Sinorhizobium meliloti acpXL mutant lacks the C28 hydroxylated fatty acid moiety of lipid A and does not express a slow migrating form of lipopolysaccharide.

Lipid A is the hydrophobic anchor of lipopolysaccharide (LPS) in the outer membrane of Gram-negative bacteria. Lipid A of all Rhizobiaceae is acylated with a long fatty acid chain, 27-hydroxyoctacosanoic acid. Biosynthesis of this long acyl substitution requires a special acyl carrier protein, AcpXL, which serves as a donor of C28 (omega-1)-hydroxylated fatty acid for acylation of rhizobial lipid A (Brozek, K.A., Carlson, R.W., and Raetz, C. R. (1996) J. Biol. Chem. 271, 32126-32136). To determine the biological function of the C28 acylation of lipid A, we constructed an acpXL mutant of Sinorhizobium meliloti strain 1021. Gas-liquid chromatography and mass spectrometry analysis of the fatty acid composition showed that the acpXL mutation indeed blocked C28 acylation of lipid A. SDS-PAGE analysis of acpXL mutant LPS revealed only a fast migrating band, rough LPS, whereas the parental strain 1021 manifested both rough and smooth LPS. Regardless of this, the LPS of parental and mutant strains had a similar sugar composition and exposed the same antigenic epitopes, implying that different electrophoretic profiles might account for different aggregation properties of LPS molecules with and without a long acyl chain. The acpXL mutant of strain 1021 displayed sensitivity to deoxycholate, delayed nodulation of Medicago sativa, and a reduced competitive ability. However, nodules elicited by this mutant on roots of M. sativa and Medicago truncatula had a normal morphology and fixed nitrogen. Thus, the C28 fatty acid moiety of lipid A is not crucial, but it is beneficial for establishing an effective symbiosis with host plants. acpXL lies upstream from a cluster of five genes, including msbB (lpxXL), which might be also involved in biosynthesis and transfer of the C28 fatty acid to the lipid A precursor.

Soil bacteria of the family Rhizobiaceae, known as rhizobia, form a nitrogen-fixing symbiosis with leguminous plants. The rhizobia-legume symbiosis is a highly specific and complex developmental process in which both partners undergo differentiation in a concerted way. Bacteria induce the formation of nodules on roots of their host plants and colonize the root tissue as intracellular nitrogen-fixing bacteroids (1). Rhizobial cell surface polysaccharides play an important role during this process (2). LPS 1 constitutes an integral part of the outer cell membrane of Gram-negative bacteria. It is therefore intimately involved in the formation of the plant-bacterial interface during symbiosis (3).
Among Rhizobiaceae, a complete LPS structure has only been resolved for Rhizobium etli CE3 and Rhizobium leguminosarum (4 -8). Actually, these related bacterial species have identical lipid A core structures and different O antigens (5). The LPS of Sinorhizobium meliloti has been characterized only partially. Its chemical composition has been analyzed many times, but its structure is still unknown (9 -13). The LPS of S. meliloti strains 1021 and 102F51 reveals a high content of acidic sugars, namely galacturonic acid, glucuronic acid, and Kdo (3-deoxy-D-manno-octulosonic acid) (11,13). In addition, the LPS of S. meliloti strain 102F51 contains DHA (3-deoxy-2heptulosaric acid), considered a component of the O chain (11,13). However, DHA could be also a component of the outer core.
The LPS of Escherichia coli and other Enterobacteriaceae is also known as endotoxin because of its diverse pathophysiological effects displayed during the infection of animal hosts, such as cytokine production, inflammation, and shock (14,15). Lipid A is primarily responsible for the endotoxic activity of LPS. Modifications of lipid A may significantly change its toxicity. The lipid A of E. coli and related Enterobacteriaceae is composed of a backbone of two glucosamines replaced by two phosphate groups at positions 1 and 4Ј, and four residues of R-3hydroxymyristic acid (C14) in ester and amide linkage at positions 2, 2Ј, 3, and 3Ј (Fig. 1). The hydroxymyristic residues at positions 2Ј and 3Ј are further replaced by laurate (C12) and myristate (C14) forming acyloxyacyl moieties (14).
Lipid A of rhizobia displays a number of remarkable differences from enteric lipid A. Regardless of significant variations of lipid A structures within the family of Rhizobiaceae there is one common feature that distinguishes their lipid A from that of Enterobacteriaceae: the presence of a long fatty acid chain, 27-hydroxyoctacosanoic acid (27-OH-C28:O) (4). It was detected in various members of this family, including species of Rhizobium, Sinorhizobium, Mesorhizobium, Bradyrhizobium, and Agrobacterium (4,54,55). Fig. 1 shows one of six lipid A components identified in R. etli. All six components share the following structural features: the absence of phosphate groups and the presence of a galacturonic residue at the 4Ј position and of a long acyloxyacyl moiety as a secondary acyl substitution of the N-linked hydroxyacyl chain (7,8). Variation of R. etli lipid A species concerns mainly the proximal glucosamine residue and its acyl substitutions (7,8).
Recently the structure of Sinorhizobium sp. NGR234 lipid A was established (56). It appeared to be a remarkable "hybrid" sharing features with enteric lipid A and lipid A of R. etli. A similarity to enteric lipid A consisted in the presence of two phosphate substitutions of glucosamine disaccharide. A similarity to lipid A of R. etli was observed in the type and location FIG. 1. Structures of lipid A from E. coli, R. etli, and S. meliloti. Lipid A of E. coli consists of a glucosamine backbone, two phosphate substitutions, and six acyl substitutions, two of which form acyloxyacyl moieties. Lipid A of the msbB mutant of E. coli lacks an acyloxyacyl residue containing myristic acid. Lipid A of R. etli lacks both phosphate groups but carries instead a single galacturonic residue. The characteristic feature of R. etli lipid A is a single amide-linked acyloxyacyl residue containing a long fatty acid chain (27-OH-28:0). Lipid A of S. meliloti consists of a glucosamine backbone with two phosphate substitutions and four acyl substitutions. No acyloxyacyl residue was detected (10). Based on the presence of the acpXL-msbB (lpxXL) gene cluster in the S. meliloti genome, we propose that lipid A of this bacterium should carry a long acyl residue in the same position as lipid A of R. etli and Rhizobium sp. NGR234. Secondary acyl substitutions (acyloxyacyl residues) are indicated by broad lines.
of an acyloxyacyl residue and in the heterogeneity of lipid A species concerning the presence or absence of an acyl substitution at position 3 (56).
The structure of S. meliloti lipid was characterized 14 years ago and probably needs confirmation with modern methods of analysis. Similar to the lipid A of enteric bacteria, lipid A of S. meliloti had a glucosamine backbone with two phosphate substitutions (10) (Fig. 1). No secondary acyl substitutions were found, and a C28 hydroxylated fatty acid chain was identified as a primary ester-linked substitution of glucosamine (10).
Interestingly, S. meliloti and R. leguminosarum lipid A were reported to have the same endotoxic activity on animal hosts as lipid A of enteric bacteria (10,16). The role of bacterial LPS in symbiotic interactions with plants seems to be profoundly different from its role in pathogenic interactions with animals. There are several lines of evidence that LPS of rhizobia may suppress host defense responses in plant cells and promote symbiotic interaction (2,17,18). Studies with rhizobial mutants producing modified lipid A may shed more light on mechanisms of LPS interaction with plant cells. To our knowledge, all LPS mutants of rhizobia studied so far contained modifications either in the core or in the O antigen, and rhizobial mutants affected in lipid A expression have not been analyzed yet.
Recently, the complete genome sequence of S. meliloti 1021 was determined (19). Sequence annotation of the S. meliloti chromosome revealed a number of genes whose products are conserved among Gram-negative bacteria and control early steps in lipid A biosynthesis, namely lpxABCDK, fabZ, and kdtA (3,14,20). In addition to the conserved lipid A biosynthesis genes, a gene cluster containing the acpXL gene was found. This cluster may be responsible for the formation of the C28 acyl substitution of lipid A. The acpXL gene was first identified because of the partial sequencing of a protein purified from R. leguminosarum. This protein mediated the incorporation of a C28 fatty acid chain into a lipid A precursor in vitro (21). Up to now, the role of the acpXL gene in vivo was not investigated. To provide experimental evidence for the predicted function of the acpXL gene of S. meliloti, we constructed an acpXL mutant in strain 1021. We demonstrate here that the acpXL mutation results in a structural modification of LPS and affects the properties of free living and endosymbiotic bacteria. Based on the results of LPS composition analysis as well as results of electrophoretic and immunological comparisons of mutant and wild type LPS, we reconsider here the structure of S. meliloti LPS.

EXPERIMENTAL PROCEDURES
Bacterial Strains and Growth Conditions-Strains and plasmids used in this work are described in Table I. S. meliloti strains were incubated at 30°C either on LB or TY or mannitol-glutamate-salts medium (28 -30). E. coli strains were incubated at 37°C on LB medium. Antibiotics were added at the following concentrations (g⅐ml Ϫ1 ): streptomycin (Sm) 600, neomycin (Nm) 120, tetracycline (Tc) 8 for S. meliloti; ampicillin (Ap) 100, kanamycin (Km) 50, tetracycline 5 for E. coli. The deoxycholate (DOC) sensitivity of S. meliloti strains was tested on LB plates containing 0.1% sodium deoxycholate.
Construction of S. meliloti acpXL Mutant-A 188-bp intragenic fragment of the 285-bp coding sequence of acpXL was amplified using Taq polymerase and the primer pair acpXLup (5Ј-TCGACAAGGTTGC-CGATATT) and acpXLdn (5Ј-GCCTTCGTTGACTTCCTGAG). The PCR product was first inserted into vector pGEM-T-Easy, then recovered using the endonucleases SphI and SalI and ligated into the mobilizable vector pK18mob2. The resulting plasmid pLSacpXL was transferred to S. meliloti recipient strains by S17-1-mediated conjugation. Transconjugants were selected on LB plates containing neomycin and streptomycin. To verify that the genomic acpXL sequence was disrupted by the plasmid, the transconjugants were analyzed by PCR using the primer pair acpXLupK (5Ј-GCGTGACAGCTACATTCGAC) and acpXLdnK (5Ј-CGATCTTGGCACAAAGGTTC). This resulted in amplification of a DNA fragment including the duplicated target DNA and the integrated vector.
Complementation of Mutant L994 -To clone the acpXL gene, a 448-bp genomic fragment including the entire open reading frame and the flanking sequences, 144 bp upstream from the start codon and 5 bp downstream from the stop codon, were amplified by PCR using primers acp11Xup (5Ј-AATCTAGATCACGAAGTAACCCCGCT) and acp11Kdn (5Ј-AAЈGGTACCGCTTCAGGCTTTCCTCAA). The underlined nucleotides represent XbaI and KpnI restriction sites, respectively, added for convenience of cloning. The amplified DNA product was cloned into the broad host range vector pPHU231 (27) in the orientation opposite to the lacZ promoter, to allow expression of acpXL from its own promoter presumably located in the 5Ј-upstream DNA sequence. The resulting plasmid, designated pXK11a, was used for complementation. Conjugational transfer of pXK11a into S. meliloti acpXL mutant L994 was accomplished as described above. Transconjugants were selected on LB plates containing tetracycline and streptomycin.
Purification of LPS-Preparative purification of LPS was performed from S. meliloti strains 1021 and L994. S. meliloti strains were grown on TY plates at 30°C for 2 days. Bacterial cells were harvested from plates and washed twice with 0.9% NaCl. The cell pellet was resuspended in an equal volume of distilled water, and LPS was extracted after a hot phenol-water extraction procedure (31). The material from the water phase was treated with DNase, RNase, and proteinase K, dialyzed against water for 2 days, and freeze dried. Then LPS was dissolved in water to a final concentration of about 40 mg/ml and centrifuged for 20 h at 100,000 ϫ g. The pellet consisting of LPS micelles was dissolved in water and freeze dried.
Fatty Acid Analysis-The determination of fatty acids and hydroxy fatty acids was done by gas-liquid chromatography (GC) and mass spectrometry (MS) as described (32) with modifications. Briefly, 200 g of purified LPS was supplemented with 10 g of 17:0-Me as internal standard. The samples were dissolved in 1 ml of 4 M HCl and heated for 4 h at 100°C in a tightly closed Reacti-Vial (Pierce). 1 ml of 5 M NaOH was added, and the mixture was incubated at 100°C for a further 30 min. After the addition of 3 ml of water, the pH was adjusted to 3 using ϳ300 l of HCl. The fatty acids were extracted with 1 ml of chloroform. The extraction was repeated three times, and the extracts were pooled and dried in a steam of nitrogen. Methylation of the carboxyl groups was carried out using diazomethane in ether (0.1 ml), freshly prepared using 1-methyl-3-nitro-1-nitroguanidin (Aldrich), and the MNNG diazomethane generator (Aldrich). The ether was evaporated by a steam of nitrogen, and fatty acid methyl esters carrying free hydroxy groups were replaced by 40 l of BSTFA (Pierce) in 4 h at 65°C to produce the particular methyltrimethylsilyl esters. As a further reference for GC-MS, octacosanoic acid (Sigma, O 4004) was subjected to the same derivatization procedure. GC-MS was carried out on a Thermo Trace gas chromatographer equipped with a 30-m SPB-50 (Supelco) and a Polaris electron impact ion trap mass spectrometer (Thermo). The following temperature program was used: 80°C for 3 min, then 5°C/min up to 300°C and holding for 10 min. MS spectra were compared with reference substances and the NIST data base. Glycosyl Composition Analysis-Analysis of neutral sugars and uronic acids was performed as described previously (33).
Analysis of Crude LPS Preparations and Immunodetection of LPS-LPS from S. meliloti strains was extracted using triethylamine/EDTA or SDS/proteinase K methods (34,35). LPS samples were fractionated by SDS-PAGE. After electrophoresis in 16.5% acrylamide gels, LPS was visualized by the silver staining method of Tsai and Frash (36). LPS was electroblotted from SDS-PAGE to an Immobilon-P (polyvinylidene difluoride) membrane (Millipore), pore size 0.45 m, and detected using polyclonal antiserum as described (37).
Generation of Polyclonal Antiserum-S. meliloti strain 2011 (23) was grown in mannitol-glutamate-salts medium overnight, washed twice with phosphate-buffered saline, pH 7.2, and incubated for 5 min at 65°C. A final titer of 10 10 cells was used for immunization of rabbits. The antiserum was produced following the 2-month standard immunization protocol (www.seqlab.de/customer_services.html) by SEQLAB, Sequence Laboratories Göttingen GmbH (Germany). The resulting antiserum 2605 showed high reactivity to surface antigens of numerous S. meliloti strains, including 1021.
Adsorption of the Antiserum with Whole Cells of S. meliloti-Coincubation of the antiserum with whole cells of a specific strain results in binding of antibodies to epitopes of cell surface polysaccharides. If the antiserum is incubated with mutant cells lacking a particular epitope, antibodies recognizing this epitope will remain unbound and may be detected by an immunoblot with material of a wild type strain. Preadsorption of the antiserum was performed basically as described previously (37) with modifications. Because adsorption of antibodies after incubation with strains 2011 and 1021 was incomplete, bacterial cells were boiled for 3 min in phosphate-buffered saline, cooled down, and mixed with the antiserum. Adsorption was performed for 2 h at room temperature with shaking. This resulted in a complete and specific adsorption of antibodies.
Plant Growth Conditions and Assays-Medicago sativa cv. Europe and Medicago truncatula genotype Jemalong seeds were surface sterilized by concentrated sulfuric acid for 10 min and washed several times with large volumes of sterile water. Seeds were germinated on 0.7% water agar plates at room temperature, in the dark. Seedlings were grown on nitrogen-free medium (38). One day after planting, the seedlings were inoculated with overnight cultures of S. meliloti strains that were washed in distilled water. Each plate containing three M. sativa or two M. truncatula seedling was inoculated with 200 l of cell suspension (10 8 -10 9 cells). Plants were grown at 20°C under a 16-h light:8-h dark cycle for 30 -40 days.
To confirm that mutant L994 did not revert during the passage through root nodules, we isolated bacteria from nodules and tested them for the expression of the neomycin antibiotic resistance marker of the integrated plasmid pLSacpXL. Nodules elicited by mutant L994 were harvested from roots of both host plants, at least three nodules from each variant of inoculation. Detached nodules were surface-sterilized with 70% ethanol for 1 min, washed with sterile water, and crushed in 0.85% saline buffer. Bacterial suspensions, recovered from nodules, were diluted and streaked on LB plates containing neomycin and streptomycin or only streptomycin.
To distinguish between nodules occupied by the competitor strain L753, carrying ␤-glucuronidase gene (GUS ϩ ), and the tester GUSstrains, roots of alfalfa were washed with phosphate buffer and incubated in the dark for 3-4 h at 37°C in 50 mM sodium phosphate buffer, pH 7.5, containing 1 mM 5-bromo-4-chloro-3-indolyl-␤-D-glucuronide and 1% SDS. Nodules occupied by a GUS ϩ strain stained blue; nodules occupied by a GUSstrain remained white or pink.

RESULTS
The acpXL-msbB Gene Cluster May Direct the Biosynthesis of the C28 Fatty Acid Chain and Its Incorporation into the S. meliloti Lipid A-All genes of the acpXL-msbB cluster are oriented in the same direction and probably comprise three transcriptional units: (i) acpXL; (ii) SMc04277, SMc04275, and SMc04270; and (iii) SMc04270 and msbB (Fig. 2). A similar gene cluster was recently identified in R. leguminosarum (39), accession AF510733. acpXL of R. leguminosarum is a specialized acyl carrier protein for a C28 acyl chain and is involved in attachment of this long fatty acid chain to (Kdo) 2 -lipid IV A , a key lipid A precursor common to different Gram-negative bacteria (21). The LpxXL protein of R. leguminosarum is an AcpXLdependent C28 acyltransferase (39). Both S. meliloti MsbB and R. leguminosarum LpxXL are similar to E. coli MsbB (LpxM), an ACP-dependent myristoyltransferase that catalyzes the addition of the secondary myristoyl chain to (Kdo) 2 -(lauroyl)-lipid IV A (40). Four other hypothetical proteins encoded by the acpXL-msbB gene cluster show similarity to E. coli enzymes of fatty acid biosynthesis. Smc04277 is similar to 3R-hydroxymyryistoyl-ACP dehydratase (accession P21774), E value 3e-11; SMc04275 and Smc04273 are similar to 3-oxoacyl-ACP synthase II or polyketide synthase (P39435), E values 5e-08 and 6e-65, respectively; Smc04270 is similar to threonine 3-dehydrogenase (P07913), E value 3e-25, and many other zincbinding dehydrogenases. The fact that the acpXL-msbB gene cluster of S. meliloti and the acpXL-lpxXL gene cluster of R. leguminosarum show conservation implies that all proteins encoded by the cluster are involved in the same biosynthetic pathway leading to synthesis and incorporation of C28 acyl chain into lipid A. None of the genes comprising the acpXL-msbB gene cluster was ever targeted by a mutation. We decided to focus on the analysis of S. meliloti acpXL gene, expecting that its function should be similar to that of R. leguminosarum acpXL.
acpXL Mutant of S. meliloti Does Not Express Smooth LPS-An acpXL mutant of the S. meliloti strains 1021 was constructed as described under "Experimental Procedures." LPS from strain 1021 and the acpXL mutant L994 was extracted using the SDS/proteinase K method (35) and analyzed by SDS-PAGE. The cell extract of strain 1021 showed two LPS bands, the upper LPS-1 and the lower LPS-2 band (Fig. 3). Compared with the LPS ladder of Salmonella enterica the molecular mass of LPS-1 and LPS-2 may be estimated as 9 kDa and 4.5 kDa, respectively (data not shown). It is thought that LPS-2 contains rough LPS (R-LPS), whereas LPS-1 contains smooth LPS (S-LPS). Cell extract of mutant L994 only revealed the fast migrating R-LPS (Fig. 3).
The type and yield of LPS from Rhizobium mutants may vary depending on the extraction method (41). Therefore, LPS from the strains of interest was also extracted using the triethylamine/EDTA method (34) and hot phenol-water procedure (31). Regardless of the method of LPS isolation, mutant L994, in contrast to its parent 1021, manifested only the fast migrating band of LPS (data not shown).

C28 Acyl Chain of Lipid A and S. meliloti Lipopolysaccharide
To confirm that the observed "rough" phenotype was caused by the acpXL mutation, we performed complementation analysis. Plasmid pXK11a carrying the acpXL gene was constructed as described under "Experimental Procedures." Upon introduction of pXK11a into L994, transconjugants manifested the same LPS profile as strain 1021 (Fig. 3). The positive result of the complementation analysis evidenced that (i) the 144-bp noncoding upstream region of acpXL was sufficient for normal expression of the gene; (ii) the rough phenotype of mutant L994 was solely caused by the acpXL mutation; and (iii) the mutation within acpXL had no polar effect on the downstream genes.
Lipid A of an acpXL Mutant Lacks the C28 Acyl Chain-To test whether the acpXL mutation affected C28 acylation of lipid A, the LPS extracted from S. meliloti strains 1021 and L994 was subjected to fatty acid analysis using GC-MS. A peak corresponding to a C28 fatty acid was found in the LPS of the wild type strain (Fig. 4). However, in the LPS of the mutant this peak was completely absent (Fig. 4). In addition, we found that the level of 3-OH-14:0 in wild type LPS was 2.6-fold higher than in the mutant LPS (Table II). Interestingly, in Sinorhizobium sp. NGR234, the highest variation of the lipid A structure was connected to presence or absence of this hydroxylated fatty acid (56). Probably, in S. meliloti lipid A, 3-OH-14:0, which, similarly to NGR234, was identified as ester-linked residue (10), also represents the most variable portion of the molecule. The levels of four other fatty acids were comparable in the LPS of both strains (Table II).

LPS of acpXL Mutant L994 and Parental Strain 1021 Have a Similar Sugar Composition-
The finding that mutant L994 was missing slow migrating LPS could be interpreted as an inability of the mutant to express O antigen. To reveal sugars comprising S. meliloti O antigen, we compared the glycosyl composition of LPS derived from strains 1021 and L994. Surprisingly, we found that LPS of both strains had the same glycosyl composition, characterized by the presence of two neutral sugars and two uronic acids ( Fig. 5 and Table II). The only difference between 1021 and L994 was in the level of glucuronic acid. It was 2-fold higher in LPS of 1021 than in LPS of L994. However, this difference was not sufficient to account for an extrastructural domain of LPS.
LPS of Mutant L994 Showed the Same Immunological Properties as LPS of Strain 1021-As shown in Fig. 6, both LPS bands from cell extracts of 1021 reacted with the polyclonal antiserum. The LPS-2 band from L994 also reacted with the antiserum. However, no other immunoreactive material was revealed in the LPS preparation of L994 (Fig. 6A). Thus, in agreement with the results of LPS silver staining, mutant L994 produced only one form of LPS that was serologically similar to LPS-2 of parental strain 1021.
To test whether the antiserum contained antibodies specific to LPS-1 (recognizing O antigen), we adsorbed this antiserum with whole cells of mutant L994. It was expected that cells of the mutant would bind LPS-2-specific antibodies but not LPS-1-specific antibodies. However, when the preadsorbed antiserum was used for immunostaining LPS from S. meliloti 1021, neither LPS-1 nor LPS-2 was detected (Fig. 6B). A failure to reveal antibodies specifically reacting with LPS-1 might be explained by the following possibilities: (i) both forms of LPS had identical antigenic epitopes; (ii) the antiserum did not contain antibodies directed against all structural modules of LPS; and (iii) adsorption of antibodies by bacterial cells was nonspecific.
We cannot exclude the first two possibilities. However, we were able to check the third one, preadsorbing the antiserum by cells of a S. meliloti mutant that had altered antigenic properties. To that aim we used lpsB mutant Rm6963, known to have no or weak reactivity with antibodies against wild type LPS of S. meliloti (24). This mutant is defective in LPS core biosynthesis and produces a truncated core oligosaccharide (24). As shown in Fig. 6C, the antiserum preincubated with   (24). Therefore, we decided to test whether our acpXL mutant would also have such a phenotype. S. meliloti strains 1021, L994, and Rm6963, were plated on solid LB medium containing 0.1% DOC. After 2 days of incubation at 28°C, strain 1021 manifested normal growth, whereas L994 and Rm6963 did not grow at all. Thus, regardless of the different nature of LPS modifications, both mutants had defects in the outer membrane permeability barrier.
Symbiotic properties of S. meliloti strains were analyzed in plant assays with aseptically grown M. sativa and M. trunca-tula plants. Mutant L994, similarly to strain 1021, was able to form nitrogen-fixing nodules on the roots of both host plants. However, in contrast to the parental strain, L994 was impaired in promoting shoot growth. Mutant Rm6963, which was also tested in this experiment, manifested the expected phenotypes: delayed nodulation of M. sativa and formation of nonfixing (ineffective) nodules on M. truncatula (42).
There remained the possibility that the acpXL mutant was unstable inside nodules. The mutant could have lost the inserted DNA region and reverted to the wild type. To rule out this possibility, bacteria were reisolated from nodules and plated on solid LB medium with and without neomycin. The number of recovered bacteria varied from nodule to nodule in a range from 10 4 to 10 6 . However, the number of colony-forming units determined for each nodule population did not depend on the presence of neomycin in the medium (data not shown), indicating that all nodule isolates retained the neomycin R marker. To assess whether the acpXL mutation affected survival of rhizobia inside nodules, we also determined the titer of viable bacteria inside nodules, occupied by strain 1021. As in case of the mutant, cell titer varied from 10 4 to 10 6 , implying that the acpXL mutation was not deleterious for endosymbiotic rhizobia. These results confirmed that mutant L994 was stable during its passage through root nodules and that the observed symbiotic phenotype was caused by the acpXL mutation.
To find out why alfalfa plants inoculated with L994 manifested reduced shoot growth compared with plants inoculated with 1021, we compared nodulation rate and the competitive ability of the strains. Nodulation rate was assayed by counting the numbers of nodules at particular time points after inoculation. As shown in Fig. 7, mutant L994 manifested delayed nodulation compared with parental strain 1021. To study the competitive ability of the strains, we employed S. meliloti strain L753, which can be easily detected inside root nodules because of the expression of an eglC-gusA fusion. 2 Results of two coinoculation experiments, where 1021 and L994 were tested against L753, showed that L994 was a weaker competitor than 1021 (Table III). Taken together, the obtained results show that the acpXL mutation caused attenuation of the outer membrane permeability barrier and quantitatively impaired symbiotic performance of S. meliloti.

DISCUSSION
From Acyl "Roots" to Glycosyl "Tops": What Is the Link between the Missing C28 Acyl Chain and the Missing O Antigen?-The presence of a C28 (-1) hydroxylated acyl chain is a taxonomically important feature of Rhizobiaceae and related C28 Acyl Chain of Lipid A and S. meliloti Lipopolysaccharide bacteria belonging to the ␣-2 subclass of Proteobacteria (4).
Brozek and coauthors (21) purified the AcpXL protein of R. leguminosarum and demonstrated that it facilitated the transfer of the C28 acyl chain to the lipid A precursor. This enzymatic activity was also found in cell extracts of R. etli and S. meliloti (21). Up until now, AcpXL has only been characterized biochemically but not genetically. In this study, we constructed S. meliloti acpXL mutant L994. Fatty acid composition analysis of lipid A from parental strain 1021 and mutant L994 confirmed that AcpXL of S. meliloti is involved in C28 acylation of lipid A. Recently, Geiger and Lopez-Lara (43) reported that the S. meliloti genome had the capacity to encode six acyl carrier proteins including AcpXL. In the light of this notion, it is important that none of the acyl carrier proteins can replace AcpXL in C28 acylation of lipid A.
Mutant L994 revealed a number of interesting phenotypes, including the lack of slow migrating LPS, LPS-1. However, cells of L994 expressed the same antigenic epitopes as cells of 1021, and LPS extracted from both strains revealed the same glycosyl composition. Because results of LPS immunodetection and its sugar composition analysis did not agree with results of SDS-PAGE demonstrating a clear difference between LPS of 1021 and L994, we decided to revise the conventional view on LPS of S. meliloti and looked for a noncontradictory explanation of our observations. Although slow migrating LPS of S. meliloti was usually considered as smooth LPS, there is no convincing evidence that it consists of lipid A core plus O antigen. The structure of S. meliloti O antigen is unknown, and even with the help of monoclonal antibodies it was impossible to demonstrate that LPS-1 possesses a unique structural domain not present in LPS-2 (44). Following this line of reasoning, we considered the possibility that LPS-1 might be an aggregated form of LPS-2 (presumed rough LPS) stabilized by hydrophobic interactions between C28 acyl chains of lipid A. Consequently, S. meliloti mutant L994, missing the C28 acyl moiety of lipid A, would express only a fast migrating form of LPS. Usually, LPS aggregates are resolved under standard conditions of SDS-PAGE or DOC-PAGE. However, it cannot be excluded that structural peculiarities of S. meliloti LPS may preclude resolving of the aggregates because aggregation properties of LPS molecules depend greatly on their structure (45). Remarkably, the molecular mass of the slow migrating form of LPS (LPS-1) is approximately two times higher than that of the fast migrating form of LPS (LPS-2), 9 kDa versus 4.5 kDa.
Interestingly, Gudlavalleti and Forsberg (56) demonstrated that S-LPS and R-LPS of Sinorhizobium sp. NGR234 differed in their acylation pattern. Both forms of LPS contained lipid A with a long fatty acid chain, but there was a difference in the number of acyl substitutions. Thus, S-LPS was enriched in triand tetraacylated lipid A species, whereas R-LPS was enriched in penta-and tetraacylated species and contained no triacylated species at all. It is obvious that a long acyl chain would have stronger influence on physical properties of underacylated LPS molecules than on fully acylated ones. Therefore, if our speculation is true, underacylalated LPS molecules containing long acyl substitution would self-aggregate most efficiently. It may be argued that LPS aggregates should have been dispersed under dissociative conditions used for separation of LPS. However, it cannot be excluded that a minor portion of LPS (S-LPS constituted only 4%) was not dispersed completely.
Although we cannot provide experimental evidence to our hypothesis, we would like to mention three groups of facts that also point to an elusive nature of S. meliloti O antigen and may support our speculation that slow migrating LPS lacks O antigen and presents an aggregated form of fast migrating LPS.
First, in SDS-PAGE, LPS-1 of S. meliloti is often fainter than LPS-2, and sometimes it is nonreproducible (12,34,44,46). Such an irregular mode of expression may be expected if LPS-1 is not a distinct form of LPS but just an aggregate of LPS-2. A similar explanation may clarify why the S. meliloti lpsB mutant expresses two LPS bands instead of one. This mutant has a defect in LPS core biosynthesis which accounts for faster migration of both bands relative to LPS-1 and LPS-2 of the wild type (47). However, in R. leguminosarum and enteric bacteria such a defect would result not only in truncation of rough LPS but also in the complete absence of smooth LPS because the O antigen cannot be added to the truncated core. Consequently, the case of the S. meliloti lpsB mutant was considered as exceptional (47). However, in the light of our speculation that the upper LPS band is just an aggregate, there will be no need to seek reasons why alteration of the core did not prevent attachment of an O antigen.
Second, Reuhs et al. (44), who considered LPS-1 as smooth LPS, i.e. lipid A core plus O antigen, pointed out that the O antigen of S. meliloti strains, in contrast to the O antigens of typical Gram-negative bacteria, was nonspecific and nonimmunogenic. Another peculiar feature of S. meliloti LPS is an immunodominance of its core (12,24,44). In this work, we also found that antiserum raised against whole cells of S. meliloti contained antibodies reacting with rough LPS (likely core) but contained no antibodies reacting with the O antigen. In S. meliloti, K antigens but not O antigens determine the serotypes of individual strains (44). A similar situation is known for the mucosal pathogens Haemophilus influenzae and Neisseria meningitidis (48 -50). These bacteria produce a specific type of LPS, named lipooligosaccharide, which lacks an O antigen but is decorated with nonrepeating oligosaccharide branches (51). S. meliloti may also produce lipooligosaccharide because S. meliloti mutants deficient in the production of both exopolysaccharides and capsular polysaccharide (e.g. exoB rkpC) do not have as rough a morphology as rough mutants of E. coli.
Third, recent analysis of a collection of S. meliloti LPS mutants generated by random Tn5 mutagenesis revealed mutants with diverse LPS profiles. 3 However, none of these mutants showed a rough phenotype in SDS-PAGE. Consequently, not a single gene was assigned to a function in O antigen biosynthesis. To date, acpXL mutant L994 is the only rough mutant ever found in S. meliloti. The existence of L994 demonstrates that rough S. meliloti mutants are viable. a The ratio of strains in the initial inoculum was determined by plating serial dilutions of both strain cultures separately before mixing. Cell titers of all inocula ranged from 10 8 to 10 9 colony-forming units. 60 -80 nodules were analyzed for each variant of inoculation.

C28 Acyl Chain of Lipid A and S. meliloti Lipopolysaccharide
To summarize, our results and the data from other authors support the idea that S. meliloti does not possess an O antigen. There remains an alternative possibility that the S. meliloti O antigen is very unusual, and it is only attached to the lipid A core that is acylated with the C28 fatty acid chain, but not to the lipid A core missing this long acyl substitution.
The C28 Acyl Chain of Lipid A Does Not Play a Specific Role during Symbiosis of S. meliloti with Its Host Plants-Since the time of the discovery of a long acyl substitution in LPS of Rhizobia (4) it has been intriguing to know, if this C28 acyl substitution plays a role during symbiosis with host plants. To address this question we studied the symbiotic properties of S. meliloti acpXL mutant and found that it was able to form nitrogen-fixing nodules on M. sativa and M. truncatula but showed a delay in nodulation rate and reduced competitive ability.
The importance of S. meliloti LPS for the establishment of symbiosis with host plants was first demonstrated using a lpsB mutant of 2011, Rm6963 (24,42). This mutant was unable to form an effective symbiosis with M. truncatula and evoked the defense response of plant cells (42). Mutant L994 resembles Rm6963 by its DOC sensitivity, delayed nodulation, and impaired competition on M. sativa plants. However, the acpXL mutant, in contrast to the lpsB mutant, retains wild type cell surface antigens, as was demonstrated by immunoblotting, and also retains the ability to induce nitrogen-fixing nodules on the roots of M. truncatula. This comparison of mutants L994 and Rm6963 shows that in relation to symbiosis, attenuation of the outer membrane permeability barrier does not cause such dramatic consequences as alteration of cell surface antigens. In addition, it is worthwhile to mention that the host plants were very tolerant of the "rough" phenotype of L994. This fact again raises the question of whether strain 1021 possess an O antigen because genuine rough mutants that lost the O antigen usually had more severe phenotypes (3).
The C28 Fatty Acid Is Likely to Form a Secondary Acyl Substitution of S. meliloti Lipid A: Critical Role of Secondary Acyl Substitutions for the Immune Response-Analysis of the S. meliloti genome sequence revealed that the acpXL gene is located upstream from a cluster of genes that probably are also involved in the biosynthesis of C28 hydroxylated fatty acid and its incorporation into lipid A. It is especially interesting that this cluster includes the msbB gene, which was initially annotated as a gene encoding a late acyltransferase, e.g. the acyltransferase that transfers an acyl chain to Kdo 2 -lipid IV A , a key lipid A intermediate carrying primary acyl substitutions and two Kdo residues. Recent studies of Basu and coauthors (39) showed that the MsbB of S. meliloti had a much closer counterpart than MsbB of E. coli, namely LpxXL, "late" C28 acyltransferase of R. leguminosarum. In the light of this finding, it makes sense to rename msbB of S. meliloti as lpxXL and to reconsider the structure of S. meliloti lipid A. Indeed, it is likely that the MsbB protein of S. meliloti functions in the same manner as LpxXL of R. leguminosarum, generating an Nlinked acyloxyacyl substitution of lipid A. If this is the case, S. meliloti, similarly to R. etli, R. leguminosarum, and Sinorhizobium sp. NGR234, should carry a long hydroxylated fatty acid chain as a secondary but not as a primary substitution of lipid A (Fig. 1). Additionally, it is important that Sinorhizobium sp. NGR234 and S. meliloti are closely related bacterial species, and therefore their lipid A structures are expected to be similar. Earlier no acyloxyacyl moiety was detected in S. meliloti lipid A (10). However, it is worthwhile to reexamine the structure of S. meliloti lipid A because a failure in detection of a secondary acyl linkage might have been caused by methodological problems. Recent studies with lipid A of R. etli showed that an acyloxyacyl moiety might be detected only by NMR procedures that enabled analysis of lipid A molecules in their intact state, avoiding chemical modifications and/or fragmentation (7,8).
According to our knowledge, no mutants deficient in AcpXL or C28 acyltransferase have been described so far. However, mutants that produce lipid A missing a secondary acyl substitution have already been isolated in a number of human pathogens. A remarkable feature of such lipid A is its reduced toxicity in comparison with wild type lipid A. In particular, msbB mutants of E. coli and S. enteridis which lack myristoyltransferase activity and produce pentaacylated lipid A instead of hexaacylated molecule (Fig. 1) had a dramatically attenuated ability to stimulate an inflammatory response and did not cause lethality of animal hosts (40,52,53). As we found, the C28 acyl substitution of S. meliloti lipid A had no specific symbiotic function. However, given a key role of secondary acyl substitutions in eliciting host immune responses, it may be relevant to yet unknown mechanisms of plant-microbe interaction.