Lectin–sugar interaction Calculated versus experimental binding energies Dirk Neumann 1 , Oliver Kohlbacher 2 , Hans-Peter Lenhof 2 and Claus-Michael Lehr 1 1 Department of Biopharmaceutics and Pharmaceutical Technology, and 2 Center for Bioinformatics, Saarland University, Saarbru ¨ cken, Germany Although a steadily increasing number of protein–ligand docking experiments have been performed successfully, there are only few studies concerning protein–sugar inter- actions. In t his study, w e investigate t he interaction o f w heat germ agglutinin (WGA) w ith N-acetylglucosamine and a number of its derivatives and predict the binding free ener- gies using fl exible docking techniques. To assess the quality of our predictions, we also determined those binding free energies exper imentally in cell-binding studies. The predicted binding site, ligand orientation, and details of the binding mode are i n p erfect agreem ent with t he known crystal structure of WGA with a sialoglycopeptide. Furthermore, we obtained a n excellent linear correlation of our predicted binding free energies with both our own data and experi- mental data from the literature [ Monsigny, M ., Roche, A.C., Sene, C., Maget Dana, R. & Delmotte, F. (1980) Eur. J. Biochem. 104, 147–153.]. In both cases, predicted energies were within 1.0 k JÆmol )1 of the experimental value. These results illustrate the usefulness o f docking-based method s for the qualitative and quantitative prediction of protein–car- bohydrate interactions. The insights gained f rom such the- oretical studies may b e used to complement t he results from the still s carce crystal s tructures. Keywords: protein–carbohydrate interactions; flexible docking; wheat germ agglutinin; binding free energy; Caco-2. With the f ast development o f biotechnology a nd genetic engineering in the last decade, the importance o f p roteins f or diagnostic and therapeutical purposes has considerably increased. Lectins b inding s pecifically to carbohydrate residues have been used primarily in cell histology. Due to their high specificity, they offer new options for drug targeting and drug delivery systems [1] and have even been discussed as potential drugs themselves [ 2]. In order to employ lectins successfully in such systems and to overcome possible d rawbacks, s uch as immunogenity or toxicity, a deeper understanding of the sugar–lectin interaction is crucial. Wheat germ agglutinin (WGA) is a well-known lectin specific for N-acetylglucosamine and its derivatives. It has been used in vitro both in transfection studies [3] a nd as a targeting carrier protein f or a c hemotherapeutic p rodrug [4], showing its potential in pharmaceutical applications. T he binding of this lectin to various cell lines [5–7] and ligands [8–12] was investigated by different methods. Despite this multitude of experimental data, there are only few accounts of theoretical predictions of protein– sugar interaction energies. Bradb rook et al.[13]modeled the heat of formation of the protein–sugar interaction by molecular d ynamics s imulations, while Liang et al.[14]and Pathiaseril & Woods [15] used free-energy simulations to calculate t he differences in binding free energies. A lthough the results of the free-energy simulations are encouraging, the number of ligands investigated so far was small and the correlation between experimental and calculated data was poor. To overcome the d rawbacks of most docking and evaluation procedures, which may overestimate the import- ance of outliers, we used a two-step p rocedure for docking. First, we localized the binding site using a global docking and a statistical approach. Then, we performed a second, local docking to determine the correct orientation and exact binding mode. When compared to the known crystal structure o f WGA in complex with a sialoglycopeptide [16], our method correctly predicted the binding site, ligand orientation and all c ontacts relevant for the binding mode. Furthermore, we wanted to compare experimental d ata obtained from in vitro studies with data resulting from docking calculations. If these data correlate r easonably well, the prediction of binding free energies DG bind for similar compounds is possible. The e xtensive knowledge g ained by such investigations is essential when developing carbohy- drate mimetics or lectin mimetics. We obtained a good linear correlation (r ¼ 0.96) between calculated and experimental data, which underlines the usefulness of the atomic models used herein for t he understanding of protein–carbohydrate interactions. MATERIALS AND METHODS Binding energies and constants f or the interaction of WGA and various ligands have been reported p reviously [8–10,12]. The thermodynamic parameter we selected for comparison wastheminimumconcentrationofligandneededtoinhibit WGA induced agglutination of r ed blood cells [17]. T his concentration i s proportional t o the equilibrium constant K for the binding of WGA w ith the respective ligand. Correspondence to D. Neumann, Saarland University, Bldg. 8.1, 66123 Saarbru ¨ cken, Germany. Fax: + 49 0681 302 4677, Tel.: + 49 0681 302 3140, E-mail: ganain@mpi-sb.mpg.de Abbreviations: WGA, wheat germ aggluti nin. (Received 9 October 2 001, rev ised 10 January 2002, accepted 22 January 2002) Eur. J. Biochem. 269, 1518–1524 (2002) Ó FEBS 2002 In addition to the data from literature, we determined the concentration of ligand needed to inhibit the binding of fluorescently labeled WGA (WGA*) to Caco-2 BBe cell monolayers by 50% ( IC 50 ). The Caco-2 BBe cloned cell line was purch ased from the A merican Type Culture C ollection (ATCC), Manassas, Virginia, USA. The rhodamine labele d WGA was obtained from Linaris, Wertheim-Bettingen, Germany. All ligands were from Sigma-Aldrich Chemie GmbH, Schnelldorf, Germany. The cells were cultured in 75-cm 2 flasks (Greiner Labortechnik GmbH, Frickenhau- sen, Germany) at 37 °Cin5%CO 2 in 25 m M Dulbecco’s modified Eagle’s medium (DMEM, Sigma-Aldrich) con- taining 10% fetal bovine serum, 1% nones sential amino acids. Approximately 16 0 00 cells were seeded per well in 96-well plates (Greiner Labortechnik). The medium was changed three times a week. Monolayers were used for experiments five days postconfluence. Inhibition of WGA binding to fixed Caco-2 BBe cells Caco-2 BBe cells were cultured on 96-well plates to 5 days postconfluence. Because the amounts of WGA found in Caco-2 cells increase with time even at temperatu res as low as 4 °C (Fig. 1), we used fi xed c ells. The monolayers were washed with Ca 2+ -free buffer, fixed in buffer c ontaining 2.5% glutaraldehyde and washed again. Solutions contain- ing both 18 lgÆmL )1 fluorescently labeled WGA and inhibitor at varying concentrations were given to the monolayers. The cells were incubated at room temperature for 1 h before washing them again to remove unbound lectin. The remaining fluorescence was measured and plotted v s. the natural logarithm of concentration o f the ligand. The point-of-inflection of a sigmoidal fit of t his c urve gave the concentration needed to in hibit 50% of maximum binding (IC 50 , see Fig. 2). Theoretical study of binding to WGA The binding energies between the flexible ligands and the large acceptor molecule were calculated using AUTODOCK 3.0 [18–20], an atom-based docking simulation program. To accomplish a quick energy evaluation, AUTODOCK precal- culates g rid-based m olecular affinity potentials. For e ach atom type in the substrate molecule, a different three- dimensional g rid map is generated b y assigning the energy of interaction of a single probe atom to the grid point. Similarly, a grid for the electrostatic potential is calculated. The ligand is d efined by a r igid root, from which rotatable bonds sprout. The program features different search methods (simulated annealing, genetic search algorithms, and loc al searc h), from which w e s elected the Lamarckian genetic algorithm. This algorithm starts the docking process with a random population o f a limited number of individ- uals. T hese individuals represent molecules with uniformly distributed random values for torsion angles, quaternions, and translation vectors. The values for torsion angles, quaternion, and translation vectors represent the genes of an individual and can be inherited by the upcoming population. Some of the ligands undergo a local search b efore the energy of each individual i s calculated to determine how many offspring i t will produce i n the following generation. Finally, a two-point crossover and mutations are performed on random members of t he population r esulting in new ligand positions and conformations. Ad ditionally, e litism was used to let some individuals with the best energies survive unchanged into the next g eneration. Ligand preparation For our docking experiments we chose s everal well-known ligands, for which concentrations to inhibit red blood cell agglutination induced by WGA have been reported in literature or are accessible by experiment: a-N-acetyl- neuraminic acid (NeuNAc, CAS: 131-48-6), N-acetyl-b- D - galactosamine (GalNAc, CAS: 1421 5-68-0), N-acetyl-b- D - glucosamine (GlcNAc, CAS: 7512-17-6), methyl-N-acetyl- b- D -glucosamine (Me-GlcNAc, CAS: 3946-01-8), N,N¢- diacetylchitobiose [the b(1–4)-dimer of GlcNAc, CAS: 35061-50-8], p-nitrophenyl-N-acetyl-a- D -galactosaminide Fig. 1. Amounts of WGA* found in Caco-2 BBe cell monolayers in dependence of incubation time and temperature. The cells were incuba- ted with 100 lL buffer containing WGA* at a concentration of 50 lgÆmL )1 . S imilar curves w ere observed for lower a nd higher lectin concentrat ions. Fig. 2. Inhibition of the binding of fluorescently labeled WGA* to Caco-2 BBE cells by GlcNAc. The point-of-inflection corresponds to 50% inhibition (IC 50 ). Ó FEBS 2002 Lectin–sugar interaction (Eur. J. Biochem. 269) 1519 (CAS: 23646-68-6, a-NPhGalNAc), p-nitr ophenyl-N- acetyl-b- D -galactosaminide (CAS: 14948-96-0, b-NPhGal- NAc), p-nitrophenyl-N-acetyl-a- D -glucosaminide (CAS: 10139-02-03, a-NPhGlcNAc) , and p-nitrophenyl-N-acetyl- b- D -glucosaminide (CAS: 3459-18-5, b-NPhGlcNAc). NeuNAc was modeled as the a-anomer, because this form is predominant in solution [ 21] and only this conformation was reported to bind t o WGA [22 ]. Galactose (Gal), which is known not to bin d to WGA, was used as a con trol for our calculations. All ligand structures were built using HYPERCHEM 6.01 [23] assuming pyranosyl ring-conformations. These starting structures were optimized with a series of restricted Hartree–Fock (RHF) quantum-mechanical calculations of increasing accuracy. I n a first step, the ge ometries of the built structures were optimized using the semiempirical PM3 method [24,25]. The resulting structures were then optimized using ab initio methods subsequently employing the small 3-21G basis set [26] and the larger 6-31G* [27] basis set. For the final conformations we calculated 6– 31G* electrostatic potential derived c harges (ESP charges). The semiempirical computations were performed using MOPAC [28], while the ab initio calculations were performed using GAMESS [29]. Receptor The s tructure of WGA was taken from the crystal structure of WGA in complex with a sialoglycopeptide available at the Protein DataBank [30] (PDB accession no. 2CWG). Water a nd ligands were removed from t he crystal structure. To this initial structure hyd rogen atoms were added and their positions optimized using the AMBER force field (parm94) [31]. These calculations were performed with t he HYPERCHEM 6.01 program [23]. In order to simplify the evaluation, the lectin was rotated such that the principal axes of the protein coincided with those of the Cartesian system. Docking In order to m odel hydrogen bon ds, polar and nonpolar hydrogens were u sed f or both ligand and receptor r esulting in two different hydrogen maps. Parameters for pairwise atomic interaction e nergies were taken from the AUTODOCK 3.0 User G uide, while the elec trostatic interaction e nergy was calculated using a sigmoidal distance-dependent dielectric function [32]. The well depths for the original Lennard-Jones, electrostatic, hydrogen bonding, and d es- olvation potential fu nctions were scaled according to Morris et al. [20] in order to improve the accuracy of the prediction of binding free energies. During the docking process, the t orsion angles for ring s ystems were kept fixed, while all other groups were freely rotatable to allow the formation of as many hydrogen bonds as possible. Using the Lamarckian genetic algorithm, each docking run started with a random population of 10 individuals per dihedral. T he number of energy evaluations was 30 000 per individual. All other docking parameters were set accord- ing to Morris et al . [20]. Each docking run resulted in exactly one ligand conformation, which of all generated conformations provided the best interaction energy with WGA. To assess the suitability of AUTODOCK for determining the binding site for carbohydrates, both GlcNAc a nd the N,N¢- diacetylchitobiose (GlcNAc-dimer) were subjected to 1000 docking runs using large grid maps of 200 · 14 5 · 145 points with a grid-point spacing of 0.575 A ˚ .Thesegrid maps were centered on the middle of the wheat germ agglutinin dimer and encompassed the whole l ectin and an additional shell of 20 A ˚ . For the calculation of the binding energies and to elucidate the binding modes, grid maps with 101 · 101 · 101 points a nd a g rid-point spacing o f 0.375 A ˚ were used. The maps were centered on the S BC binding site of WGA. The ligands were subjected to 100 docking runs resulting in 100 conformations. Determination of the binding site To exclude outliers from further evaluation, we gathered statistical i nformation about the most p robable binding site for each ligand. Because the exact orientation in the binding pocket is not crucial information for determining the l ocation of the binding site, only the position of the center of gravity of the six-membered sugar ring was evaluated instead of the positions of all atoms of each docked ligand. Furthermore, the distance from ligand to protein center was neglected, because WGA offers only shallow binding sites, so the ligand i s always met on the surface o f t he protein. This simplification allows switching from three Cartesian coordinates to the two angles o f p olar coordinates with t he protein center serving as the o rigin of the system. The binding site was l ocalized by using a two dimensional Gauss distribution analysis. The frequency of occurrence for a ligand docking to a s pecific site was c alculated by applying a density function to the polar an gles / and h: P B ðu; hÞ¼ 1 N X N i¼1 e ÀðuÀu i Þ 2 þðhÀh i Þ 2 r 2 Here, N is the total n umber of runs, (/ i ,h i ) are the ligand coordinates of a run, and r is the half-width of the peak and was set to 5°. Employing this kind o f density function to a set of docking runs results in a binding site map, where each point (/,h) i n two-dimensional polar angle space is assigned a value P B (/,h), which c an be seen as a likelihood of a binding site being at this position (although it is not normalized). Plotting of P B , for example using contour lines, results in what we call a binding site map ( Fig. 3B). In this kind of map, binding sites clearly show as high peaks. The polar angle coordinates of a peak correspond to the l ocation of the binding site with respect t o the protein center. For the docking of GlcNAc and i ts dimer to WGA, t he highest peaks observed coincide with the positions of the sialic acid residue of the T 5 sialoglycopeptide from t he crystal structure 2CWG [16], which was used for compar- ison (Fig. 3B). The C 2 symmetry shows up well in the binding site map: subtracting 180° from the / coordinate of a peak leads to a peak of s imilar h eight, which corresponds to the symmetry-related binding site. To determine the binding free energies for comparison with experimental data, we c onsidered all docking runs ending within a distance b elow 5° from the h ighest peak 1520 D. Neumann et al. (Eur. J. Biochem. 269) Ó FEBS 2002 observed. Of these conformations from the global-local search, the one with the lowest binding free energy was subjected to 100 local docking runs allowing a maximum translation i n x, y,andz direction of ± 1.5 A ˚ . For a t ypical distance betwee n ligand and lectin center of 2 5 A ˚ , a change of 5° in a polar angle translates to a displacement of the ligand coordinates of approximately 2 .1 A ˚ . RESULTS Experimental IC 50 values for the binding of WGA to fixed Caco-2 BBe cells Using the techniques described i n t he previous section, we determined the IC 50 for GalNAc, GlcNAc, NeuNAc, a-NPhGlcNAc, b-NPhGalNAc, N,N¢-diacetylchitobiose, a-NPhGalNAc, and b-NPhGlcNAc.Wewereableto obtain a good sigmoidal fit for each ligand ( similar t o the one seen in Fig. 2). Table 1 contains the corresponding IC 50 values and their s tandard deviations obtained from t hree repeated experiments. Theoretical prediction of the binding site In order to test the ability of AUTODOCK to locate the binding site, GlcNAc and N,N¢-diacetylchitobiose were subjected to 1000 docking runs in large grid maps covering the whole lectin p lus a shell o f 20 A ˚ . The binding site maps for these two ligands accurately predict all four binding sites of WGA: the four highest peaks coincide with the coordi- nates for the binding sites of the sialoglycopeptide in the crystal structure (see Fig. 3 B). For a more accurate prediction of th e various ligands’ conformations and binding energies, each ligand was docked in a smaller grid centered on the binding site S BC [33], w hich Wright or iginally called t he primary b inding site [22]. We considered only the conformer corresponding to the lowest interaction energy from all i ndividuals produced by the docking runs for further evaluation. This conformer was t hen subjected to a local search, which resulted in more robust energies, especially for the highly flexible ligands N,N¢-diacetylchitobiose and N-acetylneuraminic acid. Binding mode As the binding free energies are highly dependent on the correct prediction of the bound structure of a ligand and an accurate prediction of all interactions involved, we carefully compared the predicted binding mode of N-acetyln euram- inic acid with the known crystal structure [16]. The root mean square deviation of the positions of all heavy atoms of the docked N-acetylneuraminic acid, including the fl exible glycerol side chain, from the c oordinates r eported in the crystal structure is 1.8 A ˚ , which is well within the 2.0 A ˚ resolution of the crystal structure (Fig. 4). All interactions described i n t he original publication by Wright [22] are also present in the predicted s tructure. We then examined and compared the binding character- istics of the various ligands based on t he structures obtained from the best docking results. The b inding of the ligands is primarily mediated by hydrogen bonds formed between C ¼ O and Ser62 II , N-H and Glu115 I ,C 3 -O and Ser43 II , C 4 -OH and Ser43 II and C 4 -O and Ser114 I ,andahydro- phobic interaction of the methyl g roup with Tyr73 II as well Table 1. IC 50 values ± s tandard deviation obtained by inhibiting the b inding of fluorescently labeled W GA to fixed Caco-2 BBe cells by various ligands. All values are giv en in m M . Ligand IC 50 ±SD( M ) Hemagglutination ( M ) a GalNAc 2.1 · 10 )1 ± 0.97 · 10 )1 2.0 · 10 )1 GlcNAc 2.0 · 10 )2 ± 0.23 · 10 )2 1.0 · 10 )2 NeuNAc 1.3 · 10 )2 ± 0.09 · 10 )2 6.0 · 10 )2 a-NPhGlcNAc 5.9 · 10 )4 ± 0.9 · 10 )4 – b-NPhGalNAc 5.3 · 10 )4 ± 2.8 · 10 )4 3.0 · 10 )3 N,N¢-Diacetylchitobiose 4.1 · 10 )4 ± 0.3 · 10 )4 4.0 · 10 )5 a-NPhGalNAc 4.0 · 10 )4 ± 0.4 · 10 )4 1.0 · 10 )3 a From Monsigny et al. [17]. Fig. 3. Comparison of (A) scatter and (B) Gauss plot for the docking of GlcNAc to WGA. (B) T he higher the probability to find a ligand at a certain position, the h igher the observed peak. The highest peaks coincide with the locations of the binding sites of the NeuNAc residue as observed in the crystal structure. The S BC (ÔprimaryÕ)andS AD (ÔsecondaryÕ) bindin g si te s a re located at A (/ ¼ 126°, h ¼ 28°), A¢ (/ ¼ ) 54°, h ¼ 28°)andB(/ ¼ ) 10°, h ¼ ) 53°), B¢ (/ ¼ 170°, h ¼ ) 53°), resp ectively. Ó FEBS 2002 Lectin–sugar interaction (Eur. J. Biochem. 269) 1521 as a stacking o f t he glycopyranoside r ing with Tyr66 II .This stacking is a feature often observed in protein–saccharide interactions [34]. A similar situation is observed for the binding of NeuNAc, for which our calculations suggest a lower binding affinity than for GlcNAc due to the highly charged carboxylic group. In comparison to the crystal structure, the calculations show the six-membered ring deeper in the binding pocket, which a llows for a more pronounced stacking of the NeuNAc w ith the Tyr66 II and establishing an ad ditional hydrogen bond with Glu115 I .All other i nteractions are preserved except f or the hydrophobic interaction between the NeuNAc and Tyr64 II .Inthecaseof GalNA c, the C 4 atom does not establish any distinct hydrogen bonds: the difference in bin ding energy between GalNAc and GlcNAc i s about 1.8 kJÆmol )1 . Nonpolar groups bound to C 1 may establish a hydro- phobic contact with a tyrosine residue (Tyr64 II ). The GlcNAc-dimer and all p-nitrophenyl d erivatives e xert a second stacking with Tyr64 II with a distance between the centers of the ring s ystems of 5.5 A ˚ and 3 .8 A ˚ , r espectively. The GlcNAc-dimer forms additional hydrogen bonds resulting in a difference in binding free energy between GlcNAc and the GlcNAc-dimer of about 3.6kJÆ mol )1 .The sugar rings of t he p-nitrophenyl derivatives are tilted with respect to the positio n of GlcNAc resultin g in variations in the hydrophobic and electrostatic interactions of the p-nitrophenyl group with the tyrosine residue Tyr64 II .The p-nitrophenyl group of a-NPhGalNAc points slightly downwards, which leads to a good stacking with Tyr64 II residue and a favourable electrostatic interaction with the hydroxyl group of Tyr64 II residue. As for t he b-p-nitrophe- nyl derivatives, the aromatic rings are parallel to Tyr64 II , but the p-nitro groups point away from the hydroxyl group exerting a lesser electrostatic interaction w ith the hydroxyl group of this residue than the a-NPhGalNAc, which explains the lower calculated affinity to WGA. The aromatic ring of a-NPhGlcNAc points downward but is off-center, which leads t o a decreased hydrophobic i nterac- tion. In addition, the N O 2 group o f this ligand is even more distant f rom the hydro xyl group of the T yr64 II residue than those of the b-p-nitrophenyl derivatives. The hydrophobic interaction of Me-GlcNAc with the Tyr64 II is less pro- nounced than for the p-nitrophenyl derivatives and there is no additional electrostatic interaction. This results in a calculated binding free energy only slightly lower than that of a-NPhGlcNAc. Binding energies The binding free energies from our docking calculations were plotted vs. the natu ral logarithms of minimum concentration of ligand needed to inhibit red blood cell agglutination b y WGA as reported by Monsigny et al.[17]. The highest binding free energy calculated was )12.47 kJÆmol )1 for the interaction of WGA with Gal, which served as a control. An excellent linear correlation between calculated and experimental data with a slope of 0.66 ± 0.08 k JÆmol )1 ; a correlation coefficient r ¼ 0.96 (Fig. 5) was obtained. Plotting our calculated interaction energies vs. the natural logarithm of the corresponding IC 50 determined by inh ibiting th e binding of WGA* to fixed Caco-2 BBe cells resulted in a linear correlation with a slope of 0.64 ± 0.11 kJ Æmol )1 and a correlation coefficient r ¼ 0.92 (Fig. 6). Closer examination reveals that the deviation fromthelinearfitislargestfor(GlcNAc) 2 . B ecause our prediction for the binding energies of th e other ligand s fits nicely to t he values of Monsigny et al. [17], it s eems likely that the observed IC 50 value for (GlcNAc) 2 is an outlier. The inhibition concentration we should expect to see is very low (approximately 2.0 · 10 )5 molÆL )1 ), so binding to the surface or possibly even t he inside of the cells might cause t he observed increment in inhibitory concentration for this ligand without impacting t he values for the less potent inhibitors. Assuming (GlcNAc) 2 to be an outlier, the correlation coefficient improves to r ¼ 0.97. Except for G lcNAc 2 , a ll data points were within ± 0.6 k JÆmol )1 of the linear fit. Fig. 4. Comparison of crystal structure and docking results for the binding of N-acetylneuraminic acid to WGA. Theligandisrenderedas sticks, the lectin surface is rendered in blue. (A) The NeuNAc acid residue from the c rystal structure of WGA i n complex with a s ialo- glycopeptide is shown. The ligand exhibits a stacking with Tyr 64 II and Tyr66 II . (B ) The docked ligand fits snugly into the b ind ing pocket of WGA. In comparison with the crystal structure, the ligand establishes more hy drogen bonds and a more p ronounced stacking w ith Tyr66 II , while no stac king with T yr64 II is observed. 1522 D. Neumann et al. (Eur. J. Biochem. 269) Ó FEBS 2002 CONCLUSIONS Because our goal is the development of carbohydrate mimetics or lectin mimetics for pharmaceutical purposes, we want to establish a method to correlate theoretical binding free energies with data from in vitro experiments. The fixed Caco-2 BBe cells allow the inve stigation the inhibition of the binding o f WGA to cell s urface receptors and to determine relative binding affinities. The main advantage of fixed cells is that the a mount of WGA i nside the cells does not increase with increasing incubation time; nonetheless, analogous experiments may be e asily carried out using live cells. Ligands with little inhibitory power have to be used in high concentrations, which may eventually lead to considerable volume changes. To overcome this drawback, we used a low lectin concentra- tion, which is not feasible for l ectins with smaller affinities to this cell line, because the fluorescence determination of these bound lectins is very difficult. However, we consider fixed cells a useful tool to determine relative binding energies especially with regard to deve loping lectinomimet- ics for pharmaceutical pur poses. The accuracy of the predicted binding free energies came as a surprise to us because most docking programs still have problems p redicting important contributions to the binding free energy, especially the e ntropic c ontribu- tions. Still, we could show a linear correlation between calculated binding energies and data f rom literature as well as our o wn experiments . It might b e argued, that this linear correlation is based on the entropy–enthalpy compensation, which was reported for the binding of GlcNAc and its oligom ers t o WGA and its B- domain [12, 34]. The entropy-enthalpy compensation leads to a linear relation between binding free energy and heat of binding. This relation is usually interpreted such that hydrogen bonding and van der Waals forces are considered to contribute to the formation of the complex, while rigidification of the ligand and lectin side chains as well as solvation effects have been discussed a s c ounteracting the binding. On the other hand, the parameters of AUTO- DOCK were explicitly modified by Morris et al.[20]to enable the calculation of binding free energies by including solvation parameters, which probably account for a significant fraction of the entropic e ffects. Our calculated difference in binding free energy between GlcNAc and N,N¢-diacetylchitobiose is 3.6 kJÆmol )1 , which is close to the experimental value of 5.9 kJÆmo l )1 as determined by microcalorimetry [12]. The good correlation of experimental and calculated data emphasizes the suitability of the empirical binding free energy function of AUTODOCK 3.0aswellastheatomic approach to calculating interaction energies. Thus, we should be able not only t o reproduce, but also to predict t he relative interaction energies of d iffering small ligands with a lectin. Finally, we presented a new and easy method to determine the most probable docking sites by plotting docking probabilities vs. polar coordinates of t he ligand, allowing quick visualization o f data from docking calcula- tions. These binding site maps offer a much clearer o verview than the commonly used scatter plots and make it much easier to identify the binding sites unambigously. The importance o f outliers is lessened because the evaluation is based on the number of ligands docking to a certain site. Moreover, small ligands may bind in different orientations and conformations into the binding pocket. Usin g only t he coordinates o f the center of gravity, the exact orientation of the ligand is not evaluated, so this evaluation method is superior to clustering procedures based on the root mean square deviation of all atom coo rdinates, when it comes to determining a binding site. In summary, our study shows that theoretical methods based on flexible docking, if carefully applied, can yield interesting insights into the details of the protein– carbohydrate interactions. Although less reliable than ÔhardÕ, experimental d ata, the results of such prediction are a valuable complement to the still rather limited number of crystal structures. Whether the approach is transferable to other lectins as well will be the subject of further studies. Fig. 6. Correlation be tween our calculated energies and our experi- mental IC 50 for i nhibiting the binding of fluorescently la beled wheat germ agglutinin to Caco-2 BBe cells. (GlcNAc) 2 was not in cluded in the fit shown. Fig. 5. Correlation b etween c alculated e nergies and minimum concen- trations needed to inhibit horse red blood cell agglutination induced by wheat germ agglutinin [17]. A linear correlation between binding energies and inhibitory c oncentrations was shown. Ó FEBS 2002 Lectin–sugar interaction (Eur. J. Biochem. 269) 1523 ACKNOWLEDGEMENTS D. N. wishes to thank the ÔLandesgraduiertenfo ¨ rderung des SaarlandesÕ for giving him the opportunity to conduct this research. O. K . was supported by the DFG research cluster ÔInformatikmethoden zur Analyse und Interpretation grosser g enomischer DatenmengenÕ,grant LE952/2-1. REFERENCES 1. Lehr, C.M. (2000) Lectin-mediated drug delivery: the second generation of bio adhe sives. J. Controlled Release 65 , 19–29. 2. Beuth, J., Ko, H.L., Gabius, H.J . & Pulverer, G. (1991) Influence of treatment with the immunomodulatory effective dose of the beta-galactoside-specific lectin from mistletoe on t umor coloniza- tion in BALB/c-mice for two experimental m odel systems. In Vivo 5, 29–32. 3. Yin, W. & Cheng, P W. (1994) L ectin conjugate-directed gene transfer to airway epithelial cells. Bio chem. Biophys. R es. Commun. 205, 826–833. 4.Wirth,M.,Fuchs,A.,Wolf,M.,Ertl,B.&Gabor,F.(1998) Lectin-mediated drug targeting: p reparation, b inding ch aracter- istics and antiproliferative activity of wheat germ agglutinin conjugated doxorubicin in CACO-2 cells. Pharm. Res. 15, 1031– 1037. 5. Aubery, M ., Reynier, M., Lopez, M ., Ogier Denis, E., Font, J. & Bardin, F. (1990) WGA binding to the surface of two a utologou s human melanoma cell lines: different expression of sialyl and b-acetylglucosaminyl residues. Cell. Biol. Int. R eport 14, 275–286. 6. Gabor, F., Stangl, M. & Wirth, M. (1998) Lectin-mediated bioadhesion: binding characteristics of plant lectins on the enterocyte-like cell lines Caco-2, HT-29 and HCT-8. J. Controlled Release 55 , 131–142. 7. Lotan, R., Sharon, N. & M irelman, D. (1975) Interaction o f wheat-germ aggl utinin with bacterial cells and cell-wall polymers. Eur. J. Biochem. 55, 257–262. 8. Nagata, Y. & Burger, M.M. (1974) Wheat germ agglutinin. Molecular characteristics and specificity for sugar binding. J. Biol. Chem. 249 , 3116–3122. 9. Privat,J.P.,Delmotte,F.&Monsigny,M.(1974)Protein–sugar interactions. Association o f beta-(1,4) linked b-acetyl- D -glucosa- mine oligomer derivatives with wheat germ agglutinin (lectin). FEBS Lett. 46, 224–228. 10. Midoux, P ., Grivet, J.P., Delmotte, F . & Monsigny, M. ( 1984) The binding of monosaccharides to wheat germ agglutinin: fluores- cence and NMR investigations. Bioc hem. Biophys. Res. Commun. 119, 603–611. 11. Kronis,K.A.&Carver,J.P.(1985)Thermodynamicsofwheat germ agglutinin–sialyloligosaccharide interactions by proton nuclear magnetic re sonance. Biochemistry 24, 834–840. 12. Bains, G., Lee, R.T., Lee, Y.C. & Freire, E. (199 2) Microcalori- metric study of wheat germ agglutinin binding to b-acetyl- glucosamine a nd its o ligomers. Biochemistry 31 , 12624–12628. 13. Bradbrook, G.M., Forshaw, J.R. & Perez, S. (2000) Structure/ thermodynamics relationships of lectin-saccharide complexes: the E rythrina corallodendron case. Eur. J. Biochem. 267, 4545– 4555. 14. Liang, G., Schmidt, R.K., YuH A., Cumming, D.A. & B rady, J.W. (1996) Free energy simulations studies o f the b inding specifity of mannose-bind ing protein. J . Phys. Chem. 100, 2 528–2534. 15. Pathiaseril, A. & Woods, R.J. (2000) Relative energies of binding for a ntibody – carbohydrate-antigen complexes computed from free-energy simulations. J. Am. Chem. Soc. 12 2, 331–338. 16. Wright, C.S. & Jaeger, J. (1993) Crystallographic r efinement and structure analysis of the complex of whe at germ agglutinin with a bivalent sialoglycopeptide from glycophorin A. J. Mol. Biol. 232, 620–638. 17. Monsigny, M., Roche, A.C., Sene, C., Maget Dana, R. & Delmotte, F . (1980) S u gar–lectin i nte raction s: h ow does wheat- germ agglutinin bind sialo glycoconjugates? Eur. J. Bioc hem. 104, 147–153. 18. Goodsell, D.S. & Olson, A.J. (1990) Automated docking of substrates to proteins by sim ulated annealing. Proteins 8, 195–202. 19. Morris, G.M., Goodsell, D.S., Huey, R. & Olson, A.J. (1996) Distributed automated docking of flexible ligands to proteins: parallel applications of AutoDock 2.4. J. Comp. Aided Mol. Des. 10, 293–304. 20. Morris, G.M., Goodsell, D.S. & Halliday, R.S. (1998) Automated docking using a Lamarckian genetic algorithm and an empirical binding free e nergy function. J. Comput. Chem. 19, 1639–1662. 21. Brown, E.B., Brey, W.S. & Weltner, W. Jr (1974) Cell-surface carbohydrates and their interactions. I. NMR of b-acetyl neuraminic acid. Biochim. Biop hys. Acta 399, 124–130. 22. Wright, C.S. (1980) Crystallographic elucidation of the saccharide binding m ode in wheat ge rm agglutinin and i ts biological signifi- cance. J. Mol. Biol. 141, 267–291. 23. Reference w ithdrawn . 24. Stewart, J.J.P. (1989) Optimization of parameters for semi- empirical methods I. Method. J. Comput. Chem. 10, 209–220. 25. Stewart, J.J.P. (1989) Optimization of parameters for semiempir- ical methods II. Applications. J. Comput. Chem. 10, 221–264. 26. Binkley, J.S., Pople, J.A. & Hehre, W.J. (1980) Self-consistent molecular o rbital methods. 21. Small split-valence basis sets for first-row elements. J. Am. C he m. Soc. 102 , 939–947. 27. Hariharan, P.C. & Pople, J.A. (1973) Polarization functions for the 6-31G* and 6-31G ** basis sets. Theoret. Chim. Acta 28, 213–222. 28. Stewart, J.J.P. (1990) MOPAC: a s emiempirical molecular orbital program. J. Com put Aided Mol. Des. 4, 1–105. 29. Schmidt, M.W., Baldridge, K.K., Boatz, J.A., Elbert, S.T., Gordon, M.S., Jensen, J .H., Koseki, S., Matsunaga, N., Nguyen, K.A., Su, S.J., Windus, T.L., Dupuis, M. & Montgomery, J.A. (1993) G eneral atomic and molecular e lectronic structure system. J. Comput. C he m. 14, 1347–1363. 30. Berman, H.M., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T.N., Weissig, H., Shindyalov, I.N. & Bourne, P .E. (2000) The Protein Data Bank. Nucleic Acids Res. 28, 235–242. 31. Cornell, W.D., Cieplak, P., Bayly, C.I., Gould, I.R., Merz, K.M., Ferguson, D.M., Spellmeyer, D.C., Fox, T., Caldwell, J.W. & Kollman, P.A. (1995) A second generation force field for the simulation of proteins, nucleic acids, and organic molecules. J. Am. Che m. Soc. 117 , 5179–5197. 32. Mehler, E.L. & Solmajer, T. (1991) Electrostatic effects in proteins: comparison of dielectric and charge models. Protein Eng. 4, 903–910. 33. Kronis,K.A.&Carver,J.P.(1985)Wheatgermagglutinindimers bind sialyloligosaccharide s at four sites i n solution: p roton nuclear magnetic resonance t emperature studies at 360 MHz. Biochemistry 24, 826–833. 34. Espinosa, J.F., Asensio, J.L., G arcia, J.L., Laynez, J ., Bruix, M., Wright, C ., Siebert, H.C., Gabius, H.J., Canada, F.J. & Jimenez Barbero, J. (2000) N MR investigations of protein–carbohydrate interactions binding studies an d refined three-dimensional solution structure of the complex between the B domain of wheat germ agglutinin and N,N¢,N¢-triacetylchitotriose. Eur. J. Biochem. 267 , 3965–3978. 1524 D. Neumann et al. (Eur. J. Biochem. 269) Ó FEBS 2002 . electrostatic interaction. This results in a calculated binding free energy only slightly lower than that of a-NPhGlcNAc. Binding energies The binding free energies. more robust energies, especially for the highly flexible ligands N,N¢-diacetylchitobiose and N-acetylneuraminic acid. Binding mode As the binding free energies