The Origin of 8-Amino-3,8-dideoxy-d-manno-octulosonic Acid (Kdo8N) in the Lipopolysaccharide of Shewanella oneidensis

Background: The lipopolysaccharide of Shewanella oneidensis contains a unique sugar, Kdo8N.

Results: The Kdo8N biosynthetic genes were identified and characterized.

Conclusion: Kdo8N is synthesized from Kdo via a two-step mechanism utilizing O₂ and glutamate.

Significance: These results expand the known lipid A biosynthetic enzymes, which are important for various human health applications, and present evidence for a potentially novel class of alcohol oxidases.

Abstract

Lipopolysaccharide (LPS; endotoxin) is an essential component of the outer monolayer of nearly all Gram-negative bacteria. LPS is composed of a hydrophobic anchor, known as lipid A, an inner core oligosaccharide, and a repeating O-antigen polysaccharide. In nearly all species, the first sugar bridging the hydrophobic lipid A and the polysaccharide domain is 3-deoxy-d-manno-octulosonic acid (Kdo), and thus it is critically important for LPS biosynthesis. Modifications to lipid A have been shown to be important for resistance to antimicrobial peptides as well as modulating recognition by the mammalian innate immune system. Therefore, lipid A derivatives have been used for development of vaccine strains and vaccine adjuvants. One derivative that has yet to be studied is 8-amino-3,8-dideoxy-d-manno-octulosonic acid (Kdo8N), which is found exclusively in marine bacteria of the genus Shewanella. Using bioinformatics, a candidate gene cluster for Kdo8N biosynthesis was identified in Shewanella oneidensis. Expression of these genes recombinantly in Escherichia coli resulted in lipid A containing Kdo8N, and in vitro assays confirmed their proposed enzymatic function. Both the in vivo and in vitro data were consistent with direct conversion of Kdo to Kdo8N prior to its incorporation into the Kdo8N-lipid A domain of LPS by a metal-dependent oxidase followed by a glutamate-dependent aminotransferase. To our knowledge, this oxidase is the first enzyme shown to oxidize an alcohol using a metal and molecular oxygen, not NAD(P)⁺. Creation of an S. oneidensis in-frame deletion strain showed increased sensitivity to the cationic antimicrobial peptide polymyxin as well as bile salts, suggesting a role in outer membrane integrity.

Introduction

Bacteria of the genus Shewanella are ubiquitous marine organisms known for their remarkable metabolic capabilities (1). They have attracted substantial interest for industrial applications, such as bioremediation of heavy metal contamination, generation of electrical power from biomass, and synthesis of ω-3 polyunsaturated fatty acids (1, 2). More recently, Shewanella species have also been identified as opportunistic pathogens, typically from exposure of broken skin to the marine environment (3–5).

Like nearly all Gram-negative bacteria, Shewanella possess both an inner and an outer membrane in which the outermost layer of the outer membrane is primarily composed of lipopolysaccharide (LPS). LPS contains a hydrophobic lipid anchor known as lipid A, an inner core oligosaccharide, and a repeating O-antigen domain that is highly variable between species (6, 7). The sugar 3-deoxy-d-manno-octulosonic acid (Kdo)³ is the first sugar added to lipid A, and laboratory strains of Escherichia coli cannot survive without synthesizing the minimal LPS substructure Kdo₂-lipid A (Fig. 1) unless compensatory mutations are introduced. The biosynthesis of lipid A is largely conserved across Gram-negative organisms. Many species, such as E. coli, add two Kdo sugars with a single Kdo transferase, whereas others, including Shewanella, add a single Kdo that is then phosphorylated by a separate enzyme (Fig. 2A). In strains of Shewanella, including the model organism Shewanella oneidensis, Kdo is further modified in which the C8 hydroxyl group is converted to a primary amine (8-amino-3,8-dideoxy-d-manno-octulosonic acid (Kdo8N); Fig. 2A) (8–10). Here, we report the identification of a three-gene cluster in S. oneidensis that is responsible for biosynthesis of Kdo8N. When heterologously expressed in E. coli, >75% of the extracted LPS contains Kdo8N. Expression of all three genes was required for incorporation of Kdo8N into E. coli lipid A, allowing for a biosynthetic pathway to be proposed in which Kdo is directly converted to Kdo8N followed by incorporation into lipid A. Purification of the enzymes and generation of an in vitro assay were also consistent with the proposed biosynthetic pathway. Furthermore, chromosomal deletion of the Kdo8N biosynthetic genes in S. oneidensis resulted in increased sensitivity to compounds known to perturb the outer membrane, such as polymyxin B and bile salts, suggesting that the presence of Kdo8N may increase the integrity of the Shewanella outer membrane.

FIGURE 1.

Truncated pathway for Kdo₂-lipid A biosynthesis in E. coli. Both inner core Kdo sugars are added to lipid IV_A by a single Kdo transferase enzyme (KdtA) using the sugar-nucleotide donor CMP-Kdo. Schematic illustrations represent acyl chains (black lines), glucosamine (black ovals), and Kdo (blue ovals).

FIGURE 2.

Kdo8N and its proposed biosynthetic cluster. A, chemical structure of the Kdo moiety in LPS from E. coli (Kdo₂) and S. oneidensis (Kdo8N 4-phosphate) where R represents the lipid A anchor. B, S. oneidensis gene cluster and potential functions in Kdo8N biosynthesis.

EXPERIMENTAL PROCEDURES

General Methods

Unless otherwise stated, all reagents were obtained from Sigma and were reagent grade or better. Primers were synthesized by Integrated DNA Technologies, Inc. (Coralville, IA). All primers, plasmids, and strains are listed in Table 1. Bacterial cultures were maintained in LB containing 10 g/liter tryptone, 5 g/liter yeast extract, and 10 g/liter NaCl supplemented with ampicillin (100 μg/ml), gentamicin (15 μg/ml), or diaminopimelate (100 μg/ml) as needed.

TABLE 1.

Primers, plasmids, and strains used in this study

Genomic Database Searching

The genome of S. oneidensis was searched using NCBI/BLAST with the known aminosugar biosynthetic enzyme GnnB (11) from Acidithiobacillus ferrooxidans as a query sequence. The neighboring genes on the chromosome were examined using the gene entry on the NCBI website.

Cloning and Molecular Biology

The SO_2476, SO_2477, and SO_2478 genes were amplified from S. oneidensis genomic DNA (ATCC). The genes of unknown function were renamed kdnA (SO_2476) and kdnB (SO_2477), whereas SO_2478 was presumed to be a CMP-Kdo synthase (KdsB; Fig. 1) on the basis of homology (45% identical/58% similar to E. coli KdsB). PCR was performed using the KOD polymerase (EMD Biosciences) and the primers listed in Table 1. DNA fragments were generated for all three genes (Primers 1 and 6), each individual gene (combinations of Primers 1 and 2, Primers 3 and 4, and Primers 5 and 6), and for two gene constructs (combinations of Primers 1 and 4 and Primers 3 and 6). Each DNA fragment contained a 5′ XbaI site and a 3′ XhoI site as well as a ribosome binding site derived from pET21b. DNA fragments were digested with the restriction enzymes XbaI and XhoI (New England Biolabs), purified by a DNA binding spin-column (Qiagen), and ligated into pWSK29 (12) that had been similarly digested and purified using T4 DNA ligase (New England Biolabs). Plasmids, including a pWSK29 vector control, were transformed in E. coli strain WBB06 using the transformation and storage solution method (13) and grown in LB supplemented with ampicillin and isopropyl β-d-thiogalactopyranoside. The resulting plasmids and strains are listed in Table 1.

Chromosomal Deletion of kdnA/kdnB in S. oneidensis

An in-frame deletion of SO_2476 and SO_2477 was created as described previously (14). Briefly, ∼500 bp of sequence directly upstream and downstream of the target genes (SO_2476/kdnA and SO_2477/kdnB) was amplified by PCR using the S. oneidensis genomic DNA and Primers 5O and 5I and Primers 3O and 3I, respectively (Table 1). The primers were designed with a complementary sequence to facilitate overlap extension, and the products were combined and amplified again by PCR using Primers 5O and 3O, creating ∼1 kb of sequence complementary to 500 bp upstream and downstream of the target genes and with a NotI restriction site on each end. The fragment was digested by NotI (New England Biolabs), purified as described above, and ligated into the plasmid pDS3.1 (15), which had been similarly digested. This plasmid contains a gentamicin resistance marker and a SacB sucrose sensitivity marker and utilizes an R6K origin of replication that can only replicate in pir⁺ strains. The ligated plasmid, named pΔ7677, was transformed into E. coli strain EC100D by electroporation, and colonies that grew on 15 μg/ml gentamicin were screened for the correct construct by colony PCR. Plasmid DNA was purified by a miniprep kit (Qiagen) and electroporated into the helper strain β2155 (16) in the presence of 100 μg/ml diaminopimelate and 15 μg/ml gentamicin. The pΔ7677-containing β2155 strain was then allowed to transfer the plasmid to S. oneidensis by conjugation as described (17). Cells were plated on LB without diaminopimelate containing gentamicin, and S. oneidensis cells were screened by PCR for chromosomal insertion. The plasmid was then resolved by selection on LB plates containing no NaCl and 10% sucrose. Deletion of the target genes was confirmed by PCR using Primers 5O and 3O and Primers 1 and 4.

Lipid Extraction

Strains of E. coli WBB06 (18) harboring pWSK29-derived plasmids are listed in Table 1. Each strain was grown to saturation overnight in 5 ml of LB supplemented with ampicillin. The cultures were then diluted 1:100 into 100 ml of LB supplemented with ampicillin and 1 mm isopropyl β-d-thiogalactopyranoside. Cultures were grown at 37 °C to an A₆₀₀ of 1.0 and then harvested by centrifugation. Cells were washed once with 5 ml of phosphate-buffered saline (PBS) and then frozen at −80 °C. For lipid extraction, cells were thawed and resuspended in 5 ml of PBS. The aqueous mixture was then converted to a one-phase Bligh-Dyer mixture (19) by addition of 12.5 ml of methanol and 6.25 ml of chloroform. The cells were incubated at room temperature for 1 h and then centrifuged at 2,000 × g for 30 min. The supernatant was then converted to an acidic two-phase Bligh-Dyer mixture by addition of 6.25 ml each of 0.1 n HCl and chloroform. The lower (chloroform) phase was removed, and the upper phase was washed once with 12.5 ml of chloroform followed by removal of the lower phase. The combined chloroform fractions were neutralized with pyridine (∼1 drop/10 ml), dried under a nitrogen stream, and stored at −20 °C.

Lipid Analysis by Thin-layer Chromatography (TLC) and Mass Spectrometry

Dried samples were dissolved in 1 ml of 4:1 CHCl₃:MeOH and dispersed by sonication in a water bath. For TLC, samples were spotted on a glass-backed 10 × 10-cm silica TLC plate and developed in 25:15:3.5:4 CHCl₃:MeOH:CH₃COOH:H₂O. Plates were sprayed with 10% sulfuric acid in ethanol and charred on a hot plate. Separately, samples for mass spectrometry were diluted ∼20-fold in 4:1 CHCl₃:MeOH supplemented with 1% piperidine. Samples were immediately infused at 5 μl/min and analyzed in the negative ion mode. Spectra were obtained on a QStar XL quadrupole time-of-flight mass spectrometer (ABI/MDS-Sciex, Toronto, Canada) in direct injection mode equipped with an ESI source and analyzed using Analyst QS v.1.1.

Purification and Activity Assays of the kdnA and kdnB Gene Products

His-tagged expression constructs were made as follows. Each gene was amplified from genomic DNA to incorporate a 5′ XbaI site and a 3′ XhoI site (KdnA) or a 5′ NdeI site and a 3′ BamHI site (KdnB). The PCR products were digested with the appropriate enzymes, purified, and ligated with T4 DNA ligase into pET21b (KdnA) or pET16b (KdnB) vectors, which had been similarly digested. Plasmids were confirmed by DNA sequencing (Eton Biosciences) and transformed into chemically competent E. coli strain C41 (20). For KdnA, 2 liters of LB were inoculated at 37 °C and grown to an A₆₀₀ of 0.6, then induced with 1 mm isopropyl β-d-thiogalactopyranoside, and grown overnight. KdnB was expressed similarly, but the temperature was reduced to 25 °C at induction. After harvesting the cell pellet by centrifugation, cells were resuspended in 0.1 m Tris, pH 8.0, 150 mm NaCl, 2 mm tris(2-carboxyethyl)phosphine. Cells were lysed by two passages through a French press, and then the cell debris was removed by centrifugation at 20,000 × g for 1 h. The cell-free extract was loaded onto a nickel-nitrilotriacetic acid column equilibrated in the same buffer, washed with buffer plus 10 mm imidazole, and eluted with buffer plus 250 mm imidazole. Pure fractions of either KdnA or KdnB were pooled, concentrated, and dialyzed to remove imidazole. Pure KdnA was diluted to 50 μm in 0.1 m Tris, pH 8.0, 150 mm NaCl, 2 mm tris(2-carboxyethyl)phosphine, 200 μm pyridoxal 5′-phosphate (PLP) and incubated on ice for 1 h to allow for incorporation of the cofactor. Excess PLP was removed by buffer exchange, and the enzyme was stored in aliquots at −80 °C. A sample of the purified KdnB enzyme was subjected to inductively coupled plasma mass spectrometry (MS) analysis (Keck Elemental Geochemistry Laboratory, University of Michigan). Pure apo-KdnB was prepared by dilution to ∼50 μm in 10 mm Tris, pH 8.0, 100 mm dipicolinic acid, 100 μm EDTA on ice for 1 h. The enzyme was then desalted over a PD-10 column equilibrated with 10 mm Tris, pH 8.0 and stored in aliquots at −80 °C.

Assays were conducted using ¹⁴C-labeled Kdo, which was synthesized from ¹⁴C-labeled pyruvate (American Radiolabeled Chemicals Inc., St. Louis, MO) as described previously (21). A typical assay mixture contained 0.1 m Tris, pH 8.0, 1 mm DTT, ∼5,700 cpm [¹⁴C]Kdo, 50 μm Kdo (unlabeled), 100 mm l-glutamic acid, and an appropriate amount of enzyme. KdnB was supplemented with MnSO₄ (1 eq), or 1 mm MnSO₄ was added to the assay as indicated in the appropriate figure legends. Reactions were quenched by spotting on glass-backed silica TLC plates and run in a tank containing 6:2:1 isopropanol:aqueous ammonia:water. TLC plates were exposed to a storage phosphor screen (GE Healthcare) for ∼24 h, then scanned by a PhosphorImager, and quantified using ImageQuant 5.0. Where appropriate, the above assays without [¹⁴C]Kdo were analyzed for hydrogen peroxide formation using the Amplex Red Hydrogen Peroxide/Peroxidase Activity kit (Invitrogen) according to the manufacturer's instructions.

Minimum Inhibitory Concentration (MIC) Assays of Wild-type and ΔkdnAkdnB S. oneidensis

MIC assays of S. oneidensis were conducted in 96-well plates as described previously (22). Briefly, 2-fold serial dilutions of compound (50–0.024 μg/ml polymyxin or 10–0.005 mg/ml bile salts) were inoculated with wild-type or ΔkdnAkdnB S. oneidensis to an A₆₀₀ of 0.05 in a total volume of 200 μl. Cells were grown overnight at 30 °C and then stained with 50 μl of 1 mg/ml 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide for 4 h. Living cells metabolize 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide to a purple formazan, which was quantified by visual inspection. Assays were performed in quadruplicate in 96-well plates and repeated in three independent experiments.

RESULTS

Identification of the Putative Kdo8N Biosynthetic Genes

BLAST searches of the S. oneidensis genome revealed a gene with 34% identity to the known aminosugar biosynthetic enzyme GnnB (11). This gene (SO_2476) was annotated as an aspartate aminotransferase. Adjacent to this gene was a putative alcohol dehydrogenase (SO_2477) followed directly downstream by the kdsB gene (SO_2478; Fig. 2B), which encodes the enzyme responsible for activating Kdo to the sugar-nucleotide donor CMP-Kdo (Fig. 1). The annotated enzymatic functions and location of SO_2476 and SO_2477 adjacent to kdsB strongly suggested that these genes could be responsible for Kdo8N biosynthesis. We renamed these genes kdnA (SO_2476) and kdnB (SO_2477).

Expression of kdnA, kdnB, and kdsB Results in Kdo8N Biosynthesis

To assess the function of the putative biosynthetic cluster, the lipid A products were analyzed in the background of WBB06 (18), an E. coli strain that synthesizes Kdo₂-lipid A as its sole LPS species (Table 1). Strain SG1, which contains all three putative biosynthetic genes, was grown, and the lipids were extracted as described under “Experimental Procedures.” Although the vector control strain produced the expected single band of Kdo₂-lipid A (Fig. 3, lane A), TLC analysis of the lipids from the strain expressing all three genes showed a mixture of three lipid A-like species (Fig. 3, lane B). These bands consisted of Kdo₂-lipid A and two more polar (slower migrating) species, which were presumed to indicate incorporation of one or two Kdo8N moieties in place of Kdo.

FIGURE 3.

TLC analysis of lipid A from E. coli expressing potential Kdo8N biosynthetic genes. Lane A, WBB06 expressing empty pWSK29 control; lane B, WBB06 expressing kdnA, kdnB, and kdsB. Lipid A samples were extracted, run on silica TLC plates in 25:15:3.5:4 CHCl₃:MeOH:CH₃COOH:H₂O, and visualized by charring on a hot plate as described under “Experimental Procedures.” Proposed products are illustrated as acyl chains (black lines), glucosamine (black ovals), Kdo (blue ovals), and Kdo8N (red ovals).

Kdo₂-lipid A has an exact mass of 2235.32 atomic mass units and appears by ESI-MS as a doubly charged species with an m/z of 1117.66. Replacement of Kdo with Kdo8N results in a 0.985-atomic mass unit decrease; therefore, (Kdo8N)₂-lipid A has an expected mass of 2233.35 and will appear as a doubly charged species ([M − 2H⁺]²⁻) with an m/z of 1116.68, whereas a mixed Kdo/Kdo8N species would be 0.49 atomic mass unit larger. The lipid A from the strain expressing the entire gene cluster shows peaks at 1117.477 and 1116.978 consistent with incorporation of one or two Kdo8N moieties in place of Kdo, respectively (Fig. 4A). Furthermore, ESI-MS/MS fragmentation revealed no mass difference in the lipid A fragment, indicating that there is no modification to the lipid A anchor (Fig. 4, B and C). In contrast, there is a 0.434-atomic mass unit difference in the [M − 2H⁺]²⁻ m/z for the fragment ion corresponding to loss of one Kdo (1007.907 versus 1007.473), indicating that the modification is in the Kdo moiety (Fig. 4, B and C). This is further confirmed by a 0.986-atomic mass unit decrease in the Kdo fragment, corresponding to a single amine substituent, and a 1.969-atomic mass unit difference in the Kdo₂-disaccharide moiety, corresponding to two amines (Fig. 4, B and C). The cumulative evidence indicates that the genes identified are sufficient for biosynthesis of Kdo8N.

FIGURE 4.

ESI-MS of lipid A species from WBB06 expressing Kdo8N biosynthetic enzymes. Isolated lipid A samples were prepared and injected into a QStar XL mass spectrometer as described under “Experimental Procedures.” A, overlay of spectra obtained for extracted lipids from the vector control strain (blue) and all three genes, kdnA, kdnB, and kdsB (red), expanded to show Kdo₂-lipid A species. B, MS/MS of Kdo₂-lipid A from vector control strain. C, MS/MS of (Kdo8N)₂-lipid A from the strain expressing kdnA, kdnB, and kdsB. Proposed structures are shown adjacent to relevant peaks.

The Entire Gene Cluster Is Required for Kdo8N Biosynthesis

Strains containing each of the individual genes from the Kdo8N biosynthetic cluster in a pWSK29 plasmid in WBB06 were constructed (Table 1). When these strains were grown and the lipids were extracted as described above, no Kdo8N was observed by TLC (Fig. 5) or ESI-MS analysis (data not shown). Furthermore, when the putative alcohol dehydrogenase and aminotransferase genes were expressed with the CMP-Kdo synthetase gene kdsB omitted, still no Kdo8N was observed (Fig. 5). Because E. coli possesses a kdsB gene, this result suggests that the S. oneidensis kdsB gene product has a functional difference from its E. coli homologue that allows for Kdo8N biosynthesis.

FIGURE 5.

TLC analysis of lipid A from E. coli expressing individual Kdo8N biosynthetic genes. Lipid A species were extracted, run on silica TLC plates in 25:15:3.5:4 CHCl₃:MeOH:CH₃COOH:H₂O, and visualized by charring on a hot plate as described under “Experimental Procedures.” Samples were obtained from WBB06 strains expressing an empty pWSK29 vector control (lane A), kdnA (lane B), kdnB (lane C), kdsB (lane D), kdnA and kdnB (lane E), and kdnA, kdnB, and kdsB (lane F). Proposed products are illustrated as acyl chains (black lines), glucosamine (black ovals), Kdo (blue ovals), and Kdo8N (red ovals).

Recombinant Purified KdnA and KdnB Can Synthesize Kdo8N in Vitro

Bioinformatics predictions suggested NAD⁺ or NADP⁺ and an amino acid such as aspartate or glutamate as co-substrates for KdnB and KdnA, respectively. Preliminary experiments using extracts suggested that l-glutamate yielded the highest activity (supplemental Fig. S1). Interestingly, product formation occurred in extracts in the absence of added pyridine nucleotide and was not stimulated by additional NAD⁺, suggesting that the classification of KdnB as an NAD-dependent alcohol dehydrogenase was incorrect. Both KdnA and KdnB were purified to homogeneity (Fig. 6) and assayed in vitro as described under “Experimental Procedures.” Incubation of both purified enzymes with Kdo and l-Glu resulted in Kdo8N formation, confirming that Kdo is a direct substrate for the reaction (Fig. 7, lane 8). KdnB alone, which is predicted to be the first step in the reaction, did not lead to formation of a detectable intermediate (Fig. 7, lane 4), suggesting two possibilities. (a) The reaction catalyzed by KdnB is thermodynamically unfavorable and requires the second reaction catalyzed by KdnA to drive product formation, or (b) the two enzymes are active only in a complex. Because many alcohol dehydrogenases favor reduction of the aldehyde, we believe the first possibility is most likely. Inductively coupled plasma MS analysis showed that KdnB contained 0.51 ± 0.03 mol of iron/mol of enzyme as well as 0.24 ± 0.01 eq of zinc. Metal activation experiments showed highest activity with Mn²⁺ and Fe²⁺ (supplemental Fig. S2) and inhibition by EDTA (Fig. 7, lanes 3 and 7). Because Fe²⁺ is unstable in the presence of oxygen, we chose to complete our initial studies with apo-KdnB supplemented with MnSO₄. As purified, KdnA did not contain PLP and had minimal activity in the in vitro assay (not shown) but did exhibit robust activity after incubation with PLP (Fig. 7), suggesting that KdnA requires PLP for activity. When both enzymes were present, no product formation was observed if l-Glu was withheld or EDTA was included (Fig. 7, lanes 6 and 7) consistent with the proposal that KdnB is a metal-dependent enzyme, KdnA is a PLP-dependent aminotransferase that utilizes l-glutamate, and both enzymes are required for product formation.

FIGURE 6.

SDS-PAGE of KdnA and KdnB purifications. Lane L, protein molecular weight marker. Approximately 1.5 μg of KdnA and KdnB as a cell-free extract (CFE) or purified protein was loaded as indicated and run on a 4–20% Tris-HCl gel.

FIGURE 7.

In vitro activity of KdnA and KdnB. A, TLC-based assay of cell extracts and purified KdnB and KdnA. Assays were conducted using 0.1 m Tris, pH 8.0, 1 mm DTT, 50 μm Kdo, 5,700 cpm [¹⁴C]Kdo, and 1 mm MnSO₄ in the presence or absence of the following components as indicated: 15 μg of cell extract, 30 μm apo-KdnB, 30 μm PLP-KdnA, 0.1 m l-glutamate, and 2 mm EDTA in 10 μl. After 60 min, 2.5 μl of the reaction mixture was spotted and run on a TLC plate in 6:2:1 isopropanol:aqueous ammonia:water as described under “Experimental Procedures.” B, hydrogen peroxide formation catalyzed by KdnA/KdnB. H₂O₂ formation was monitored using the Amplex Red Hydrogen Peroxide kit (Invitrogen). Assays were performed in 100 μl with 0.1 m Tris, pH 8.0, 50 μm Kdo, 50 μm Amplex Red, 0.1 unit/ml horseradish peroxidase, 0.1 m l-glutamate, and one of the following: no enzyme control (open circles); no enzyme + 5 mm NAD (open squares), 10 μm apo-KdnB/PLP-KdnA (open diamonds), 10 μm Mn-KdnB/PLP-KdnA (closed triangles), 10 μm Mn-KdnB/PLP-KdnA + 5 mm NAD (closed squares), or 10 μm Mn-KdnB/PLP-KdnA + 2 mm EDTA (open triangles).

Perhaps the most surprising result was that NAD(P) was not required for Kdo8N formation. We hypothesized that the most reasonable electron acceptor was molecular oxygen. Indeed, we observed production of H₂O₂ when Mn-KdnB and PLP-KdnA were incubated with Kdo and l-Glu, whereas no H₂O₂ was observed when the reaction was performed with apo-KdnB or when EDTA was included (Fig. 7B). Furthermore, inclusion of NAD⁺ only slightly inhibited H₂O₂ formation, suggesting that O₂ is the preferred oxidant. In addition, no product inhibition was observed for NADH up to 30 mm, further suggesting that it is not a product of the reaction (supplemental Fig. S3). Therefore, KdnB appears to be an alcohol oxidase as opposed to an alcohol dehydrogenase. To our knowledge, no metal-dependent alcohol oxidases have been reported previously, suggesting that this enzyme could represent a novel class of alcohol oxidases. Further mechanistic studies are required to fully address this question.

Loss of Kdo8N Increases Susceptibility to Polymyxin and Bile Salts

The MIC values for wild-type and ΔkdnAkdnB S. oneidensis are shown in Table 2. The knock-out strain showed increased sensitivity to polymyxin B (∼3-fold) and bile salts (∼2-fold). Both of these compounds affect outer membrane permeability, suggesting that Kdo8N contributes modestly to outer membrane integrity.

TABLE 2.

MIC values for wild-type and ΔkdnAkdnB S. oneidensis

S. oneidensis strain	MICa
	Polymyxin	Bile salts
	μg/ml	mg/ml
Wild type (MR-1)	0.16 ± 0.05	1.25 ± 0
ΔkdnAkdnB	0.05 ± 0.01	0.63 ± 0

^a MIC values were determined in quadruplicate and repeated for three independent experiments. Values are expressed as the mean and S.D. of the minimum concentration that inhibited growth as described under “Experimental Procedures.”

DISCUSSION

Identification of the Kdo8N Biosynthetic Genes

Kdo8N was first observed by direct analysis of the LPS from S. oneidensis (10). This modification to Kdo had not previously been observed, and the genes responsible for its formation were completely unknown. The chemical conversion of Kdo to Kdo8N consists of replacing the C8 hydroxyl of Kdo with a primary amine. This chemistry has been observed frequently in deoxysugar modification pathways (23). One such pathway is the conversion of UDP-N-acetylglucosamine (UDP-GlcNAc) (Fig. 1) to the 3-amino derivative UDP-GlcNAc3N by the enzymes GnnA/GnnB (11) that is the origin of lipid A species containing only N-linked acyl chains in certain Gram-negative bacteria (24). This reaction is accomplished by oxidation of the 3-OH of UDP-GlcNAc to a ketone by an NAD-dependent alcohol dehydrogenase followed by transamination by a PLP-dependent glutamate aminotransferase. We reasoned that a similar pathway might be responsible for Kdo8N biosynthesis and searched the S. oneidensis genome for homologues of GnnA and GnnB. Although no significant homologues for GnnA were found, a homologue to GnnB with 34% identity was found in a gene cluster adjacent to a putative alcohol dehydrogenase and a homologue of the kdsB gene (SO_2478), the last step in CMP-Kdo biosynthesis (Fig. 1). The putative alcohol dehydrogenase showed very little similarity to GnnA, suggesting a function distinct from the GnnA/GnnB reaction. The two enzymes upstream of kdsB were renamed kdnA (SO_2476) and kdnB (SO_2477).

Expression of all three genes (kdnA, kdnB, and kdsB) was carried out in E. coli model strain WBB06, which harbors deletions in WaaC and WaaF, the glycosyltransferases necessary for initial extension of the core sugars in LPS (Table 1). As a result, the final lipid A species obtained from this strain is Kdo₂-lipid A, which partitions to the chloroform phase of a Bligh-Dyer extraction. WBB06 expressing the entire cluster was analyzed by TLC and mass spectrometry, results of which were consistent with Kdo8N formation (Figs. 3 and 4). Taken together, this evidence strongly indicates that kdnA, kdnB, and kdsB are sufficient for Kdo8N biosynthesis in S. oneidensis.

Proposed Pathway for Kdo8N

In Gram-negative bacteria, Kdo is synthesized from phosphoenolpyruvate and arabinose 5-phosphate, dephosphorylated, and then activated by conversion to a sugar nucleotide, CMP-Kdo (Fig. 1). This activated species is then transferred to a tetra-acylated lipid A precursor, known as lipid IV_A, and processed to Kdo₂-lipid A by two acyl transferases, LpxL and LpxM (6). In many species, including E. coli, a single enzyme transfers two Kdo moieties to lipid IV_A (referred to as a bifunctional KdtA), whereas in others, including S. oneidensis and Haemophilus influenzae, only a single Kdo is transferred (monofunctional KdtA) followed by phosphorylation at C4 by an additional Kdo kinase enzyme (Fig. 2A). Because Kdo exists in several forms in E. coli (free sugar, sugar nucleotide, Kdo₂-lipid IV_A, and Kdo₂-lipid A), several possible pathways exist for producing (Kdo8N)₂-lipid A. One such pathway is direct modification of Kdo₂-lipid A to (Kdo8N)₂-lipid A (or the Kdo₂-lipid IV_A precursor). Given the differences in activities between the E. coli and S. oneidensis KdtA enzymes (monofunctional versus bifunctional), this pathway, which bypasses the Kdo transferase step, might have been expected. However, when any of the genes was expressed individually, no modification was observed, suggesting that direct modification of Kdo₂-lipid A does not occur. If the lipid A was directly modified, but the oxidized intermediate was unstable, co-expression of kdnA and kdnB should be sufficient to produce (Kdo8N)₂-lipid A. However, expression of kdnA and kdnB in the absence of the S. oneidensis kdsB gene resulted in no observed modification. Therefore, it is most likely that the amine-modified species is formed prior to glycosylation of lipid IV_A, implying that the E. coli Kdo transferase can use CMP-Kdo8N as a substrate.

Formation of the sugar nucleotide CMP-Kdo is catalyzed by the enzyme CMP-Kdo synthase (CKS; gene product of kdsB). All of the WBB06-derived strains expressing S. oneidensis proteins contained a chromosomal copy of the E. coli kdsB. Therefore, if S. oneidensis CKS (SoCKS) was functionally identical to E. coli CKS, there should be no difference between the strain expressing all three genes and the strain expressing kdnA and kdnB but omitting S. oneidensis kdsB. Because all three genes were required to observe Kdo8N in the extracted lipid A, this suggests that SoCKS has a function distinct from the E. coli enzyme. This can be explained by the pathway proposed in Fig. 8 wherein Kdo is converted to Kdo8N by KdnA and KdnB presumably through an aldehyde intermediate followed by activation to CMP-Kdo8N by a CKS with expanded substrate specificity recognizing Kdo8N. This proposed biosynthetic pathway explains the failure to observe intermediates when expressing any of the single genes and explains the requirement of SoCKS to observe Kdo8N in LPS. This pathway is also consistent with the results of in vitro assays of purified KdnA and KdnB where we observed Kdo8N formation directly from Kdo without an enzyme to make CMP-Kdo. Indeed, when CTP and Mg²⁺ were included in the presence of SoCKS to allow formation of CMP-Kdo, product formation was inhibited, suggesting that CMP-Kdo is a poor substrate (supplemental Fig. S4). Furthermore, the observed activity in the absence of NAD⁺ suggests that molecular oxygen could function as the electron acceptor in the reaction, which was confirmed by detection of H₂O₂ production. Interestingly, NAD⁺ did not inhibit H₂O₂ production, suggesting that it is not a better substrate than O₂. This in turn suggests that either (a) molecular oxygen is the physiological substrate for KdnB or (b) another redox cofactor is used in S. oneidensis. Preliminary experiments showed no rate enhancement with E. coli cytochrome c or ubiquinone (not shown), but it is still possible that a specific S. oneidensis redox cofactor is used in vivo.

FIGURE 8.

Proposed pathway for Kdo8N biosynthesis and incorporation into lipid A.

In addition to S. oneidensis, Kdo8N has been observed in all species of Shewanella studied to date, and significant homologues of kdnA and kdnB exist in all those with sequenced genomes. No other species to date has been found to contain Kdo8N, but several organisms that have not been subjected to detailed analysis of their LPS contain a homologous cluster with at least 50% identity, including Thiomicrospira crunogena XCL-2, Thioalkalimicrobium cyclicum ALM1, and Chitinophaga pinensis DSM 2588. T. crunogena and T. cyclicum are obligate sulfur-oxidizing bacteria (25, 26), whereas C. pinensis is notable for its ability to degrade the polysaccharide chitin (27). All three are marine organisms, and the incorporation of Kdo8N may be helpful in some aspect of these organisms' ecological niches. Lipid A modifications that incorporate an amino substituent have previously been shown to confer resistance to cationic antimicrobial peptides (28) presumably due to unfavorable interactions with a positively charged amine. Because deletion of kdnA/kdnB in S. oneidensis led to ∼3-fold increased sensitivity to polymyxin and ∼2-fold increased sensitivity to bile salts (Table 2), we speculate that Kdo8N may contribute modestly to outer membrane integrity.

Kdo is an 8-carbon aldulosonic acid. C₉ aldulosonic acids also exist, including the neuraminic acids, which are also known as sialic acids. Naturally occurring neuraminic acids contain one or more acetamido groups, which are derived biosynthetically from UDP-GlcNAc. Like Kdo, sialic acids must be activated as CMP derivatives prior to glycosylation. In Leptospira interrogans, we observed a gene cluster similar to the kdnA/kdnB genes (34 and 48% identity, respectively) but adjacent to a CMP-neuraminic acid synthase instead of a CMP-Kdo synthase. This raises the possibility that additional amine modifications could be present in LPS or glycosylated proteins at residues other than Kdo, perhaps initiated by a specific class of alcohol oxidases that recognize aldulosonic acids. The detailed carbohydrate analysis of Leptospira LPS required to answer this question has yet to be performed.

The physiological role of Kdo8N is not clear. Besides the role in stabilizing the membrane, it is also possible that Kdo8N may alter recognition of lipid A by the MD2-TLR4 receptor complex, which senses lipid A as part of the mammalian innate immune system. Previously characterized lipid A modifications, such as dephosphorylation of the 1-phosphate in Francisella novicida, have been shown to attenuate virulence (29). Altered TLR4 activation by modified lipid A species has been shown previously to be useful for the development of vaccine strains as well as vaccine adjuvants.

Interestingly, the function of KdnB diverges from its predicted activity. Oxidation of an alcohol to an aldehyde can occur by a variety of mechanisms, all of which involve transfer of electrons from the alcohol to an acceptor (for a review of alcohol dehydrogenase/oxidase classifications, see Ref. 30). The metal-dependent alcohol dehydrogenases typically use NAD(P)⁺ as the electron acceptor. Metal-independent alcohol dehydrogenases use pyrroloquinoline quinone as a cofactor to transfer electrons to another biological electron acceptor, such as ubiquinone. Probing the sequence of KdnB in the Conserved Domain Database (31–33) suggests that KdnB is an iron-containing alcohol dehydrogenase, class 3 (Fe-ADH3), within the dehydroquinate synthase-like and iron-containing alcohol dehydrogenase (DHQ_Fe-ADH) superfamily. Proteins of this specific family have not been well characterized, but proteins within the DHQ_Fe-ADH superfamily typically use NAD or NADP. KdnB is clearly metal-dependent but does not require any redox cofactor or pyridine nucleotide and instead utilizes molecular oxygen. An enzyme that can oxidize an alcohol directly with O₂ is an alcohol oxidase, but all known examples contain a flavin cofactor (FAD or FMN). KdnB shows no similarity to these flavin-dependent oxidases. This hybrid activity wherein we observed metal-dependent alcohol oxidase activity has never been observed to our knowledge, and more detailed mechanistic studies are warranted to probe a potentially novel enzyme subfamily. Because members of the Fe-ADH3 family have not been well characterized, it is difficult to say whether KdnB should be placed in a new family or whether the Fe-ADH3 family contains a mixture of dehydrogenases and oxidases. This biosynthetic pathway is also unusual because it involves the direct modification of a free sugar rather than the sugar-nucleotide, which is generally observed (23). By uncovering the biosynthetic pathway for Kdo8N, we have set the stage for illuminating the physiological role and potential applications of this unstudied lipid A modification and present evidence for a novel class of alcohol oxidases.