LysM domains of Medicago truncatula NFP protein involved in Nod factor perception. Glycosylation state, molecular modeling and docking of chitooligosaccharides and Nod factors

Abstract

The establishment of the symbiosis between legume plants and rhizobial bacteria depends on the production of rhizobial lipo-chitooligosaccharidic signals (the Nod factors) that are specifically recognized by roots of the host plant. In Medicago truncatula, specific recognition of Sinorhizobium meliloti and its Nod factors requires the NFP (Nod factor perception) gene, which encodes a putative serine/threonine receptor-like kinase (RLK). The extracellular region of this protein contains three tandem lysin motifs (LysMs), a short peptide domain that is implicated in peptidoglycan or chitin binding in various bacterial or eukaryotic proteins, respectively. We report here the homology modeling of the three LysM domains of M. truncatula NFP based on the structure of a LysM domain of the Escherichia coli membrane-bound lytic murein transglycosidase D (MltD). Expression of NFP in a homologous system (M. truncatula roots) revealed that the protein is highly N‐glycosylated, probably with both high-mannose and complex glycans. Surface analysis and docking calculations performed on the models of the three domains were used to predict the most favored binding modes for chitooligosaccharides and Nod factors. A convergent model can be proposed where the sulfated, O-acetylated lipo-chitooligosaccharidic Nod factor of S. meliloti binds in similar orientation to the three LysM domains of M. truncatula NFP. N-glycosylation is not expected to interfere with Nod factor binding in this orientation.

Introduction

The establishment of the agronomically and ecologically important symbiosis between legume plants and rhizobial bacteria is initiated through the production of lipo-chitooligosaccharidic bacterial signals (the nodulation or Nod factors). These Nod factors are required for host specificity, bacterial infection, and root nodule organogenesis and provoke host cell responses within seconds at low concentrations (10⁻¹² to 10⁻⁹ M) (Dénarié et al., 1996). The Nod factor structure is based on scaffolds of linear β(1-4)-linked N-acetylglucosamine (GlcNAc) oligosaccharides (usually four or five residues), in which the terminal nonreducing sugar is de-N-acetylated and N-acylated. Differences in the number of sugar residues, their chemical substitutions, and the structure of the acyl chain are characteristic of the rhizobial species and are involved in host specificity. For example, the major Nod factor of Sinorhizobium meliloti, the symbiont of Medicago plants (alfalfa and related species), consists of tetra-N-acetylglucosamine, acylated with a C16:2Δ2,9 fatty acid, O-acetylated on C6 of the terminal nonreducing sugar, and O-sulfated on C6 of the reducing sugar (NodSm-IV, Ac, S, C16:2). The presence of the O-sulfate determines host specificity and is required for all symbiotic responses on Medicago (Roche et al., 1991), whereas the structure of the acyl chain and the acetylation of the terminal nonreducing sugar are important for infection and nodulation (Ardourel et al., 1994). Structural studies on this and related Nod factors suggest that they are probably monomeric at physiological concentrations and that the structure of the lipid chain influences potential interaction with the carbohydrate moiety (Groves et al., 2005). The fact that specific structures provoke responses on host legumes at low concentrations suggests that they are perceived by specific plant receptors.

Recent studies, aimed at understanding the mechanism of Nod factor perception and signal transduction, have identified legume genes that are required for symbiotic nodulation (for a review, see Geurts et al., 2005). Whereas some of these genes are required for all Nod factor responses, others determine the specificity of response to certain Nod factor structures, and these genes have been hypothesized to encode Nod factor receptors. Such genes include NFP (Nod Factor Perception) of Medicago truncatula (Ben Amor et al., 2003), NFR1 and NFR5 of Lotus japonicus (Madsen et al., 2003; Radutoiu et al., 2003), and SYM10 and SYM2 of Pisum sativum (Walker et al., 2000; Limpens et al., 2003). The recent cloning of some of these genes (Limpens et al., 2003; Madsen et al., 2003; Radutoiu et al., 2003; Arrighi, J.F. et al., submitted for publication) has shown that they encode members of a functionally uncharacterized class of plant receptor-like kinases (RLKs) with two or three lysin motifs (LysMs) in the putative ligand-binding, extracellular region (LysM-RLKs).

LysMs are protein domains of ∼40 amino acids found initially in several bacterial autolysin proteins (Joris et al., 1992; Birkeland, 1994). They have now been shown to occur in various proteins from bacteria and some eukarya but are not present in archaea (Bateman and Bycroft, 2000). They may be associated with a variety of other protein domains but only in plants are they found associated with a kinase domain (Ponting et al., 1999). As LysM-RLKs occur in nonlegume plants such as Arabidopsis thaliana and Oryza sativa (Shiu et al., 2004), they are not specific for Nod factor signaling. In bacteria, they are found in many cell wall enzymes and have been described as peptidoglycan-binding modules (Bateman and Bycroft, 2000). This hypothesis has been reinforced by work on the major autolysin of Lactococcus lactis which has shown that the C-terminal part of the protein, which contains three LysM domains, binds to bacterial peptidoglycan, probably to the glycan part (Steen et al., 2003). The fact that certain chitinases from Caenorhabditis elegans (Bateman and Bycroft, 2000), Kluyveromyces lactis (Butler et al., 1991), and Volvox carteri (Amon et al., 1998) contain such domains suggests that LysM domains may show specificity for GlcNAc-containing glycans. This suggestion is reinforced by the identification of the symbiotic LysM-RLK genes, which by extrapolation should encode Nod factor-binding receptors.

In the model legume, M. truncatula, NFP is the sole gene identified to date, which is required for all Nod factor responses (Ben Amor et al., 2003). It encodes a LysM-RLK with the typical type I membrane protein structure in which the single transmembrane-spanning helix separates the internal kinase domain from the extracellular region (Arrighi, J.F. et al., submitted for publication, GenBank accession no. DQ496250). This region contains three LysM repeats (termed LysM1, LysM2, and LysM3), separated by 11 and 16 amino acid linkers containing paired cysteine residues (CXC motifs). In this article, we examine the structure of the LysM domains of this receptor protein and their potential to interact with Nod factors from S. meliloti.

Results

Description of the overall structure of the models

Although according to the PFAM database (Bateman et al., 2004) the LysM domain has been identified in >1500 proteins, only two three-dimensional (3D) structures are available. The structure of one isolated LysM domain has been solved by nuclear magnetic resonance (NMR) using a fragment of Escherichia coli membrane-bound lytic murein transglycosidase D (MltD) (Bateman and Bycroft, 2000). Very recently, the crystal structure of Bacillus subtilis ykuD has been reported, a protein that contains one LysM domain and a putative peptidase or peptidotransferase domain (Bielnicki et al., 2006). In both structures, the LysM domain is characterized by a βααβ secondary structure with the two helices packing onto the same side of an antiparallel β-sheet. The N-terminus and C-terminus are close to each other. Alignment of the three LysM domains identified in the M. truncatula NFP sequence with the one from E. coli MltD yielded low sequence identity scores of 17, 19, and 12% for LysM1, LysM2, and LysM3, respectively. However, the repartition of hydrophobic amino acids allowed for identification of the conserved secondary structures that are characteristic of this fold (Figure 1).

Based on the conservation of the two α-helices and two β-strands, it was therefore possible to build homology models for the three NFP LysM domains. Building of the loops did not present too many difficulties since the two first ones are relatively short. The longest loop, that is, the one connecting α2 and β2, is relatively well conserved between the MltD LysM domain and the two first domains of NFP. The third domain, that is, LysM3, has a shorter loop at this position, and the model could be built keeping the same overall shape of the domain. The overall quality of the resulting models was verified by the PROCHECK program (Laskowski et al., 1993). All peptide bonds belong to allowed regions of the Ramachandran Map. Furthermore, the burying of hydrophobic amino acids observed in the NMR structure of MltD LysM (Tyr5, Leu13, His20, and Leu45) is well reproduced in the three models. Figure 2 displays the three model domains compared with the MltD NMR structure. The other highly conserved amino acid is an Asn residue, which appears to play a role in forming the turn at the end of helix 2 (Bateman and Bycroft, 2000).

Analysis of NFP glycosylation

Analysis of the NFP sequence predicted the presence of 10 potential N-glycosylation sites, eight of which occur in the LysM domains (Figure 1). One other site is located before the first LysM domain, and the remaining one is located between LysM1 and LysM2. Genetic constructs were made in which the sequence coding for the whole protein (NFP-full) or the extracellular and transmembrane regions (NFP-TM) was fused to the sequence coding for either a fluorescent protein (FP) or the TAPTAG (TT) marker. The predicted sizes of the two fusion proteins are 93.2 and 51.8 kDa, respectively, before putative signal peptide cleavage. These two proteins were expressed in transgenic M. truncatula roots and yielded proteins with apparent molecular weights (MWs) ∼17 to 22 kDa larger than the theoretical ones (Figure 3A).

The glycosylation of plant proteins depends upon their trafficking: Glycosylation and subsequent glycan modification take place in the endoplasmic reticulum and the Golgi apparatus, respectively, and lead to biantennary structures either of a high-mannose type (Man₉GlcNAc₂) or of a complex type, with the particular occurrence of core α3-fucose and β2-xylose that are characteristic of plants. Plasma membrane and secreted proteins display this glycosylation, whereas the vacuolar proteins are modified by glycosylhydrolases resulting in shorter N-glycans of the paucimannose type (Lerouge et al., 1998). For NFP, that is not expected to travel to the vacuole, the higher expected mass for each N-glycan could be 1880 Da for a high-mannose type and 2210 Da for a complex type. Within the hypothesis that the ∼20 extra kilodalton observed in plant-expressed NFP is because of glycosylation, most of the 10 N-glycosylation sites are effectively expected to carry a glycan.

To confirm the glycosylation status of NFP, membrane extracts of roots expressing the NFP-TM–TT protein were treated with peptide N-glycosidase F (PNGase F) or endoglycosidase H (Endo H). Endo H is only active on high-mannose-type N-glycans and not on complex ones, whereas PNGase F will cleave off both types, except if the GlcNAc residue linked to the asparagine is α(1,3)-fucosylated. Either PNGase F or Endo H treatment resulted in a protein with a size of ∼6 kDa smaller than the nontreated protein but still ∼16 kDa larger than predicted (Figure 3B). To eliminate the possibility that NFP may run aberrantly on denaturing gels and not according to its MW, a genetic construct was made in which the NFP sequence was fused to the sequence encoding the hemagglutinin (HA) tag, and the encoded NFP-HA protein was expressed in yeast. The yeast-expressed NFP-HA migrated at a size of ∼12 kDa larger than predicted, but treatment with PNGase F or with Endo H yielded proteins migrating at about the predicted size (73.3 kDa). This result suggests that NFP expressed in yeast is N-glycosylated with N-glycans that can be removed by treatment with PNGase F or Endo H and that de-N-glycosylated NFP does not run aberrantly on a denaturing polyacrylamide gel. Together with the plant results, these data suggest that plant-expressed NFP is highly N-glycosylated and probably carries two types of glycans. First, about two to four residues appear to be N-glycosylated with high-mannose-type glycans that can be removed by treatment with Endo H and PNGase F. Second, the additional size of the treated protein in plants compared to yeast is consistent with 6–8 sites being N-glycosylated with complex-type N-glycans, which would be α(1,3)-fucosylated on the proximal GlcNAc residue and resistant to Endo H and PNGase F. Confirmation of this hypothesis would require purification of the protein and analysis of released N-glycans.

Among the 10 potential glycosylation sites, eight are located within the LysM domains—two in LysM1 and three each in LysM2 and LysM3 (Figure 1). The corresponding asparagine residues have been identified in the models, and none of them is located in a buried environment (Figure 2). The models that are proposed here are therefore in agreement with extensive glycosylation of the LysM domains, as demonstrated by experimental data.

Prediction of chitooligosaccharide-binding sites in LysM domains

Analysis of the shape and available surface of LysM domains from the MltD and NFP models suggests structures that can be described as triangle-based pyramids (in the orientation of Figure 2). The larger flat face is bordered by the two helices and the long loop between α2 and β2 (Face A). A smaller one is limited by helix α1 and strand β1 (Face B). The third face is more variable and does not present favorable binding regions, and the base corresponds to the N-terminal and C-terminal extremities of the domain. Accessible surface calculations performed with the MOLCAD program together with representation of the hydrophobic potential demonstrated that both Face A and Face B have amphiphilic characters that are compatible with oligosaccharide binding.

The docking simulations were performed with AutoDock 3.0, a program that has been demonstrated to be well adapted for protein/carbohydrate interactions (Rockey et al., 2000; Wen et al., 2005). The advantage of this method is that it takes into account the flexibility of the ligand, which is of primary importance for oligosaccharides both at the level of glycosidic linkages and at the level of the exocyclic groups (hydroxyl, hydroxymethyl, etc.). Docking was initially performed with chitotetraose on each of the three LysM domains of NFP and the MltD LysM domain. When parameters were set to obtain 50 different docking solutions, some of the solutions, as expected, converged in the same binding orientation, therefore leading to different predictions for the binding of chitotetraose to each of the LysM domains. The most populated binding modes clustered in Face A or Face B of the domains (Figure 4A) and very sparsely in other regions (data not shown). Energies of binding present some differences, but they are not systematically in favor of one site or another, so it is difficult to take them into account at this stage.

Docking of Nod factors on LysM domains

A Nod factor molecule corresponding to the major Nod factor of the M. truncatula bacterial symbiont was submitted to the same docking simulation. This Nod factor differs from chitotetraose in bearing a sulfate on position 6 of the reducing GlcNAc and an O-acetyl on position 6 of the terminal nonreducing sugar, and the N-acetyl group on this same sugar is replaced with an N-linked C16:2Δ2,9 fatty acid. The lipid moiety was considered as flexible with the exception of the two double bonds. Only the three modeled LysM domains of the NFP protein were considered as targets. Again, several binding possibilities were predicted. However, among the lower energy ones, one binding mode, located in Face A, was almost identical to the three domains. The common binding site is of interest because of the conservation between the three domains (Figure 4B).

The tetrasaccharide moieties lie on the flat area of Face A and interact mostly through van der Waals bond with amino acids of the long loop between helix α2 and strand β2. One hydrogen bond is established between the hydroxyl groups O6 of one of the inner GlcNAc residues and amino acids Glu59, Ser121, and His183 of this long loop. In domains LysM1 and LysM2, the sulfate group of the ligand is in close proximity to a lysine residue (Lys52 and Lys114). Lysine residues are flexible, and the distance is short enough such that local conformational flexibility allows for a direct salt bridge between the NH₃⁺ of the lysine and the sulfate group. The O-acetyl group at position 6 of the nonreducing GlcNAc is located in a hydrophobic pocket of each of the three LysM domains, made mainly by amino acids of the loop between strand β1 and helix α1 (Pro31 and Leu34 in LysM1, Phe97 in LysM2, and Trp157 in LysM3).

In this model, the lipid moiety is predicted to wrap around the LysM domain and to specifically interact with a patch of amino acids that are located at the base of strand β2. The amino acids that would play a special role in the recognition of the acyl chain of Nod factors are Leu63, Ile 64, and Pro65 in LysM1; Leu124, Pro126, and Leu127 in LysM2; and Phe185 in LysM3.

In respect of the N-glycosylation sites discussed above, none of them interferes with Nod factor binding, with the only exception of Asn184 in LysM3 that is close to the second GlcNAc residue, counting from the residue that carries the lipid. Nevertheless, the Asn184 side chain is not buried by the ligand. A N-glycan at this position would interact with the Nod factor ligand but would not necessarily prevent it from binding.

Discussion and Conclusions

Although LysM domains are widely distributed and well characterized, only two structures are available (Bateman et al., 2004; Bielnicki et al., 2006), and therefore molecular modeling is currently the only tool that can help in visualizing the structure of LysM domains from other proteins and their potential interaction with ligands. The three M. truncatula LysM domains of NFP were built using the NMR structure from the MltD fragment. Although sequence identities to the MltD domain are very low, we propose that the βααβ architecture is well conserved in the plant LysM domains, while some conformational variations occur in the long loop that connects helix α2 to strand β2. The proposed models are reinforced by the very recent structure of the ykuD protein from B. subtilis, which has a similar overall LysM structure to MltD. However, the conformation of this particular loop is rather different from the corresponding one in the MltD protein, and the average thermal factors for that part of the structure are higher than for the rest of the domain, consistent with higher intrinsic mobility of the loop (Bielnicki et al., 2006).

The proposed conservation in structure between the plant domains and the bacterial ones is consistent with the binding of structurally related ligands. For NFP, genetic studies suggest that the ligand is the lipo-chitooligosaccharidic Nod factor of S. meliloti, whose structure has previously been studied by NMR and molecular modeling (Gonzalez et al., 1999; Groves et al., 2005). This advantage, in comparison with the macromolecular peptidoglycan ligand proposed for the bacterial LysM domains, has permitted ligand docking studies to be performed with the plant domains. We have explored several possibilities for the binding of chitooligosaccharides and Nod factors to the models. The solutions retained in Figure 4B are in agreement with structural characteristics of chitooligosaccharide binding where hydrophobic contact plays an important role (Asensio et al., 1995). Validation of the docking of ligands to the models is not available at the present time, and the models of complexes that are proposed in the present article should therefore be considered as preliminary. More biochemical and structural studies will be needed for a complete characterization of the binding mode.

By expressing NFP from a strong promoter in a homologous, recombinant system (M. truncatula roots), we have shown that it encodes a protein that is ∼17 to 22 kDa larger than the predicted size and is N-glycosylated (Figure 3). The 10 predicted N-glycosylation sites in the extracellular region are all surface exposed (Figure 2), and studies on the protein using different N-glycosidases suggest that probably most of them are used, with between two and four sites being N-glycosylated with high-mannose-type glycans and the others with complex-type N-glycans (Figure 3). As glycosylation can affect ligand binding, the position of these sites was related to the binding model, and it appears that N-glycosylation of none of the sites in the LysM domains (Figure 2) would interfere with ligand docking to the favored orientations shown in Figure 4B. However, the N-glycan on Asn184 of LysM3 could somewhat interact with the Nod factor ligand.

One exciting possibility provided by the models and the docking procedure is the observation that in principle, each LysM domain could bind one Nod factor molecule. The number and structure of LysM domains that are present in the receptor may be of importance for Nod factor signaling, together with the manner in which the domains are interconnected. Proteins encoded by the putative orthologous genes of other legumes (SYM10 of pea and NFR5 of L. japonicus) also contain three such domains (Madsen et al., 2003). A recent study of the AcmA autolysin from L. lactis has demonstrated that the presence of three LysM domains is needed for optimal peptidoglycan binding and biological functioning (Steen et al., 2005). For NFP, and its orthologs, the function of three LysM domains could be either to accommodate Nod factors with variant structures, as suggested by the considerable sequence variation between the three domains, or to amplify the signal for optimal transduction. More biochemical and structural studies of the protein are required to establish whether this protein functions as a Nod factor-binding protein and to characterize the binding site.

RLKs are known to function as dimers, and two pieces of evidence suggest that NFP may form a heterodimer with another LysM-RLK. First, in the legume L. japonicus, the NFR1 gene, encoding a different LysM-RLK, is also required for early Nod factor responses, suggesting that it acts at the same level in the signaling cascade as NFR5 (Radutoiu et al., 2003). Second, NFP, and its putative orthologs, contains an aberrant kinase domain and appears not to be kinetically active (Arrighi, J.F. et al., submitted for publication), consistent with NFP interacting with another protein to transduce the Nod factor signal. Although to date no M. truncatula gene, apart from NFP, has been shown to be necessary for the first Nod factor responses, the gene LYK3, which is highly homologous to NFR1 and required for Nod factor-dependent rhizobial infection (Limpens et al., 2003), is a good candidate to encode an NFP partner. Since 3D structures have been solved only for single LysM domains, it is not possible at the present time to model the architecture that a multi-domain receptor will adopt. The models presented here, however, highlight the structural similarities between bacterial and plant LysM domains and have identified certain residues whose potential role in Nod factor signaling may be functionally tested. Moreover, the observation that the protein is highly N-glycosylated in M. truncatula roots suggests that post-translational modifications must be taken into account when analyzing the structure–function of this protein.

Materials and Methods

Cloning and expression in hairy roots

The binary plasmid pGr29-35SOM-EGFP-CaMVT (Navarro-Gochicoa et al., 2003) based on the pGreen0029 vector (Hellens et al., 2000) was used to make a fusion of the extracellular and transmembrane (NFP-TM) region of NFP, fused to the TAPTAG (TT) marker, a fusion of calmodulin-binding protein and protein A (Rigaut et al., 1999). The enhanced green fluorescent protein (EGFP) was first replaced by TAPTAG, and then polymerase chain reaction (PCR) was used to introduce NcoI and SmaI sites at the start codon (ATG) and at codon 282 (ATC—located 30 amino acids after the TM region as predicted with the TMHMM algorithm) of the complete open-reading frame of NFP, respectively, before cloning into the appropriate sites in the vector. The NFP-full–FP fusion was produced by eliminating the internal NcoI site by site-directed mutagenesis, and PCR was then used to introduce NcoI and SmaI sites at the start and termination codons, respectively, before cloning into the appropriate sites in the vector. The two plasmids (coding for NFP-TM–TT and NFP-full–FP) were introduced into Agrobacterium rhizogenes ArquaI and used to produce kanamycin-resistant roots as described (Boisson-Dernier et al., 2001). Different lines were tested of the same construct by western blot and maintained on solid media, as described previously. All lines tested of the same construct expressed fusion proteins of indistinguishable MW.

Cloning and expression of NFP in yeast

For NFP-HA expression in Saccharomyces cerevisiae, PCR was used to amplify NFP cDNA and to add SfiI restriction sites just after the predicted signal peptide cleavage site according to the PSORT algorithm (codon GCT) and just before the stop codon. The PCR fragments were digested with SfiI and cloned into yeast plasmid pDL2xN-SUC (Dualsystem Biotech AG, Zurich, Switzerland) opened with SfiI. Transformation, selection, and maintenance of the resulting strains were performed following standard procedures and as described in the Dualsystem Biotech membrane kit manual.

Analysis of glycosylation

To prepare the plant microsomal fraction, frozen plant material was crushed in a mortar and resuspended in 25 mM Tris–HCl buffer pH 8.5, 5 mM MgCl₂, 470 mM sucrose, 10 mM β-mercaptoethanol and protease inhibitors, 1 mM phenylmethylsulfonyl fluoride, and 1 mg/mL of each of leupeptin, aprotinin, antipain, chymostatin, and pepstatin (Sigma-Aldrich Corporation, St Louis, MO). The homogenate was centrifuged at 4500g for 15 min, and the resulting supernatant was centrifuged again at 45,000g for 60 min. The membrane pellet was then resuspended in 25 mM Tris–HCl buffer pH 7, 1 mM MgCl₂, 1 mM CaCl₂, and 250 mM sucrose and protease inhibitors. Protein concentration was measured with the Bradford standard procedure. Proteins were then precipitated with trichloroacetic acid (TCA) (10% w/v) and washed twice in 90% acetone. The pellet was resuspended in 25 mM Tris–HCl buffer pH 7 by sonication.

Yeast homogenates were prepared as described in the Dualsystem membrane kit manual. Briefly, cells were lysed by NaOH/TCA treatment and resuspended in 3% sodium dodecyl sulfate (SDS). Proteins were diluted to 0.2% SDS and TCA precipitated, as described previously.

Plant or yeast proteins (5 µg) were then incubated in 25 mM Tris–HCl buffer pH 7 with either 2500 U PNGase F (Sigma) or 25 mU Endo H (Roche Diagnostics, Meylan, France) for 30 min at 37°C.

After separation by SDS–PAGE on 8% (w/v) acrylamide gels and transfer onto a nitrocellulose membrane (western blotting), proteins were detected either with polyclonal anti-green fluorescent protein (GFP) antibodies (Molecular Probes, Invitrogen SARL, Cergy Pontoise, France) for NFP-TM-TT and NFP-full–FP followed by peroxidase-linked anti-rabbit antibodies (GE Healthcare Europe, Orsay, France) or with anti-HA conjugated with peroxidase (Roche) for yeast-expressed NFP-HA. Peroxidase activity was revealed by chemiluminescence using the enhanced chemiluminescence kit (Amersham Biosciences). MWs were estimated by comparison with proteins of similar size (PageRuler Prestained Protein Ladder; Euromedex, Mundolsheim, France).

Homology modeling

Multiple alignment of LysM domains was performed with the ClustalX program (Thompson et al., 1994). The homology modeling COMPOSER program (Blundell et al., 1988) of the Sybyl software (SYBYL Tripos Associates, St. Louis, MO) was used to build the domain models. Structurally conserved regions (SCRs in COMPOSER) were built from the NMR structure of a LysM domain from E. coli MltD (Bateman and Bycroft, 2000) (code 1E0G in the Protein Data Bank; Berman et al., 2000). Loops were modeled using the most similar fragments in the Sybyl library of 3D structure of proteins, and their geometry were optimized using the Tripos force field (Clark et al., 1989). The stereochemistry of the resulting models was checked using the PROCHECK program (Laskowski et al., 1993), and backbone conformations lying outside the allowed regions of Ramachandran Map were further optimized. Hydrogen atoms were added on all atoms, and partial atomic charges were derived using the Pullman procedure. Connolly surfaces have been calculated using the MOLCAD program (Waldherr-Teschner et al., 1992).

Docking of ligand

Docking of oligosaccharides and Nod factors was performed using the AutoDock 3.0 program (Morris et al., 1998). AMBER force field charges were assigned to all protein atoms. The chitotetraose and the Nod factor coordinates used for the starting models were taken from a previous study (Navarro-Gochicoa et al., 2003). Grids of probe atom interaction energies and electrostatic potential were generated around the whole protein by AutoGrid program present in AutoDock 3.0 with a spacing of 0.5 Å in a box of 80 × 80 × 80 ų around each of the modeled domain. All ligands’ torsion angles, excluding intra-ring bonds, double bonds, and N-acetyl exocyclic groups, were allowed to rotate freely, considering therefore 19 and 26 rotatable bonds for the tetrasaccharide and the Nod factor, respectively. For each interacting pair, one job of 50 docking runs was performed using a starting population of 50 individuals for the tetrasaccharide and 200 individuals for the Nod factor and an energy evaluation number of 2 × 10⁶. For clustering of solution, a root mean square tolerance of 2.5 Å was used for the tetrasaccharide and of 3.0 Å for the Nod factor.

Final optimization

For the interaction of the Nod factors with each of the three LysM domains, the most interesting docking modes predicted by AutoDock 3.0 (Face A) were submitted to a further energy minimization with the use of the Tripos force field (Clark et al., 1989) with parameterization previously developed for protein–carbohydrate interactions (Imberty et al., 1999). The aim of this additional step was to take into account the flexibility of the protein side chains that is not considered in the AutoDock 3.0 procedure. Therefore, only the protein backbone was considered as rigid, whereas the ligand and the side chains were fully optimized using the Powell minimizer. Among the lower energy complexes for each domain, three were selected since they have very similar geometrical features and could be compared between the domains.

Conflict of interest statement

None declared.

Acknowledgments

We thank Clare Gough (Laboratoire des Interactions Plantes-Microorganismes [LIPM], Toulouse) and Loïc Faye (Université de Rouen) for helpful discussions and Sylvie Camut (LIPM, Toulouse) for excellent technical assistance. The authors are grateful to Jean-François Arrighi and Clare Gough for sharing the NFP sequence before publication. This work was supported by the European Community’s (EC’s) Human Potential Programme as a Research Training Network on Oligosaccharide Signalling in Plants (contract no. HPRN-CT-2002-00251 [SACC-SIG-NET]) and by a grant from the Agence Nationale pour la Recherche (NT05-4_42720 [NodBindsLysM]). Lonneke Mulder benefited from a Young Researcher’s fellowship on the EC contract.

References

Author notes

2Laboratoire des Interactions Plantes-Microorganismes, INRA-CNRS, BP 52627, 31326 Castanet-Tolosan Cedex, France; and 3Centre de Recherches sur les Macromolécules Végétales, CNRS (affiliated with Université Joseph Fourier), 601 rue de la Chimie, BP 53, 38041 Grenoble Cedex 9, France