Biochemical and structural characterization of a mannose binding jacalin-related lectin with two-sugar binding sites from pineapple (Ananas comosus) stem

A mannose binding jacalin-related lectin from Ananas comosus stem (AcmJRL) was purified and biochemically characterized. This lectin is homogeneous according to native, SDS-PAGE and N-terminal sequencing and the theoretical molecular mass was confirmed by ESI-Q-TOF-MS. AcmJRL was found homodimeric in solution by size-exclusion chromatography. Rat erythrocytes are agglutinated by AcmJRL while no agglutination activity is detected against rabbit and sheep erythrocytes. Hemagglutination activity was found more strongly inhibited by mannooligomannosides than by D-mannose. The carbohydrate-binding specificity of AcmJRL was determined in some detail by isothermal titration calorimetry. All sugars tested were found to bind with low affinity to AcmJRL, with Ka values in the mM range. In agreement with hemagglutination assays, the affinity increased from D-mannose to di-, tri- and penta-mannooligosaccharides. Moreover, the X-ray crystal structure of AcmJRL was obtained in an apo form as well as in complex with D-mannose and methyl-α-D-mannopyranoside, revealing two carbohydrate-binding sites per monomer similar to the banana lectin BanLec. The absence of a wall separating the two binding sites, the conformation of β7β8 loop and the hemagglutinating activity are reminiscent of the BanLec His84Thr mutant, which presents a strong anti-HIV activity in absence of mitogenic activity.


Results and Discussion
Purification of A. comosus lectin. As shown in Fig. 1a, the affinity column of D-mannose-Agarose separated the total protein soluble fraction from A. comosus stem into two fractions, a flow-through fraction and a retained fraction. The flow-through fraction contained mainly the cysteine proteases as revealed by their apparent molecular weight of 24 kDa (Fig. 1b, line 3 of Inset 1) and the protease activity measured after reduction of their catalytic cysteine residues by dithiolthreitol. The retained fraction was eluted by D-mannose and contained molecular species with apparent molecular weight of 14 kDa (Fig. 1b, lane 4 of Inset 1). N-terminal sequencing up to the 7 th amino acid (SGLVGLG) clearly identified the protein band as jacalin-like lectin from A. comosus (SwissProt entry: Q53J09_ANACO), although the N-terminal methionine residue documented in the SwissProt sequence was not detected. To further improve its purity, the retained fraction was subjected to FPLC-gel filtration chromatography. The sharp and symmetrical peak obtained after this step (Fig. 1b) suggested a homogenous protein preparation. In agreement, both native (Fig. 1b, Inset 2) and SDS-PAGE experiments (Fig. 1b, lane 5 of Inset 1) reflected the high purity and homogeneity of A. comosus lectin. N-terminal sequencing also argued for protein homogeneity as only one sequence was obtained. Besides, accurate molecular mass determination by ESI-Q-TOF-MS revealed one major peak of 15389.0 ± 1.5 Da (Supplementary Fig. 1). This value should be compared Lectins are known to undergo oligomerization in solution. The molecular mass of native A. comosus mannose jacalin-related lectin (AcmJRL), as estimated using FPLC-gel filtration chromatography, was ~30 kDa, suggesting that AcmJRL was a homodimer in solution. To check if the AcmJRL was a glycoprotein, it was subjected to carbohydrate analysis by periodic acid oxidation. The results, shown in lines 3-6 of Fig. 1c,d, clearly demonstrated that AcmJRL was not glycosylated, as also indicated by the experimental mass determination. AcmJRL resembles thus other mannose specific jacalin-related lectins that were not glycosylated, such as artocarpin 40 and heltuba 41 , while the galactose specific lectin jacalin 40 was glycosylated.
Hemagglutinating activity. Hemagglutinating assays revealed that AcmJRL showed no hemagglutination activity of rabbit and sheep erythrocytes, in the absence or presence of 50 mM of the divalent ions Mg 2+ , Mn 2+ Zn 2+ and Ca 2+ . In contrast, AcmJRL readily hemagglutinates rat erythrocytes ( Supplementary Fig. 2) with a specific hemagglutinating activity of 62.5 µg/mL, a value higher compared with that (8.9 µg/mL) of the lectin purified from Chinese leek seeds 42 and the mannose specific lectin purified from mulberry seeds 43 . Other mannose binding lectins were reported to be unable to hemagglutinate sheep erythrocytes 44,45 , but the striking difference in the hemagglutinating activity of AcmJRL compared to the majority of mannose binding lectins described to date is the inability of AcmJRL to hemagglutinate rabbit erythrocytes 21,41,46,47 . However, comparing agglutination activity of lectins between laboratories has to be taken with caution, as the results can be highly dependent on the erythrocyte preparation and experimental protocol.
Lectin oligomerization was described as a possible factor enhancing both affinity and selectivity by generating multiple sites for carbohydrate binding 48,49 . Banerjee et al. 50 reported a homodimeric lectin from Allium sativum, ASAL, which loses its hemagglutinating property when engineered to a stable monomeric form of the lectin, mASAL. These authors suggested that the loss of the hemagglutination activity results from a smaller number of carbohydrate binding sites in the monomeric lectin form that consequently causes a loss of the multivalency essential for hemagglutination. Size-exclusion chromatography clearly showed that AcmJRL is a homodimeric lectin in solution. The absence of hemagglutinating activity of rabbit erythrocytes cannot thus be explained by the absence of oligomerization. Furthermore, many mannose specific lectins displaying hemagglutination activity against rabbit erythrocytes are described as dimeric in solution 21,41,45,46,51 .
The results of hemagglutination inhibition assays of AcmJRL (62.5 µg/mL final concentration) indicated that the mannopentaose αMan(1,3) (αMan(1,6))-αMan(1,6) (αMan(1,3))Man exerted the highest inhibitory effect, with a MIC value of 0.32 mM followed by the mannotriose αMan(1,3) (αMan(1,6))Man (0.47 mM) and by the dimannose Man-α-1,3-Man (0.94 mM). Furthermore, minimum inhibitory concentration required for both methyl-α-D-mannopyranoside and methyl-α-D-glucopyranoside was 1.56 mM while for D-glucose and GlcNAc the MIC value was 3.12 mM. D-mannose presented the lowest inhibitory effect, with a MIC value of 6.25 mM. These data (summarized in Supplementary Table 1) demonstrate that AcmJRL displays preference for oligosaccharides compared to monosaccharides. Other mannose specific lectins exhibited similar behavior. These are, e.g., the lectins from Phlebodium aureum rhizomes and Centrolobium microchaete seeds and heltuba 19,47,52 . Equilibrium unfolding experiments. The coincidence of the curves (Fig. 2) obtained by intrinsic fluorescence and Far-UV CD measurements, in the presence of various concentrations of guanidinium chloride (GdmCl), indicates that secondary and tertiary structures are destabilized simultaneously. These results suggest that the protein unfolds in an apparent two-state transition (N 2 ⇌ 2U), where only the native dimer and the unfolded monomer are significantly populated. Analysis of the data in Fig. 2, according to a two-state model for a homodimeric protein, allowed computation of the characteristic thermodynamic parameters, i.e. (ΔG°(H 2 O) N U = 59.1 ± 0.1 kJ·mol −1 ) and m NU = −9.4 ± 0.3 kJ·mol −1 ·M −1 (for further details, see Table 1). Similar results were  71 , and the solid line was drawn using the mean values of the parameters obtained by both techniques (Table 1). Data are presented as the fractional change in signal as a function of GdmCl concentration.

Isothermal titration calorimetry (ITC).
In order to study the interactions of AcmJRL with potential ligands, the lectin was titrated with mono-and oligosaccharides, using ITC ( Supplementary Fig. 3). This system records the heat generated by association of the binder with its ligand and following progressive saturation, the binding enthalpy, the affinity constant and the stoichiometry are derived from a Wiseman plot (Fig. 3a, lower panel). The various sugars used here were selected from those already investigated for the mannose binding jacalin-related lectin artocarpin 27,54 . All sugars tested were found to bind with low affinity to AcmJRL, with  dissociation constants in the mM range. For such low affinity systems, full saturation by the ligand cannot be reached because of technical limitations. Accordingly, only a partial binding isotherm is recorded (Fig. 3a) and one of the fitted variables has to be kept constant, usually the binding stoichiometry 55 . In the present case, on the basis of two mannose molecules bound to the X-ray structure of AcmJRL, the thermodynamic parameters of sugar binding were calculated for a fixed stoichiometry n = 2. Furthermore, the fitted binding enthalpy ΔH°b was checked independently by direct injections of AcmJRL aliquots into a saturating concentration of saccharide (20 × Kd) as described 56 ( Supplementary Fig. 4). Both fitted and experimental enthalpies were found to agree within 5-10%. AcmJRL does not bind galactose and it is therefore not related to galactose-binding lectins. In contrast, as shown in Table 2, the association constant K a for the various sugars follows the same trend than that reported for artocarpin 57 , with a weak affinity for monosaccharides and increasing affinity from mannose to di-, tri-and penta-mannooligosaccharides. Data from Table 2 showed that AcmJRL binds to the mannopentaose with a 10-fold higher affinity compared to mannose. The same behavior was observed with the hemagglutination inhibition assays. Interestingly, AcmJRL differs by its ability to bind GlcNAc, in contrast to artocarpin. However, the K a values for AcmJRL are one order of magnitude lower than those reported for artocarpin and suggest that the tested sugars are potentially not the physiological ligands of AcmJRL. The large fitting errors on K a for mannooligosaccharides may also indicate that both observed binding sites in the crystal structure are not fully equivalent for sugar binding. Fitting the ITC data with n = 1 reduces the error on K a but provides binding enthalpy values much higher than those reported for lectins (Supplementary Table 2). This again indicates that two sites contribute to saccharide binding, although differently. Bridging between the two binding sites by a branched glycan has been recently reported for JRLs 58,59 . A similar bridging in AcmJRL with mannopentaose will be discussed thanks to the structures presented below. ITC data cannot discriminate such binding mode. The enthalpic and entropic contributions to ΔG°b show that the association is enthalpy-driven by favorable interactions such as H-bonds or van der Waals contacts 60 . Furthermore, these parameters exhibit strong enthalpy-entropy compensation (Fig. 3b) characterized by a linear relationship. Such behavior primarily indicates that sugar recognition and binding by AcmJRL occur via the same mechanism for the tested saccharides. The compensation phenomenon is expected to originate from increased bonding (more negative ΔH°b) at the expense of increased order (more negative TΔS°b as shown by monosaccharides), whereas solvent reorganization at the binding surfaces tend to render the entropic term more favorable, as shown for mannooligosaccharides 60 . The majority of mannose binding jacalin-related lectins has been described to occur as dimeric and tetrameric forms in solution 15,38,41,61 , though a monomeric form was also reported 62 . This diversity in quaternary structures highlights the important role of these oligomeric assemblies for carbohydrate recognition events, both in term of affinity and specificity 29,48 . Interestingly AcmJRL showed more comparable but still lower saccharide binding affinity when compared to two mJRLs from banana (BanLec and plantain) that were also described as dimeric forms in solution 63 .
Overall structure of AcmJRL. The crystal structure of AcmJRL was determined to 1.8 Å resolution using as search model the banana lectin (pdb code 1 × 1 V) 31 , the structure from the Protein Data Bank with the highest sequence identity (42%). This structure belongs to the P2 1 2 1 2 1 space group and is characterized by R work and R free of 14.6% and 18.3% respectively (see Table 3 for additional model refinement statistics). The asymmetric unit contains four monomers. Each monomer starts at Ser2 and ends at Tyr145, with a well-defined carboxy terminal group, in agreement with the sequence reported in the UniProtKB database (Q53J09_ANACO) except for the absence of the N-terminal methionine also not detected by N-terminal sequencing.
AcmJRL adopt the expected β-prism fold (Fig. 4a,b) with the three characteristic Greek key four-stranded β-sheets (β1-2 + β11-12, β3-6 and β7-10) and all the connecting loops well defined in the electron density. Two densities were identified and attributed to citrate molecules. One is located in the carbohydrate-binding site 2 of monomer D and the second interacts with the β7β8 and β9β10 loops (Greek key 3) of monomer B.
The most likely biological assembly identified by the PDBePISA server (http://www.ebi.ac.uk/pdbe/pisa/) corresponds to the four monomers present in the asymmetric unit. Monomers A and C form a tail-to-tail interaction similar to monomers B and D and the AC and BD dimers are aligned side-by-side with an approximate angle of 45° ( Fig. 4c and d). The strongest interactions are between monomers A and B, and C and D; they mostly involve strands β1 and β10 (Fig. 5) with an average buried surface of 826 Å 2 and ΔG of −11.5 kcal/mol as calculated by PISA. On the other hand the A-C and B-D interactions have an average buried surface of 695 Å 2 and ΔG of about −6.2 kcal/mol. This tetrameric form of AcmJRL differs from the dimeric form observed experimentally by size exclusion chromatography. The main cause of this discrepancy is likely the high protein concentration used for  Outlier regions (%) 0.5 0.7 0.7 Table 3. Crystallographic data and model refinement statistics. a Values in parenthesis refer to the highest resolution shell). The same feature was observed for BanLec, which forms dimers in solution but is also found as a tetramer in the crystal structure 31 . Despite this similarity, the quaternary structures of AcmJRL and BanLec are only partially similar. The side-by-side interaction involving strands β1 and β10 is indeed conserved with some amino acids substitution (V5I, L7V, V117L, V119L and S121A) approximately keeping the size and hydrophobic characteristic that drives this interaction (Fig. 5). The tail-to-tail interaction is however not conserved and generates a different tetramer than the one observed in BanLec. The same is also true for artocarpin, jacalin and the mJRLs from Morus nigra and Maclura prolifera 27,33,64,65 , who all share the same side-by-side dimerization as AcmJRL, but have a common tail-to-tail interaction different than AcmJRL and BanLec. The mJRL from Helianthus tuberosus (heltuba) 23 forms donuts shaped octamers made of four basic dimers similar to AcmJRL, and the mJRL from Calystegia sepium (calsepa) 66 only forms dimers different than the AcmJRL ones (Fig. 5). Finally, for the mJRL from Ipomoea batatas (Ipomoelin), the dimer is similar to calsepa, but the tetramer is stabilized by the swapping of a 10 amino acids N-terminal extension 67 .
Saccharide binding sites. Two structures of AcmJRL in complex with D-mannose (AcmJRL:Man) and methyl-α-D-mannopyranoside (AcmJRL:MeαMan) were obtained at 2.75 Å and 1.9 Å respectively. These structures belong to the P6 4 22 space group but with significantly different c axes (257.3 Å and 204.4 Å respectively) indicating different packing. In both cases, the asymmetric unit contains two molecules. They form the side-by-side dimer in AcmJRL:Man and the tail-to-tail dimer in AcmJRL:MeαMan. In either case, the tetramer observed in the AcmJRL structure can be reconstructed using symmetry mates, the interactions between tetramers are different in the two crystals. The overall structures of the AcmJRL:Man and AcmJRL:MeαMan complexes are similar to the AcmJrl structure, with rms deviation between carbon α of the four monomers of 0.75 Å and 0.7 Å respectively.
In the AcmJRL:Man complex, four extra densities were interpreted as D-mannose molecules, two at the same positions in each monomer. One interacts with Greek key 1 (site 1) and one with Greek key 2 (site 2) (Figs 5 and  6a). In site 1, D-mannose oxygens are within hydrogen binding distance of the two carboxyl oxygens of Asp136 (O4 and O6), the hydroxyl of Ser133 (O5), and the backbone nitrogen of Gly15 (O3), Ser133 (O5) and Leu134 (O6). In site 2, the binding mode is similar to site 1, and the D-mannose oxygens interact similarly with the two carboxyl oxygens of Asp39 (O4 and O6) and the backbone nitrogen of Gly62 (O3), His36 (O5) and Ala37 (O6).
In the AcmJRL:MeαMan complex, a methyl-α-D-mannopyranoside molecule was modeled in each of the two binding-sites 1. The interactions with AcmJRL were all conserved except the hydrogen bond between O5 and the hydroxyl group of Ser133, which rotated 120° as a result of steric hindrance with the methyl group of this ligand (Fig. 6b). For the binding site 2, a methyl-α-D-mannopyranoside molecule was only modeled in monomer A. The hydrogen bonds between the ligands and the protein in the AcmJRL:Man complex where conserved in the AcmJRL:MeαMan complex, despite a slight rotation of the ligand potentially due to the close proximity of the ligand methyl and the side chain of His36. An extra density is also present in the binding site 2 of monomer B but the quality was not sufficient to properly fit the ligand. This is likely a consequence of an insufficient occupation of this site due to a steric hindrance with the methyl-α-D-mannopyranoside bound in the binding site 1 of a symmetric molecule in the crystal.
Comparison of binding sites of the different mJRLs. The structure of AcmJRL is the second structure of a mJRL reported with two carbohydrate-binding sites per monomer after BanLec. The binding mode of D-mannose and methyl-α-D-mannopyranoside in the site 1 is similar to the one observed in the other structures of complexes between mJRLs (BanLec, artocarpin, Morus nigra mJRL and heltuba) and D-mannose or mannose-containing oligosaccharides with only subtle shifts of the β1-2 and β11-12 loops, which contribute to  1 Figure 5. Multiple sequence alignment of relevant mJRLs. The alignment of AcmJRL, Banlec, artocarpin, heltuba and calsepa was obtained using Clustal W and improved thanks to the superimposition of the corresponding structures. Arrows above the alignment represent the β-strands of AcmJRL. The residues involved in D-mannose binding in site 1 and 2 are highlighted in yellow and orange, respectively. Residues binding the 3 rd D-mannose in Banlec are highlighted in green. Residues involved in side-by-side dimerization and higher order oligomerization (octamer for heltuba and tetramer for others) are highlighted in grey and blue respectively. The four residues forming the "wall" between the binding sites of AcmJRL and Banlec are showed in red. the binding 23,27,64,68 . In the case of the mannose-binding site 2 (Fig. 6c), the binding mode is also similar to the one observed in BanLec, but the shift in the β3-4 and β5-6 loops is slightly more pronounced (about 1.2 Å). In some BanLec structures (pdb codes 3MIU and 2BMY 31,68 ), a third saccharide is detected close to Ser33, Asn106 and His63 (between Greek key 2 and 3). This binding site has not been discussed in the related publications and is not conserved in AcmJRL.

Banlec
The affinity of AcmJRL for D-glucose and D-mannose is lower than the one reported for artocarpin and BanLec, despite a similar binding mode (same hydrogen bond network). However, except for the central aspartate, the atoms involved in these bonds are backbone nitrogens and variations in the identity of the side chains in the different mJRLs could affect the binding. Indeed, AcmJRL is characterized by a glutamate at position 14 in binding site 1 instead of a glycine in all other structures. In the AcmJRL:Man and AcmJRL:MeαMan complexes, the distance between the oxygen O4 of the ligand and Cβ of Glu14 is only 3.4 Å (Fig. 6c). This represents an unfavorable contact that could be responsible for the slightly lower affinity of AcmJRL for D-mannose and mannose-containing oligosaccharides. This glutamate is also likely to interfere with the binding in the site 1 of oligosaccharides containing a β1,3 link as observed in the structures of BanLec in complex with laminaribose and Xyl-β1,3-Man-α-O-methyl 31 . In binding site 2, no interference would however occur, making it difficult to evaluate the effect on the overall affinity of AcmJRL to reducing 3-O-β-linked sugar molecules. The physiological implication of this substitution remains also to be determined. Bridging between the two binding sites. Bridging between the two binding sites by a branched glycan has been reported for Banlec and other lectins 58,59 . Because of the similarity with Banlec, AcmJRL is expected to allow a similar bridging between its two binding sites. Mannopentaose is the largest oligosaccharide used in this study; it is made of three branches, each one being terminated by a D-mannose linked by its C1 carbon. A maximum of two mannose units are present in between these terminal saccharides, equivalent to a distance of maximum 10.5 Å between the oxygen atoms of the terminal links. In the complex structures of AcmJRL, the distance between the O1 oxygens of the two binding sites is about 14 Å, making it impossible for mannopentaose to be bound in both site in the conformations observed in the AcmJRL:Man complex structures. However, we cannot rule out that one mannose binds one site in the canonical conformation and a second terminal mannose interacts with the second binding site in a different conformation with lower affinity but still contributing to the overall binding and preventing the binding of a second molecule.
Potential therapeutic use of AcmJRL as antiviral similar to BanLec? BanLec has been shown to be a potent T cell mitogen and to avidly associate with high-mannose-type N-glycans on the envelope of HIV-1, hepatitis C virus, and influenza virus, thus blocking viral entry into cells 38 . Swanson et al. 69 were able to engineer a single amino acid mutation (His84Thr) that maintains the antiviral activity of BanLec while dramatically reducing its mitogenic activity, which constitutes a potential side effect preventing potential clinical uses of lectins. This mutation is localized in the β7β8 loop (amino acids 83 to 86) of the third Greek key and is adjacent to both carbohydrate-binding sites (Fig. 6d). The effect of this mutation was linked to the disruption of a wall separating the two binding sites, reducing multivalent interactions that drive the mitogenic activity, while keeping a loop conformation appropriate for the high binding affinity for numerous viruses.
We found that AcmJRL is closely related to BanLec. It has 42% amino acid sequence identity and is a rare mJRL with two very similar carbohydrate-binding sites. In AcmJRL, the β7β8 loop maintains a conformation very similar to the BanLec one (Fig. 6d) despite being one amino acid longer and having no sequence identity ( 85 LGAVTNL 91 in AcmJRL vs 83 YHGAVV 88 in BanLec). In BanLec, the wall between carbohydrate binding sites is mainly constituted of four residues (Tyr83, His84, Lys130 and Phe131) that are all replaced by shorter residues in AcmJRL (Leu85, Gly86, Ser133 and Leu134) (Fig. 6d). In BanLec, the His84Thr mutation induces a size reduction of this wall that is correlated with an increase of the minimal concentration required for rabbit erythrocytes hemagglutination (from 3 to 437 μg/mL) and a decrease of the mitogenic activity. In AcmJRL, the more pronounced reduction of this wall is concomitant with an absence of detection of rabbit erythrocytes hemagglutination up to a concentration of 500 μg/mL, while the affinity for D-mannose is only slightly lower than the one reported for BanLec. If the same correlation as in BanLec exists between hemagglutination and mitogenic activity, the mitogenic activity of AcmJRL would potentially be very low.
Despite the absence of sequence identity between the β7β8 loops of these two proteins, two of the four amino acids facing the carbohydrate binding sites have conserved properties and orientations (Ala86 and Val88 in BanLec vs Val88 and Leu91 in AcmJRL), the two other amino acids being Tyr83 and His84 that contribute to the wall separating the two binding sites. Swanson et al. 69 tested numerous mutations other than a threonine to replace His84, and found that His84Gly is the only one with a reduction of the mitogenic activity while preserving some antiviral potency. Interestingly, in AcmJRL, the residue equivalent to His84 in BanLec is precisely a glycine (Gly86), further suggesting a potential behavior of AcmJRL similar to the His84Thr BanLec mutant. However, two differences potentially preventing a high affinity of AcmJRL for HIV viruses remain, the absence of a residue equivalent to Tyr83 in AcmJRL, and the replacement of Gly14 by a glutamate in the carbohydrate binding site 1 of AcmJRL as discussed above. Therefore, further experimental data are required to assign a potential clinical use of AcmJRL as an antiviral agent against HIV, hepatitis C, and influenza viruses (and/or other viruses) while presenting a low mitogenic activity.

Materials and Methods
Hemagglutination activity and carbohydrate-inhibition assays. Sheep erythrocytes were from Biomeŕieux (France), rabbit erythrocytes were from Complement Technology, Inc. (USA) and rat erythrocytes from blood samples kindly provided by Profs. A.K. Cardozo and L. Ladrière from the ULB Diabetes Research Center (University of Brussels, Faculty of Medicine). Rat experimentation protocols were approved by the Animal Welfare and Ethical Committee of University of Brussels, Faculty of Medicine (CEBEA), protocol number 554N.
Hemagglutinating activity was performed in microtiter plates with V-bottom wells, using the two-fold serial dilution method, starting with a concentration of 1 mg/mL (65 µM) lectin in phosphate-buffered saline containing 150 mM NaCl. Rabbit, sheep and rat erythrocytes were washed six times with 150 mM NaCl and resuspended at a final concentration of 3% (v/v) in 150 mM NaCl. Twenty-five µL of lectin solution were mixed with 25 µL of 3% erythrocytes, serially diluted and incubated at room temperature and 37 °C for 1 h. Hemagglutination activity was also assayed in the presence of 50 mM of the divalent ions Mg 2+ , Mn 2+ Zn 2+ and Ca 2+ . In this case, Tris-HCl 50 mM buffer at pH 7.2 containing 150 mM NaCl was used to avoid precipitation of phosphate compounds. The hemagglutination activity was given as a titer (the reciprocal of the highest two-fold dilution) or expressed as the minimum protein concentration (μg/mL), which produced visible hemagglutination and denoted as Hemagglutinating Unit (HU). Specific activity was defined as the HU per milligram of protein. The carbohydrate-binding specificity was determined by the inhibition of hemagglutination of rat erythrocytes, that were only the hemagglutinated ones, with starting concentrations of 100 mM for D-mannose, methyl-α-D-mannopyranoside and methyl-α-D-glucopyranoside, 50 mM for D-glucose and N-acetyl-D-glucosamine, 15 mM for Manα1,3Man and Manα1,3(Manα1,6)Man and 10 mM for Manα1,3Manα1,6(Manα1,6)Manα1,3. The different saccharides were serially diluted from the starting concentrations. After addition of 25 µL of A. comosus lectin (187.5 µg/mL) to all microplate wells, 25 µL of each saccharide were mixed separately in the first microtiter plate well followed by two-fold serial dilution and pre-incubated Scientific REPoRTS | (2018) 8:11508 | DOI:10.1038/s41598-018-29439-x at room temperature for 1 h. Then 25 µL of 3% mouse erythrocyte suspension were added to each well and the hemagglutination activity was evaluated after a further 1 h incubation period at room temperature and 37 °C. The lowest concentration of saccharides that visibly decreased agglutination was defined as the minimum inhibitory concentration (MIC).

Chemical-induced unfolding transitions.
Chemical-induced unfolding experiments were performed in 50 mM sodium phosphate buffer, pH 7.0. GdmCl solution was prepared in the same buffer, and the pH was adjusted to 7.0 with HCl or NaOH. Unfolding was studied at 20 °C. Samples at various GdmCl concentrations were allowed to equilibrate for at least 12 h (under these conditions, equilibrium is reached throughout the transition). Unfolding transitions were obtained by monitoring the changes in intrinsic fluorescence emission (λ exc = 280 nm; λ em = 370 nm) and circular dichroism (CD) at 222 nm, using a Tecan Infinite M200Pro fluorimeter and a Jasco J-810 spectropolarimeter, respectively, both thermostatically controlled. With all samples, the data were corrected for the contribution of the solution (buffer + denaturant). Denaturant concentrations in the samples were determined from refractive index measurements 70 using a R5000 hand-held throughout. A protein concentration of 0.1 mg/mL (6.5 µM) was used for both intrinsic fluorescence and far-UV CD measurements.
Unfolding curves were analyzed on the basis of a two-state model for a homodimeric protein (N 2 ⇌ 2U), according to equations derived by Bowie and Sauer 71 , and Gittelman and Matthews 72 . This analysis is based on the assumption that the free energy change for unfolding is linearly dependent on the denaturant concentration 73

Isothermal titration calorimetry (ITC). ITC titrations were performed on a MicroCal ITC200
(GE-Malvern) equipped with a 200 µL Hastelloy sample cell and an automated 40 µL glass syringe rotating at 1000 rpm. In order to avoid buffer mismatch and the generation of dilution heats, AcmJRL was first concentrated and exhaustively diafiltrated at 4 °C against Tris-HCl 50 mM, 150 mM NaCl, at pH 7.4 using 5000 MWCO Vivaspin 15R concentrator devices. Then, the diafiltration effluent was used to solubilize the saccharides. Control experiments indicated negligible heat signals for both buffer injections into AcmJRL and saccharide injection into buffer. In standard experiments, AcmJRL (0.5-0.8 mM) was titrated by an initial 0.4 µL injection followed by 19 injections (2 µL) of saccharides (10-50 mM) at an interval of 150 s.
The data so obtained were fitted via nonlinear least squares minimization method to determine the association constant (K a ) and change in enthalpy of binding (ΔH°b) with a fixed binding stoichiometry (n = 1 and n = 2) for n equivalent binding sites, using ORIGIN 7 software v.7 (OriginLab). The Gibbs free energy of binging, ΔG°b, was calculated from K a values and the entropic term, TΔS°b, was derived from the Gibbs-Helmholtz equation using the recorded ΔH°b values.
Crystallization and soaking procedures. AcmJRL was concentrated to 17 mg/mL in a solution containing only water and crystallized at 20 °C using the hanging drop vapor diffusion method. The 4 μL hanging drops consisted of a 1:1 (v/v) mixture of the well solution and the protein solution. The crystal used to get the apo structure was obtained with a well solution composed of 15% (w/v) polyethylene glycol 3000 and 0.1 M sodium citrate buffer at pH 5.5. The crystal was transferred in a cryoprotectant solution containing 50% (v/v) 2-Methyl-2,4-pentanediol and 25% (w/v) polyethylene glycol 6000 before freezing in liquid nitrogen. For the AcmJRL:Man and AcmJRL: αMeMan complexes, the crystals were respectively obtained with well solutions of 4 M sodium nitrate in a 0.1 M Tris-HCl pH 8.5 buffer and 2 M ammonium citrate in a 0.1 M Bis-Tris propane buffer at pH 7.0. Crystal soakings were done in the original crystallization drops for 20 minutes before harvesting by adding 0.5 μL of 0.1 M D-mannose or 0.1 M methyl-α-D-mannopyranoside. In this case, the cryoprotectant solution contained 1.8 M ammonium sulfate and 45% (v/v) glycerol.

Data collection, structures solution and determination.
The diffraction data were collected on the Proxima1 beamline of the Soleil synchrotron. The data were indexed, integrated and scaled using XDS 74 . The crystal used for the apo AcmJRL structure belongs to the orthorhombic space group P2 1 2 1 2 1 , and the crystals of the two complexes to two different hexagonal P6 4 22 space group. The structure of AcmJRL was solved by molecular replacement with Phaser 75 using the structure of the banana lectin as search model (pdb code 1 × 1 V) 31 . The refinement and model building cycles were respectively performed with phenix.refine 76 and Coot 77 . A summary of the relevant statistics of the data collection and refinement is given in Table 3. Protein superimposition was performed using the Secondary Structure Matching algorithm implemented in Coot. The figures were prepared using PyMOL (The PyMOL Molecular Graphics System, Version 1.7.4.3 Enhanced for Mac OS X, Schrödinger, LLC.).
The coordinates and structure factors of the apo AcmJRL, AcmJRL:Man and AcmJRL:MeαMan structures have been deposited in the Protein Data Bank with the PDB ID codes 6FLW, 6FLY and 6FLZ, respectively.
Additional information can be found in the Supplementary Materials and Methods section.
Data availability. The data that support the findings of this study are available from the authors upon reasonable request.