Origin of Lipid A Species Modified with 4-Amino-4-deoxy- L arabinose in Polymyxin-resistant Mutants of Escherichia coli AN AMINOTRANSFERASE (ArnB) THAT GENERATES UDP-4-AMINO-4-DEOXY- L -ARABINOSE*

In Escherichia coli and Salmonella typhimurium , addition of the 4-amino-4-deoxy- L -arabinose ( L -Ara4N) moi- ety to the phosphate group(s) of lipid A is required for resistance to polymyxin and cationic antimicrobial peptides. We have proposed previously (Breazeale, S. D., Ribeiro, A. A., and Raetz, C. R. H. (2002) J. Biol. Chem. 277, 2886–2896) a pathway for L -Ara4N biosynthesis that begins with the ArnA-catalyzed C-4 (cid:1) oxidation and C-6 (cid:1) decarboxylation of UDP-glucuronic acid, followed by the C-4 (cid:1) transamination of the product to generate the novel sugar nucleotide UDP- L -Ara4N. We now show that ArnB (PmrH) encodes the relevant aminotransferase. ArnB was overexpressed using a T7 lac promoter-driven construct and shown to catalyze the reversible transfer of the amino group from glutamate to the acceptor, uridine 5 (cid:2) -( (cid:1) - L - threo -pentapyranosyl-4 (cid:1) -ulose diphosphate), glycerol) with the use of a Centricon (cid:3) (YM-10) membrane, which subjected several speed centrifugation and bicinchoninic acid procedure overexpression ArnB, ArnB-N6H, and ArnB-C6H, the initial of the hexa-histidine fusion SDS-PAGE stacking Coomassie staining

Lipopolysaccharide (LPS) 1 is an immunogenic glycolipid that is the major component of the outer leaflet of the outer mem-branes of Gram-negative bacteria (1)(2)(3). LPS consists of three domains, which are the O-specific antigen, the core oligosaccharide, and the lipid A moiety (1)(2)(3). The latter provides a hydrophobic membrane anchor and is also the active component (or endotoxin) of LPS that is responsible for many of the pathophysiological effects associated with Gram-negative sepsis (1)(2)(3). Lipid A potently activates the innate immune system, inducing a host response that includes the production of cationic antimicrobial peptides, cytokines, clotting factors, and various immunostimulatory molecules (4 -6). In severe infections, high levels of cytokines and pro-coagulant activity lead to Gram-negative septic shock, a syndrome characterized by increased vascular permeability, hypotension, multiple organ failure, and death (7).
The Kdo-lipid A portion of LPS, which is relatively conserved among diverse Gram-negative bacteria, is sufficient to support the growth of Escherichia coli or Salmonella typhimurium (2,8). Certain covalent modifications to Kdo-lipid A are induced by environmental stimuli, such as low magnesium ion concentrations and/or low pH, as would be encountered by bacteria in phagolysosomes of macrophages (9 -11). As shown in Fig. 1 for S. typhimurium, these modifications include the incorporation of palmitate (12,13), the formation of S-2-hydroxymyristate (14), the addition of phosphoethanolamine (pEtN) (15), and/or the addition 4-amino-4-deoxy-L-arabinose (L-Ara4N) moieties ( Fig. 1) (15)(16)(17). Addition of the L-Ara4N and pEtN moieties to the phosphate groups of lipid A confers resistance to polymyxin and various cationic antimicrobial peptides (15)(16)(17). Resistance is thought to occur by reduction of the net negative charge of lipid A, resulting in a reduced affinity for cationic peptides or polymyxin, thereby preventing these substances from penetrating the outer membrane.
The L-Ara4N modification of lipid A is governed by the PmrA/ PmrB two-component regulatory system (10, 18 -21), which is switched on by low pH. The transcription factor PmrA activates genes at the pmrE(ugd) and pmrHFIJKLM loci (10,22). Constitutive mutants of pmrA (pmrA c ) have been isolated in E. coli and S. typhimurium; these mutants modify their lipid A with L-Ara4N and pEtN groups under all growth conditions, and they are polymyxin-resistant (17,20,23). However, inactivation of pmrE or of any of the genes in the pmrHFIJKL operon in a pmrA c S. typhimurium mutant results in loss of polymyxin resistance and loss of the L-Ara4N modification of lipid A (10).
Sequence analyses of the various genes in the pmr loci have led us to propose a pathway for the biosynthesis of the L-Ara4N moiety and its attachment to lipid A (22). Our hypothesis (Fig.  2) is supported by the following observations. 1) ArnT (PmrK) is an inner membrane enzyme that utilizes the novel glycolipid undecaprenyl phosphate-␣-L-Ara4N as the donor of the L-Ara4N moiety to modify lipid A (16,17). 2) ArnA (PmrI) is a dehydrogenase/decarboxylase that converts UDP-glucuronic acid to a novel 4Љ-keto-pentose, designated UDP-Ara4O, which can be isolated in milligram quantities as the hydrated ketone (22). 3) Cell extracts of polymyxin-resistant (but not polymyxinsensitive) E. coli cells convert UDP-Ara4O to the putative sugar nucleotide UDP-L-Ara4N in the presence of L-glutamate (22). 4) Cell extracts of polymyxin-resistant E. coli cells can further convert the proposed UDP-L-Ara4N to a putative formyl-amino derivative in the presence of N-10-formyltetrahydrofolate (22).
We have now cloned, overexpressed, and purified to homogeneity the aminotransferase ArnB (PmrH). We demonstrate that ArnB (GenBank™ accession number AAM92146) contains a pyridoxal phosphate cofactor and catalyzes the reversible transfer of the ␣-amine moiety from L-glutamate to the C-4Љ position of UDP-Ara4O, generating UDP-L-Ara4N. The structure of UDP-L-Ara4N was validated by one-and two-dimensional NMR spectroscopy. The transamination is highly specific for glutamate as the amine donor. Our biochemical data establish ArnB as the key aminotransferase required for the biosynthesis of the L-Ara4N moiety present on lipid A molecules of polymyxin-resistant Gram-negative bacteria. Obvious ArnB orthologs are present in Yersinia pestis, Pseudomonas aeruginosa, and Burkholderia cepacia, all of which are capable of modifying their lipid A with L-Ara4N (24). The purification of ArnB permits the enzymatic synthesis of milligram quantities of UDP-L-Ara4N, which can now be used to probe for the remaining steps in the transfer of L-Ara4N to lipid A. A crystal structure of ArnB (25), recently obtained as part of a structural genomics program, is in accord with our enzymatic and biochemical studies.

EXPERIMENTAL PROCEDURES
Buffers, Reagents, and Materials-UDP-glucuronic acid, NAD ϩ , 1.0 M triethylammonium bicarbonate, pH 8.5, L-glutamate, polymyxin B sulfate, kanamycin, and HEPES were purchased from Sigma. The UDP-[U-14 C]glucuronic acid was purchased from PerkinElmer Life Sciences. Restriction enzymes (NdeI, NcoI, and XhoI) were obtained from New England Biolabs. T4 DNA ligase, various oligonucleotides, and isopropyl-1-thio-␤-D-galactopyranoside were purchased from Invitrogen. The Pfu polymerase was from Stratagene. Polyethyleneiminecellulose TLC plates were purchased from Merck. DEAE-Sepharose TM CL-6B was obtained from Amersham Biosciences AB. Protein concentrations were determined by the bicinchoninic acid method (26) using the BCA protein assay reagent (Pierce) with bovine serum albumin as the standard.
Strains and Recombinant DNA Techniques-Construction of WD101, a polymyxin-resistant mutant of E. coli W3110, was described previously (17). Preparation and transformation of competent cells by the CaCl 2 method, genomic DNA purification, and gel electrophoresis of restriction fragments were performed according to published procedures (27,28). Plasmids were purified using the QIAprep spin miniprep kit (Qiagen). Restriction enzymes, Pfu DNA polymerase, and T4 DNA ligase were used as recommended by their manufacturers. DNA sequencing was performed on an ABIprism 377 instrument at the Duke University DNA Analysis Facility.
Construction of pETArnB, pETArnB-C6H, and pETArnB-N6H-The predicted coding region for ArnB (GenBank™ accession number AAM92146) was amplified by PCR from E. coli W3110 genomic DNA with primers to the arnB 5Ј end (5Ј-CGGGATCCATGGCGGAAGGAA-AAGCAATG-3Ј) and to the arnB 3Ј end (5Ј-CGGGATCCTCGAGTTAT-TGTCCTGCGAGTTGCTG-3Ј), containing engineered NcoI and XhoI restriction sites, respectively. The Opti-Prime TM PCR Optimization kit (Stratagene) was used as described by the manufacturer. The amplification reaction (50 l) contained the following components: 5 l of 10ϫ Opti-Prime Buffer 9 (100 mM Tris-HCl, pH 9.2, 15 mM MgCl 2 , 250 mM KCl), 0.2 mM of each dNTP, 250 ng of genomic DNA, 0.5 M of each primer, and 2.5 units of Pfu DNA polymerase. After 30 cycles (94°C for 1 min, 50°C for 1 min, and 72°C for 1.5 min) followed by 1 cycle (94°C for 1 min, 50°C for 1 min, and 72°C for 10 min), a single product of the predicted size was obtained. This was isolated with the QIAquick PCR Purification kit (Qiagen), digested with XhoI and NcoI, and then ligated into the T7lac expression vector, pET28b (Novagen), previously digested with the same enzymes. The presence of the arnB gene (29) and appropriate flanking DNA was confirmed by sequencing of final construct, which was designated pETArnB.
Amplification of DNA for cloning a modified arnB gene expressing a C-terminal hexa-histidine fusion protein of ArnB (designated ArnB-C6H) utilized the arnB 5Ј primer described above for the native protein and the oligonucleotide 5Ј-CGGGATCCTCGAGTTGTCCTGCGAGTT-GCTGAAG-3Ј as the 3Ј primer. The latter contained an engineered XhoI restriction site. The PCR (50 l) contained the following: 5 l of 10ϫ Opti-Prime Buffer 11 (100 mM Tris-HCl, pH 9.2, 35 mM MgCl 2 , 250 mM KCl), 0.2 mM of each dNTP, 250 ng of genomic DNA, 0.5 M of each primer, and 2.5 units of Pfu DNA polymerase. The same temperature parameters described above were used for the PCR. The resulting product was cloned into the expression vector pET28b, as described above, to generate pETArnB-C6H. The final construct was confirmed by DNA sequencing.
Amplification of DNA for cloning a modified arnB gene encoding an N-terminal hexa-histidine fusion protein of ArnB (designated ArnB-N6H) was accomplished using the same reaction conditions as for pETArnB-C6H. The arnB 3Ј primer described above for the native protein was used together with the oligonucleotide 5Ј-CGGATCCATAT-GGCGGAAGGAAAAGCAATG-3Ј as the 5Ј primer, which also contained an engineered NdeI restriction site. The PCR product was isolated with the QIAquick PCR Purification kit (Qiagen), digested with XhoI and NdeI, and ligated into the pET28b expression vector, previ-  (2). However, as indicated by the dashed bonds, covalent modifications of the phosphate groups and/or acyl chains of lipid A may be induced under certain growth conditions (2). For instance, the phosphates may be substituted with L-Ara4N and/or pEtN groups (blue); both of these modifications are controlled by the PmrA transcription factor, which is activated by growth at low pH and is constitutively activated in polymyxin-resistant mutants (2). Additional minor lipid A derivatives are present in which the locations of the L-Ara4N and pEtN groups are reversed (not shown) or in which both phosphates are modified with the same group (15,36). Incorporation of a secondary palmitoyl chain at position 2 in the proximal unit is catalyzed by the outer membrane enzyme PagP (13,48), whereas removal of the hydroxymyristoyl chain at position 3 is carried out by PagL (49). Formation of S-2-hydroxymyristate is O 2 -dependent and requires a hydroxylase encoded by lpxO (14). PagL and LpxO are normally present in S. typhimurium (50) but not E. coli K-12 (29). With the possible exception of LpxO, enzymes that modify the acyl chains of lipid A (red) are under the control of the PhoP/PhoQ two-component system, which is activated by growth in the presence of low Mg 2ϩ concentrations (9,11,42).
ously digested with the same enzymes, to generate pETArnB-N6H. The final construct was confirmed by DNA sequencing.
Overexpression of the arnB Gene Product and Purification of ArnB-N6H and ArnB-C6H-Plasmids pETArnB, pETArnB-N6H, or pETArnB-C6H were transformed into E. coli NovaBlue(DE3) cells (Novagen). Cultures were grown overnight at 37°C from single colonies in LB broth (30) supplemented with 30 g/ml kanamycin. The overnight culture was then used to inoculate a fresh 100-ml culture of the same broth at A 600 ϳ0.01. Cells were grown at 37°C to A 600 of ϳ0.6. Expression of ArnB was induced by addition of isopropyl-␤-D-thiogalactopyranoside to a final concentration of 1 mM, and growth was continued at 30°C for 3.5 h. Cells were harvested by centrifugation at 6500 ϫ g for 15 min at 4°C. The cell pellet was resuspended in 3 ml of buffer A, consisting of 25 mM HEPES, pH 7.5, 10% glycerol, and 300 mM NaCl. Cells were broken by passage through a French pressure cell at 18,000 pounds/square inch. Cell debris was removed by centrifugation at 16,500 ϫ g for 15 min at 4°C to give crude cell-free extracts.
For the hexa-histidine fusion proteins, the crude cell-free extracts were loaded onto a 1-ml nickel-nitrilotriacetic acid column (Qiagen), which was pre-equilibrated in buffer A. The column was washed with 15 column volumes of buffer A and 15 column volumes of buffer A containing 50 mM imidazole; it was then eluted with 15 column volumes of buffer A containing 100 mM imidazole. Proteins were concentrated, and the imidazole was removed by exchanging for buffer B (consisting of 50 mM HEPES, pH 7.5, and 10% glycerol) with the use of a Centricon (YM-10) membrane, which was subjected to several cycles of low speed centrifugation and dilution in buffer B. Protein concentrations were determined using the bicinchoninic acid procedure (26).
Enzymatic Synthesis and Isolation of UDP-L-Ara4N-A 10-ml reaction mixture, containing 0.05 mg/ml pure ArnB-N6H, 1 mM UDP-Ara4O, and 10 mM L-glutamate in 50 mM HEPES, pH 7.5, was incubated at 30°C for 5.5 h. Protein was removed by passing the reaction mixture through a Centricon (YM-10) membrane. The enzymatic product was then loaded onto a 10-ml column of DEAE-Sepharose TM CL-6B, pre-equilibrated with 10 mM BisTris, pH 5.9. The column was washed with 50 ml of deionized water, followed by elution of the product with a 0 -100 mM BisTris, pH 5.9, gradient (200-ml total volume). Chromatography was monitored by absorbance of the effluent at 254 nm. Fractions (12 ml) containing the desired product, which eluted at about 30 mM BisTris, were pooled and diluted 10-fold with H 2 O. The diluted material was then loaded onto a 10-ml column of DEAE-Sepharose TM CL-6B, pre-equilibrated with 10 mM triethylammonium bicarbonate, pH 8.5. The column was washed with 50 ml of deionized water, followed by elution of the product with a 0 -100 mM gradient of triethylammonium bicarbonate, pH 8.5 (200-ml total volume). Chromatography was again monitored by absorbance at 254 nm. Fractions containing the product (which eluted at about 60 mM triethylammonium bicarbonate) were pooled and concentrated by lyophilization to give the triethylammonium salt of the sugar nucleotide. Conversion to the sodium salt was accomplished by passage over a 1-ml column of AG50W-X8 resin (sodium form), followed by lyophilization to give the pure compound (1.7 mg, 31% yield).

Enzymatic Synthesis and Purification of UDP-L-[U-14 C]Ara4N-Conversion of UDP-[U-14 C]glucuronic acid (380 mCi/mmol) to UDP-L-[U-14 C]
Ara4N was performed in 400-l reaction mixtures containing 50 mM HEPES, pH 7.5, 2 mM dithiothreitol, 3 mM NAD ϩ , 1 mM L-glutamate, 0.05 mg/ml ArnA overproducing cell-free extract (22), 0.05 mg/ml pure ArnB-C6H, and 13 M UDP-[U-14 C]glucuronic acid (2 Ci), previously dried under nitrogen and re-dissolved in a small volume of deionized water. The reaction proceeded for 220 min at 30°C and was then diluted to 1.5 ml with deionized water before loading onto a 2-ml column of DEAE-Sepharose TM CL-6B, pre-equilibrated with 10 mM BisTris, pH 6. The column was washed with 5 ml of deionized water, followed by elution of the nucleotides with a 0 -100 mM gradient of BisTris, pH 6 (40-ml total volume). Fractions containing UDP-L-[U-14 C]Ara4N were pooled, diluted 10-fold with deionized water, and loaded onto a 2-ml column of DEAE-Sepharose TM CL-6B, pre-equilibrated with 10 mM triethylammonium bicarbonate, pH 8.5. The column was washed with 5 ml of deionized water. The NAD ϩ was resolved from the desired sugar nucleotide with a 0 -100 mM gradient of triethylammonium bicarbonate, pH 8.5 (40-ml total volume). The radioactive fractions were concentrated using a Speed-Vac centrifuge and dissolved in 250 l of deionized water. The yield of UDP-L-[U-14 C]Ara4N was 74% based on the input radioactivity. The product was stored frozen at Ϫ20°C.
Nuclear Magnetic Resonance Spectroscopy-The sodium salt of the putative UDP-L-Ara4N (1.7 mg prepared as described above) was dissolved in 0.6 ml of 99% D 2 O and placed into a 5 mm NMR tube. Measurements with a 3-mm pH electrode revealed a pD value of 8.00. Proton and 13 C NMR chemical shifts were referenced relative to 2,2dimethylsilapentane-5-sulfonic acid at 0.00 ppm. 1 H NMR spectra were obtained on a Varian Inova 800 spectrometer equipped with a Sun Ultra 10 computer and a 5-mm Varian triple resonance probe. 1 H spectra were obtained with a spectral width of 8.0 kHz, a 76 o pulse flip angle (7 s), a 4.0-s acquisition time, and a 1.5-s relaxation delay, and were digitized using 96,000 points to obtain a digital resolution of 0.16 Hz/point. Two-dimensional NMR experiments were performed as described previously (17,22,32).
ArnB Enzymatic Assays-The rate of formation of UDP-L-[U- 14 C]Ara4N by ArnB was quantified using a coupled assay in which overexpressed ArnA was first used to generate UDP-[U- 14  Reactions were incubated further at 30°C. At the desired time points, a 1-l portion was withdrawn and spotted onto a polyethyleneimine-cellulose TLC plate. The spots were allowed to dry, and the plate was washed in methanol for 5 min. After drying, the plate was developed using 0.4 M LiCl in 0.25 M aqueous acetic acid. The plate was dried and exposed to a PhosphorImager screen. Remaining substrate and radioactive product were quantified with a PhosphorImager (ImageQuant software, Amersham Biosciences). ArnB specific enzymatic activity was calculated in units of nmol/min/mg.
Ultraviolet-visible Spectroscopy of ArnB-Purified ArnB-C6H (25 M) in 50 mM HEPES, pH 7.5, containing 10% glycerol was incubated at 30°C alone or in the presence of 1 mM L-glutamate in a total volume of 100 l. Spectra were recorded on a BeckmanCoulter DU 640B UVvisible scanning spectrophotometer. Experimental data were exported into Kaleidagraph (Synergy Software, Reading, PA) and plotted. The reactions were blanked against 50 mM HEPES, pH 7.5, containing the buffer alone (10% glycerol in 50 mM HEPES, pH 7.5).

Cloning, Overexpression, and Purification of ArnB as a Hexahistidine-tagged Fusion
Protein-The L-Ara4N moiety found on the phosphate groups of lipid A in polymyxin-resistant Gramnegative bacteria (like pmrA constitutive strains of E. coli or S. typhimurium) is proposed to arise via the novel sugar nucleotide, UDP-L-Ara4N (22). The transamination of the precursor UDP-Ara4O (Fig. 2), a nucleotide synthesized in polymyxinresistant (but not polymyxin-sensitive) E. coli (22), could generate UDP-L-Ara4N (Fig. 2). We have shown previously that the E. coli gene arnA encodes a protein that catalyzes the oxidative decarboxylation of UDP-GlcUA to give UDP-Ara4O (22). Furthermore, in the presence of L-glutamate, cell extracts of polymyxin-resistant E. coli can convert UDP-Ara4O to a species with the R F expected for such an amino-sugar nucleotide at the rate of ϳ2 nmol/min/mg (22). These preliminary dataareconsistentwiththeinvolvementofapyridoxalphosphatedependent enzyme that utilizes L-glutamate as the amine donor for converting UDP-Ara4O to UDP-L-Ara4N.
Upstream of arnA resides arnB (pmrH), another gene required for polymyxin resistance in S. typhimurium (10), which is predicted to encode an aminotransferase (22,33). The genetic location of arnB and the predicted sequence of its encoded protein made it a likely candidate for the proposed transaminase involved in the UDP-L-Ara4N formation (Fig. 2). To validate the function of ArnB, the full-length arnB gene was cloned into the expression vector pET28b behind an inducible T7lac promoter. Expression was carried out in E. coli NovaBlue(DE3), a polymyxin-sensitive strain that does not modify its endogenous lipid A with the L-Ara4N moiety (16). A protein of the predicted molecular mass of ArnB (ϳ43 kDa) was observed by SDS-gel electrophoresis in cell-free extracts of induced cells but was absent in uninduced controls (Fig. 3, lanes 2 and 3,  respectively). Extracts of induced cells containing pETArnB catalyzed the rapid L-glutamate-dependent conversion of UDP-Ara4O to a species with the R F expected for UDP-L-Ara4N (data not shown).
The relatively low level of ArnB expression, as judged by gel electrophoresis (Fig. 3), led us to engineer hexa-histidine sequences into the protein to facilitate purification. To this end, arnB was cloned into the pET28b expression vector to express either an N-terminal hexa-histidine fusion protein (ArnB-N6H) containing an additional 20 amino acids, or a C-terminal hexahistidine fusion protein (ArnB-C6H) containing an additional 7 amino acids. Similar to the wild-type ArnB, the hexa-histidine fusion proteins were expressed in induced E. coli NovaBlue(DE3) but were absent in uninduced controls (Fig. 3, lanes 5  and 4 and lanes 8 and 9, respectively). The ArnB-specific activity of a cell extract of induced E. coli NovaBlue(DE3)/ pETArnB-C6H is 130 nmol/min/mg. Approximately the same specific activity is seen with cell extracts overexpressing ArnB-N6H or wild-type ArnB. By using nickel chelation chromatography, the hexa-histidine ArnB fusion proteins were purified to homogeneity, as judged by SDS gel electrophoresis (Fig. 3,  lanes 6 and 7). The final specific activity of the pure proteins is 1.3 ϫ 10 3 nmol/min/mg.
Analysis of the Covalent Structure of the ArnB Product by One-dimensional 1 H NMR Spectroscopy-By using pure ArnB-N6H, 10 mol of UDP-Ara4O was converted to the putative UDP-L-Ara4N in the presence of 100 mol of L-glutamate, as described under "Experimental Procedures." The substrate and product were separated using anion exchange chromatography at pH 5.9 (0 -100 mM BisTris). The fractions containing the ArnB product were then separated from the BisTris, using anion exchange chromatography at pH 8.5 in the volatile buffer, triethylammonium bicarbonate (0 -100 mM). Fractions containing the desired product were lyophilized to give the triethylammonium salt of the sugar nucleotide, which was converted to the sodium salt by passage through a Dowex AG®50W-X8 column (sodium form).
The sodium salt (1.7 mg) of the product was dissolved in D 2 O, and its 1 H NMR spectrum was recorded, Fig. 4 (panels A and  B). The uracil and ribose moieties of the substrate, UDP-  (2,10,33). ArnA then catalyzes the oxidative decarboxylation of UDP-GlcUA to form the novel sugar nucleotide, UDP-Ara4O (22). As shown in the present study (boxed reaction), ArnB transaminates the resulting 4Љ-keto moiety in a reversible reaction to generate UDP-L-Ara4N. The ArnA protein contains a second catalytic domain with strong homology to methionyl-tRNA formyltransferase (22), which transfers the formyl group from N-10-formyltetrahydrofolate to the 4Љ amine of UDP-L-Ara4N. Based upon its homology to the yeast dolichoyl phosphate-mannose synthase (51) and recent enzymatic studies in our laboratory, 3 we suggest that ArnC (formerly PmrF) (10) transfers the L-Ara4-formyl-N unit (indicated by the black rectangle) to undecaprenyl phosphate. Following transport of the periplasmic surface of the inner membrane, we speculate that deformylation occurs prior to L-Ara4N addition to lipid A. It is well established by in vitro studies that the polytopic membrane protein ArnT transfers the L-Ara4N moiety (indicated by the blue rectangle) of undecaprenyl phosphate-␣-L-Ara4N to lipid A (16,17). Undecaprenyl phosphate-␣-L-Ara4N can be isolated from polymyxin-resistant mutants (17). The proteins involved in the proposed transport and deformylation of undecaprenyl phosphate-␣-L-Ara4-formyl-N of have not yet been identified.
Ara4O, and of the ArnB product possess virtually identical chemical shifts and coupling constants (Table I). However, the well resolved signals arising from the pyranose moiety of the UDP-Ara4O and the putative UDP-L-Ara4N show significant differences (Fig. 4, panels B versus C, and Table I). The onedimensional 1 H NMR analysis therefore suggests that ArnB alters only the pyranose moiety of UDP-Ara4O.
Characterization of the ArnB Product by Two-dimensional 1 H NMR Spectroscopy-The two-dimensional 1 H-1 H COSY NMR spectrum of the putative UDP-L-Ara4N product (Fig. 5) was used to assign the protons in the ribose, uracil, and pyranose rings. As suggested by the one-dimensional spectra, the uridine nucleoside moiety was not altered by ArnB (Table I). Close similarities between the putative UDP-L-Ara4N and the UDP-Ara4O spectra include the strong cross-peak connectivity between the uracil H-6 and H-5 doublet resonances (not shown), the H-1Ј connectivity to H-2Ј (4.39 ppm), the H-3Ј connectivity to H-4Ј (ϳ4.30 ppm), and the connectivities of H-4Ј to H-5bЈ and H-5aЈ (Table I).
The measured values of 2.1 and 2.0 Hz for the two J 4Љ,5Љ couplings (equatorial-equatorial and equatorial-axial) do not allow distinction of the axial or equatorial H-5Љ protons. Attempts to distinguish these protons for this pyranose ring directly through two-dimensional NOE experiments were unsuccessful, as no clear NOE cross-peak was observed between the axially disposed H-3Љ (Fig. 2) and either H-5Љ proton, at both 600-and 800-MHz fields. The equatorial H-4Љ showed a stronger NOE cross-peak to H-5bЉ (ϳ4.29-ppm signal) and a weaker NOE cross-peak to H-5aЉ (3.80-ppm signal), but those results do not allow firm assignments. However, the H-5aЉ (3.80 ppm) and H-5bЉ (4.29 ppm) signals are assigned to the axial and equatorial positions following previous literature values with synthetic standards (34,36), based upon the chemical shift consideration that a proton in an axial position at a given carbon atom on a pyranose ring is expected to resonate at ϳ0.5 ppm higher field than the corresponding equatorial proton attached to the same carbon (37). Furthermore, the two-dimensional NOE spectroscopy data for the ␣-anomeric L-Ara4N ring of the natural product undecaprenyl phosphate-L-␣-Ara4N (17) yielded the expected NOE cross-peaks between H-1Љ, H-3Љ and the upfield H-5Љ proton resonance, in agreement with the literature assignments, and distinctly different from what is observed with the ArnB product generated in vitro.
Evaluation of the Carbon Structure of the ArnB Product by HMQC/HMBC Spectroscopy-13 C NMR data for the ArnB enzymatic product were obtained indirectly through 1 H-detected HMQC and HMBC two-dimensional NMR experiments. The partial two-dimensional HMQC 1 H-13 C correlation map (Fig. 6, panel A) reveals three direct 1 H-13 C single-bond correlations in the anomeric region (90 -105 ppm). The anomeric H-1Љ near 5.63 ppm correlates to the carbon resonance of the pyranose at 98.2 ppm (C-1Љ). The C-1 atom of a pyranose ring that is glycosidically linked with an axial oxygen is expected to resonate in the 98 -103-ppm range, whereas a C-1 atom that is glycosidically linked with an equatorial oxygen is expected to resonate in the 103-106-ppm region (35). The 98-ppm shift of C-1Љ for the 4-amino-4-deoxy-arabinose is thus consistent with an axially disposed oxygen atom and an equatorially disposed H-1Љ, as deduced from the 1 H NMR data. The ribose H-1Ј proton signal at 6.00 ppm correlates to the anomeric carbon resonance at 90.9 ppm (C-1Ј). The uracil H-5 near 5.99 ppm correlates to the 105.3-ppm carbon signal, and the uracil H-6 at 7.97 ppm correlates to a carbon signal at 144.2 ppm (not shown).
Examination of the HMQC data for the other proton resonances of the pyranose moiety shows that the H-5bЉ and H-5aЉ proton multiplets correlate to a carbon signal at 62.4 ppm (C-5Љ), whereas the H-2Љ and H-3Љ multiplets connect to carbon resonances at 70.1 and 68.5 ppm, respectively. These carbon shift positions correspond to oxygen-substituted carbons of sugars. The nitrogen-substituted carbons of amino sugars are reported to resonate near 52-55 ppm (35,38). The H-4Љ multiplet shows a prominent cross-peak near 55 ppm. This C-4Љ chemical   Table I. FIG. 6. Partial two-dimensional HMQC and HMBC spectra of UDP-L-Ara4N. Panel A is a partial two-dimensional HMQC spectrum of UDP-Ara4N in D 2 O showing the single bond 1 H-13 C correlations arising from the uracil H-5, the ribose protons, and the pyranose protons to their respective carbons. Key cross-peaks are labeled. Panel B is the corresponding partial two-dimensional HMBC spectrum of the same UDP-Ara4N sample. Of particular importance are the multibond correlations from H-5bЉ and H-5aЉ to the 54.6-ppm carbon signal of the pyranose, verifying its assignment as C-4Љ. X is the residual HOD signal remaining after water suppression. The full 1 H and 13 C assignments are listed in Table I. shift is diagnostic evidence confirming C-4Љ as the site of the amino group substitution. The remaining sugar HMQC crosspeaks arise from protonated carbons of the ribose ring. The ribose assignments derived (Table I) agree with the literature values (35,39).
The multibond HMBC two-dimensional map (Fig. 6, panel B) allows identification of the quaternary carbons leading to full assignments (Table I) for the enzyme product. Analyses of the multibond correlations from the uracil H-6 (7.97 ppm) and H-5 (5.99 ppm) yielded assignment of the 154 and 168 ppm quaternary carbon signals to C-2 and C-4, respectively (not shown in Fig. 6, panel B), consistent with literature assignments (39). The multibond correlation from the uracil H-6 to C-1Ј was also detected (not shown), thus verifying the linkage of the base group to the ribose. The most salient feature in the sugar region is the detection of the multibond correlations from both H-5Љ protons to the 54.6-ppm carbon resonance, thus further validating C-4Љ as the nitrogen-substituted carbon (Fig. 6). The remaining multibond correlations in the sugar region verify the assignments of the arabinose and ribose carbon resonances. The 1 H and 13 C NMR assignments are summarized in Table I.
Comparison of the 13 C NMR data for the ArnB product with the starting substrate, UDP-Ara4O (22) (Table I), reveals that the uracil and ribose carbons of both compounds resonate at very similar positions, consistent with the 1 H NMR evidence that the enzymatic reaction did not alter the uridine moiety. The 13 C NMR data dramatically show that the enzyme-catalyzed amino group substitution is localized to C-4Љ, which shifts upfield by ϳ40.6 ppm, from 95.2 ppm for UDP-Ara4O in D 2 O (corresponding to the hydrated form of UDP-Ara4O) (22) to 54.6 ppm for UDP-L-Ara4N in D 2 O. The C-3Љ and C-5Љ resonances show much smaller upfield chemical shifts of 6.6 and 5.6 ppm, respectively. The chemical shift change decreases by another factor of 2 at C-2Љ to 3 ppm and by a further factor of 10 at C-1Љ to 0.3 ppm.
ArnB-catalyzed Transamination of UDP-Ara4O-Purified ArnB-C6H in the presence of various amine donors at 7.5 mM preferentially utilizes L-glutamate for the synthesis of UDP-L-Ara4N from 150 M UDP-Ara4O. The rate of transamination with L-methionine, L-glutamine, and L-alanine is measurable at 5, 2, and 1%, respectively, of the rate observed with L-glutamate (data not shown). No activity was seen with L isomers of asparagine, aspartate, glycine, lysine, ornithine, phenylalanine, or tryptophan. However, the analog L-aminoadipic acid (25) supported the transamination of UDP-Ara4O at about 50% the rate of L-glutamate (not shown).
As shown in Fig. 7, the extent of conversion of 100 M UDP-Ara4O to UDP-L-Ara4N increases as the molar ratio of amine donor to UDP-Ara4O is increased. At a 10-fold excess of L-glutamate, only 53% of the UDP-Ara4O is converted to product, whereas 87 and 91% of the UDP-Ara4O are converted to UDP-L-Ara4N with a 50-or 100-fold excess of L-glutamate, respectively. These findings suggest that that the transamination of UDP-Ara4O is reversible and thermodynamically unfavorable. The equilibrium constant in the direction of UDP-L-Ara4N formation is ϳ0.1.
The reversibility of the reaction can be demonstrated directly (Fig. 8). In the presence of a 10-or 100-fold excess of ␣-ketoglutarate, 100 M UDP-L-Ara4N is largely converted back to UDP-Ara4O (Fig. 8, panel A). No UDP-Ara4O is observed in the absence of ␣-ketoglutarate. The full time course of the reverse reaction with 1 mM ␣-ketoglutarate is shown in Fig. 8, panel B. The extent of conversion to UDP-Ara4O is consistent with the unfavorable equilibrium constant noted above.
ArnB Contains a Bound Pyridoxal Phosphate Cofactor-Pyridoxal phosphate-dependent enzymes typically have signature absorbance maxima around 430 and/or ϳ335 nm, depending on whether the pyridoxal 5Ј-phosphate or the pyridoxamine 5Јphosphate form of the cofactor is predominant (40,41). The absorbance spectrum of purified ArnB suggests that the cofactor is mostly in the pyridoxal phosphate form, presumably linked as an aldimine to an active site lysine (maximum at ϳ430 nm, Fig. 9) (40). After a 60-min incubation in the presence of excess glutamate, however, the absorbance at ϳ430 nm was greatly decreased, and a concomitant increase in the absorbance at ϳ335 nm was observed (Fig. 9). This is consistent with the conversion of the pyridoxal 5Ј-phosphate form of the enzyme to the pyridoxamine 5Ј-phosphate form (40). The results are in accord with the bioinformatic prediction that ArnB is a member of a large family of aminotransferases.
The presence of the cofactor in the pure protein, albeit mainly in its pyridoxamine 5Ј-phosphate form, was recently confirmed by x-ray crystallography (25). The authors suggest that use of methionine in some of the purification buffers may account for the fact that the pyridoxal 5Ј-phosphate form of the cofactor was not predominant in these structural studies. DISCUSSION In E. coli, S. typhimurium, and many other pathogenic Gram-negative bacteria, modification of the phosphate groups of lipid A with the L-Ara4N moiety confers resistance to polymyxin and cationic antimicrobial peptides (2,11,24,42). The L-Ara4N unit is believed to arise from UDP-glucuronic acid via the novel biosynthetic pathway shown in Fig. 2 (22). All the enzymes in Fig. 2 are required for L-Ara4N biosynthesis (22) and polymyxin resistance (10). We have shown previously that ArnA catalyzes the novel oxidative-decarboxylation of UDP-GlcUA to generate UDP-Ara4O (which can be isolated in milligram quantities) (22), and that ArnT transfers the L-Ara4N moiety from undecaprenyl phosphate-␣-L-Ara4N (a minor lipid that accumulates in polymyxin-resistant cells) to lipid A (16,17). We have also shown that, in the presence of L-glutamate, crude extracts of polymyxin-resistant (but not polymyxin-sensitive) E. coli are able to convert UDP-Ara4O to a less negatively charged sugar nucleotide product, presumed to contain a free amine group (22). This observation was the basis for proposing that the novel aminotransferase, ArnB, can convert UDP-Ara4O to UDP-L-Ara4N (Fig. 2). The arnB gene product, which is required both for L-Ara4N modification of lipid A and maintenance of polymyxin resistance in S. typhimurium, is homologous to a large family of pyridoxal-dependent aminotransferases. The crystal structure of S. typhimurium ArnB was recently determined as part of a structural genomics program (25), and the presence of the pyridoxal phosphate cofactor was validated.
We have now developed the first quantitative assay for ArnB and demonstrated that the purified E. coli enzyme is indeed a pyridoxal-dependent aminotransferase that generates UDP-L-Ara4N from UDP-Ara4O in the presence of glutamate (Fig. 2, box). E. coli ArnB, either with or without a hexa-histidine tag, was expressed using an IPTG-inducible T7lac promoter construct in E. coli NovaBlue(DE3), a polymyxin-sensitive host. In each case, the expression of a 43-kDa protein (Fig. 3) correlated with the overproduction of L-glutamate-dependent conversion of UDP-[ 14 C]Ara4O to the faster migrating nucleotide, presumed to be UDP-L-[ 14 C]Ara4N; both the native and the hexahistidine tagged proteins displayed similar specific activities (data not shown). Both of ArnB hexa-histidine fusion proteins were readily purified by standard nickel-chelation chromatography (Fig. 3), yielding about 8 mg of pure protein per liter from induced cells with a specific activity of about 1300 nmol/min/mg under standard assay conditions in the forward direction. By using purified ArnB, we prepared 1.7 mg of the putative UDP-L-Ara4N product. NMR spectroscopy (Figs. 4 -6) demonstrated unambiguously that the material synthesized by ArnB from UDP-Ara4O is indeed UDP-L-Ara4N. In particular, the location of the amine group at position 4Љ and the stereochemistry of the pyranose moiety, as judged by the analysis of the coupling constants and chemical shifts (Table I), are essentially the same as those in the L-Ara4N moiety attached to lipid A (15,36). To our knowledge, UDP-L-Ara4N is the first example of a nucleoside diphosphate 4Љ-amino pentose, featuring a pyranose ring. The only reported 4Љ-amino hexose nucleoside diphosphate is GDP-4-amino-4,6-dideoxymannose (GDP-perosamine), which is a precursor of lipopolysaccharide O-antigens in some Gram-negative bacteria (43). Several 3Љ-amino hexose nucleoside diphosphates have also been reported; these are precursors of the sugar moieties present in some macrolides and glycopeptide antibiotics (44,45).
Close orthologs of ArnB are found in all Gram-negative bacteria known to modify their lipid A with L-Ara4N (24). These include, in addition to E. coli, all strains of Salmonella, Y. pestis, P. aeruginosa, and B. cepacia. Additionally, ArnB displays somewhat lower but still very significant sequence homology to the above-mentioned aminotransferases involved in the synthesis of the perosamine present in lipopolysaccharide O-antigens, or the amino sugars found in some macrolide and glycopeptide antibiotics (45). Like ArnB, each of these aminotransferases displays a strong preference for L-Glu as the amine donor (45). Multiple sequence alignments indicate that ArnB contains a highly conserved lysine residue (188), which presumably forms a Schiff's base with the pyridoxal phosphate cofactor that is detected in the ultraviolet visible absorption spectrum at 430 nm (Fig. 9). The recently reported crystal structure of 3-amino-5-hydroxybenzoic acid synthase likewise shows the cofactor covalently bound to the analogous Lys (188) through the expected internal aldimine linkage (46). Our spectroscopic data (Fig. 9) are consistent with ArnB being isolated mainly in the pyridoxal phosphate (aldimine) form of the cofactor, presumably linked to Lys-188. In the presence of excess L-Glu (Fig. 9), the bound pyridoxal phosphate cofactor is converted to the pyridoxamine phosphate derivative. The latter would be used directly to transfer the amine group to UDP-Ara4O (Fig. 2).
The UDP-L-Ara4N (Figs. 4 -6) generated by ArnB in vitro provides a new tool with which to explore the later steps of the pathway for the modification of lipid A with L-Ara4N (Fig. 2). In cell extracts derived from polymyxin-resistant E. coli, UDP-L-Ara4N can be metabolized to what appears to be a 4Љ-formylamino derivative (22). A separate catalytic domain of ArnA catalyzes formyl group transfer from N-10-formyltetrahydrofolate to the 4Љ-amine of UDP-L-Ara4N to make the proposed nucleotide UDP-Ara4-formyl-N (Fig. 2). 3 The role of this formylated nucleotide in the modification of lipid A with L-Ara4N is not immediately obvious, given that undecaprenyl phosphate-␣-L-Ara4N (16,17) is the Ara4N donor used by ArnT for transferring the L-Ara4N moiety to lipid A.
Biosynthesis of the UDP-L-Ara4-formyl-N intermediate in the cytoplasm (Fig. 2) might compensate for the relatively unfavorable equilibrium constant for the conversion of UDP-Ara4O to UDP-L-Ara4N, which we estimate to be about 0.1 (Figs. 7 and 8). Preliminary results suggest that ArnC might actually be selective for UDP-L-Ara4-formyl-N, as proposed in Fig. 2. 3 However, deformylation would then be necessary prior to L-Ara4N transfer to lipid A from undecaprenyl phosphate-␣-L-Ara4N (16,17). It would make sense for the proposed deformylation to occur in the periplasm in order to avoid a futile cycle in the cytoplasm (Fig. 2). Furthermore, if the putative inner membrane transporter (flippase) for undecaprenyl phosphate-␣-L-Ara4-formyl-N (Fig. 2) were unable to transport undecaprenyl phosphate-␣-L-Ara4N, accumulation of the latter on the outer surface of the inner membrane would take place, thereby ensuring that ArnT has an adequate supply of its donor substrate with which to modify lipid A prior to its export to the outer membrane. The latter process occurs very rapidly (within a minute) in wild-type cells (47). The biochemical characterization of the formyltransferase activity of ArnA (22), together with the development of assays for ArnC and the putative deformylase (Fig. 2), should reveal fundamental new insights into the compartmentalization of lipid A modification enzymes and the mechanisms of lipid transport across the inner membrane.