpm for 20 min) at 4°C, washed once with 1ϫ PBS, and stored at Ϫ80°C until needed (48).
For large-scale purification of the MCR-1 protein, the bacterial pellets we harvested were resuspended in buffer A (20 mM Tris-HCl (pH 8.0), 100 mM NaCl, 5 mM DNase I, 1 mM phenylmethylsulfonyl fluoride, 2 mM MgCl 2 ) to 20% (m/v), lysed by a single passage through a French press (JN-Mini, Guangzhou, China) (at 500 p.s.i. once and 1,300 p.s.i twice), and then centrifuged at 16,800 rpm for 1 h at 4°C to collect the supernatant (31,36). This was further spun at 38,000 rpm for 1 h at 4°C to collect the precipitate. The pellet was then dissolved in buffer B (20 mM Tris-HCl (pH 8.0), 100 mM NaCl, 5% glycerol, 1% detergent DDM (m/v)) and subjected to further centrifugation at 38,000 rpm for 1.5 h at 4°C, giving MCR-1-containing supernatants that were incubated overnight with pre-equilibrated Ni-NTA-agarose beads at 4°C.

Circular dichroism analyses
To test the protein secondary structure and folding properties, MCR-1/EptA and its derivatives (TM1-EptA and TM-MCR-1) were subjected to routine CD assays. In each trial, 600 l of protein (ϳ0.2 mg/ml) in Tris buffer (20 mM Tris-HCl, 300 mM NaCl, 0.03% DDM, 10% (v/v) glycerol, pH 8.0) was placed into a quartz cylindrical cuvette with a path length of 2 mm. The CD spectra were collected on a Jasco model J-1500 spectrometer (Jasco Corp., Tokyo, Japan) by continuous wavelength scanning (in triplicate) from 200 to 260 nm at a scan rate of 50 nm/min (49) and smoothed with a Savitsky-Golay filter (50).

ICP-MS
To determine whether or not Zn 2ϩ is bound to MCR-1 and its derivatives, ICP-MS was applied. Briefly, the protein samples were loaded into an NexION TM 300ϫ ICP-MS instrument (PerkinElmer Life Sciences) switched to Collision-Cell mode, and then the mass-to-charge ratio (m/z) was monitored using the kinetic energy discrimination mode with helium as the carrier gas (51).

Liquid chromatography quadrupole time-of-flight mass spectrometry
The identities of two chimeric MCR-1 versions (TM-MCR-1 and TM1-EptA) were determined using a Waters Q-ToF API-US Quad-ToF mass spectrometer connected to a Waters nano Acquity UPLC as described earlier (52,53). The expected protein bands were cut from the SDS-PAGE and digested with trypsin (G-Biosciences, St. Louis, MO). The resultant peptides were loaded on a Waters Atlantis C-18 column (0.03-mm particle, 0.075 ϫ 150 mm) and subjected to MS/MS analyses (54). Finally, data analyses were conducted using the Waters Protein Lynx Global Server 2.2.5, Mascot (Matrix Sciences), and BLAST against the NCBI non-redundant database.

Mechanism for transferable polymyxin resistance
signal on the TLC plate was detected under Epi blue light (455-485 nm) with a gel imaging system (Bio-Rad) (34).

LC/MS
In addition to the pure lipid substrate of NBD-glycerol-3-PEA, the mixture of the MCR-1-catalyzed reaction was subjected to further analyses with the LC/MS system (Agilent Technologies 6460 Triple Quad LC/MS) (55). The analytical chromatographic column was Zorbax SB C18 (2.1 ϫ 50 mm, 3.5 m), and it was eluted with methanol, 0.1% methanoic acid (95:5) at 0.3 ml/min. MS was coupled with an electrospray ionization source, in which neutral loss ion (m/z 141) mode was used for the positive ion scanning.

Overlapping PCR and site-directed mutagenesis
To generate the chimeric EptA/MCR-1 protein, overlapping PCR was performed with appropriate primers (Table S2) (36). Site-directed mutagenesis was conducted to give the point mutants of mcr-1 (31). The PCR system was the Mut Express II fast mutagenesis kit V2 (Vazyme Biotech Co., Ltd.) with an array of specific primers for mcr-1 (Table S2).

Measurement of colistin resistance
To quantify the MIC of colistin in different mcr-1-carrying clinical strains and/or the E. coli MG1655 strain expressing the wildtype mcr-1 (and/or its mutants), liquid broth dilution tests were carried out as recommended by EUCAST with CAMHB (31,38). Briefly, overnight cultures were diluted 100-fold in fresh CAMHB medium and grown until A 600 reached 0.5. To measure MIC value, these cultures were diluted again to A 600 0.05 in CAMHB containing varied levels of colistin (ranging from 0 to 0.25, 0.5, 1.0, 2.0, 4.0, 8.0, 16.0, and 32.0 g/ml).
The level of colistin resistance was also determined using a solid LBA broth dilution test (21,31). Bacterial survival ability was detected as follows. Mid-log phase cultures diluted appropriately were spotted on LBA plates supplemented with colistin at different levels (ranging from 0 to 0.5, 1.0, 2.0, 4.0, 8.0, 16.0, and 32.0 g/ml) and kept at 37°C overnight. Of note, the pBAD24-borne expression of mcr-1 and its derivatives required the addition of 0.2% arabinose as an inducer into either CAMHB medium or LBA plates.

Extraction and purification of LPS-lipid A
The crude LPS was prepared as described by Caroff et al. (56) with modifications. Briefly, overnight cultures on the LBA plates supplied with 8 g/ml colistin were collected for LPS isolation. Bacterial cells (ϳ20 mg) were washed with 10 ml of Tris-HCl (30 mM, pH 8.0), centrifuged at 6,000 rpm for 10 min, and resuspended in 0.4 ml of Tris-HCl (30 mM, pH 8.0) containing 20% sucrose. After a 0.5-h incubation with 40 l of lysozyme (1 mg/ml, 100 mM EDTA, pH 7.3) on ice, the bacterial samples were kept at Ϫ80°C for 0.5 h and then thawed at room temperature. Following two rounds of the process described above, the sample was resuspended in 3 ml of EDTA (3 mM, pH 7.3) and subjected to sonication-aided lysis using a probe tip sonicator at a constant duty cycle for 2 min at 50% output (0.5 s/burst). Bacterial lysate was spun for 15 min at 6,000 rpm, and the resultant supernatant was further centrifuged at 16,000 rpm for 1 h to precipitate the crude LPS.
Then the crude LPS was freeze-dried and redissolved in the solution of 30 mM Tris-HCl (pH 8.0) with 0.2% SDS. To remove nucleic acid contamination, DNase I (25 g/ml) and RNase A (100 g/ml) were added to the solution and incubated for 2 h at 37°C, which was followed by the removal of protein contaminants via treatment with proteinase at 37°C for 1 h. To cleave the Kdo linkage, the crude LPS was heated in 10 mM sodium acetate buffer (pH 4.5) with aqueous 0.2% SDS at 100°C for 1 h and then precipitated with acidified ethanol (100 l) for the removal of SDS (56), which was followed by centrifugation (5,000 rpm, 5 min). Finally, the precipitate was subjected to two rounds of washing with 100 l of 95% ethanol and a final round of wash with 1 ml of ethanol, giving purified lipid A (44).
The purity of lipid A was judged by the method of sensitive sliver staining following separation with SDS-PAGE (57). The brief protocol was described as follows: First, the lipid A sample was dissolved in LUG buffer and separated with a SDS-PAGE (10%) gel under a constant current (10 mA) until the blue front was in the middle of the gel. Second, the gel was soaked in fixing solution consisting of 25% methanol and 7.5% acetic acid for 2 h and washed twice in 7.5% acetic acid (30 min each). Third, the gel was washed three times (30 min each), after a 5-min soak in periodic acid solution containing 7.5% acetic acid. Finally, the gel was soaked in AgNO 3 solution for 10 min, rinsed with water three times (5 min each), and developed in fresh developing solution until bands were appropriately visible. Development was terminated by rinsing for 1 h in a 1% acetic acid solution.

MALDI-TOF mass spectrometry
The chemical structures of lipid A were analyzed by MALDI-TOF/TOF MS (Bruker, ultrafleXtreme) in negative-ion mode with the linear detector (11,58). Lipid A fractions were dissolved in 20 l of chloroform/methanol (2:1) solution and mixed with 2,5-dihydroxybenzoic acid matrix in the solution of chloroform/methanol/water (3:1.5:0.25) (20 mg/ml) in a ratio of 1:1 (44). Finally, 1 l of lipid A solution was loaded onto the MALDI sample plate, giving the MS spectrum. In this case, each spectrum was derived from an average of 500 shots and 50% laser power (12,46,59).

Structure modeling and molecular docking
The 3D structure of lipid A phosphoethanolamine transferase (EptA) from N. meningitidis (PDB entry 5FGN) (34) was obtained from the RCSB PDB database (60). The structure of MCR-1 was modeled by software Swiss-Model (https:// swissmodel.expasy.org/interactive) 4 (61), using the structure of EptA (PDB accession number 5FGN) as the template. Although MCR-1 shows 35.65% identity to EptA, its modeled structure possesses a coverage score of 96% (aa 11-541) relative to EptA. In this case, the score of GMQE (global model quality estimation) is 0.7, and the value of QMEAN (which provides a global and local absolute quality estimate on the modeled structure (62)) is Ϫ3.75, suggesting a qualified structural prediction. The ready-to-dock 3D structure of PE (ID ZINC32837871) and the headgroup of PE (ID ZINC02798545) was obtained from the ZINC database (63).
The potential binding modes of the PE molecule and headgroup of PE to EptA and MCR-1 was modeled by an in silico molecular docking method using UCSF DOCK 6 software (version 6.7) (64). Concretely, protein structure was processed for molecular docking using UCSF Chimera software (65). Solvent molecules were removed. Hydrogens were added, and charges were assigned using the chimera tool Dock Prep. The active-site zinc ion of EptA was assigned a charge of ϩ2.0. The molecular surface of the protein structure was generated using the chimera tool Write DMS. Sets of overlapping spheres that represent a negative image of the surface of the protein structure were calculated by the program Sphgen_cpp. The preferred orientation of PE in EptA was searched in 20 Å of space around the complexed ligand DDM in 5FGN. A set of spheres located in the 20-Å space around DDM was calculated by the program Sphere_select. The accessory program GRID in the DOCK 6 software package was used to compute the van der Waals potential grid and electrostatic potential grid for energy scoring. Finally, the program DOCK6.7 was used to dock ligand to the designated binding pocket with default parameter configuration (Table S3). Considering that the PE molecule contains many rotatable bonds and is thus more flexible, an anchor-andgrow algorithm was set for the conformational search in docking studies. This type of flexible ligand docking allows the ligand to structurally rearrange in response to the receptor.
In principle, the cavity space of MCR-1 might be dynamic enough to hold its real PE lipid substrate with two flexible acyl chains. However, the PE molecule failed to dock into the modeled MCR-1 structure. It seems likely that the modeled MCR-1 structure might occur on a constrictive state, which consequently cannot provide enough cavity space to hold the flexible acyl chains of the PE molecule. Based on the structural shape of the PE molecule and the volume of the cavity on the protein, we speculated that the headgroup of the PE molecule contributed mainly to the binding specificity of the PE molecule to the cavity on the protein. Therefore, only the headgroup of the PE molecule was used for docking into the MCR-1 structure. This strategy can reduce the interference of the long tail group (two acyl chains) of the PE molecule for docking analysis. The two-dimensional all ligand-protein interaction diagrams were generated using LigPlotϩ (66). Hydrogen bonds are shown as green dotted lines, whereas the spoked arcs represent protein residues making non-bonded contacts with the ligand.