Structure and carbohydrate recognition by the non-mitogenic lectin horcolin

Abstract

Lectins are sugar-binding proteins that have shown considerable promise as antiviral agents because of their ability to interact with envelope glycoproteins present on the surface of viruses such as HIV-1. However, their therapeutic potential has been compromised by their mitogenicity that stimulates uncontrolled division of T-lymphocytes. Horcolin, a member of the jacalin family of lectins, tightly binds the HIV-1 envelope glycoprotein gp120 and neutralizes HIV-1 particles, but is non-mitogenic. In this report, we combine X-ray crystallography and NMR spectroscopy to obtain atomic resolution insights into the structure of horcolin and the molecular basis for its carbohydrate recognition. Each protomer of the horcolin dimer adopts a canonical β-prism I fold with three Greek key motifs and carries two carbohydrate binding sites. The carbohydrate molecule binds in a negatively charged pocket and is stabilized by backbone and side chain hydrogen bonds to conserved residues in the ligand binding loop. NMR titrations reveal a two-site binding mode and equilibrium dissociation constants for the two binding sites determined from 2D lineshape modeling are 4-fold different. Single-binding-site variants of horcolin confirm the dichotomy in binding sites and suggest that there is allosteric communication between the two sites. An analysis of the horcolin structure shows a network of hydrogen bonds linking the two carbohydrate binding sites directly as well as through a secondary binding site, and this coupling between the two sites is expected to assume importance in the interaction of horcolin with high-mannose glycans found on viral envelope glycoproteins.

Introduction

Lectins are a ubiquitous class of proteins found in bacteria, viruses, plants and animals that bind to carbohydrates without enzymatically modifying them (1). Lectins interact with mono and oligosaccharides found on viral and cell surface glycoproteins through their carbohydrate recognition domains, where the lectin-glycan complex is stabilized via hydrogen bonds, van der Waals contacts, hydrophobic interactions and metal coordination linkages (2–5). Considering the importance of protein-carbohydrate interactions in several biological processes such as receptor-mediated endocytosis, immune response, signal transduction, viral replication and cell-cell adhesion, lectins have been extensively explored for their therapeutic use over the past few decades (6–14).

The mannose-specific jacalin-related lectin (mJRL) family recognizes and binds terminal as well as internal mannose residues of N-glycans that are covalently bonded to asparagine residues in glycoproteins. They have a typical β-prism topology containing four-stranded Greek key motifs that form each face of the prism and harbour up to three sugar binding sites per protomer (15–19). Members of the mJRL family such as jacalin and banana lectin (BanLec) associate with high-mannose N-glycans on the HIV envelope glycoprotein and block viral entry (20–23). However, their therapeutic potential has been marred by their potent lymphocyte mitogenic activity (24, 25) that is believed to result from a direct interaction between the lectin and the lymphocyte cell-surface glycoprotein CD4 (26). Such uncontrolled stimulation of mitosis in T-cells not only causes systemic inflammation (25), but may actually result in increased viral transmission.

BanLec has two carbohydrate binding sites per protomer and binds high-mannose glycans in a bidentate mode (22), where both sites engage with the same glycan molecule. The mitogenicity and antiviral activity of BanLec have been decoupled by the H84T mutation, which disrupts π-π stacking between Y83 and H84, making the two glycan binding sites independent and incapable of bidentate ligand binding (27, 28). Though H84T BanLec is not mitogenic, there are conflicting reports on its ability to neutralize HIV particles, with IC₅₀ values varying between 1 and 46 nM (wild type (wt) IC₅₀ = 0.9 nM) (22, 27). It is likely that high mannose chains on the HIV gp120 envelope glycoprotein also need to bind in bidentate fashion for effective inhibition of the virus (29). In support of this hypothesis, tetrameric BanLec molecules, which have eight glycan binding sites, have 5 to 77-fold lower IC₅₀ values against HIV strains compared to dimeric mutants of BanLec, which have only four (22).

Horcolin is a dimeric lectin from Hordeum vulgare (barley) belonging to the mJRL family made up of 146 amino acids in each of its protomers. Isothermal titration calorimetry studies have shown that it exhibits monovalent binding to mannose, mannobiose and mannotriose, while higher mannooligosaccharides with five, seven and nine mannose units interact with horcolin in a bivalent mode (30). The binding between horcolin and HIV gp120 is very tight, with an equilibrium dissociation constant (K_d) of 14.4 nM, and horcolin also has the ability to neutralize HIV-1. Crucially, horcolin does not cause cell-cycling in splenocytes, indicating that it does not have detectable mitogenic activity in contrast to some of its other family members such as jacalin and BanLec (30).

In this report, we provide high resolution crystal structures of apo- and methyl-α-D-mannose-bound horcolin. These structures demonstrate horcolin to be a dimeric protein with a subunit structure resembling other β-prism I fold lectins and carrying two carbohydrate binding sites (CBS1 and CBS2). We then use multidimensional NMR spectroscopy in conjunction with isotope labeling and perdeuteration to probe the thermodynamics and kinetics of mannose recognition. NMR 2D lineshape analysis reveals that the two binding sites in horcolin have four-fold different affinities for mannose. The dichotomy in the ligand binding sites is confirmed by the use of single-binding-site variants of horcolin, which also suggest an element of allostery between CBS1 and CBS2.

Results

The overall structure of apo-horcolin

As the first step towards understanding the non-mitogenic horcolin at the molecular level, we solved its X-ray crystallographic structure in the ligand-free (apo) state at a resolution of 1.16 Å (Fig. 1, Table 1). Apo-horcolin has previously been shown to be a dimer in solution (30). It crystallizes in the P22₁2₁ space group with 12 molecules (or six biological dimers) in the asymmetric unit (ASU). The dimeric arrangement is robust and the six crystallographically independent dimers in each ASU superpose with a root-mean-square deviation (RMSD) in Cα positions ranging from 0.19 - 0.60 Å. Each molecule contains 12 β-strands arranged in the form of three antiparallel β-sheets (Fig. 1A, 1B). These three Greek key motifs come together to form a β-prism I fold with pseudo-three-fold symmetry characteristic of several highly conserved carbohydrate binding proteins (15). The core of this fold is stabilized by interactions between conserved hydrophobic residues (16, 17, 31) such as L96, F116, I126, F129 and I140.

Figure 1. The crystal structure of apo-horcolin.

A) The domain organization of horcolin, showing the three Greek key motifs in green, blue and purple colours and the residues at the dimer interface in orange. B) The subunit structure perpendicular to the pseudo three-fold axis. The three Greek keys are depicted in the same colours as in panel A. C) The dimeric structure of the one of the six crystallographically independent dimers of apo-horcolin (left) and rotated by approximately 90° about a horizontal axis (right). The Greek keys have the same colours as in panels A and B. The C-termini of the protomers for the dimer orientation on the right-hand side are hidden below the magenta Greek keys and their approximate locations are indicated by cyan lines. D) Cartoon representation of the horcolin dimer (grey) with residues burying more than 15 Ų surface area shown as coloured spheres (green: carbon, red: oxygen, blue: nitrogen). Amino acid labels in black and orange refer to residues in different protomers of the horcolin dimer.

Table 1. Data collection and refinement statistics.

	Apo-horcolin	Me-α-Man
Wavelength (Å)	0.99	0.99
Resolution range (Å)	59.8 - 1.16 (1.23 - 1.16)	37.1 - 1.2 (1.2 - 1.2)
Space group	P22121	P3121
Unit cell
a	69.18	42.83
b	94.42	42.83
c (Å)	238.36	139.20
α, β, γ (°)	90, 90, 90	90, 90, 120
Total reflections	2615682 (313746)	91658 (8786)
Unique reflections	513816 (69425)	45961 (4402)
Multiplicity	5.1 (4.5)	2.0 (2.0)
Completeness (%)	96.8 (90.2)	96.74 (94.20)
I/(σI)	8.9 (2.5)	10.47 (2.22)
Wilson B-factor (Å2)	11.8	11.4
R merge	0.09(0.42)	0.04 (0.30)
R meas	0.10(0.47)	0.05 (0.43)
R pim	0.04 (0.20)	0.04 (0.30)
CC1/2	0.99 (0.86)	0.99 (0.78)
CC*	0.99 (0.962)	0.999 (0.938)
Reflections used in refinement	513590 (46914)	45943 (4403)
Reflections used for Rfree	25277 (2324)	2326 (236)
R work	0.19 (0.23)	0.21 (0.28)
R free	0.20 (0.24)	0.23 (0.28)
CC work	0.95 (0.78)	0.95 (0.78)
CC free	0.94 (0.78)	0.95 (0.73)
Ramachandran favoured (%)	98.1	98.6
Ramachandran allowed (%)	1.9	1.4
Ramachandran outliers (%)	0.00	0.00
Average B-factor (Å2)	16.0	15.0

Statistics for the highest-resolution shell are shown in parentheses.

The inter-subunit interactions at the dimer interface originate primarily from residues located in β1 (K3-W10) and β10 (G113-V120) and result in the burial of 1600 Ų surface area (800 Ų per protomer, Fig. 1C,1D). The two β strands β1 and β10 from different subunits of the dimer pack in an antiparallel manner (Fig. 1C). Backbone hydrogen bonds between K6 (CO) and L121 (N), and G8 (N) and P119 (CO), as well as an interesting side chain-backbone hydrogen bond between W10 (Nε1) and R117 (CO), stabilize the dimer interface. The packing of hydrophobic residues such as W10, I114 and L121 at the inter-subunit interface also contributes to the dimerization free energy (Fig. 1D).

Structure of the horcolin-mannose complex and the nature of lectin-sugar interactions

In order to determine the molecular determinants of carbohydrate recognition by horcolin, we co-crystallized horcolin with methyl-α-D-mannose. The methyl group in methyl-α-D-mannose is a passive cap and does not interact with horcolin (see below), methyl-α-D-mannose is hereafter referred to as mannose for simplicity. The horcolin-mannose complex crystallized in the P3₁21 space group with one molecule in the asymmetric unit. Horcolin does not show major conformational changes concomitant to ligand binding (Fig 2A), with an RMSD of 0.215 Å between the Cα atoms of apo and mannose-bound forms. Each subunit of the biological dimer has two mannose binding sites, just as in the case of the homologous banana lectin. The structure of the complex was determined to 1.2 Å resolution and the electron densities for mannose molecules at both the carbohydrate binding sites (CBS1 and CBS2) are well-defined (Fig. 2B).

Figure 2. The crystal structure of methyl-α-D-mannose-bound horcolin.

A) Overlay of the cartoon representations of apo (yellow) and mannose-bound (blue) horcolin. Mannose molecules are shown in sphere representation and coloured according to atom type. B) Electron density for mannose at CBS1 (top) and CBS2 (bottom) in the simulated annealing F_o-F_c omit map. The map is contoured at 5σ level. C) A zoomed-in view of the sugar binding site, showing the elements of the structure important for mannose recognition: LBL1 (red), LBL2 (blue), the GG loops (green) and the secondary binding site (cyan). The conserved D39 and D138 residues whose side chains form hydrogen bonds with the bound mannose are shown as sticks.

The location of the first carbohydrate binding site (CBS1) is common to all lectins in this family and is formed from the conserved residues located on Greek key I, involving two loop regions (GG loop 1 (G11 - G15) and the ligand binding loop 1 (LBL1, G134 - A139)) (Fig. 2C). The second carbohydrate binding site (CBS2) is situated on Greek key II, formed from residues in the second ligand binding loop between strands β3 and β4 (LBL2, residues G35-D39), as well as and the loop linking strands β5 and β6 (GG loop 2 (G60 - T65)). Both ligand binding loops contain the conserved GXXXD sequence that is involved in sugar recognition in lectins belonging to this family (17, 30). Backbone atoms from the conserved G15 residue in the GG loop and from G134, A135 and F136 in LBL1, as well as side chain atoms of D138, interact with mannose in CBS1, while the equivalent residues G64, G35, A36, I37 and D39 establish contacts with the mannose moiety in CBS2 (Fig. 2C,3). The residues lining the binding pockets in CBS1 and CBS2 generate negatively charged depressions that house the mannose residues, with CBS1 having a larger negative electrostatic potential than CBS2 (Fig. 3A). The predominant non-covalent interactions at both binding sites are hydrogen bonds between the lectin and mannose. These hydrogen bonds remain similar at both sites despite the differences in amino acid sequence between the two GXXXD motifs (GAFLD in CBS1 versus GAIVD in CBS2), primarily because most of the hydrogen bonds involve protein backbone atoms (Table 2) (Fig. 3B, 3C). However, there is a water-mediated hydrogen bond observed in the crystal structure of the complex between O3 and O4 atoms of mannose at CBS1 with D90 OD1 that is absent in CBS2 (Fig. 3B).

Figure 3. Interactions mediating sugar recognition by horcolin.

A zoomed-in view of the electrostatic potential surface of mannose-bound horcolin, showing the negatively charged clefts that house the mannose ligands. Protein-sugar interactions in the complex of horcolin with mannose in CBS1 (B) and CBS2 (C), identified using a distance cut-off of 3.5 Å between the two electronegative atoms and an angular constraint of 60°. The main difference between the interactions at the two sites is the water-mediated interaction involving the O atom of Wat976, the side chain of D90 and the hydroxyl groups of mannose (O3 and O4), which exists only in CBS1.

Table 2. Details of the interactions at CBS1 and CBS2 of horcolin with methyl-α-D-mannose.

Mannose-bound horcolin	Site I		Site II
Sugar/water hetero atom	Protein/water hetero atom	Distance (Å)	Protein hetero atom	Distance (Å)
O1	-	-	-	-
O2	-	-	-	-
O3	G15 N	2.8	G64 N	3.0
	Wat976	3.2
O4	G15 N	3.2	G15 N	3.5
	D138 OD2	2.7	D39 OD2	2.6
	Wat976	2.8
O5	A135 N	3.0	A36 N	3.0
O6	G134 N	3.4	G35 N	3.2
	A135 N	3.1	A36 N	3.0
	F136 N	2.9	I37 N	3.0
	F136 O	3.2	I37 O	3.2
	D138 OD1	2.7	D39 OD1	2.7
	D138 OD2	3.5	D39 OD2	3.6
Wat976	D90 OD1	2.9

Structure and dynamics of apo-horcolin in solution

As the next step, we assessed the atomic resolution structural features of apo-horcolin in solution using multidimensional NMR spectroscopy. Figures 4A and S1 show the ¹H-¹⁵N HSQC spectrum of ¹⁵N-labeled apo-horcolin. The chemical shift dispersion of the lectin resonances in both the ¹⁵N and ¹H dimensions is large, confirming that horcolin is folded in solution. Since horcolin is a 30 kDa dimer, perdeuteration was necessary to obtain backbone resonance assignments using a combination of HNCACB, HN(CO)CACB, HNCO and HN(CA)CO (32) 3D datasets acquired on a 600 MHz NMR spectrometer (see Materials and Methods). Chemical shift assignments were further verified using the 3D HNN experiment (33), as well as using i/i+1 sequential H^N-H^α correlations in 3D ¹⁵N-edited NOESY spectra. 130 out of the expected 138 ¹H-¹⁵N correlations were assigned and the overall backbone assignment (including N, H^N, C^α, C^β, CO and H^α atoms) is 91 % complete (Table S1).

Figure 4. The structure of horcolin in solution matches the crystal structure.

A) The ¹H-¹⁵N HSQC spectrum of U-¹⁵N labeled apo-horcolin. The assignments of selected residues present at the lectin-mannose interface are indicated on the spectrum and color-coded as in Figure 2C. B) Chemical shift-derived probabilities of α-helical (grey), β-strand (orange) and random coil (cyan) secondary structures as a function of residue number. C) Backbone ϕ/ψ dihedral angles obtained from the crystal structure (yellow) and from TALOS (red) plotted on a Ramachandran map. Grey dots represent ϕ/ψ angles obtained from high-resolution protein crystal structures (resolution > 1.9 Å). Only TALOS dihedral angle predictions with high confidence are included in this plot. D) A contact map depicting H^N-H^N contacts within 5 Å in the crystal structure (pink, lower triangle) and H^N-H^N NOE cross-peaks (purple, upper triangle) observed in ¹⁵N-edited and H^N-H^N NOESY datasets. E) NMR chemical shift-derived squared order parameters (S²) as a function of residue obtained using TALOS. The secondary structure of horcolin derived from the crystal structure is indicated in green in panels B, D and E.

Backbone chemical shifts are excellent probes of protein secondary structure (34). We used the chemical shifts of horcolin as inputs to the TALOS software package (35) in order to determine the backbone secondary structure of the protein in solution. The residue-specific locations of β-strands (orange bars) and loops (cyan line) predicted by TALOS match very well with the crystal structure, and there is no significant α-helical conformation detected anywhere (grey line) in horcolin (Fig. 4B). TALOS-derived backbone dihedral angles ϕ and ψ (Fig. 4C, red circles) predominantly fall in regions of the Ramachandran map expected for β-sheets (ϕ,ψ: -90°,120°), while a few instances of Type I (ϕ,ψ (i+1): -60°,-30°; ϕ,ψ (i+2): -90°,0°) and Type II (ϕ,ψ (i+1): -60°,120°; ϕ,ψ (i+2): 80°,0°) are also found. The overall distribution of dihedral angles in solution (Fig. 4C, red circles) matches well with ϕ/ψ angles seen in the crystal structure of apo-horcolin (Fig. 4C, yellow circles), confirming that the backbone secondary structure is very similar in the crystalline and solution states.

In order to evaluate whether the overall arrangement of β-strands in solution agrees with the three-Greek key β-prism I fold of the crystal structure, we measured ¹H^N-¹H^N NOEs using ¹⁵N-edited NOESY and ¹H^N-¹H^N NOESY pulse sequences on both protonated and perdeuterated (and back-exchanged) ¹⁵N-labeled horcolin. Figure 4D shows a contact map where the experimentally obtained ¹H^N-¹H^N NOEs are plotted as contacts in the upper triangle. The ¹H^N-¹H^N contacts shorter than 5 Å observed in the crystal structure of apo-horcolin are indicated in the lower triangle for comparison. β-strands appear as anti-diagonal streaks in these plots. The pattern of contacts in the upper and lower triangles agrees very well, showing that the topology of the β-strands, as well as the tertiary structure of horcolin, are retained in solution.

The crystal structure of apo-horcolin shows well-defined electron density for all residues from 3-146, including regions of the loop connecting the 12 β-strands. The average B-factors of apo-horcolin are low (16 Ų), indicating the amplitude of dynamics is low throughout horcolin in its high-resolution structure. In order to characterize the dynamics in solution state, we calculated squared order parameters (S²), which are measures of the amplitude of fast timescale dynamics, using chemical shift information within the TALOS program (36). S² values (Fig. 4E) are uniformly high across the entire sequence of horcolin and range only between 0.68 and 0.95, demonstrating that horcolin displays very little mobility on the fast timescale even in loop regions. The rigid overall structure of horcolin in solution is consistent with the well-defined electron density and low B-factors in the crystalline state. Taken together, the solution NMR data agrees very well with observations from the crystal structure of apo-horcolin, demonstrating that apo-horcolin adopts a rigid 12-stranded β-prism I fold comprised of three Greek key motifs.

Mannose binding to horcolin

The utility of horcolin as a potential antiviral agent relies crucially on its ability to recognize mannose residues present in the complex glycans of glycoproteins present in the envelopes of viruses such as HIV-1. We next used ¹H-¹⁵N HSQC-detected NMR titrations to characterize the binding of mannose to horcolin. Mannose was titrated into a solution containing 0.54 mM horcolin (initial concentration), with final mannose concentrations ranging from 0 - 500 mM. ¹H-¹⁵N HSQC spectra acquired at each mannose concentration showed significant chemical shift changes in a number of resonances including G15, G60, G61 and G64 (Fig. 5A). Interestingly, peaks undergoing large chemical shift perturbations also displayed pronounced exchange broadening, with the intensity decreasing at intermediate mannose concentrations and subsequently increasing again. The HSQC correlation from G64, which shows the largest chemical shift perturbation (0.58 ppm) upon mannose binding, disappears entirely in the sixth HSQC spectrum corresponding to a mannose concentration of 2 mM, while exhibiting intense correlations in spectra at the beginning and the end of the titration. In contrast, resonances from residues such as D38 and D139, which show small chemical shift perturbations (D39: 0.078 ppm, D138: 0.048 ppm), have uniform intensities during the titration. These observations suggest that mannose binding to horcolin occurs at the boundary of the fast-intermediate chemical shift exchange timescale, with the overall exchange broadening determined by the chemical shift difference between the exchanging states, as well as their relative populations.

Figure 5. Two-site binding of mannose to horcolin observed through ¹H-¹⁵N HSQC spectra.

A) Trajectories of selected resonances of horcolin observed in HSQC spectra during a mannose titration. Mannose concentrations range from 0 mM (blue) to 500 mM (pink). The approximate directions along which chemical shifts move are indicated by arrows for residues with significant CSPs. B) Chemical shift perturbations upon addition of mannose, calculated as in Materials and Methods and plotted on the crystal structure of mannose-bound horcolin. The colour scheme used in the plot is indicated alongside. The two bound mannose moieties are shown as spheres that are coloured according to atom type. C) CSPs calculated as in panel B and plotted as black bars as a function of residue. Key mannose-binding motifs are indicated on the plot as coloured rectangles. The secondary structure of horcolin derived from the crystal structure is indicated in green above panel C. D) Regions of the HSQC spectra containing residues G15 and G64, with experimental data shown in cyan and fits in orange contours. Fits were obtained by modeling the experimental data using a sequential binding scheme as described in the main text. Stars represent the fit chemical shifts of P, PL1 and PL12 for each residue and their positions are the same in all four panels.

In order to quantify the effect of mannose binding, we calculated residue-specific chemical shift perturbations using ¹H-¹⁵N HSQC spectra corresponding to the beginning (0 mM mannose) and end (500 mM mannose) of the titration (Fig. 5B,5C). The overall magnitude of chemical shift changes is small, with an average value of 0.061 ± 0.090 ppm over 130 assigned residues. The largest chemical shift differences between the free and bound forms in the ¹⁵N and ¹H dimensions are 2.24 and 0.545 ppm, respectively. Among residues in β-strands, the average CSP is 0.04 ppm, indicating that horcolin undergoes minimal change in secondary structure upon mannose binding, consistent with the close similarity between the crystal structures of apo- and mannose-bound horcolin.

Figures 5B and 5C show the chemical shift perturbations occurring in horcolin after binding mannose. All the perturbations larger than 0.1 ppm are localized to residues present in key mannose-recognition motifs identified from the crystal structures of apo- and mannose-bound horcolin, namely the two ligand-binding loops, the two GG loops and the secondary binding site. In addition, a number of residues, such as G15 (Fig. 5A) and F136 (Fig. S2A), show pronounced curvatures in their chemical shift trajectories through the titration, unequivocally demonstrating the presence of two binding sites for mannose on every horcolin protomer.

Modeling HSQC lineshapes using the Bloch-McConnell equations

NMR lineshapes in the intermediate-fast exchange timescale are exquisitely sensitive to exchange rate constants and chemical shift differences between the exchanging states (37). Lineshape analysis has been extensively used to determine rate constants for conformational interconversion in small molecules (38, 39), as well as to study folding transitions (40, 41) and enzyme-catalyzed Pro cis-trans isomerization (42). Since mannose binding to horcolin occurs in the intermediate-fast regime on the chemical shift timescale, simple fast exchange equations cannot be used to determine accurate dissociation constants (43). In order to quantitatively model the titration data, we used the recently developed TITAN software package (44), which numerically integrates the Bloch-McConnell equations (45) at each titration point (i.e. for each protein and ligand concentration) to directly fit the peak lineshape observed in the corresponding HSQC spectrum.

Regions of the HSQC spectra of horcolin acquired at mannose concentrations ranging from 0 to 500 mM were chosen to include at least one peak with a chemical shift perturbation larger than 0.1 ppm, as well other resonances with intensities overlapping with the peak of interest. A total of 17 residues, including peaks necessary for accounting for overlapped intensity, were selected for data fitting.

Since the crystal structure of mannose-bound horcolin, as well as the curved NMR chemical shift perturbation trajectories confirm the existence of at least two mannose binding sites, we first chose the two-binding-site non-cooperative model (NC), given by the equations

$$P+L\underset{k_{off1}}{\overset{k_{on1}}{\rightleftharpoons }}PL1$$

$$P+L\underset{k_{off2}}{\overset{k_{on2}}{\rightleftharpoons }}PL2$$

$$PL1+L\underset{k_{off2}}{\overset{k_{on2}}{\rightleftharpoons }}PL12$$

$$PL2+L\underset{k_{off1}}{\overset{k_{on1}}{\rightleftharpoons }}PL12$$

for modeling the titration. In the above equations, P is horcolin, L is mannose and the dissociation constants for binding at the two sites are given by:

$$K_{d1}=\frac{k_{off1}}{k_{on1}},$$

$$K_{d2}=\frac{k_{off2}}{k_{on2}}.$$

Global fits of the titration data for 17 residues and 22 mannose concentrations to the NC model did not converge, likely because the chemical shifts of PL1 and PL2 cannot simultaneously be unambiguously located from the NMR spectra. In order to reduce the number of residue-specific parameters, we chose a cluster of four residues that includes G15, G64 for model analysis. Residues in this cluster respond appreciably to both binding events; G64 changes markedly in chemical shift upon binding of the first mannose, while G15 moves significantly when the second mannose binds, making this cluster an ideal choice for data fitting. Global modeling of this cluster gives K_d values of 59 ± 2 mM (K_d1) and 7.7 ± 0.1 mM (K_d2) for CBS1 and CBS2 respectively (Table S2). However, there are small systematic deviations between the spectra and the fits of G15 (Fig. S3). Since G15 is maximally responsive to mannose binding at CBS1, this discrepancy suggests that K_d1 may be somewhat different from the fit value of 59 mM.

We next evaluated whether the two-site sequential binding mechanism was sufficient to model our titration data. This model is defined by the equations:

$$P+L\underset{k_{off1}}{\overset{k_{on1}}{\rightleftharpoons }}PL1$$

$$PL1+L\underset{k_{off2}}{\overset{k_{on2}}{\rightleftharpoons }}PL12$$

While the number of global kinetic parameters in both the sequential and the NC models is four, there are six fewer residue-specific parameters (R_1N, R_1H, R_2N, R_2H, ϖ_N, ϖ_H) per residue because PL2 does not feature in the equations. Here, R_1i, R_2i and ϖ_i are the longitudinal and transverse relaxation rate constants and the chemical shift of nucleus i (i ∈ N,H) in PL2. Since PL2 is not directly visible in the HSQC spectra during the mannose titration, these parameters are poorly defined and removing them from the fitting routine should improve the convergence characteristics.

However, the sequential binding model is an approximation to the two-site NC mechanism and is a good approximation only when the amount of PL2 is small throughout the titration. In order to determine whether the sequential model was appropriate for our data, we simulated the populations of P, PL1, PL2 and PL12 through the titration using the K_d values obtained from fits of the data to the NC model (Fig. S4). The simulations show that the maximum fraction of PL2 during the titration is 0.070, with an average over the entire titration of 0.039, clearly indicating that the contribution of the P ↔ PL2 ↔ PL12 pathway to the formation of doubly-bound horcolin is minimal and justifying the use of a P ↔ PL1 ↔ PL12 sequential model for analyzing the titration data.

HSQC-detected titration data for 17 residues and 22 mannose concentrations were then globally fit to the sequential two-site binding model to extract values of K_d1, k_off1, K_d2, k_off2 and the chemical shifts of P, PL1 and PL12, as detailed in Materials and Methods. A representative overlay of the experimental and fit HSQC data (Figs. 5D and S5) for the region including G15 and G64 demonstrates that the chosen model faithfully reproduces the measured intensities and confirms the robustness of the fitting procedure.

Table 3 lists the K_d and k_off values obtained from the TITAN fitting routine. The robustness of parameters estimated from a ligand binding titration depends crucially on the extent to which the binding sites are saturated with ligand. Accordingly, we calculated the amount of doubly bound horcolin (PL12) and the occupancy of carbohydrate binding sites at the end of the titration from the K_d1 and K_d2 of wt horcolin. The percentage of PL12 is 95 % and the occupancy factor is 98 %, confirming that horcolin is virtually saturated with mannose at the end of the titration. In order to further validate the results from TITAN modeling, we chose F136, which shows significant chemical shift differences upon binding of mannose to both CBS1 and CBS2, but which was left out of the global fitting routine because of extensive peak overlap in adjoining regions of the HSQC spectrum. Using the thermodynamic and kinetic parameters for the horcolin-mannose interaction (Table 3), we simulated the F136 lineshape as a function of mannose concentration (Fig. S2). The simulated data agree well with the experimentally observed chemical shift trajectory, showing the reliability of the global parameters recovered from the fit.

Table 3. Equilibrium dissociation constants (K_d), as well as on- (k_on) and off-rate constants (k_off) for the binding of mannose to wt, D138A and D39A horcolin.

	WT	D138A	D39A
Kd1 (mM)	19.9 ± 0.1	-	55.6 ± 0.2
koff1 (s-1)	(2.4 ± 0.2) x 103	-	(16.4 ± 0.7) x 103
kon1 (M-1s-1)	(1.19 ± 0.08) x 105	-	(2.9 ± 0.1) x 105
Kd2 (mM)	4.71 ± 0.02	3.2 ± 0.3	-
koff2 (s-1)	(2.5 ± 0.1) x 103	(5.0 ± 0.4) x 103	-
kon2 (M-1s-1)	(5.3 ± 0.2) x 105	(1.6 ± 0.1) x 106	-

The binding affinities of the two ligand binding sites for horcolin are both weak and fall in the millimolar range (K_d1 = 19.9 mM and K_d2 = 4.71 mM, Table 3), consistent with previous observations on mannose-binding by most lectins from the mJRL family (4). TITAN fits also provide an estimate of the dissociation rate constant (k_off) values at each binding site, which was used in conjunction with K_d to determine the on-rate constants (k_on). kon values fall between 1 - 6 x 10⁵ M^-1s^-1 for the two sites and are two orders of magnitude smaller than the diffusion-limited rate constant in water (~3 x 10⁸ M^-1s^-1, (46)). This indicates that every collision of mannose with horcolin is not productive and that considerable reorientation may be required before native contacts are established between the hydroxyl moieties in mannose and the binding site residues of horcolin. There may also be transient occlusion of the binding site by the motion of neighbouring loop residues, though this seems less likely given the high order parameter values of the loop regions observed in solution (Fig. 4C).

Interestingly, the binding affinity of horcolin for mannose is different between the two sites, with mannose binding four-fold more tightly to CBS2 (K_d2 = 4.71 mM) than CBS1 (K_d1 = 19.9 mM). This four-fold difference can be extracted reliably from NMR titrations and their modeling using TITAN, as seen from the bootstrapped (47) parameter distribution and χ² analysis (Fig. S6). The difference in K_d values stems almost entirely from the k_on value between the two sites ((1.19 ± 0.08) x 10⁵ M^-1s^-1 for CBS1 vs (5.3 ± 0.2) x 10⁵ M^-1s^-1 for CBS2), while the dissociation off-rate constant is virtually identical. This suggests that the interactions that hold the mannose in the two binding sites are very similar and need comparable activation free energies to be disrupted, but that mannose has more difficulty (slower on-rate constant) in entering and engaging with CBS1 than with CBS2.

In order to dissect the contribution of the two mannose binding events to the overall CSPs seen in wt horcolin (Fig. 5C), we plotted the TITAN-derived chemical shift changes for the first ( P → PL1) and the second ( PL1 → PL12) binding events (Fig. S7). The largest changes in chemical shift during P → PL1 are seen in the second GG loop (G60-T65), as expected from mannose binding to CBS2, while the PL1 → PL12 transition causes significant changes in the first GG loop (G11-G15), which makes contacts with mannose at CBS1.

Finally, a comparison of the parameters obtained from the sequential and NC mechanisms is provided in Table S2. Both models consistently reveal a measurable difference in binding affinities between the two sites on horcolin, with a 4.2-fold higher affinity for CBS2 from the sequential model and a 7.7-fold higher affinity from the NC model. We have chosen to use the results from the sequential model in all subsequent sections because we get good quality fits with a global analysis of all 17 residues, as well as statistics on all affinities and rate constants. The NC model, on the other hand, does not yield fits that are of as good quality, the chemical shifts of PL1 and PL2 vary arbitrarily, and global fits do not converge because of ambiguities in residue-specific chemical shift and relaxation parameters.

Structure-guided mutagenesis for validating the difference in binding affinities

The NMR titrations of horcolin with mannose reveal a dichotomy in the molecular recognition of mannose by the two ligand binding sites in horcolin, with CBS2 displaying a four-fold higher affinity than CBS1. To the best of our knowledge, this is the first time that a difference in dissociation constants in the two binding sites has been characterized within the jacalin family of lectins. While statistical measures including χ² surfaces and bootstrap distributions confirm that the four-fold difference in K_d values is significant, we desired to further validate this result using site-directed mutagenesis. The structure of mannose-bound horcolin shows that the pattern of protein-ligand interactions is very similar at both CBS1 and CBS2, with most of the hydrogen bonds involving backbone donor or acceptor atoms. However, there is one conserved Asp residue in each site (D138 in LBL1 and D39 in LBL2) whose side chain hydrogen bonds with the mannose at that particular site. Accordingly, we chose to mutate D39 and D138 individually to Ala in order to disrupt these side chain-ligand hydrogen bonds and create single-binding-site variants of horcolin.

NMR chemical shift perturbations observed upon mutagenesis are small with an average value of 0.08 ppm (D138A) and 0.13 ppm (D39A) across the entire protein (Fig. S8). The largest changes are localized to the site of the mutation itself and confirm that D39A and D138A horcolin share the same structure as the wt protein. Figures 6A and S9 show ¹H-¹⁵N HSQC-detected NMR titration data of D138A and D39A horcolin, where the mannose concentration is varied from 0 to 167 mM over 18 points for D138A and 0 to 914 mM over 21 points for D39A horcolin. Unlike in wt horcolin, all residues that show chemical shift perturbations upon mannose addition in both mutants trace a linear trajectory from the initial to the final resonance position through the titration, confirming that the stoichiometry of binding is 1:1 in both cases (Fig. 6A and S9). Figures 6B and 6C shows the perturbations in D39A and D138A horcolin when they each bind mannose. As expected from the single-binding-site nature of D39A and D138A horcolin, mannose binding to CBS2 in D138A horcolin does not cause CSPs in LBL1 (Fig. 6B), which directly contacts the mannose bound to CBS1 in wt horcolin. Similarly, CSPs for a number of residues in LBL2 are reduced significantly when mannose binds to CBS1 in D39A horcolin (Fig. 6C).

Figure 6. Single-binding-site variants of horcolin confirm the difference in binding affinities between CBS1 and CBS2.

A) Experimental (left) and fit (right) G61 peaks during the mannose titration of D138A (top row), D39A (middle row) and wt (bottom row) horcolin. Each colour represents a different mannose concentration ranging from blue (0 mM) to pink (wt: 500 mM, D138A: 167 mM and D39A: 914 mM). Residue-specific CSPs of D138A (B) and D39A (C) horcolin plotted as black bars. Important mannose-binding motifs are shown on the plot as coloured rectangles.

In order to determine the binding thermodynamics and kinetics of glycan recognition, we fit selected resonances with CSP values larger than 0.1 ppm to a two-state model

$$P+L\underset{k_{off}}{\overset{k_{on}}{\rightleftharpoons }}PL$$

using the TITAN software package. A total of 16 residues for D39A and 14 residues for D138A horcolin were chosen for data fitting. Representative G61 resonances of wt, D39A and D138A in the mannose titration are shown along with the fits from TITAN, and the fits agree very well with the data (Fig. 6A). Both mutants show narrow K_d and k_off bootstrap distributions and sharp χ² surfaces (Fig. S6), confirming that the equilibrium and dissociation rate constants can be obtained reliably from the mannose titrations. The K_d for D138A horcolin, where binding occurs only at CBS2, is 3.2 ± 0.3 mM (corresponding to 98 % mannose-bound D138A horcolin at the end of the titration), while the K_d of 55.6 ± 0.2 mM for D39A horcolin (94 % bound form at the end of the titration) which retains binding only at CBS1, is almost 20-fold smaller. The k_on values of both single-binding-site mutants differ by 5.3-fold, and this ratio is similar to the k_on values for CBS1 and CBS2 in wt horcolin. However, the k_off values, which are comparable for the two binding sites in the wt protein, vary by a factor of 3.3 in the mutants, increasing the difference in K_d values between CBS1 and CBS2. Taken together, the single-binding-site mutant data thus unequivocally demonstrate that mannose binding at CBS2 occurs more tightly than at CBS1.

Discussion

Horcolin is a mannose-specific lectin from barley that binds HIV glycoproteins with high affinity and shows therapeutic potential as an antiviral agent, as it lacks the mitogenic activity that is a common attribute of most other lectins. In this report, we use a combination of X-ray crystallography and solution NMR spectroscopy to characterize the structure of horcolin and its binding to mannose at atomic resolution.

Horcolin is a dimer in both the apo- and mannose-bound forms and adopts a β-prism I fold composed of three Greek key motifs. Perhaps the best studied mannose-specific β-prism I fold lectin is banana lectin (16, 17, 31) (Fig. 7A), which is evolutionarily close to horcolin and shares 39 % sequence identity with it. A superposition of banana lectin and horcolin subunit structures gives an RMSD of 1.3 Å, whereas the dimeric structures superpose with an RMSD of 2.2 Å. While most of the reported β-prism I fold lectins are dimeric or tetrameric (48), there is one example of unusual heptameric quaternary association (49) and another instance where octameric assembly has been observed (50). The octamer in the latter case is made up of banana lectin-type dimers. The dimer interface in horcolin is composed of two β-strands (β1 and β10) from adjacent subunits that pack in an antiparallel fashion. This dimeric arrangement is similar to several other β-prism I fold lectins, emphasizing the importance of these interactions in quaternary organization. Horcolin shows highest sequence identity with Orysata lectin (44%), followed by banana lectin (39%), AcmJRL (38%) and Heltuba (37%). All these lectins are evolutionarily close to horcolin and all of them are of plant origin. The subunit structure of horcolin is similar to those of banana lectin, AcmJRL and other β-prism I fold lectins and lectin domains of known structure, with RMSD values between the Cα atoms of horcolin and Orysata, banana lectin and AcmJRL of 1.60 Å, 1.30 Å and 1.81 Å, respectively.

Figure 7. Hydrogen bonding interactions link the two carbohydrate binding sites.

A) Overlay between horcolin (pink) and banana lectin (PDB ID:1XIV, red), both bound to α-methyl-D-mannose. The mannose units in horcolin and banana lectin are depicted in blue and green colours respectively. B) Top view of a monomer of mannose-bound horcolin showing hydrogen bonds connecting LBL1 (red), LBL2 (blue), the GG loop (green) and the secondary binding site (V93). The rest of horcolin is depicted in grey. Mannose molecules are represented as yellow balls and sticks.

In β-prism I fold lectins, a conserved GXXXD motif is involved in carbohydrate recognition. This motif was originally characterized in artocarpin (Artocarpus integrifolia agglutinin), which has a single carbohydrate binding motif per subunit located in the C-terminal Greek key I (51). Banana lectin, on the other hand, shows two primary binding sites located in Greek keys I and II (Fig. 7A), while the residues at the corresponding positions in Greek key III serve as a common secondary site for recognition of mannooligosaccharides. In BanLec, the two LBLs are comprised of ¹²⁹GDFID¹³³ and ³⁴GDVVD³⁸ on Greek key I and II (16), while in horcolin, LBL1 and LBL2 are ¹³⁴GAFLD¹³⁸ on Greek key I and ³⁵GAIVD³⁹ on Greek key II (30). A recently characterized mJRL from pineapple (AcmJRL) also has two mannose binding loops- ¹³¹GSLVD¹³⁵ on Greek key I and ³⁵AHAID³⁹ on Greek key II, where the second mannose recognition motif contains Ala in place of Gly which is found in all other members of the family (18). In all three lectins with two LBLs, horcolin, BanLec and AcmJRL, the backbone NH of the first three residues in each LBL participate in hydrogen bonding interactions with O6 of mannose. The fourth residue is a hydrophobic amino acid that is not involved in any interaction, while the last residue is Asp, which forms two hydrogen bonds through its side-chains with O4 and O6 of mannose. Thus, mannose shows a similar mode of binding with the LBLs of these lectins. Cooperativity between a pair of carbohydrate binding site has been observed in a very few instances, that too typically when cross-linked complexes form between lectins and multivalent oligosaccharides (52, 53). However, there is no evidence for differences in affinities and rates of monosaccharide binding to a lectin with two very similar binding sites.

Our results illustrate the potential of HSQC-based titrations, in conjunction with NMR lineshape modeling, for elucidating the thermodynamics, kinetics and mechanism of protein-ligand interactions. Here, the analysis of lineshapes in the ¹H-¹⁵N HSQC-detected mannose titration clearly demonstrates the presence of two mannose binding sites in horcolin with K_d values of 4.71 and 19.9 mM. While previous ITC-based measurements had suggested a stoichiometry of two and affinity in the millimolar range, it was not possible to discriminate between the affinities of the two sites (30). Since the exchange between free and ligand-bound forms of horcolin lies in the intermediate-fast timescale, NMR lineshapes are also sensitive to the kinetics of conformational exchange and enabled us to characterize the k_off and k_on values for mannose binding to horcolin. Moreover, the high solubility of mannose made it possible to saturate the binding of wt and mutant horcolin, despite the weak millimolar ligand binding affinities of these proteins.

The thermodynamics and kinetics of mannose recognition by the single-binding-site variants D138A and D39A horcolin suggest that there is allosteric coupling between the two CBSs. The binding affinities of CBS2 and CBS1 for mannose in wt horcolin are 4.71 and 19.9 mM, corresponding to a binding free energy difference ΔΔG⁰ = 0.85 kcal/mol. While the affinity for CBS2 is similar in D138A horcolin, the affinity of CBS1 decreases by 2.5-fold (K_d = 55.6 mM) in the D39A variant, corresponding to a ΔΔG⁰ between the two single-site binding events of 1.7 kcal/mol. There are at least two mechanisms for this decrease in binding affinity of CBS1 between wt and mutant horcolin. First, it is possible that mannose binding by wt horcolin at CBS2 allosterically alters the CBS1 to increase its capacity for recognizing mannose. This mechanism will introduce cooperativity in the binding of mannose to horcolin, and while we do not need a cooperative two-site model to fit the titration data, it is likely that our NMR datasets are not sufficiently sensitive to pick up this cooperative behaviour superimposed on a set of already weak binding affinities. Second, the perturbation of CBS2 consequent to the D39A mutation may be transmitted allosterically to CBS1, spatially rearranging CBS1 and reducing its affinity for mannose. Both these mechanisms suggest the existence of molecular communication between CBS1 and CBS2, and this is supported by the observation that mutation at one site affects sugar binding rate constants at the other. For example, k_on and k_off at CBS2 increase by 3- and 2-fold respectively when D138 at CBS1 is mutated to Ala, while k_on and k_off at CBS1 increase by 2.4- and 7-fold when D39 at CBS2 is mutated to Ala. Interestingly, CSPs occurring at the secondary binding site upon mannose binding in wt horcolin disappear in both single-site-binding mutants, suggesting that the simultaneous presence of sugar residues at both sites might be necessary to engage the secondary binding site.

In order to probe the molecular determinants for the communication between CBS1 and CBS2, we examined the interactions established by residues at the two sugar binding sites. Figure 7B clearly shows the presence of a network of hydrogen bonds linking LBL1 and LBL2 directly as well as through the GG loops and the secondary binding site, providing a molecular pathway for the allosteric transmission of signals between the two sites. This cross-talk between the two CBSs of horcolin is expected to have implications for the binding of higher order mannose-rich glycans such as those present in the envelopes of viral glycoproteins.

Conclusions

The absence of detectable mitogenicity in horcolin, coupled with its strong and specific recognition of gp120, as well as its ability to neutralize HIV infection, highlight the potential of horcolin as a therapeutic anti-HIV agent. Here, we use X-ray crystallography and multidimensional NMR spectroscopy to elucidate the structural and glycan recognition features of horcolin. Horcolin adopts a β-prism I fold with three Greek key motifs that carry two carbohydrate binding sites. Using a combination of NMR 2D lineshape modeling and site-directed mutagenesis, we show that the two sites exhibit at least four-fold difference in binding affinities and communicate with each other through a network of hydrogen bonds linking the two sites. Our results pave the way for a molecular level understanding of the recognition of high mannose glycans and glycoproteins by horcolin, and possibly for a dissection of the molecular determinants of mitogenicity in lectins.

Experimental procedures

Protein expression and purification

The codon-optimized DNA sequence for the horcolin gene (Uniprot ID: Q5U9T2), either carrying or lacking a C-terminal hexa-His purification tag, was subcloned in the pET-22b(+) vector between the NdeI and XhoI restriction sites. In order to generate D39A and D138A horcolin mutants, site-directed mutagenesis was carried out using the Quikchange method with the help of overlapping primers. Additionally, a non-cleavable C-terminal 6x-His tag was inserted in the mutant horcolin for purification purposes.

Horcolin was overexpressed in BL21(DE3) cells transformed with the pET-22b(+) plasmid containing the horcolin gene insert. Cells were grown either in LB broth (unlabeled horcolin) or in M9 minimal media (¹⁵N, ¹³C or ²H-labeled horcolin). 1 g/L of ¹⁵NH₄Cl and 3 g/L [¹H,¹³C] glucose were added to the M9 media as the sole nitrogen and carbon sources respectively to obtain uniformly ¹⁵N and ¹³C labeled proteins. U-²H,¹³C,¹⁵N-labeled horcolin was derived from cells grown in M9 media prepared in D₂O and using ¹⁵NH₄Cl and [²H,¹³C]-glucose as the nitrogen and carbon sources. When the cell culture reached an OD600 of 0.8, overexpression of horcolin was induced with 0.2 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) and the culture was incubated at 18 °C for 18-20 h. Subsequently, cells were harvested by centrifugation at 7000 rpm for 20 min, 4 °C and washed with 1x phosphate buffer saline (PBS). The cells were then resuspended in lysis buffer (pH 7.4) containing 1x PBS and 1 mM PMSF, sonicated on ice for 30 min (2 s on and 4 s off cycles) at 50 % amplitude and centrifuged twice at 14000 rpm for 45 min at 4°C.

For wt horcolin without a hexa-His tag, the supernatant was incubated with 5 mL of mannose-sepharose beads and allowed to bind on an end-rocker for 2 h at 4 °C. The beads were then loaded on a column and washed with 1 L of 1x PBS. Horcolin was eluted in fractions of 2 mL with a 100 - 300 mM gradient of D-mannose spanning 10 column volumes. Pure protein fractions were dialyzed against 1x PBS and stored at 4 °C. For His-tagged wt horcolin or the single-binding-site mutants, the supernatant after lysis and centrifugation was loaded onto a Ni-NTA column pre-equilibrated with the lysis buffer. After washing the column with 1x PBS buffer containing 20 mM imidazole, pure protein was eluted with buffer (pH 7.4) containing 50 mM Tris, 200 mM NaCl, and 250 mM imidazole. Pure protein fractions were dialyzed against buffer (pH 7.4) containing 25 mM Tris, and 50 mM NaCl to remove the imidazole. Back-exchange of deuterated amides in U-²H,¹³C,¹⁵N-labeled horcolin was carried out by denaturing perdeuterated horcolin in 1x PBS containing 6 M guanidine hydrochloride (GdmCl). The protein was then refolded by step-wise dialysis using successive GdmCl concentrations of 6, 3, 1.5, 0.75 and 0 M.

Purity of the protein was checked using SDS-PAGE on a 15% Tris-glycine gel. NMR samples were made in 10% (vol/vol) D₂O buffer (pH 7.4) containing 25 mM Tris, 50 mM NaCl, 1 mM EDTA and 0.03% NaN₃. The protein yield was typically 7 mg/L of cell culture in both LB and M9 media.

All experiments involving a carbohydrate reported in this manuscript have been carried out using methyl-α-D-mannose, which has been referred to as mannose throughout the manuscript.

Wt horcolin lacking the C-terminal hexa-His tag was used for all crystallization and NMR studies. Since the single-binding-site variants did not bind mannose strongly enough for efficient purification, a hexa-His tag was used for D39A and D138A horcolin. NMR spectra of wt horcolin without and with the hexa-His tag show only minor chemical shift perturbations, confirming that the C-terminal hexa-His tag does not induce structural perturbations in horcolin (Fig. S10A). Moreover, titration data acquired with His-tagged and tagless horcolin constructs agree very well, demonstrating that the purification tag does not alter mannose binding characteristics of horcolin (Fig. S10B).

Crystallization and data collection

Crystallization experiments involving apo-horcolin were carried out employing the microbatch under oil method using a 10 mg/mL protein solution in PBS buffer pH 7.4 (20 mM PBS and 150 mM NaCl). Crystals were obtained in the Crystal Screen I crystallization condition No.10 from Hampton Research consisting of 0.2 M ammonium acetate, 0.1 M sodium acetate trihydrate pH 4.6 and 30 % (w/v) polyethylene glycol 4000. The crystals of apo-horcolin grew to an average size of 0.3 × 0.3 × 0.3 mm over the course of 10 days at 22 °C in the trigonal space group P22₁2₁ with 12 molecules in the asymmetric unit. Crystals of methyl-α-D-mannose bound horcolin were obtained by co-crystallization using the microbatch under oil method. A 300 μL of protein solution at a concentration of 10 mg/mL in Tris buffer (pH 7.4) was mixed with 300 μL of methyl-α-D-mannose solution (100 mM stock concentration) in the same buffer.

X-ray diffraction data from apo- and mannose-bound crystals were collected at 100 K on the XRD2 beamline at a wavelength of 0.99 Šat the Elettra Synchrotron, Trieste, Italy using a Pilatus 6M detector. Diffraction images were processed with iMOSFLM (54) and merged using SCALA (55) in the CCP4 program suite (Collaborative Computational Project, Number 4 1994). The TRUNCATE software package (56) from the CCP4 suite was used to convert intensities into structure factor amplitudes. The Matthews coefficient was estimated to be 1.90 ų Da^-1 and 35% solvent content with 12 subunits for apo-horcolin and 2.18 ų Da^-1 and 44% solvent content with single subunit for the complex (57).

Structure solution, refinement and validation

Simultaneous attempts were made to solve the structure of apo-horcolin by the molecular replacement (MR) method (58, 59) using the dimer and subunit structures of Orysata lectin (PDB code: 5X5F), banana lectin (PDB code: 1X1V) and Heltuba (PDB code: 1C3M) as search models. MR calculations involving Orysata dimer resulted in a reasonable Log Likelihood Gain (LLG) and Translation function Z-scores (LLG=215 and TFZ=12), consisting of 12 subunits/six dimers in the ASU. Apo-horcolin subunit coordinates were used as a search model for the complex involving methyl-α-D-mannose. Refinement cycles were performed with REFMAC5 (60). A few rounds of rigid body refinement were performed first, followed by positional and B-factor refinement. Model building was done using Coot (61). Tight non-crystallographic symmetry (NCS) was employed during the initial stages of refinement and progressively relaxed during the final stages. Initially, ligand molecules were located/identified on the basis of the difference map and their positions were confirmed using simulated omit maps (62). Ligand molecules were modelled into the electron density where appropriate, when the R-factors were approximately 0.23. Oxygen atoms of water molecules were identified on the basis of peaks with heights >3σ in Fo–Fc maps and greater than 1σ in 2Fo–Fc maps. The structures were validated using PROCHECK (63) and MolProbity (64). A summary of the relevant statistics of the data collection and refinement is given in Table 1.

Analysis of structures

PyMol (65) was used for structure analysis and generating figures. Interatomic and hydrogen bond distances were calculated using CONTACT in the CCP4 program suite, or PyMol. Structural superpositions were made using PyMol or the ALIGN software (66). Buried surface area was calculated employing the PISA software package (55) or PyMol.

PDB references

Apo-horcolin (7V4Z)

Horcolin in complex with Methyl-α-D-mannose (7V4S)

NMR spectroscopy

NMR spectra were acquired on 14.1 T Agilent (600 MHz ¹H Larmor frequency) or 16.5 T Bruker (700 MHz) spectrometers equipped with a ¹H/¹³C/¹⁵N triple resonance single-axis gradient cryoprobe (600 MHz spectrometer) or a room temperature TXI probe (700 MHz Bruker spectrometer). All NMR measurements were carried out at 25 °C. NMR datasets were processed and visualized using NMRPipe (67) and NMRFAM-Sparky (68) software packages respectively.

Backbone assignments

Backbone ¹H^N, ¹⁵N, ¹³Cα, ¹³Cβ, ¹³CO and ¹Hα resonances of horcolin were assigned using experiments carried out on samples of 1 mM U-¹⁵N, 1 mM U-¹³C,¹⁵N or 0.6 mM U-²H,¹³C,¹⁵N horcolin (after back-exchanging amide deuterons to protons). A combination of the following datasets was used for assignment purposes: (U-²H,¹³C,¹⁵N horcolin): HNCACB, HN(CO)CACB, HNCO, HN(CA)CO, HNN. (U-¹³C,¹⁵N horcolin): HNCA, HN(CO)CA, HBHA(CO)NH, HACA(CO)NH. (U-¹⁵N horcolin): HNHA, ¹⁵N-NOESY-HSQC (140 ms mixing time) and ¹H^N-¹H^N NOESY (150 ms mixing time) (32). Backbone ¹H^N, ¹⁵N assignments of D39A and D138A horcolin were transferred from the assignments of wt horcolin using a combination of ¹⁵N-edited NOESY-HSQC (150 ms mixing time) and ¹⁵N-edited TOCSY-HSQC (53.5 ms mixing time) datasets.

NMR titrations

¹H-¹⁵N-HSQC spectra of horcolin were acquired at increasing concentrations of methyl-α-D-mannose. In the titration series for wt horcolin, the initial concentration of horcolin was 0.54 mM and mannose was added to final mannose (protein) concentrations of 0 (0.54), 0.1 (0.539), 0.3 (0.538), 0.5 (0.537), 1 (0.536), 2 (0.535), 4 (0.533), 6 (0.530), 8 (0.529), 10 (0.528), 12 (0.527), 14 (0.526), 19 (0.523), 26 (0.520), 36 (0.515), 51 (0.507), 64 (0.5), 100 (0.481), 150 (0.454), 300 (0.374), 400 (0.321), 500 (0.267) mM (22 points). In the D138A horcolin titration series, the initial protein concentration was 0.79 mM and the total ligand (protein) concentrations were 0 (0.79), 0.2 (0.784), 0.4 (0.783), 0.8 (0.781), 2 (0.78), 3 (0.778), 6 (0.776), 9 (0.773), 12 (0.77), 16 (0.768), 19 (0.766), 22 (0.763), 28 (0.758), 39 (0.75), 55 (0.737), 78 (0.719), 98 (0.704), 167 (0.658) (18 points). In the D39A mutant titration series, the initial protein concentration was 0.63 mM and the total ligand (protein) concentrations were 0 (0.627), 0.2 (0.623), 0.3 (0.622), 0.6 (0.62), 1 (0.619), 2 (0.618), 5 (0.616), 7 (0.615), 10 (0.613), 12 (0.612), 15 (0.61), 17 (0.609), 22 (0.605), 31 (0.6), 44 (0.593), 63 (0.581), 78 (0.571), 132 (0.542), 299 (0.464), 550 (0.371), 914 (0.272) (21 points).

To determine the combined and weighted backbone amide chemical shift perturbation (CSP) upon addition of mannose, we used the following equation:

$$\Delta \delta =\sqrt{\Delta H^2+{\left(0.159\Delta N\right)}^2}$$

where ΔH and ΔN are the residue-specific differences in ¹H and ¹⁵N chemical shifts extracted from ¹H-¹⁵N HSQC spectra at the beginning and the end of the titration.

Fitting and simulating NMR titration profiles

Two dimensional lineshape modeling of the titration data was performed within the software package TITAN (44), which numerically simulates, in Liouville space, the evolution of magnetization during a pulse sequence in the presence of chemical exchange. All ¹H-¹⁵N-HSQC spectra of a titration series were processed identically using an exponential window function in both dimensions (4 Hz and 8 Hz line broadening in ¹H and ¹⁵N respectively). Residues that showed a chemical shift perturbation larger than 0.1 ppm upon mannose binding, and whose peaks were not severely overlapped with neighbouring resonances, were chosen for lineshape analysis. For the titration of D138A and D39A horcolin, lineshapes of 14 and 16 residues respectively were globally fit to a two-state single-site binding model. The following 14 residues were used in the fitting of D138A titration series: V5, M30, G61, E66, E79, V84, F87, D90, V93, R106, G113, G134, F136, and one unassigned residue. The following 16 residues were used in the fitting of D39A titration series: V5, G15, M30, F43, N50, G60, G61, G64, E66, F87, T100, G131, G134, F136 and two unassigned residues. The wt titration dataset was fit to a three-state sequential model. The chemical shifts of all three states were fit first and then fixed for the rest of the fitting procedure. 17 residues (G14, G15, A40, I41, S55, G56, G60, G61, G64, E66, E79, F87, D90, T107, T112, A133 and one unassigned residue) were globally fit in the wt titration series. The errors in the fit parameters were determined by bootstrapping and are obtained as the standard deviation from the mean of 100 bootstrap replicas. Simulations of peak lineshapes were carried out within the TITAN software using parameters derived from the lineshape fitting routine described above.