Oxidative decarboxylation of UDP-glucuronic acid in extracts of polymyxin-resistant Escherichia coli. Origin of lipid a species modified with 4-amino-4-deoxy-L-arabinose.

Addition of the 4-amino-4-deoxy-l-arabinose (l-Ara4N) moiety to the phosphate groups of lipid A is implicated in bacterial resistance to polymyxin and cationic antimicrobial peptides of the innate immune system. The sequences of the products of the Salmonella typhimurium pmrE and pmrF loci, both of which are required for polymyxin resistance, recently led us to propose a pathway for l-Ara4N biosynthesis from UDP-glucuronic acid (Zhou, Z., Lin, S., Cotter, R. J., and Raetz, C. R. H. (1999) J. Biol. Chem. 274, 18503-18514). We now report that extracts of a polymyxin-resistant mutant of Escherichia coli catalyze the C-4" oxidation and C-6" decarboxylation of [alpha-(32)P]UDP-glucuronic acid, followed by transamination to generate [alpha-(32)P]UDP-l-Ara4N, when NAD and glutamate are added as co-substrates. In addition, the [alpha-(32)P]UDP-l-Ara4N is formylated when N-10-formyltetrahydrofolate is included. These activities are consistent with the proposed functions of two of the gene products (PmrI and PmrH) of the pmrF operon. PmrI (renamed ArnA) was overexpressed using a T7 construct, and shown by itself to catalyze the unprecedented oxidative decarboxylation of UDP-glucuronic acid to form uridine 5'-(beta-l-threo-pentapyranosyl-4"-ulose diphosphate). A 6-mg sample of the latter was purified, and its structure was validated by NMR studies as the hydrate of the 4" ketone. ArnA resembles UDP-galactose epimerase, dTDP-glucose-4,6-dehydratase, and UDP-xylose synthase in oxidizing the C-4" position of its substrate, but differs in that it releases the NADH product.

Mutations in the pmrE/ugd gene or in the pmrF operon render S. typhimurium incapable of making L-Ara4N-modified lipid A (20,23), and abolish polymyxin resistance in pmrA constitutive strains. The genes in the pmrF operon, pmrHFI-JKLM, are co-transcribed under PmrA control, and except for pmrM, are required for the maintenance of polymyxin resistance (23). The pmrE gene and the pmrF operon are also present in E. coli (20), and appear to be activated in polymyxin-resis-tant mutants or in wild-type cells exposed to ammonium metavanadate (14). Based upon sequence similarity of the deduced proteins of the pmrE and pmrF loci to enzymes of known function, a putative biosynthetic pathway that starts with UDP-glucuronic acid (UDP-GlcA) was recently proposed by two independent laboratories as the source of the L-Ara4N substituent of lipid A (Fig. 2) (14,26). Although the initial steps of the pathway have not yet been validated, we recently showed that arnT (previously pmrK) encodes an inner membrane enzyme that utilizes the novel glycolipid, undecaprenyl phosphate-L-Ara4N, to modify lipid A (27,28). The existence of an undecaprenyl phosphate intermediate implies that L-Ara4N transfer to lipid A occurs on the periplasmic side of the inner membrane (Fig. 2).
We now show that extracts of a polymyxin-resistant E. coli mutant (but not wild type) can convert UDP-GlcA to the novel sugar nucleotide, uridine 5Ј-␤-L-threo-pentapyranosyl-4Љ-ulose diphosphate (UDP-L-Ara4O) (Fig. 2). The latter is subsequently transaminated to make UDP-4-amino-4-deoxy-L-arabinose (UDP-L-Ara4N) (Fig. 2). Furthermore, UDP-L-Ara4N is formylated in vitro in the presence of N-10-formyltetrahydrofolate, but the biological significance of this reaction is unclear (Fig.  2). The UDP-GlcA C-4Љ-dehydrogenase activity of ArnA was demonstrated with overexpressed protein, made by cloning arnA behind the T7 promoter. ArnA (previously designated PmrI or Orf3) (20,23) by itself catalyzed both the C-4Љ oxidation and C-6Љ decarboxylation of UDP-GlcA to generate UDP-L-Ara4O (Fig. 2). The latter was obtained in 6 mg quantities almost entirely in the form of its 4Љ-hydrate in aqueous solution, as judged by one-dimensional and two-dimensional NMR spectroscopy. The data support our proposed scheme for L-Ara4N biosynthesis and transfer to lipid A (Fig. 2), but are at variance with Baker et al. (26) and Gunn (29) who suggested that an additional enzyme (PmrJ) catalyzes the C-6Љ-decarboxylation.
ArnA is homologous to other enzymes that oxidize the C-4Љ position of certain UDP sugars, such as UDP-galactose epimerase (30), dTDP-glucose-4,6-dehydratase (31), and the eucaryotic enzyme that converts UDP-GlcA to UDP-xylose (32,33). ArnA is unique, however, in releasing the 4Љ ketone and the NADH products without reutilizing the latter for a subsequent reduction step. ArnA, furthermore, contains a second catalytic domain that is highly homologous to methionyl-tRNA formyltransferase (34). Although the function of the formyltransferase domain of ArnA remains unclear, it is responsible for the UDP-L-Ara4N formylation observed in vitro (Fig. 2).
Synthesis of [␣-32 P]UDP-glucose-The synthesis of [␣-32 P]UDP-glucose was analogous to that of [␣-32 P]UDP-GlcNAc (35), and accomplished in a reaction containing the following: [␣-32 P]UTP (250 Ci, 800 Ci/mmol), 0.5 mM glucose-1-phosphate, 1 mM MgCl 2 , 5 mM DTT, 100 mM Tris-HCl (pH 8.0), UDP-glucose pyrophosphorylase (3 units), and inorganic pyrophosphatase (12.5 units) in a total volume of 65 l. The reaction was allowed to proceed at 30°C for 1 h, diluted to 1.5 ml 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 followed by elution of the product with a 0 -100 mM triethylammonium bicarbonate (pH 8.5) gradient (40 ml total volume). The eluted material was concentrated using a Speed-Vac centrifuge, and was re-dissolved in 200 l of deionized water. The yield was generally greater than 75% based on recovery of radioactivity.
Synthesis of [␣-32 P]UDP-glucuronic acid-The conversion of [␣-32 P]UDP-glucose to [␣-32 P]UDP-glucuronic acid proceeded in the following reaction: 50 mM Tris (pH 8.7), 2 mM DTT, 6 mM NAD, [␣-32 P]UDPglucose (100 Ci), 3 mM MgCl 2 , and 75 g of UDP-glucose dehydrogenase in 150 l total volume for 60 min at 30°C. The reaction was diluted to 1.5 ml 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 and 5 ml of 50 mM triethylammonium bicarbonate (pH 8.5). The product was eluted with 30 ml of 100 mM triethylammonium bicarbonate (pH 8.5). The eluted material was concentrated by lyophilization, and re-dissolved in deionized water. The yield was generally greater than 75% based on recovery of radioactivity.
Preparation of Cell Extracts-Typically, bacteria were grown in LB medium (36) at 37°C to late log phase, and harvested at an A 600 ϭ 1.0 by centrifugation at 6500 ϫ g for 15 min. Cells from a 100-ml culture were resuspended in 5 ml of 50 mM HEPES (pH 7.5), 2 mM DTT, and 10% glycerol, passed through a French pressure cell at 18,000 psi, and then clarified by centrifugation at 16,500 ϫ g for 30 min at 4°C. These cell-free extracts were stored at Ϫ80°C. Protein concentrations were determined by bicinchoninic acid using the Pierce assay reagent with bovine serum albumin as the standard (37).
In Vitro Assays for Detecting [␣-32 P]UDP-glucuronic Acid Conversion to Novel ␣-32 P-Labeled Sugar Nucleotides-Assays contained 50 mM HEPES (pH 7.5), 2 mM DTT, 10 M [␣-32 P]UDP-glucuronic acid (8 ϫ 10 5 cpm/nmol), 0.5 mg/ml protein, and when appropriate, 3 mM NAD, 3 mM NADP, 1 mM L-glutamate, and/or 1 mM 5,10-methenyltetrahydrofolate in a total volume of 50 l, as indicated. Reactions were incubated at 30°C, and at the desired time points 1 l was withdrawn and spotted  (20,25). Minor lipid A species may be 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 (25). Addition of the palmitoyl chain is catalyzed by the outer membrane enzyme PagP (59), removal of the hydroxymyristoyl chain at position 3 is carried out by PagL (60), and formation of 2-hydroxymyristate requires a novel hydroxylase homologue (61), encoded by lpxO. PagL and LpxO occur only in S. typhimurium, but can function when transferred to E. coli. Substituents that are thought to be under PhoP/PhoQ control (15) are shown in red.
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 0.25 M acetic acid with 0.4 M LiCl. The plate was dried and exposed to a PhosphorImager screen. Radioactive enzymatic products were identified and quantified with a Phosphor-Imager (ImageQuant software, Molecular Dynamics).
Synthesis of N-5,N-10-Methenyltetrahydrofolate-The conversion of N-5-formyltetrahydrofolate to N-5,N-10 methenyltetrahydrofolate was carried out as described (38). In 1.5 ml of deionized water containing 1% (v/v) 2-mercaptoethanol, 7.5 mg of N-5-formyltetrahydrofolate was dissolved, and the pH was adjusted to 1.9 with 0.1 N HCl. The material was diluted to 2.2 ml with deionized water, and the reaction was allowed to proceed at room temperature. Generation of N-5,N-10 methenyltetrahydrofolate was followed by measuring the increase in absorbance at 355 nm. Upon completion, the reaction mixture was stored at Ϫ20°C.
Recombinant DNA Techniques-Preparation of competent cells, transformation by the CaCl 2 method, genomic DNA purification, and electrophoresis were performed according to published procedures (39). Plasmids were purified using the QIAprep spin miniprep kit (Qiagen). Restriction enzymes, Pfu DNA polymerase, and T4 DNA ligase were used as recommended by the manufacturer. DNA sequencing was performed on an ABIprism 377 instrument at the Duke University DNA Analysis Facility.
Construction of pETArnA-The predicted coding region for ArnA (PmrI) in the pmrF operon (23) was amplified by polymerase chain reaction from E. coli W3310 genomic DNA (40) with primers to the arnA 5Ј end (5Ј-CGGGATCCATATGAAAACCGTCGTTTTTGCCT-3Ј) and to the arnA 3Ј end (5Ј-CGGGATCCTCGAGTCATGATGGTTTATCCGTA-AG-3Ј), containing engineered NdeI and XhoI restriction sites, respectively. The Opti-Prime TM PCR Optimization kit (Stratagene) was used as described by the manufacturer. Amplification in a 50-l reaction mixture with 10 ϫ Opti-Prime Buffer 9 (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 cloned Pfu DNA polymerase for 30 cycles (94°C for 1 min, 52°C for 1 min, and 72°C for 2.5 min) followed by one cycle of 94°C for 1 min, 52°C for 1 min, and 72°C for 15 min resulted in a single product of the expected size. The PCR product was isolated with the QIAquick PCR Purification kit (Qiagen), digested with XhoI and NdeI, and ligated into a T7 expression vector, pET24b (Novagen), previously digested with the same enzymes. The fidelity of the Pfu DNA polymerase was confirmed by DNA sequencing of the cloned arnA gene in pETArnA.
Overexpression of the ArnA Gene Product-The plasmid pETArnA was transformed into E. coli NovaBlue(DE3) cells (Novagen). Cultures were grown overnight at 37°C from single colonies in LB broth supplemented with 30 g/ml kanamycin. The overnight culture was then used to inoculate 100 ml of the same LB broth, and cells were grown at 37°C to A 600 of 0.6. Expression was induced by the addition of isopropyl-␤-Dthiogalactopyranoside to a final concentration of 1 mM, and growth was continued at 30°C for 3.5 h. Cells were harvested by centrifugation at 6,500 ϫ g for 15 min at 4°C. The cell pellet was resuspended in 3 ml of buffer containing 50 mM HEPES (pH 7.5), 10% glycerol, and 2 mM DTT, and lysis was achieved by passage through a French pressure cell at 18,000 psi. Cell debris was removed by centrifugation at 16,500 ϫ g for 15 min at 4°C to give crude cell-free extracts. Expression was monitored by SDS-polyacrylamide gel electrophoresis.
Enzymatic Synthesis and Isolation of Uridine 5Ј-(␤-L-Threo-pentapyranosyl-4Љ-ulose diphosphate) (Compound 1)-A 15-ml reaction mixture containing 0.1 mg/ml of the recombinant ArnA crude cell-free extract, 1 mM UDP-glucuronic acid, and 6 mM NAD, in buffer consisting of 50 mM HEPES (pH 7.5) and 2 mM DTT, was allowed to proceed for 8 h at 30°C. Protein was removed by passing the reaction mixture through a Centricon TM (YM-10) membrane. The enzymatic product was 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

FIG. 2. Biosynthesis of UDP-L-Ara4N from UDP-GlcA in polymyxin-resistant E. coli and S. typhimurium.
The pathway starts with the conversion of UDP-glucose to UDP-GlcA by the well characterized dehydrogenase, Ugd (PmrE). ArnA (previously Orf3 or PmrI) (20, 23) catalyzes the oxidative decarboxylation of UDP-glucuronic acid to form the novel sugar nucleotide, UDP-L-Ara4O. A separate enzyme to catalyze the decarboxylation is not needed. ArnB (previously Orf1 or PmrH) (20, 23) then catalyzes the further transamination to make UDP-L-Ara4N. Based upon its homology to dolichoyl phosphate-mannose synthase of yeast, we suggest that ArnC (PmrF) transfers the L-Ara4N moiety to undecaprenyl phosphate (14), forming the novel compound undecaprenyl phosphate-␣-L-Ara4N, which has recently been isolated from polymyxin-resistant mutants (28). After translocation to the outer surface of the inner membrane, ArnT (Orf5, PmrK, or YfbI) transfers the L-Ara4N unit to lipid A (27). The ArnA protein has a second catalytic domain with strong similarity to methionyl-tRNA formyltransferase (34). This may account for the efficient transfer of a formyl group from N-10-formyltetrahydrofolate to UDP-L-Ara4N in our in vitro system (see below), but the significance of this modification is unclear. It may be that the true substrate for ArnC is the formylated derivative of UDP-L-Ara4N, but an assay for ArnC is not yet available.
washed with 40 ml of deionized water followed by elution of the product with a 0 -100 mM triethylammonium bicarbonate (pH 8.5) gradient (200 ml total volume). Chromatography was monitored by measuring the absorbance at 254 nm, and fractions containing the desired product, the putative UDP-L-Ara4O, 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 AG TM 50W-X8 resin (sodium form), followed by lyophilization to give the pure compound (6 mg, 68% yield).
Nuclear Magnetic Resonance Spectroscopy-The putative UDP-L-Ara4O (Na ϩ salt, total of 6 mg) and the UDP-glucuronic acid (34 mg) were each dissolved in 0.6 ml of 99% D 2 O in 5-mm NMR tubes. Measurements with a 3-mm pH electrode revealed pD values of 6.51 and 5.44 for the ArnA product and the UDP-GlcA samples, respectively. All 1 H and 13 C NMR chemical shifts were referenced to 2,2-dimethylsilapentane-5-sulfonic acid at 0.00 ppm. The 31 P NMR spectra were referenced to 85% phosphoric acid at 0.00 ppm. 1 H and 13 C NMR spectra were obtained on a Varian Inova 600 spectrometer equipped with a Sun Ultra 5 computer and a 5-mm Varian triple resonance probe. One-dimensional 1 H spectra were obtained with a spectral width (SW) of 6.5 kHz, a 72°pulse flip angle (6 s), a 6.3-s acquisition time, and a 1.2-s relaxation delay (RD), and digitized using 82000 points to obtain a digital resolution of 0.159 Hz/pt. 1 H-decoupled one-dimensional 13 C spectra were recorded with a 35 kHz SW, a 54°p ulse flip angle (6 s), and a 3.1-s repeat time, and digitized into 91000 points to give a digital resolution of 0.777 Hz/pt. Two-dimensional COSY data were recorded in absolute value mode with 2048 points, 1-s RD, and 32 scans per increment; 512 time increments were collected and zero-filled to 2048 points with sine-bell weighting in both dimensions before Fourier transformation, followed by symmetrization of the two-dimensional matrix.
The two-dimensional NOESY data were recorded in hypercomplex phase-sensitive mode with two sets of 256 time incremented spectra, 32 scans per increment, a 1-s delay and a mixing time of 500 ms. The final two-dimensional matrix was 2048 ϫ 2048 with Gaussian weighting in both dimensions, and was not symmetrized.
Proton-detected single bond 1 H-13 C two-dimensional chemical shift correlation spectra were recorded using the HMQC method (41) with 13 C decoupling during acquisition. Two sets of 200 time increments were obtained in hypercomplex phase-sensitive mode with 2048 points in t 2 ; 64 scans were recorded per time increment, and the RD was 1.2 s. The two-dimensional data were processed using Gaussian functions and zero-filled to a final size of 2048 ϫ 2048.
The two-dimensional HMBC multiple bond correlation spectra (41,42) were recorded in hypercomplex phase-sensitive mode without 13 C decoupling during acquisition. Two sets of 200 time increments were recorded with 2048 points in t 2 . The RD was 1.2 s, the filter delay was set to an average 1 J CH of 140 Hz, and 96 transients were obtained per increment. The long-range 1 H-13 C couplings were allowed to evolve for a delay of 83 ms (6 Hz optimization). The data were processed using a shifted Gaussian weighting function along f 2 and a cosine weighting function along f 1 , and zero-filled to a final size of 2048 ϫ 2048.
The one-dimensional 1 H and 31 P NMR spectra were obtained on a Varian Unity 500 spectrometer equipped with a Sun Ultra 5 computer and a 5-mm Varian inverse probe. 1 H spectra at 500 MHz were obtained with a 4.5 kHz SW, 67°pulse flip angle (6 s), 9.4-s acquisition time, and 2-s RD, and digitized using 84000 points to obtain a digital resolution of 0.107 Hz/pt. 31 P spectra at 202 MHz were obtained with a 12 kHz SW, 59°pulse flip angle (13 s), 2.4-s acquisition time, and 1.6-s RD, and digitized using 25000 points to obtain a digital resolution of 0.97 Hz/pt. Selective inverse decoupling difference spectroscopy (43,44) was implemented by detecting the 1 H NMR spectrum at high digitization while selectively exciting a phosphorus signal with 1 s of low-power (estimated 0.2 mW, ϳ4 db on the Unity 500 spectrometer) continuous wave on-and off-resonance irradiation during the acquisition period. To improve subtraction in the generation of one-dimensional difference 31 P-decoupled 1 H NMR spectra, the samples were thermally equilibrated in the magnet for over 1 h before recording data, and the 1 H difference spectra were obtained in an interleaved manner with 4 scans accumulated for each free induction decay and looping around n times to achieve a good signal-to noise ratio with 4n scans per free induction decay.  (28) were tested for their ability to catalyze the oxidative decarboxylation and transamination of [␣-32 P]UDP-GlcA (Fig. 2), using a combination of TLC and PhosphorImager analysis to detect the products. In the presence of exogenous NAD, extracts of the polymyxin-resistant E. coli mutant WD101 (28) rapidly converted [␣-32 P]UDP-GlcA to a faster migrating species designated compound 1 (Fig. 3, lane 5). The relatively slow migration of the substrate versus the product suggests that 1 is less negative than [␣-32 P]UDP-GlcA. Formation of 1 was absolutely dependent upon the presence of added NAD, but did occur with NADP at a slow rate (Fig. 3, lane 6). An extract of the polymyxinsensitive wild-type strain W3110 (the parent of WD101) did not metabolize [␣-32 P]UDP-GlcA in the presence of NAD or NADP (Fig. 3, lanes 1-3). We propose that the biosynthesis of 1 is catalyzed by ArnA(PmrI/Orf3) of the PmrF operon (Fig. 2), given the distant sequence similarity of ArnA to dTDP-glucose-4,6-dehydratase and UDP-galactose epimerase (14).

Enzymatic
The formation of the 4Љ-keto intermediate (Fig. 2) should facilitate 6Љ-decarboxylation, which might be spontaneous or enzyme catalyzed. In either case, compound 1 provides the substrate for an aminotransferase (proposed to be ArnB/PmrH/ Orf1) (14) that generates [␣-32 P]UDP-L-Ara4N (Fig. 2). In accordance with this idea, crude extracts of mutant WD101 efficiently converted [␣-32 P]UDP-GlcA to an even faster migrating species, designated compound 2 (Fig. 4, lane 3), when L-glutamate was present in addition to NAD. The R F of 2 in this solvent system was about the same as that of UDP-glucosa- mine, which has the same net charge as [␣-32 P]UDP-L-Ara4N. Presumably, the NAD is first used to convert [␣-32 P]UDP-GlcA to compound 1, given that [␣-32 P]UDP-GlcA by itself is not converted directly to 2 in the presence of L-glutamate alone (Fig. 4, lane 4). A matched extract of a polymyxin-sensitive wild-type E. coli strain was completely inactive in the formation of 2 (Fig. 4, lane 1).
L-Glutamate likely functions as the co-substrate for a pyridoxal phosphate-dependent aminotransferase (Fig. 2). L-Glutamine also supports this conversion, but less efficiently than L-glutamate. L-Alanine and L-aspartate are inactive (data not shown). The natural cosubstrate of the putative aminotransferase is probably L-glutamate, which would be converted to ␣-ketoglutarate during the formation of UDP-L-Ara4N (Fig. 2).
Formylation of [␣-32 P]UDP-L-Ara4N in Extracts of WD101-ArnA, which is predicted based upon its sequence to catalyze the NAD-dependent conversion of UDP-GlcA to 1, possesses an additional N-terminal domain of unknown function that displays significant similarity to methionyl-tRNA formyltransferases (34). Given this peculiar observation, we investigated the possibility of a formyltransferase that acts on [␣-32 P]UDP-L-Ara4N in extracts of WD101 by including exogenous N-5,N-10-methenyltetrahydrofolate in addition to NAD and glutamate in the assay system (Fig. 5). The role of a formylated-nucleotide intermediate in the biosynthesis of L-Ara4Nmodified lipid A is unclear a priori, because formylated L-Ara4N moieties have not been reported in association with lipid A, or elsewhere (14, 25, 44 -47). Nevertheless, when exogenous N-5,N-10-methenyltetrahydrofolate was included in our system, an additional more slowly migrating substance, designated compound 3, was formed in significant quantities (Fig. 5, lane 4). The low R F of 3 (Fig. 5) suggested formylation of the free amino group in 2 (Fig. 2), but O-formylation could not be excluded. The formylation reaction was highly specific for N-5,N-10-methenyltetrahydrofolate, which at pH 7.5 rap-idly rearranges to N-10-formyltetrahydrofolate (38). The latter is the actual formyl donor (Fig. 2). The commercially available analog, 5-formyltetrahydrofolate, was inactive (Fig. 5, lane 5). The substrate requirements for the formation of 3 indicated that UDP-GlcA and 1 (Fig. 5, lanes 1 and 2) were not utilized directly by the formyltransferase. As in the case of the other transformations described above, extracts of wild-type cells did not catalyze formyl transfer to 2 (data not shown).
Cloning and Overexpression of arnA-Although the metabolism of [␣-32 P]UDP-GlcA observed in extracts of E. coli WD101 (Figs. 3-5) supported the proposed pathway for L-Ara4N biosynthesis and transfer to lipid A (Fig. 2), the actual enzymes catalyzing these transformations had not been identified, and data establishing the structures of the proposed biosynthetic intermediates (14) were lacking. To prove that arnA encodes the NAD-dependent dehydrogenase that converts UDP-GlcA to compound 1 (Fig. 3), the full-length arnA gene was cloned and expressed in the pET24b vector behind a T7 RNA polymerase promoter, inducible with isopropyl-␤-D-thiogalactopyranoside (48). Expression was carried out in E. coli NovaBlue(DE3), a polymyxin-sensitive strain that does not modify its own lipid A with the L-Ara4N moiety (27). A protein of the predicted molecular mass of ArnA, 73 kDa, was observed by SDS-gel electrophoresis in cell-free extracts of induced cells, but was absent in uninduced controls (Fig. 6A).
The induced cell-free extracts derived from cells containing pETArnA catalyzed very rapid NAD-dependent conversion of UDP-GlcA to 1; this activity was not present in matched extracts derived from cells harboring the vector pET24b (data not

FIG. 4. Conversion of [␣-32 P]UDP-GlcA to compound 2 in the presence of NAD and glutamate.
Cell-free extracts (0.5 mg/ml) of wild-type E. coli (Pm s ) or WD101 (Pm r ) were assayed at 30°C for 90 min with 10 M [␣-32 P]UDP-GlcA. Next, a 1-l portion of each reaction mixture was spotted onto a polyethyleneimine-cellulose plate, which was developed and analyzed as described in the legend to Fig. 3. Lane 1, wild-type E. coli (Pm s ) with 3 mM NAD and 1 mM L-glutamate. Lanes 2-4, WD101 (Pm r ) with 3 mM NAD, with 3 mM NAD plus 1 mM L-glutamate, or 1 mM L-glutamate, respectively. Lane 5, no enzyme control with 3 mM NAD and 1 mM L-glutamate.

FIG. 5. Conversion of [␣-32 P]UDP-GlcA to compound 3 in the presence of NAD, glutamate, and N-10-formyltetrahydrofolate.
Cell-free extracts (0.5 mg/ml) of E. coli mutant WD101 (Pm r ) were assayed at 30°C for 180 min in the presence of 10 M [␣-32 P]UDP-GlcA. Next, a 1-l portion of each reaction mixture was spotted onto a polyethyleneimine-cellulose plate, which was developed and analyzed as described in the legend to Fig. 3. As indicated for lanes 1-5, 3 mM NAD, 1 mM L-glutamate, and/or various formyl donors (each at 1 mM) were also included in the assay. The N-5,N-10-methenyltetrahydrofolate is converted in situ to N-10-formyltetrahydrofolate by means of a 10-min preincubation (38). Formation of compound 3 from 2 is absolutely dependent upon the presence of N-10-formyltetrahydrofolate. Lane 6 is the no enzyme control.
shown). The extracts of cells containing the induced pETArnA plasmid showed ϳ100-fold higher ArnA specific activity than comparable preparations of WD101 (Fig. 6B).
Given these results, the predicted genes of the pmrF operon (14,20) were renamed to emphasize the enzymatic activities they encode (27), in accordance with the proposal of Reeves et al. (49). Although these genes are necessary for maintenance of polymyxin resistance, they are in fact specifically involved in L-Ara4N biosynthesis and transfer of lipid A (hence "Arn"), with ArnA catalyzing the first reaction of the pathway (Fig. 2).
Evaluation of the Covalent Structure of Compound 1 by Onedimensional 1 H NMR Spectroscopy-Compound 1 was isolated on a 6-mg scale using the overexpressed recombinant ArnA protein, as described under "Experimental Procedures." The full 1 H NMR spectra of UDP-glucuronic acid and 1 are shown in Fig. 7, A and B, respectively. The spectra reveal the presence of a uracil and ribose moiety in both compounds with virtually identical chemical shifts and coupling constants (Table I and Fig. 7). The remaining 1 H NMR signals, which are well resolved and arise from the pyranose, differ in the two compounds (Table I and Fig. 7). The spectrum of 1 is simplified in the 3.6 -4.0 ppm region, where three distinct doublets are now seen (Fig. 7B) when compared with the more complex multiplets of the UDP-GlcA (Fig. 7A). The 1 H NMR spectra thus demonstrate that the enyzmatic reaction catalyzed by ArnA altered the glucuronic acid moiety of UDP-GlcA but left the uridine portion of the molecule unchanged. 31 P NMR Spectroscopy of Compound 1-31 P NMR analysis (not shown) confirmed the presence of an unaltered diphosphodiester linkage in compound 1 versus UDP-GlcA from the observation of two distinct 31 P NMR signals near Ϫ11.9 and Ϫ10.2 ppm with ϳ20.7 Hz 31 P-31 P coupling. Subtraction of two 1 H NMR spectra obtained with on-and off-resonance selective decoupling of the Ϫ10.2 ppm 31 P signal in compound 1 revealed decoupling changes only at ribose H-5Ј and H-4Ј (Fig. 7C), clearly assigning the Ϫ10.2 ppm 31 P signal as the phosphate group linked to ribose C-5Ј. The 1 H NMR difference spectrum obtained upon selective decoupling of the Ϫ11.9 ppm 31 P resonance (Fig. 7D) reveals decoupling changes only at the anomeric H-1Љ and the H-2Љ of compound 1, assigning the Ϫ11.9 ppm signal to the phosphate group linked to C-1Љ.
Analysis of Compound 1 by Two-dimensional 1 H NMR Spectroscopy-The two-dimensional COSY analysis (Fig. 8) confirms the identity of the ribose and uracil rings in the two compounds. The cross-peak connectivity from H-1Ј to H-2Ј, the degeneracy of H-2Ј and H-3Ј (ϳ4.40 ppm), the connectivity from H-3Ј to H-4Ј (ϳ4.31 ppm), and the connectivities of H-4Ј to H-5aЈ and H-5bЈ are the same in both compounds. The uracil H-6 and H-5 crosspeaks are likewise the same in both (not shown).
The small J 1Љ2Љ couplings (3.3 Hz) and the large J 2Љ3Љ coupling (9.8 Hz) of the pyranose moiety in both the substrate and the product (Table I) demonstrate that these sugars have an equatorially disposed H-1Љ, and contain axially disposed H-2Љ and H-3Љ protons (50,51). Additional two-dimensional NOESY experiments with 1 (not shown) reveal strong intramolecular NOE cross-peaks between H-1Љ and H-2Љ, further supporting the idea of an equatorial H-1Љ and an axial H-2Љ in an ␣-glucopyranosyl configuration (50). The 9.8 Hz J 2Љ3Љ coupling implies that H-3Љ is axial and trans-relative to H-2Љ, as in UDP-glucuronic acid. The H-5aЉ (3.58 ppm) and H-5bЉ (3.96 ppm) signals of compound 1 (Figs. 7B and 8B) were assigned to the axial and equatorial positions, respectively, based upon the consideration that a proton in an axial position on a given pyranose carbon usually resonates ϳ0.5 ppm higher than the corresponding equatorial proton (51). However, attempts to validate the assignments of the two H-5Љ protons of compound 1 independently with a two-dimensional NOESY experiment were unsuccessful, as no NOE cross-peak was observed between the axially disposed H-3Љ and either H-5aЉ or H5bЉ.
Because the ketone group of a hexulose may be hydrated (52), it is possible that both the ketone form and its corresponding hydrate co-exist in solution. The 1 H NMR spectrum of CDP-3,6-dideoxy-D-glycero-4-hexulose in D 2 O (52) does in fact show two sets of hexulose sugar resonances, whereas our product shows only one. Furthermore, compound 1 is very stable, given that identical 1 H NMR spectra were recorded over the course of three months in D 2 O solution, and repeated integrations showed no evidence for deuterium exchange at the axial H-3Љ. The 1 H NMR data indicate only one predominant form of compound 1 in D 2 O, most likely the hydrate, given the negligible rate of deuterium exchange at H-3Љ.
Evaluation of Compound 1 by 13 C NMR Spectroscopy-To distinguish between the ketone and/or hydrate forms of compound 1, and to validate its structure, natural abundance 13 C NMR spectra were recorded for both UDP-GlcA and 1 (Figs. 9,   A and B, respectively). Similar to the 1 H NMR results, the 13 C spectra show that the uracil and ribose moieties are the same in both compounds ( Fig. 9 and Table I). The 13 C spectra differ in the locations of the pyranose sugar resonances between 67 and 77 ppm, and in the appearance of a new carbon signal at 95.1 ppm in compound 1 (Fig. 9B). The 95.1 ppm signal is in the anomeric region, indicating an additional carbon substituted with two oxygens, as would be expected for the hydrated form of the ketone. Furthermore, the glucuronic acid carbonyl of the substrate at 178.8 ppm (Fig. 9A) is entirely absent in 1 (Fig.  9B), confirming that 6Љ-decarboxylation has occurred.
The 95.1 ppm carbon signal of compound 1 is further shown to arise from a quaternary ketal carbon by the two-dimensional HMQC 1 H-13 C correlation map (Fig. 10A). There are only three direct 1 H-13 C single-bond correlations in the anomeric region (90 -105 ppm). The anomeric H-1Љ (ϳ5.6 ppm) correlates to the carbon resonance of the pyranose near 98 ppm (C-1Љ) (Fig. 10A), as in UDP-GlcA (not shown), consistent with an axially disposed oxygen atom and an equatorially disposed H-1Љ (50). The anomeric ribose H-1Ј proton near 6.01 ppm correlates to the carbon resonance at 90.9 ppm (C-1Ј), while the uracil H-5 near 6.00 ppm correlates to the carbon signal near 105.2 ppm. The fourth carbon signal in the anomeric region near 95.1 ppm has no proton partner in the two-dimensional spectrum (Fig. 10A), consistent with the presence of a quaternary ketal carbon.
Examination of the two-dimensional HMQC 1 H-13 C data for  (Table   I), the assignments of which agree with literature values (50).
The multibond HMBC map in the sugar region of 1 (Fig. 10B) further verifies the above assignments. Multibond correlations are clearly observed from the H-3Љ and the H-5aЉ doublets to the 95.1 ppm carbon resonance, unequivocally demonstrating TABLE I NMR spectroscopy of UDP-GlcA and UDP-L-Ara4O 31 P chemical shifts of UDP-glucuronic acid are Ϫ10.1 (␣-P) and Ϫ11.8 ppm (␤-P), respectively. 31 P-31 P coupling ϭ 20.6 Hz. 31 P chemical shifts of UDP-L-Ara4O are Ϫ10.2 (␣-P) and Ϫ11.9 ppm (␤-P), respectively. 31   that the latter must arise from C-4Љ. Taken together, the NMR data are fully consistent with the identification of 1 as the hydrated form of the novel sugar nucleotide, uridine 5Ј-␤-Lthreo-pentapyranosyl-4Љ-ulose diphosphate, designated UDP-L-Ara4O (Fig. 2). DISCUSSION The covalent modification of the phosphate groups of lipid A with L-Ara4N moieties confers resistance to polymyxin and cationic anti-microbial peptides in E. coli and S. typhimurium (17,18,20,25,27,28). The L-Ara4N unit is thought to arise by a novel pathway from UDP-GlcA. The proposed scheme (Fig. 2) was initially developed based upon bioinformatic analyses (14,26) of the gene products encoded by the pmrE and pmrF loci, which are required for L-Ara4N biosynthesis and polymyxin resistance.
All three enzymes are absent in extracts of W3110, the polymyxin-sensitive, wild-type strain of E. coli from which the pmrA constitutive mutant WD101 is derived (28). These findings are consistent with the fact that PmrA activates the transcription of the L-Ara4N biosynthesis/polymyxin-resistance genes of the pmrE and pmrF loci (20,23).
The E. coli arnA gene was cloned and overexpressed in E. coli NovaBlue(DE3) behind an inducible T7 promoter (Fig. 6). The massive overproduction of the NAD-dependent conversion of UDP-GlcA to compound 1 in association with the synthesis of the expected protein of 74 kDa (Fig. 6) demonstrates that arnA is the structural gene for the enzyme. Preparation of 6 mg of compound 1 with the ArnA overproducing extracts greatly facilitated detailed structural analysis by NMR spectroscopy (Figs. 7-10). Our results demonstrate unequivocally that compound 1 is the proposed UDP-L-Ara4O intermediate (Fig. 2), which exists almost entirely as the 4Љ-hydrate in aqueous solution (Fig. 7). In the case of CDP-3,6-dideoxy-D-glycero-D-glycero-4-hexulose (a precursor of CDP-paratose) (52), distinct 13 C NMR signals were observed at 210 ppm (the 4Љ-keto form) and at 90 ppm (the 4Љ-hydrate), whereas the ketone could not be detected by NMR spectroscopy in our product (Fig. 9). The purified UDP-L-Ara4O is stable in water, however, and can be converted to compounds 2 and 3 in milligram quantities in the presence of cloned, purified ArnB and ArnA, respectively, facilitating the validation of the proposed structures shown in Fig. 2. 2 The NMR analysis (Figs. 7-10) demonstrates that ArnA is the only enzyme required for the conversion of UDP-GlcA to UDP-L-Ara4O (Fig. 2). Although it is possible that the C-6Љ 2 S. Breazeale and C. R. H. Raetz, manuscript in preparation. decarboxylation is nonenzymatic following C-4Љ oxidation, we favor the view that C-6Љ oxidation is also catalyzed by ArnA. The fact that ArnA alone is sufficient to catalyze both the C-4Љ oxidation and the C-6Љ decarboxylation of UDP-GlcA is at variance with the biosynthetic scheme proposed by Baker et al. (26) and Gunn (29), who suggest that PmrJ is the decarboxylase. Since PmrJ is not homologous to other proteins of known function, we cannot yet assign a role for PmrJ in our scheme (Fig.  2). However, the genetic evidence does clearly demonstrate that PmrJ is needed for polymyxin resistance (23).
ArnA represents the first reported enzyme that catalyzes the oxidative decarboxylation of UDP-GlcA with the concomitant release of NADH, which can be detected spectrophotometrically when purified ArnA is incubated with UDP-GlcA and NAD (data not shown). UDP-xylose synthase, the structural gene of which was recently identified and cloned based on its intriguing functional similarity to ArnA, 3 likewise catalyzes the C-4Љ oxidation and the C-6Љ decarboxylation of UDP-GlcA (32,33), but the ketone intermediate is immediately reduced by NADH to generate UDP-xylose (32,33). Several other types of enzymes catalyze the oxidation at C-4Љ of related sugar nucleotides, and display limited sequence similarity to the dehydrogenase domain of ArnA. The best characterized of these is UDP-galactose epimerase (30,53), which interconverts UDPgalactose and UDP-glucose. In this case, the presumed 4Љketopyranose intermediate is rotated within the active site and immediately reduced by the adjacent NADH molecule (30), inverting the configuration at C-4Љ. Lastly, there is a large family of sugar nucleotide-4,6-dehydratases that catalyze the oxidation of the pyranose C-4Љ to activate the C-5Љ proton for elimination of water between C-5Љ and C-6Љ (31,54). Like the UDP-galactose epimerases, these dehydratases reutilize the NADH that is generated during the initial C-4Љ oxidation to reduce C-6Љ to a methyl group following elimination of water (31). It may be interesting to study UDP-L-Ara4O as an inhibitor or as an alternative substrate for some of these enzymes.
The function of the formylated derivative of UDP-L-Ara4N (compound 3, Fig. 5) is unclear. Formylated L-Ara4N moieties have not been found in association with lipid A or other natural products (14, 25, 44 -47). Undecaprenyl phosphate-L-Ara4N, the donor substrate in the ArnT catalyzed modification of lipid A (Fig. 2), has recently been purified from WD101 (28). Extensive characterization by NMR spectroscopy and mass spectrometry did not reveal the presence of a formyl group in this compound (28). However, the existence of such a modification might have been overlooked, given the scheme that was used to purify undecaprenyl phosphate-L-Ara4N (28). If the formylated derivative of UDP-L-Ara4N that can be generated in vitro (Fig.   3 T. Doering, personal communication. The carbon signal at 95.1 ppm is consistent with a quaternary ketal carbon, as it shows no connectivity to a proton. Panel B is a partial two-dimensional HMBC spectrum of 1 under the same conditions, demonstrating the multibond 1 H-13 C correlations from the uracil H-5, ribose, and pyranose protons. Of special interest are the multibond correlations from H-3Љ and H-5aЉ to the 95.1 ppm carbon signal, confirming its assignment as C-4Љ. X indicates an impurity or residual HOD resonances. 5) is an obligatory intermediate in the biosynthesis of L-Ara4Nmodified lipid A, a novel deformylase must also be required at some later stage. One possibility is that compound 3 (rather than UDP-L-Ara4N) is the actual physiological substrate of ArnC (Fig. 2). An in vitro assay for ArnC has not yet been developed to test this possibility, and furthermore, a candidate gene encoding a putative deformylase that would have to function prior to the formation of undecaprenyl phosphate-L-Ara4N has not been identified (14,26). However, we have recently shown that the formylation of UDP-L-Ara4N to make compound 3 is catalyzed by purified ArnA, which is a bifunctional protein with a dehydrogenase domain that synthesizes UDP-L-Ara4O (compound 1) and a methionyl-tRNA formyltransferase-like domain that generates compound 3. 2 The sequence similarity of the dehydrogenase domain of ArnA to the epimerases and the dehydratases is not sufficient to permit homology modeling based on available x-ray structures (53,54). On the other hand, the homology of the formyltransferase domain of ArnA to the methionyl-tRNA formyltransferases (34) is much greater, and may be amenable to modeling. Many procaryotic ArnA sequences are in fact currently annotated as methionyl-tRNA formyltransferases. It will be important to determine the x-ray structure of ArnA to evaluate its mechanism of action in relationship to the UDPgalactose epimerases and the sugar nucleotide 4,6-dehydratases, and to assess the interaction of the dehydrogenase and formyltransferase domains.
Orthologs of E. coli and S. typhimurium ArnA are present in the genomes of all pathogenic bacteria known to modify their lipid A with the L-Ara4N moiety (14,45,(55)(56)(57), including all types of Salmonella, Yersinia pestis, Pseudomonas aeruginosa, and Burkholderia cepacia. The existence of full-length ArnA homologues in these organisms indicates that the Ara4N pathway for lipid A modification and polymyxin resistance is likely to be conserved (Fig. 2). The best eucaryotic homologues of ArnA are present in Arabidopsis thaliana, but since they correspond only to the dehydrogenase domain of ArnA, they are likely to encode plant UDP-xylose synthases (32,58).