Accumulation of a Polyisoprene-linked Amino Sugar in Polymyxin- resistant Salmonella typhimurium and Escherichia coli STRUCTURAL CHARACTERIZATION AND TRANSFER TO LIPID A IN THE PERIPLASM*□S

Polymyxin-resistant mutants of Escherichia coli and Salmonella typhimurium accumulate a novel minor lipid that can donate 4-amino-4-deoxy-L-arabinose units (L-Ara4N) to lipid A. We now report the purification of this lipid from a pss pmrA mutant of E. coli and assign its structure as undecaprenyl phosphate-L-Ara4N. Approximately 0.2 mg of homogeneous material was isolated from an 8-liter culture by solvent extraction, followed by chromatography on DEAE-cellulose, C18 reverse phase resin, and silicic acid. Matrix-assisted laser desorption ionization/time of flight mass spectrometry in the negative mode yielded a single species [M H] at m/z 977.5, consistent with undecaprenyl phosphate-L-Ara4N (Mr 978.41). P NMR spectroscopy showed a single phosphorus atom at 0.44 ppm characteristic of a phosphodiester linkage. Selective inverse decoupling difference spectroscopy demonstrated that the undecaprenyl phosphate group is attached to the anomeric carbon of the L-Ara4N unit. Oneand twodimensional H NMR studies confirmed the presence of a polyisoprene chain and a sugar moiety with chemical shifts and coupling constants expected for an equatorially substituted arabinopyranoside. Heteronuclear multiple-quantum coherence spectroscopy analysis demonstrated that a nitrogen atom is attached to C-4 of the sugar residue. The purified donor supports in vitro conversion of lipid IVA to lipid IIA, which is substituted with a single L-Ara4N moiety. The identification of undecaprenyl phosphate-L-Ara4N implies that L-Ara4N transfer to lipid A occurs in the periplasm of polymyxinresistant strains, and establishes a new enzymatic pathway by which Gram-negative bacteria acquire antibiotic resistance.

As demonstrated in the preceding article (25), a membranebound donor, proposed to be undecaprenyl phosphate-␣-L-Ara4N 1 (Fig. 1) 2 based upon bioinformatic considerations (26,27), is required for the modification of lipid A with 4-amino-4deoxy-L-arabinose (L-Ara4N) units in polymyxin-resistant mutants of Escherichia coli and Salmonella typhimurium. A novel L-Ara4N transferase, encoded by arnT (previously designated orf5 or pmrK) (28,29), catalyzes L-Ara4N transfer to lipid A-like molecules in vitro when membranes of polymyxin-resistant mutants are employed as the source of the L-Ara4N donor (25). The formation of L-Ara4N and its transfer to lipid A are induced by activation of the transcription factor PmrA, which may occur by mutation (30 -32), by activation of PhoP (33,34), or by exposure of cells to mildly acidic pH, ferric ions, or metavanadate (26,35,36). Attachment of the positively charged L-Ara4N moiety to lipid A is critical for resistance to the antibiotic polymyxin and to certain cationic antimicrobial peptides present inside phagocytic cells (37,38).
We now report the purification and structural characterization of a novel, minor lipid that accumulates in polymyxinresistant mutants of E. coli and S. typhimurium. The purified lipid functions as a donor of L-Ara4N residues in the ArnTcatalyzed modification of lipid A in vitro. MALDI/TOF mass spectrometry and high resolution NMR spectroscopy strongly support the proposal (25,26) that the donor lipid has the structure undecaprenyl phosphate-␣-L-Ara4N (Fig. 1). The unambiguous demonstration of an undecaprenyl-linked intermediate indicates that lipid A modification with the L-Ara4N moiety occurs on the periplasmic surface of the inner membrane. Modification of lipid A with L-Ara4N units may provide a new biochemical marker for lipid A flip-flop (39) across the inner membrane.

EXPERIMENTAL PROCEDURES
Materials-32 P i and [␥-32 P]ATP were obtained from PerkinElmer Life Sciences. Silica Gel 60 (0.25-mm) thin layer plates were purchased from EM Separation Technologies. Tryptone and yeast extract were from Difco. Triton X-100 and bicinchoninic acid were from Pierce. CDCl 3 , CD 3 OD, and D 2 O were purchased from Aldrich. All other chemicals were reagent-grade and were purchased from either Sigma or Mallinckrodt.
Bacterial Strains-The strains used in the present study are described in Table I. Bacteria were usually grown at 37°C in LB medium (40), which contains 10 g of NaCl, 10 g of tryptone, and 5 g of yeast extract per liter. When needed, cultures were supplemented with 100 g/ml ampicillin, 12 g/ml tetracycline, 30 g/ml chloramphenicol, or 30 g/ml kanamycin.
Preparation of Cell-free Extracts and Membranes-Typically, 100-ml cultures of bacteria were grown to A 600 ϭ 1.0 at 37°C and harvested by centrifugation at 7,000 ϫ g for 15 min. All steps were carried out at 4°C. Cell pellets were resuspended in 50 mM Hepes, pH 7.5, at a protein concentration of 5-10 mg/ml, and broken by passage through a French press at 18,000 pounds/square inch. The crude extract was cleared by centrifugation at 7,000 ϫ g for 15 min. Membranes were prepared by ultracentrifugation at 149,000 ϫ g for 60 min, followed by resuspension and a second ultracentrifugation step to remove all cytosolic components. The final membrane pellet was resuspended in 50 mM Hepes, pH 7.5, at a protein concentration of 5-10 mg/ml. The supernatant from the first 149,000 ϫ g centrifugation step was subjected to a second ultracentrifugation to remove residual membrane particles. All samples were stored as aliquots at Ϫ80°C. Protein concentrations were determined with bicinchoninic acid (44) using bovine serum albumin as the standard.
L-Ara4N Transferase (ArnT) Assay Conditions-The L-Ara4N transferase (ArnT) was assayed under optimized conditions in a 10-l reac-tion mixture containing 50 mM Mes, pH 6.5, 0.2% Triton X-100, and 10 M of [4Ј-32 P]lipid IV A (20,000 cpm/nmol) (25). Reactions were incubated at 30°C at the indicated protein concentration and times and were stopped by spotting 5-l portions onto a Silica Gel 60 TLC plate. The substrate and product(s) were separated by developing the plate with chloroform, pyridine, 88% formic acid, water (50:50:16:5, v/v), and quantified using a Molecular Dynamics PhosphorImager equipped with ImageQuant software. ArnT-specific activity was calculated from the percentage of substrate converted to product and expressed as nmol/ min/mg (25).
Construction of a Polymyxin-resistant Mutant of E. coli K-12-Wildtype E. coli K-12 cells (W3110) were treated with 50 g/ml of the mutagen N-methyl-NЈ-nitro-N-nitrosoguanidine (40). Approximately 10 7 cells were plated on LB agar containing 2 g/ml polymyxin B sulfate (45). At this plating density, no growth of untreated cells was observed, but with mutagenesis, polymyxin-resistant colonies were recovered at a frequency of about 10 Ϫ7 . Several such colonies were purified twice on LB agar containing 10 g/ml polymyxin B sulfate. The polymyxin resistance phenotype of a representative isolate was moved into a wild-type background (W3110) by P1vir transduction (40) by selecting directly for polymyxin resistance at 10 g/ml on LB agar. To validate the chromosomal location of the resistance gene in the transductant (designated WD101), pmrA ϩ (basR) of E. coli was re-introduced into WD101 by co-transduction with a linked Tn10 transposon at 37°C, using a P1vir lysate prepared on strain AKK211 (zjd-2211::Tn10). Several of the tetracycline-resistant transductants of WD101 (2 of 14) generated in this manner had lost their polymyxin resistance, when tested for growth of single colonies on LB agar containing 10 g/ml polymyxin. This finding supports the idea that a mutation in pmrA is responsible for the polymyxin resistance of WD101. To validate this hypothesis, however, the pmrA and pmrB genes were amplified together by subjecting WD101 chromosomal DNA to polymerase chain reaction, using Pfu DNA polymerase (Stratagene) according to the manufacturer's instructions. Both genes were sub-cloned into the vector pT7Blue-3 (Novagen) using blunt end ligation at the EcoRV site. The sequences of the forward and reverse primers were 5Ј-CAGGCTGCG-GATGATATTCTGC-3Ј and 5Ј-GTTTAACTACCGTGTTCAGCGTG-3Ј respectively. The sequences of both genes were then determined to locate mutation(s). Two amino acid residues were altered in PmrA of WD101, demonstrating that this mutant is a likely pmrA constitutive (pmrA C ). No changes were found in the pmrB sequence.
Construction of a pmrA C Mutant of E. coli K12 Lacking Phosphatidylethanolamine-The presence of phosphatidylethanolamine interferes with the purification of undecaprenyl phosphate-L-Ara4N, since both compounds are zwitterionic phospholipids. The E. coli K-12 strain AD90 contains a kan insertion in the phospholipid biosynthetic gene pss, resulting in the complete absence of phosphatidylethanolamine in its membranes (46). AD90 cells harboring the temperature-sensitive  (46). Growth of AD90 cured of its covering plasmid after a shift to 42°C is slow and requires supplementation of the medium with 50 mM MgCl 2 (46). A tetracycline-resistant derivative of WD101 was selected by transducing zjd-2211::Tn10 from AKK211 into WD101, as described above, and retrieving one of the polymyxin-resistant recombinants. The desired strain, WD102 (pmrA C zjd-2211::Tn10), was used to generate another P1vir lysate with which the pmrA C allele could be transferred into AD90/pDD72 by selecting for tetracycline resistance (12 g/ml) at 30°C on LB agar, followed by screening of individual colonies for polymyxin resistance at 10 g/ml as the unselected marker. A pmrA C zjd-2211::Tn10 transductant of AD90/pDD72 was then cured of its covering plasmid at 42°C, as described previously (46), generating the phosphatidylethanolamine-deficient strain WD901 (pss93::kan pmrA C zjd-2211::Tn10). Like AD90, WD901 grew slowly on LB broth in the presence of 50 mM MgCl 2 .
Extraction and Separation of Phospholipids from 32 P i -Labeled Cells-Log phase cells were uniformly labeled with 5 Ci/ml of 32 P i in 5 ml of LB broth (or LB broth containing 50 mM MgCl 2 when appropriate) at a starting A 600 of 0.05. Cells were grown in a rotary shaker at 37 or 30°C as indicated and harvested when A 600 reached ϳ1.0. The cells were collected using a clinical centrifuge and washed with 5 ml of phosphate-buffered saline, pH 7.4. To extract the 32 P-labeled phospholipids, the cell pellet was resuspended in 3 ml of a single-phase Bligh/ Dyer mixture (47), consisting of chloroform/methanol/water (1:2:0.8, v/v). After mixing and incubating for 60 min at room temperature, the insoluble material was removed by centrifugation, and the supernatant containing the 32 P-labeled phospholipids was transferred to a clean glass tube. The supernatant was converted to a two-phase Bligh/Dyer system consisting of chloroform/methanol/water (2:2:1.8 v/v) by adding 1.2 ml of chloroform, 0.42 ml of methanol, and 1.17 ml of water. The phases were separated by low speed centrifugation, and the lower phase was removed. The sample was dried under a stream of N 2 and redissolved in a small volume of chloroform/methanol (4:1, v/v). The 32 P-labeled phospholipids (100,000 cpm/lane) were separated on Silica Gel 60 TLC plates in the solvent chloroform/methanol/water/NH 4 OH (65:25:3.6:0.4, v/v). The plate was dried and exposed to a Phosphor-Imager screen overnight to visualize the 32 P-containing bands.
Extraction of Lipid A from 32 P i -Labeled Cells-32 P i -Labeled cells were grown, harvested, and extracted as described above with 3 ml of a single phase Bligh/Dyer mixture. The insoluble residue, which contains the 32 P-labeled lipid A still covalently bound to lipopolysaccharide, was recovered by centrifugation and subjected to hydrolysis at 100°C in 12.5 mM sodium acetate buffer, pH 4.5, in the presence of 1% SDS to cleave the Kdo-lipid A linkage (26,48). The released 32 P-labeled lipid A species were extracted by the Bligh/Dyer method, and a portion was spotted onto a Silica Gel 60 TLC plate (ϳ10,000 cpm/lane) that was developed in the solvent chloroform, pyridine, 88% formic acid, water (50:50:16:5, v/v). The plate was dried and exposed to a PhosphorImager screen overnight to visualize the 32 P-lipid A species.
Large Scale Purification of the Putative Donor Lipid Undecaprenyl Phosphate-L-Ara4N-A 200-ml culture of WD901 (pss93::kan pmrA C zjd-2211::Tn10) was grown for ϳ36 h in LB broth containing 50 mM MgCl 2 to A 600 of ϳ2.0. This culture was used to inoculate four 2-liter cultures, each in a 6-liter Erlenmeyer flask, at a starting A 600 of 0.05. The cells were grown in a rotary shaker at 37°C until A 600 reached ϳ2.0, harvested by low speed centrifugation at 4°C, and resuspended in a total of 120 ml of phosphate-buffered saline, pH 7.4. The cell suspen- sion was divided equally into six 125-ml Corex tubes. The content of each tube was converted into a single phase Bligh/Dyer system by addition of 25 ml of chloroform and 50 ml of methanol. After mixing and 60 min of incubation at room temperature, the insoluble material was removed by centrifugation at 4000 ϫ g for 15 min. The supernatants were combined (570 ml final volume) and then divided equally into eight clean 125-ml Corex centrifuge tubes. The content of each tube was converted to a two-phase Bligh/Dyer system by adding 18.75 ml of chloroform and 18.75 ml of water. After mixing, the phases were separated by centrifugation at 4000 ϫ g for 15 min. The lower phases, containing the phospholipids and the putative L-Ara4N donor lipid, were pooled and dried by rotary evaporation.
The residue was dissolved in 60 ml of chloroform/methanol/water (2:3:1 v/v) and subjected to a 30-s sonic irradiation in a bath apparatus. The sample was then applied to a 20-ml DEAE-cellulose column in the acetate form at room temperature, pre-equilibrated, and washed with 200 ml of chloroform/methanol/water (2:3:1 v/v) as described previously (26,49,50). Fractions of 5 ml were collected. After the flow-through, the column was washed with another 60 ml of chloroform/methanol/water (2:3:1 v/v). The wash was followed by four separate 60-ml elution steps, using the 2:3:1 system with the aqueous phases containing either 60, 120, 240, or 500 mM ammonium acetate in ascending order. For each fraction emerging from the column, a 30-l sample was spotted onto a TLC plate, and the lipids were separated using chloroform/methanol/ water/NH 4 OH (65:25:3.6:0.4, v/v). The presence or absence of the donor lipid and other phospholipids was detected by charring with 10% sulfuric acid in ethanol. The putative donor lipid, which was initially identified by its presence as a minor component among the lipids of a pmrA C mutant and its absence in wild type, emerged in the flowthrough and the first 25 ml of the wash fractions, consistent with its net charge of zero ( Fig. 1). Fractions containing the donor lipid were pooled (85 ml) and converted to a two-phase Bligh/Dyer system (see above). The phases were separated by low speed centrifugation, and the lower phase, which contained the donor lipid, was dried by rotary evaporation.
Reverse phase C 18 chromatography was used as the second step of the purification. Solvent A, consisting of acetonitrile/water (1:1, v/v), and solvent B, consisting of isopropanol/water (85:15, v/v), were used in various ratios, and all solutions contained 10 mM tetra-butyl ammonium dihydrogen phosphate. A 4.0-ml C 18  An additional fractionation on silicic acid was necessary to obtain pure donor lipid. The above residue was dissolved in 10 ml of chloroform/methanol (95:5, v/v) and loaded onto a 1-ml acid-washed silicic acid column (Bio-SilA from Bio-Rad), equilibrated in chloroform/methanol (95:5, v/v). After the flow-through, the column was washed with 5 ml of chloroform/methanol (95:5, v/v). The lipid was eluted using 5 ml of chloroform/methanol (80:20, v/v), and the appropriate fractions (1 ml each) were dried by rotary evaporation. To remove tetra-butyl ammonium dihydrogen phosphate carried over from the C 18 column, the solvent was converted into a two-phase Bligh/Dyer system (as described above). The lower phase, containing the donor lipid, was washed 3 times with fresh upper phase. Although a small amount of residual tetrabutyl ammonium dihydrogen phosphate remained behind, as judged by NMR studies (see below), about 98% was removed by a single two-phase partitioning.
Mass Spectrometry-Spectra of the purified donor lipid were acquired in the negative and positive linear modes using a time of flight matrix-assisted laser desorption ionization (MALDI/TOF) mass spectrometer (Kompact MALDI 4, Kratos Analytical Manchester, UK), equipped with a nitrogen laser (337 nm), 20-kV extraction voltage, and time-delayed extraction. Each spectrum was the average of 50 laser shots. The instrument was operated at a resolution of about Ϯ1 atomic mass units for compounds with M r ϳ2000. Saturated 6-aza-2-thiothymine in 50% acetonitrile and 10% tribasic ammonium citrate (9:1, v/v) served as the matrix in both the positive ion and the negative ion modes. The lipid samples were dissolved in chloroform/methanol (4:1), mixed 1:1 with the specified matrix, and dried at room temperature prior to analysis.
NMR Spectroscopy-Approximately 0.2 mg of the purified donor lipid was dissolved in 0.6 ml of CDCl 3 /CD 3 OD/D 2 O (2:3:1, v/v) in a 5-mm NMR tube. NMR spectra were obtained at 25°C using Varian Inova 800, Inova 600, or Unity 500 spectrometers, as indicated, equipped with Sun Ultra 10 computers and 5-mm Varian probes. The 2 H signal of CD 3 OD was used for a field frequency lock. Use of the CDCl 3 /CD 3 OD/ D 2 O solvent system introduces four solvent resonances. The signals from CH 3 OD (3.3 ppm) and CHCl 3 (7.6 ppm) do not overlap with the donor lipid resonances. The HOD (4.6 ppm) and CD 3 OH (4.8 ppm) signals are removed with a presaturation sequence. 1 H NMR spectra at 800 MHz were obtained with 8.2-kHz spectral width, 73°pulse flip angle (7 s), 5.0 s acquisition time, and 1.2-s relaxation delay. The spectra were digitized using 82,000 points yielding a digital resolution of 0.2 Hz/point. 1 H NMR spectra at 600 MHz were obtained with 5.7-kHz spectral width, 63°pulse flip angle (6 s), 5.6-s acquisition time, and 1.2-s relaxation delay and were digitized using 64,000 points to give a digital resolution of 0.18 Hz/point. 1 H NMR spectra at 500 MHz were obtained with 4.5-kHz spectral width, 74°pulse flip angle (7.5 s), 5.0-s acquisition time, and 1.2-s relaxation delay and were digitized using 45,000 points to obtain a digital resolution of 0.20 Hz/point. 31 P NMR spectra at 202 MHz (500 MHz field) and selective inverse ( 31 P) decoupling difference spectra were obtained as described previously (51)(52)(53). Two-dimensional-COSY, HMQC, and HMBC analyses were performed at the 800-and 600-MHz fields, based on the experiments described previously (51) at 500 MHz.

Isolation and Characterization of a Polymyxin-resistant Mu-
tant of E. coli K-12-To facilitate the purification of the proposed undecaprenyl phosphate-L-Ara4N donor lipid, a new polymyxin-resistant mutant of E. coli K-12, designated WD101, was isolated by chemical mutagenesis of W3110. WD101 formed single colonies on agar containing 10 g/ml polymyxin. The resistance phenotype was located near the pmrA/pmrB locus at minute 93 on the E. coli chromosome (30), as judged by P1vir transduction. Polymerase chain reaction-based cloning and DNA sequencing of the pmrA/B cluster of WD101 showed that two amino acids were altered in the transcription factor PmrA (A42T and G53E). No mutations were present in pmrB.
TLC analysis of 32 P-labeled lipid A species isolated from WD101 showed extensive modification with polar moieties, when compared with wild-type W3110 (Fig. 2, left panel). The modified lipid A components of WD101 were generally more hydrophilic, indicating the presence of additional L-Ara4N and/or pEtN substituents (26,53), as judged by TLC analysis (Fig. 2, left panel) and mass spectrometry (not shown). Lipid A species containing an extra palmitoyl group were not detected in large amounts in WD101, consistent with the activation of pmrA but not phoP. These findings are in accord with earlier studies (32) of polymyxin-resistant mutants of E. coli, which were not characterized by DNA sequencing.
Accumulation of a Novel Minor 32 P-Labeled Lipid in WD101 and in Certain Strains of S. typhimurium-The major phospholipids of WD101 consist of phosphatidylethanolamine, phosphatidylglycerol, and cardiolipin, as in the wild-type W3110 (Fig. 2, right panel). However, a minor 32 P-labeled substance, designated "L-Ara4N Donor," accumulates in WD101 to about 0.2-0.4% of the total (Fig. 2, right panel).
The presence or absence of this putative L-Ara4N donor lipid was further investigated using well characterized strains of S. typhimurium containing mutations in the phoP/phoQ or pmrA/ pmrB systems (28,29,33,54). As in WD101, the putative donor lipid accumulates to the highest levels in S. typhimurium strains in which PmrA is constitutively active (pmrA C ) (Fig. 3,  lane 4). However, in contrast to E. coli W3110 (Fig. 2), the lipid is also detectable in wild-type S. typhimurium (Fig. 3, lane 1). This finding is consistent with the fact that lipid A of wild-type S. typhimurium grown on LB broth contains some L-Ara4Nmodified species, albeit at lower levels than in pmrA C mutants (26,53).
The putative donor lipid was absent in a S. typhimurium pmrA Ϫ mutant (Fig. 3, lane 5) and was also missing in pmrA C mutants containing deletions in either the pmrE or pmrF genes (data not shown). The latter are required for the modification of lipid A with the L-Ara4N moiety and for the maintenance of polymyxin resistance (28). Furthermore, phoP/phoQ regulation was evident in an otherwise wild-type background, given the increase in the minor lipid in a PhoP C setting (Fig. 3, lane 2) and its complete absence in a phoP Ϫ mutant (Fig. 3, lane 3). Because its presence or absence was regulated in the same manner as reported previously for L-Ara4N addition to lipid A in both E. coli (30,32) and S. typhimurium (28,29,55), we decided to purify and characterize the compound.
Construction a pmrA C E. coli Mutant (WD901) Lacking Phosphatidylethanolamine-Purification of the putative donor lipid from WD101 or from a pmrA C mutant of S. typhimurium was not feasible because it could not be separated from phosphatidylethanolamine, the major phospholipid of these bacteria (56), in various large scale chromatography steps (data not shown). To solve this problem, the pmrA C allele of WD101 was transferred into E. coli AD90/pDD72 (46) by P1vir transduction, generating WD901/pDD72. After being cured of the covering plasmid pDD72, these strains lack phosphatidylethanolamine because of an insertion mutation in their chromosomal copy of the pss gene. As shown in Fig. 4, lane 2, the presence of the pmrA C mutation in WD901/pDD72 resulted in the accumulation of the putative L-Ara4N donor lipid. Once cured of pDD72, the donor lipid was still produced by WD901, despite the complete absence of phosphatidylethanolamine (Fig. 4, lane 4). Neither AD90/pDD72 nor AD90 contained this minor lipid (Fig.  4, lanes 1 and 3).
To make certain that the pss mutation had no effect on the transfer of the L-Ara4N moiety to lipid A, 32 P-labeled lipid A species were prepared from AD90 (pss Ϫ ) and WD901 (pss Ϫ pmrA C ) in the presence or absence of the pss ϩ covering plasmid pDD72. As judged by TLC analysis, the pattern of lipid A modifications of WD901/pDD72 (Fig. 5, lane 2) was identical to that of WD101 (pss ϩ pmrA C ) (Fig. 2, left panel). Upon removal of the pss ϩ covering plasmid pDD72, however, the three lipid A species containing the pEtN moiety disappeared (Fig. 5, lane  4), whereas incorporation of the L-Ara4N moiety was unaffected. The lipid A species of AD90 grown with or without the covering plasmid (Fig. 5, lanes 1 and 3) were essentially the same as wild type (Fig. 2). These findings demonstrate that FIG. 5. Formation of pEtN-modified lipid A species requires phosphatidylethanolamine in E. coli. The 32 P-labeled lipid A species from the indicated strains were isolated as described under "Experimental Procedures," separated by TLC using the solvent chloroform, pyridine, 88% formic acid, water (50:50:16:5, v/v), and visualized with a PhosphorImager. A small amount of lipid A 4Ј-monophosphate may arise as a decomposition product during pH 4.5 hydrolysis at 100°C to remove the Kdo moiety. phosphatidylethanolamine is the source of the pEtN moieties attached to lipid A in pmrA C mutants. About 70% of the lipid A of WD901 consisted of a species with a single L-Ara4N substituent (Fig. 5, lane 4), presumably attached to the 4Ј-phosphate position (52).
The relative resistance of AD90 and WD901 to polymyxin could not be evaluated, because 50 mM Mg 2ϩ (required for growth of both strains) interfered with the disc diffusion assays (data not shown).
Purification of the Putative L-Ara4N Donor Lipid from WD901-A 100-mg sample of WD901 phospholipids (Fig. 6,  lane 1) was applied to a 20-ml DEAE-cellulose column equilibrated in chloroform/methanol/water (2:3:1, v/v). As expected, phosphatidylglycerol and cardiolipin bound to the column, whereas the putative donor lipid emerged in the run through (Fig. 6, lane 2). This finding is consistent with the proposed zwitterionic character of the donor lipid (Fig. 1). If pss ϩ cells had been employed, phosphatidylethanolamine would also have been present in the run through.
The DEAE-cellulose run-through fraction was applied to a 4.0-ml C 18 reverse phase column, prepared in an acetonitrile/ water/isopropyl alcohol system containing tetra-butyl ammonium dihydrogen phosphate, and eluted with increasing amounts of isopropyl alcohol (Fig. 6, lane 3). The final step was chromatography on a 1-ml silicic column, yielding a homogenous preparation (Fig. 6, lane 4) as judged by TLC analysis.
The apparent R f of the donor lipid increased slightly after the reverse phase and silicic acid steps (Fig. 6, lanes 3 and 4), possibly arising from interaction with contaminating tetrabutyl ammonium dihydrogen phosphate. When most of this material was removed by two-phase Bligh/Dyer partitioning, the purified lipid (ϳ0.2 mg) migrated exactly like the desired component of the original sample (Fig. 6, lane 5).
MALDI/TOF Mass Spectrometry of the Purified Lipid-When analyzed in the negative mode, the purified material yielded a major ion at m/z 977.5 (Fig. 7A), consistent with [M-H] Ϫ for the proposed structure (Fig. 1), undecaprenyl phosphate-L-Ara4N (M r ϭ 978.41). No peak was observed at m/z 1057.4 atomic mass units, the [M-H] Ϫ expected for a hypothetical undecaprenyl diphosphate-L-Ara4N. Analysis of the purified sample in the positive ion mode produced a major peak at m/z 767.9 atomic mass units, which could be interpreted as arising from the fragment [undecaprenol ϩ H] ϩ (Fig. 7B). The molecular weight of undecaprenol is 767.3.
Mass spectrometry (Fig. 7) indicates the presence of 11 isoprene units in the purified donor lipid. Keeping in mind that the isoprene unit contains 2 methyl groups (one cis and one trans in relation to the methine proton), a total of 12 methyl groups are expected ( Fig. 1 and Table III). After correcting for the overlap of the C-2 methylene proton signal arising from residual tetra-butyl ammonium dihyrogen phosphate (designated ϫ at the top of Figs. 9B and 10B), the 1 H NMR signals between 1.6 and 1.75 ppm integrate to 12 methyl groups. The methyl signal at 1.74 ppm (d, long range J ϭ 1.0 Hz) (see Fig.  1 in the Supplemental Material) corresponds to the methyl protons that are in cis configuration relative to the methine proton of the ␣ isoprene residue (Figs. 1, 9B, and 10B). The methyl signals at 1.70 and 1.69 ppm integrate to 6 and 2 methyl groups, respectively, corresponding to the 7 cis-methyls of the interior isoprene units and the cis-methyl of the isoprene unit (Figs. 1, 9B, and 10B and Table III). The methyl signals at 1.63 and 1.61 ppm (Figs. 9B and 10B) integrate to 1 and 2 methyl groups, respectively, and are ascribed to the three trans-methyls (as defined in relation to their respective methine protons) of the -2, -1, and isoprene units ( Fig. 1 and Table III).
Characterization of the Purified Lipid by Two-dimensional 1 H NMR Spectroscopy-To our knowledge, intact samples of undecaprenyl phosphate derivatives isolated from natural sources have not been studied previously by two-dimensional NMR methods because of their instability in the available solvents (18,19). However, as noted in our previous work (51) with E. coli lipid A, many natural lipids, including undecaprenyl phosphate-L-Ara4N, are stable in CDCl 3 /CD 3 OD/D 2 O (2: 3:1, v/v) for weeks and display sharp, well resolved resonances.
The positions of the individual protons of the arabinose sugar were derived from a two-dimensional COSY analysis (Fig. 8C, Table II The two-dimensional COSY also revealed a strong crosspeak between the CH and proximal CH 2 groups of the ␣ isoprene unit (Fig. 8C). The CH and proximal CH 2 of the ␣ isoprene unit each showed a weak cross-peak to the resolved methyl signal at 1.74 ppm (see Fig. 1 in Supplemental Material). These weak cross-peaks arise from four-and five-bond long range couplings and strengthen the assignment of the 1.74 ppm methyl signal to the methyl group of the ␣ isoprene unit. The major unresolved polyisoprene CH signals (ϳ 5.1 ppm) similarly show strong cross-peaks to the upfield CH 2 resonances and weaker (long range) cross-peaks to the upfield CH 3 signals (see Fig. 1 in the Supplemental Material).
The low field shift of H-1 (4.90 ppm) and the large J 1,2 coupling constant (6.4 Hz) indicate that H-1 of the L-Ara4N residue is in the axial position (57), so that the undecaprenyl phosphate chain must be situated equatorially. The large J 2,3 coupling (8 Hz 1 through H-5Ј), the methine and proximal methylene protons of the ␣ isoprenyl unit, and the bulk unresolved isoprenyl methine protons. This spectrum was obtained using a presaturation sequence to remove the HOD (4.8 ppm) and CD 3 OH (4.5 ppm) resonances, giving a clear sugar region. The large signal at 3.34 ppm (ϫ) arises from residual CD 2 HOD, and the multiplet at 3.20 ppm (ϫ) arises from the C-1 methylene protons (the ones closest to the N atom) of the butyl chains of residual tetra-butyl ammonium dihydrogen phosphate carried over from the reverse phase step of the purification. B, the selective inverse decoupled difference 1 H NMR spectrum obtained at 500 MHz with on and off resonance 31 P decoupling of the Ϫ0.44 ppm phosphorus signal shows that the anomeric carbon of L-Ara4N is linked via a phosphodiester to the proximal isoprene unit. C, the partial two-dimensional 1 H-1 H COSY analysis at 800 MHz at 25°C shows the connectivities between the key sugar and isoprenyl 1 H resonances seen in A.
between H-1, H-3, and H-5, supporting the idea that these are all axial protons on the same face (below plane in Fig. 1) of the sugar. The equatorial H-4 (below plane) showed NOE crosspeaks to H-3 (axial, below plane) and to both H-5 protons but with a stronger NOE cross-peak to H-5 (axial, below plane) and a weaker NOE to H-5Ј (equatorial, above plane) (see Fig. 2 in the Supplemental Material). The large coupling (12.8 Hz) be-tween H-5 and H-5Ј (Table II) is typical of geminal methylene protons within constrained pyranose rings (57).
Selective 31 P-Decoupled 1 H Difference Spectroscopy-The linkage between the undecaprenyl chain and the arabinose sugar unit was investigated. One-dimensional 31 P NMR spectroscopy revealed a single phosphorus resonance near Ϫ0.44 ppm, consistent with a phosphodiester linkage (see Fig. 3 in the Resonances designated ϫ are due to residual solvents or other impurities. B, this expansion shows the partially resolved isoprenyl CH 2 and CH 3 groups and their directly bonded carbon atoms. The cis, trans, and adjacent configurations are defined in Table III. The butyl chains of tetra-butyl ammonium dihydrogen phosphate, carried over from the reverse phase chromatography step of the purification, give rise to multiplet 1 H resonances at 3.20, 1.66, 1.44, and 1.04 ppm (see Supplemental Material Fig. 1). The tetra-butyl ammonium dihydrogen phosphate resonance (ϫ) at 1.66 ppm partially overlaps with two of the undecaprenyl CH 3 signals of the donor lipid. The donor lipid sample used for the HMQC and HMBC experiments also displayed CH 3 (1.19 ppm) (not shown) and CH 2 (3.63 ppm) signals arising from small amounts of an ethanol impurity in the sample, which resulted in masking of the L-Ara4N H-2 by the ethanol CH 2 signal in A. This impurity is not present in the experiment shown in Fig. 8, in which an additional two-phase Bligh-Dyer partitioning was carried out to remove the contaminating ethanol.  13 C NMR assignments and coupling constants (J, Hz) of the L-Ara4N moiety in the purified donor lipid 1 H chemical shifts (ppm from internal TMS) are from one-dimensional 1 H NMR spectra with a digital resolution of 0.2 Hz per point. Coupling constants (J H,H , Hz) and (J 1,P , Hz) were obtained from 31 P-decoupled 1 H NMR spectra. 13 C chemical shifts (ppm from internal TMS) are estimated from 2D HMQC spectra.   Supplemental Material) (58). Next, subtraction of two 1 H NMR spectra obtained with on and off resonance selective decoupling of the Ϫ0.44 ppm phosphate signal (51, 52) revealed simultaneous decoupling changes at the anomeric H-1 signal of the L-Ara4N moiety and at the proximal CH 2 signal of the ␣ isoprene residue (Fig. 8B), thereby establishing the presence of a single phosphate group linking the arabinose C-1 carbon and the proximal CH 2 carbon of the ␣ isoprene unit of the undecaprenyl chain (Fig. 1).
Evaluation of the Carbon Structure of the Donor Lipid by HMQC Spectroscopy-To confirm the assignments derived from the 1 H NMR analysis, 13 C data for the donor lipid (ϳ0.2 mg) were obtained indirectly through 1 H-detected HMQC and HMBC two-dimensional NMR experiments. The partial twodimensional HMQC 1 H-13 C correlation map (Fig. 9A) reveals six direct 1 H-13 C single-bond correlations in the sugar region (Table II). The L-Ara4N H-1 signal reveals the anomeric carbon resonance at 99.5 ppm (C-1). The H-5 and H-5Ј multiplets correlate to a single carbon signal at 63.0 ppm (C-5), whereas the H-2 and H-3 multiplets connect to carbon resonances at 72.4 (C-2) and 70.8 (C-3) ppm, respectively, as predicted for oxygen-substituted carbon atoms of sugars. However, nitrogensubstituted carbons of amino sugars resonate near 50 -55 ppm (51,59). The H-4 multiplet shows a prominent cross-peak near 51 ppm, confirming C-4 as the site of the amino group substitution. Fig. 9A also shows the direct bond correlations from the major unresolved methine proton signals of the undecaprenyl chain to unresolved olefinic carbon signals near 126 ppm, and from the methine and the proximal methylene protons of the ␣ isoprene unit to carbon resonances at 122.8 and 63.5 ppm, respectively.
The bulk CH 2 protons of the undecaprenyl chain yield four distinct 1 H-13 C HMQC cross-peaks ( Fig. 9B and Table III). Based upon the multibond correlations discussed below, the major carbon peak at 27.2 ppm is assigned to the proximal CH 2 groups of the bulk isoprene units (i.e. the ones adjacent to a methine proton as shown in Table III), whereas the 33.0 ppm peak of about equal intensity is assigned to those CH 2 groups of the bulk isoprene units that are trans relative to a methine proton ( Fig. 1 and Table III). The smaller carbon cross-peak at 40.5 ppm arises from the cis-CH 2 groups of the -1 and -2 isoprene units (Fig. 1). The small cross-peak at 32.8 ppm arises from the trans (distal) CH 2 group of the ␣ isoprene unit ( Fig. 1 and Table III).
The CH 3 protons yield six distinct HMQC peaks (Fig. 9B). The cis-CH 3 protons of the ␣ isoprene unit (1.74 ppm), six of the seven cis-CH 3 protons of the interior ␤ to Ϫ3 isoprene units (1.70 ppm), and the remaining interior cis-CH 3 group that overlaps with the cis-CH 3 of the unit (1.69 ppm) yield distinct carbon cross-peaks at 23.8, 24.1, and 26.3 ppm, respectively (also see Table III). The three trans methyl groups of the -2, -1, and isoprene units yield three distinct carbon crosspeaks at 16.8, 16.8, and 18.2 ppm, respectively ( Fig. 9B and Table III).
Evaluation of the Carbon Structure of the Donor Lipid by HMBC Spectroscopy-The HMBC multibond correlations in the sugar region (Fig. 10A) verify the L-Ara4N assignments derived from the COSY and HMQC experiments. For example, H-5Ј at 3.96 ppm shows distinct multibond correlations to C-4 (51.0 ppm), C-3 (70.8 ppm), and C-1 (99.5 ppm). The HMBC correlations also yield a complete analysis of the ␣ isoprene unit. The CH 2 OP proton signal at 4.44 ppm (also see Fig. 8A and Table III) shows distinct multibond correlations to carbon resonances at 122.8 ppm (the methine carbon of the ␣ isoprene unit) and 142.0 ppm (the quaternary carbon of the ␣ isoprene unit). The methine proton signal at 5.41 ppm yields multibond carbon peaks at 23.8 ppm (the cis-CH 3 group of the ␣ isoprene unit) and at 32.8 ppm (the distal trans CH 2 of the ␣ isoprene unit). Scrutiny of the cross-peaks from 1.74 ppm methyl proton signal (Fig. 10B) shows the corresponding multibond correlations to 32.8, 122.8, and 142.0 ppm, thus verifying the assign-  Hz) of the undecaprenyl chain in the donor lipid The abbreviations used are: adj, adjacent in relation to a methine proton of an isoprene unit; likewise, cis and trans are used throughout to designate the configurations of various groups in relation to a methine proton of an isoprene unit, as shown above.
The abreviations used are: adj, adjacent in relation to a methine proton of an isoprene unit; likewise, cis, and trans are used throughout to designate the configurations of various groups in relation to a methine proton of a isoprene unit, as shown above. ** Proton-phosphorus coupling constant. ment of the 1.74 ppm signal as the cis-methyl group of the ␣ isoprene unit. Similarly, the major unresolved CH proton signals (ϳ5.1 ppm) of the undecaprenyl chain yield multibond correlations to various CH 3 and CH 2 carbon resonances between 16 and 41 ppm (Fig. 10A). The major CH 2 and CH 3 proton signals show multibond correlations to methine carbon peaks near 126 ppm and to quaternary carbon signals near 136 ppm (Fig. 10B). The trans-CH 3 signals of the -1 and -2 units show a distinct connectivity to the cis-CH 2 carbons at 40 ppm, thus verifying assignment of the upfield shifted CH 2 proton signals (at 2.0 ppm) as cis-CH 2 groups of the -1 and -2 units of the undecaprenyl chain. The interior ␤ to Ϫ3 cis-CH 3 signals at 1.70 and 1.69 ppm show multibond correlations exclusively to the CH 2 carbons at 33.0 ppm, and not to the CH 2 carbons at 27.0 ppm, thus yielding the assignment of the 33.0 ppm CH 2 carbon signals to the CH 2 groups trans from the CH proton (Table III). Finally, multibond correlations are seen in Fig. 10B from the trans CH 2 proton signals at 2.06 ppm to the adjacent CH 2 carbon signals at 27.0 ppm, and from the adjacent CH 2 proton signals at 2.08 ppm to the trans CH 2 carbon signals at 33.0 ppm. This cross-peak pattern is the reverse of that observed in the direct correlation experiment (Fig. 9B).
The 1 H and 13 C NMR assignments derived for the sugar and the undecaprenyl moieties of the donor lipid are summarized in Tables II and III. The 13 C NMR assignments derived for the to -2 units of the undecaprenyl chain are in excellent agreement with the 13 C NMR assignments of farnesol, ␤-farnesene, ␤-springene, eleganediol, and other linear terpenes (60,61). Taken together with the coupling constants and the mass spectrometry, the results provide unequivocal proof for the novel glycolipid structure proposed in Fig. 1, i.e. undecaprenyl phosphate-␣-L-Ara4N.
Reconstitution of L-Ara4N Transferase Activity in Vitro with Purified Undecaprenyl Phosphate-␣-L-Ara4N-Transfer of the L-Ara4N moiety to lipid IV A to form lipid II A proceeds rapidly when ArnT is overexpressed in the polymyxin-resistant E. coli host BLR(DE3) (Fig. 11) (25). However, when ArnT is overexpressed in the E. coli K-12 strain NovaBlue(DE3), transferase activity is not observed (Fig. 11). Like all other E. coli K-12 strains, NovaBlue(DE3) does not synthesize the L-Ara4N donor lipid (data not shown) and is sensitive to polymyxin (25). However, addition of purified undecaprenyl phosphate-␣-L-Ara4N to membranes of E. coli K-12 NovaBlue(DE3) cells overexpressing arnT reconstitutes robust lipid II A formation in vitro in a concentration-dependent manner, whereas donor lipid addition to membranes from the vector control cells NovaBlue(DE3)/ pET21 does not (Fig. 11). Efficient reconstitution was also seen FIG. 10. Partial HMBC spectra of the purified donor lipid. A, the expansion shows multibond 1 H-13 C correlations from the L-Ara4N group and the ␣ isoprenyl methine and proximal methylene unit. Noise streaks arise from HOD and ethanol signals at 4.56 and 3.6 ppm. The major HOD signal has been suppressed with a presaturation pulse. The broad envelope signal at 4.53 ppm is a "probe-hump" artifact from a small residual water magnetization being detected at an off resonance position by the leads of the probe coil. ϫ indicates residual solvent or impurity resonances. B, this expansion shows the region of the partially resolved bulk isoprenyl CH 2 and CH 3 groups. ϫ indicates residual solvent or impurity resonances. when using membranes of E. coli K-12 NovaBlue(DE3) overexpressing S. typhimurium arnT (data not shown). ArnT activity, which was absent in a pmrE Ϫ or pmrF Ϫ derivative of a pmrA C mutant of S. typhimurium, was likewise recovered by the addition of purified donor lipid to membranes (data not shown). These findings, in conjunction with product analysis presented in the accompanying article (25), demonstrate conclusively that the addition of the L-Ara4N moiety to lipid A requires undecaprenyl phosphate-␣-L-Ara4N as the donor substrate.

DISCUSSION
The lipid A moiety of lipopolysaccharide of certain Gramnegative bacteria can be modified with the positively charged sugar L-Ara4N (58,(62)(63)(64)(65)(66) in a regulated manner (26,28,55). By masking of the negative phosphate groups of lipid A, the L-Ara4N residue prevents binding of cationic antimicrobial peptides and polymyxin to the outer membrane, rendering the bacteria resistant to killing by these substances (30 -32, 37). In the intracellular pathogen S. typhimurium, modification with the L-Ara4N unit may occur on both the 1-and the 4Ј-phosphate groups of lipid A (53,58). Modification of the 4Ј-position is predominant under most conditions (52, 53) but appears to be Kdo-dependent.
As was suggested by our bioinformatic analysis of the pmrF operon (26), a polyisoprene-linked intermediate is required for the transfer of the L-Ara4N moiety to lipid A. Here we have presented the identification, purification, and biophysical characterization of this novel substance, which is now properly described as undecaprenyl phosphate-␣-L-Ara4N or undecaprenyl 4-amino-4-deoxy-␣-L-arabinopyranosyl phosphate. This material was first detected as a minor phospholipid that accumulates in a polymyxin-resistant mutant of E. coli (Fig. 2). In S. typhimurium, the lipid is also present in wild-type cells, but its level is increased in pmrA C and PhoP C mutants. It is absent in pmrA Ϫ strains (Fig. 3).
The purification to near-homogeneity of the donor lipid required as a first step the construction of a special pmrA C mutant of E. coli lacking phosphatidylethanolamine to simplify all subsequent procedures (Figs. 4 -6). MALDI/TOF mass spectrometry and high resolution NMR spectroscopy were then used for the actual structure determination. As noted in previous NMR studies of lipid A, the mixture CDCl 3 /CDOD/D 2 O (2:3:1, v/v) is ideal for NMR analyses of small lipid samples (51,52,67), because there is no measurable decomposition for weeks under these conditions. The data (shown in Figs. 7-10, Tables II and III, and see Figs. 1-3 in the Supplemental Material) confirm unequivocally that the L-Ara4N donor substrate has the structure undecaprenyl phosphate-␣-L-Ara4N. The proposed structure and conformation of the sugar is supported by its coupling constants and chemical shifts, summarized in Table II, which are in accord with a synthetic standard of L-Ara4N (57). The attachment of the nitrogen atom to position 4 of the sugar is established by the HMQC experiment (Fig. 9A). The size and configuration of the polyisoprene chain are confirmed by mass spectrometry (Fig. 7) and by the HMQC/HMBC analyses (Figs. 9 and 10).
Reconstitution of the purified lipid donor with the cloned E. coli or S. typhimurium L-Ara4N transferase (ArnT) resulted in robust in vitro conversion of the precursor lipid-IV A to lipid-II A (Fig. 11), a compound that contains a single L-Ara4N moiety on its 1-phosphate residue (50,52,58). ArnT is a complex enzyme with 12 predicted trans-membrane helices (25) and is a member of an ancient family of glycosyltransferases that includes the eucaryotic protein mannosyltransferases (23,68). All the enzymes of this family appear to use polyisoprene-linked sugars as their donor substrates (23,68). However, prior to the present work, identification of undecaprenyl phosphate-␣-L-Ara4N and in vitro systems for detecting the transfer of L-Ara4N to lipid A or lipid A precursors had not been reported.
Many interesting questions regarding the possible translocation of the donor lipid across the inner membrane for utilization by ArnT in the periplasm remain unanswered. The relevant transporter has not been identified. It is unlikely that Wzx (RfbX), the putative O-antigen flippase (5,69), is responsible for the translocation of undecaprenyl phosphate-␣-L-Ara4N into the periplasm, because Wzx proteins are thought to transport undecaprenyl diphosphate oligosaccharides (70). Nevertheless, wzx-deficient mutants should be tested for their sensitivity to polymyxin. Most importantly, perhaps, in vitro systems for measuring flip-flop of undecaprenyl phosphate-␣-L-Ara4N across lipid bilayers need to be developed, but first, efficient enzymatic, radiochemical, and chemical methods for making undecaprenyl phosphate-␣-L-Ara4N must be devised. A new enzymatic system for the conversion of UDP-glucuronic FIG. 11. Reconstitution in vitro of ArnT activity with purified undecaprenyl phosphate-␣-L-Ara4N. ArnT was assayed using 10 M [4Ј-32 P]lipid IV A as the acceptor substrate under standard assay conditions for 5 min with 0.5 mg/ml of membrane protein, except that the Triton X-100 concentration was increased to 0.7% to dissolve the undecaprenyl phosphate-␣-L-Ara4N. The source of the membranes, as indicated, was from the polymyxin-resistant BLR(DE3)/pLysS or the polymyxin-sensitive NovaBlue(DE3)/ pLysS E. coli host strains, containing either the pET21 vector or the hybrid pArn-TEc plasmid (25). The purified donor substrate was added in increasing concentrations ranging from about 1.5 to 150 M. In the left panel, 150 M donor was included in the indicated reaction. The products were separated by TLC using the solvent chloroform, pyridine, 88% formic acid, water (50:50:16:5, v/v) and were detected with a PhosphorImager. acid to UDP-L-Ara4N has recently been identified in extracts of polymyxin-resistant mutants of E. coli, 3 and should facilitate the preparation of undecaprenyl phosphate-␣-L-Ara4N.