Rational design for crystallization of LG-vitamin D complex (Crystal Growth & Design)

Abstract

β-Lactoglobulin (LG) is a major milk whey protein containing primarily a calyx for

vitamin D3 binding, although the existence of another site beyond the calyx is controversial.

Using fluorescence spectral analyses in the previous study, we showed the binding stoichiometry for vitamin D3 to LG to be 2:1 and a stoichiometry of 1:1 when the calyx was

“disrupted” by manipulating the pH and temperature, suggesting that a secondary vitamin D binding site existed. To help localize this secondary site using X-ray crystallography in the present study, we used bioinformatic programs (Insight II, Q-SiteFinder, and GEMDOCK) to identify the potential location of this site. We then optimized the occupancy and enhanced the electron density of vitamin D3 in the complex by altering the pH and initial ratios of vitamin D3/LG in the cocrystal preparation. We conclude that GEMDOCK can aid in searching for an extra density map around potential vitamin D binding sites. Both pH (8) and initial ratio of vitamin D3/LG (3:1) are crucial to optimize the occupancy and enhance the electron density of vitamin D3 in the complex for rational-designed crystallization. The strategy in practice may be useful for future identification of a ligand-binding site in a given protein.

Introduction

Bovine β-lactoglobulin (LG) is a major protein in milk comprising about 10−15%.¹ Because of its thermally unstable and molten-globule nature, LG has been studied extensively for its physical and biochemical properties.^2-7 Some essential functions of LG including hypocholesterolemic effect,⁸ retinol transport,^9,10 and antioxidant activity^11-14 have been reported. According to the crystal structure, LG comprises predominantly a β-sheet configuration containing nine antiparallel β-strands from A to I (Figure 1A).^15-17 Topographically, β-strands A-D form one surface of the barrel (calyx), whereas β-strands E-H form the other. The only α-helical structure with three turns is located at the COOH-terminus, which follows the β-strand H beyond the calyx.¹⁸ A remarkable feature of

the calyx is its ability to bind hydrophobic molecules such as retinol, fatty acids, and vitamin D3 (Figure 1B).^19-22

The location of vitamin D binding sites has been controversial, yet most evidence points toward the calyx. It has been postulated that another site exists,^23-25 but it could not be verified by the crystal structure using the vitamin D2-LG complex.¹⁰ In a previous study, we first conducted a ligand binding assay using the fluorescence changes by retinol, palmitic acid, and vitamin D3 to address the existence of another site for vitamin D binding. This data demonstrated that the maximal binding stoichiometry of vitamin D3 with LG was 2:1, whereas that of retinol or palmitic acid was 1:1 (Table 1),²⁶ suggesting that there was another

switching off the gate of calyx at low pH (2-6)¹⁵ to further substantiate the “two site hypothesis”. As expected, the binding capability of retinol and palmitic acid diminished under this condition, but it retained the vitamin D binding with a vitamin D3 to LG stoichiometry of 1:1 (Table 1).²⁶ We also used a strategy to denature the conformation of the calyx by heat treatment⁷ and conducted the binding assay after the calyx was thermally

“disrupted”. Under this condition, the binding stoichiometry of vitamin D3 to heated LG (100 °C for 16 min) was found to be 1:1 (Table 1).²⁶ These previous binding experiments²⁶ suggest that there is a secondary site for vitamin D3 binding distinct from the calyx.

In the present study, we located the secondary vitamin D3 binding site of LG, which has been a controversy among many researchers, using bioinformatic analysis to narrow the region of potential binding sites for crystallographic verification. We first utilized well-known programs (ActiveSite_Search Insight II and Q-SiteFinder) to predict the ligand-binding site in search of a possible location for vitamin D binding with the geometric or energetic criteria. The Insight II²⁷ is based on the size of surface cavities of a given protein without a specified ligand; it searches the location and extent of the pocket according to the geometric criteria. The Q-SiteFinder,²⁸ however, defines a binding pocket only by energy calculations using a methyl probe for van der Waals interactions with a given protein.

We next used a more accurate docking program, GEMDOCK, which was developed in our laboratories,²⁹ to dock a specific ligand with a given protein based on a nonbiased search for

their interactions. GEMDOCK is specifically designed for a flexible ligand that may best fit into possible binding sites. Under this condition the docked ligand conformations are generated. All three programs require an established 3D structure of a given protein. We show that Insight II and Q-SiteFinder predicted three and six major potential regions, respectively, available for any nondefined ligand binding on LG. Docking using GEMDOCK predicted six possible sites for vitamin D3 binding. Following the analyses of Insight II, Q-SiteFinder, and GEMDOCK cross-docking, we identified that there were two potential secondary sites for vitamin D binding located near the C-terminal α-helical region. This led

us to study the crystal structure of LG-vitamin D3 complex to further identify the secondary vitamin D binding site, if any.

Furthermore, we cocrystallized the LG-vitamin D3 complex which was prepared at pH 7 with a vitamin D3/LG ratio of 2 according to the previous crystallographic study.²⁰ A search for an extra electron density for vitamin D3 around potential secondary binding sites was based on the predicted data obtained from bioinformatic analysis. We found a weak extra electron density that was located near the C-terminal α-helical region. With respect to

cocrystallization, the maximum occupancy of the ligand should provide a better opportunity in growing high-quality crystals of the ligand-protein complex. In general, the affinity, solubility, and concentrations of added hydrophobic ligands would influence the ligand

occupancy at equilibrium. For 90% occupancy, the amount of added ligands must be greater than the amount of protein so that the free ligands at equilibrium are not depleted.³⁰

In this work, we reported a rationally designed approach for preparing the complex of LG and vitamin D3 at various pH and vitamin D3/LG ratios to optimize the occupancy of vitamin D3 and improve the electron density of the secondary binding site. Finally, we identified an exosite for vitamin D binding to be located near the α-helix and β-strand I of LG

using a crystal prepared at pH 8 with a vitamin D3/LG ratio of 3:1. The biological significance of the revealed exosite for vitamin D binding in milk LG was also discussed.

Materials and methods

Materials

LG was purified from fresh raw milk using 40% saturated ammonium-sulfate, followed by a G-150 column chromatography of the supernatant as described previously.⁴ Vitamin D3

(cholecalciferol) was purchased from Sigma (St. Louis, MO).

Predicting possible binding site by ActiveSite_Search Insight II and Q-SiteFinder

To identify the possible binding sites, the ActiveSite_Search Insight II²⁷ and Q-SiteFinder²⁸ software were used, and the prediction was performed according to the standard protocol. For ActiveSite_Search Insight II, we used the size within 50 Å as a cutoff site for the smallest cavity (in grid points). For Q-SiteFinder, the parameters for identified

protein residues involved in the van der Waals interactions with the methyl probe, 5.0 Å was used. The coordinates of possible sites predicted by Insight II and Q-SiteFinder were saved in PDB format and depicted using the Swiss-Pdb Viewer program.³¹

Docking analysis between ligands and LG by GEMDOCK

We extrapolated the 3D structure of retinol (PDB ID 1GX8, LG-retinol complex), palmitic acid (1GXA, LG-palmitic acid complex), and vitamin D3 (modified from 25(OH)-vitamin D3; 1MZ9, cartilage oligomeric matrix protein-vitamin D3) from the Protein Data Bank (PDB). Energy minimization of these three compounds was performed using a SYBYL program package. An established 3D structure of LG (1GX8) was used as the target for cross-docking. Since GEMDOCK is able to search the whole protein for exploring the binding site of a given ligand, we presumed those atoms near the charged surface and hydrophobic areas of LG were the potential binding sites for vitamin D3. The predicted area(s) was then compared with the established binding pocket (calyx) for palmitic acid and retinol. Following the removal of all structured water molecules in LG molecule, GEMDOCK search was then conducted according to the procedures previously described.²⁹ Using an empirical energy function which consists of the electrostatic, steric, and hydrogen bonding potentials for docking, GEMDOCK²⁹ seemed to be more accurate than some conventional approaches, such as GOLD and FlexX, based on a diverse data set of 100

demonstrated when screening the ligand database for antagonist and agonist ligands of the estrogen receptor (ER).³³

In this work, we used an empirical scoring function to estimate interaction energies between LG and ligands. The parameters used in the flexible docking included the initial step size (σ = 0.8 and ψ = 0.2), family competition length (L = 2), population size (N = 1000), and recombination probability (pc = 0.3). For each docked ligand, optimization was stopped when either the convergence was below a certain threshold value or the iterations exceeded the maximal preset value of 80. Therefore, GEMDOCK produced 3600 solutions in one generation and terminated when it exhausted to 324 000 solutions for each docked ligand.

Crystallization

Purified LG was concentrated to 20 mg/mL in 20 mM acetate (pH 4), cacodylate (pH 6), HEPES (pH 7), or Tris buffer (pH 8). Vitamin D3 stock solution prepared as 50 mM in 100% ethanol was added to LG solution to give a molar ratio of 3:1, 2:1, or 1:1 with a final ethanol concentration less than 7% and incubation for three hours at 37 °C. Crystallization of the LG-vitamin D3 complex was achieved using the hanging-drop vapor-diffusion method at 18 °C with 2 μL hanging drops containing equal amounts of LG-vitamin D3 complex and a reservoir solution (0.1 M HEPES containing 1.4 M trisodium citrate dehydrate, pH 7.5).

Crystallographic data collection and processing

The crystals were mounted on a Cryoloop (0.1-0.2 mm), dipped briefly in 20% glycerol as a cryoprotectant solution, and frozen in liquid nitrogen. X-ray diffraction data at 2.1-2.2 Å resolution were collected at 110 K using synchrotron radiation on the Taiwan contracted beamlines BL12B2 at SPring-8 (Harima, Japan) and BL13B at NSRRC (Hsinchu, Taiwan).

The data were indexed and processed using a HKL2000 program.³⁴

Results and discussion

Although a secondary vitamin D binding site of LG has been proposed from some physicochemical experiments,23-25, 35, 36 its existence and location have remained elusive and controversial. Several studies have clearly demonstrated that the binding stoichiometry between retinol or palmitic acid and LG is 1 where the central calyx of LG is responsible for retinol and palmitic acid binding,^21-23 but whether the binding of vitamin D to LG is 1 or 2 remains uncertain. Wang et al. proposed that LG possesses two potential binding sites for vitamin D: one is in the calyx formed by a β-barrel and the other is near an external hydrophobic pocket between the α-helix and the β-barrel.^{24, 25} In the previous study, we first

showed that the relative binding of vitamin D to LG was 56% of the maximal binding at a vitamin D3/LG ratio of 1:1 and the binding stoichiometry of vitamin D3 to LG was 2:1 relative to that 1:1 of retinol or palmitic acid verified using extrinsic fluorescence emission,

support the view that LG comprises two vitamin D binding sites. Second, we monitored the vitamin D3 binding of LG by utilizing a unique structural change property of LG at various pH to further substantiate the “two site hypothesis”. The EF loop of LG is known to act as a gate over the calyx;¹⁵ at pH values lower than 6 the loop is in a “closed” position. Of remarkable interest, LG still retained about 30% of the maximal binding for vitamin D3 at pH below the transition (2-6) (Table 1).²⁶ It suggested that there might be another vitamin D binding site that is independent of pH. Our previous study has shown that thermally denatured LG (heated 100 °C for 5 min) is unable to bind to retinol and palmitic acid owing to the deterioration of the calyx.⁷ We tested the hypothesis whether the “secondary binding site” for vitamin D (if any) still existed after heating. Since LG is a molten globule with a heating transition between 70 and 80 °C,^{5, 7} we monitored the vitamin D3 binding with LG preheated at different temperatures including one that could denature the calyx structure.

Interestingly, it showed a dramatic and sharp decrease in retinol, palmitic acid, and vitamin D3

binding near the LG transition temperature. At the temperature above 80 °C, it almost completely abolished the retinol and palmitic acid binding, while still retaining 40% of the vitamin D3 binding even at 100 °C heating for 16 min (Table 1).²⁶ To confirm it further, we monitored the binding of heated LG (100 °C for 16 min) with various amounts of vitamin D3, and a stoichiometry of 1:1 was observed between heat-denatured LG and vitamin D3 using the same fluorescence quenching analysis.²⁶ Taking the previous pH and thermal experiments

together (Table 1), we concluded that a thermally stable site (defined as an exosite) beyond the calyx exists for vitamin D3 binding. In the present study, we located the secondary vitamin D3 binding site of LG using bioinformatic analysis to narrow the region of potential binding sites, the search of extra electron density around the potential binding sites of the cocrystal prepared on previous crystallographic study,²⁰ and a rationally designed crystallographic approach for obtaining cocrystals with sufficient quality and ligand occupancy.

Structural analysis of the binding pockets of LG using Insight II and Q-SiteFinder

We attempted to identify the regions of LG that might be available for the interaction with any nondefined molecules using well-known programs (Insight II²⁷ and Q-SiteFinder²⁸), which are based on the size of the binding pocket and the binding energy between the methyl group and the pocket of a given protein, respectively, for prediction of a ligand binding site.

The former and latter predicted that there were 3 and 10 possible “binding sites” for any ligand, respectively. Table 2 and Figure 2A show the ranking and the location of six major predicted sites. Essentially both programs predicted that site 1 is exactly located at the calyx with the ranking superior to any others. This site is also known for all the vitamin D, retinol, and palmitic acid binding. Location of other predicted sites 2 (among C-terminal loop, β-strands C and D), 3 (among the pocket C-terminal α-helix, β-strands F, G, H, and A), 4 (at

the side α-helix and β-strands I), 5 (near loop H), and 6 (between CD and DE loops) is

depicted in Figure 2A.

Docking analysis of the interaction between LG and retinol or palmitic acid using

GEMDOCK

Because both Insight II and Q-SiteFinder are not able to perform the specific interaction between an assigned ligand and a given binding site, we next used a more accurate docking program, GEMDOCK,²⁹ for molecular docking to search for a possible number of vitamin D binding sites and to further assess the interaction of a specific ligand with each site.

GEMDOCK appeared to be more accurate than comparative approaches, such as GOLD and FlexX, on a diverse data set of 100 protein-ligand complexes³² for docking and two cross-docking experimental sets.²⁹ The screening accuracies of GEMDOCK were also better than GOLD, FlexX, and DOCK on screening the ligand database for antagonist and agonist ligands of estrogen receptor (ER).³³ GEMDOCK is a better docking tool for assessing the interaction of a specific ligand with each site of LG. We generated 30 docked conformations of each retinol or palmitic acid with whole LG. Based on the docking protocol,²⁹ we considered that a correct binding-mode was reproducible when the root-mean-square deviation (rmsd) between the best energy-scored conformation and crystal coordinates was less than 2 Å.³² In order to meet this criterion, we then evaluated the feasibility of GEMDOCK by cross-docking retinol and palmitic acid into the calyx or site 1 of LG (PDB

code: 1GX8), known to be an established site for the binding of retinol or palmitic acid. Our data reveal that these two ligands could be docked exactly into the calyx with a rmsd less than 2.0 Å. The average fitness of docked energy of retinol and palmitic acid was -86.3 and -75.7 kcal/mol (Table 3), respectively. Following the analyses of all the docking data, we found that none of the other sites 2-6 were fitted by either retinol or palmitic acid (<10%

probability), except for palmitic acid interacting with site 2 (46% of total 30-docked conformations) (Table 3). For example, none of the docked conformations of retinol and palmitic acid could fit into sites 3, 4, 5, and 6. This result is almost consistent with the current knowledge that the calyx (site 1) is the major pocket for interacting with these two ligands. It is thus feasible to use a similar strategy for studying the interaction between vitamin D3 and LG.

Docking analysis of the interaction between LG and vitamin D3 using GEMDOCK Each docked conformation of vitamin D3 (n = 30) obtained from GEMDOCK was then evaluated for a possible vitamin D binding ability. Table 2 shows that vitamin D3 could fit into the calyx with about 50% of the probability. Interestingly, the docked energy of vitamin D3 into the calyx site 1 was similar to that of retinol. The result suggests that GEMDOCK is suitable in conducting interactions between vitamin D3 and LG. It also pointed out site 2 (Figure 2B) as a possible second binding site for palmitic acid (46%) and vitamin D3 (20%).

the calyx is the only available binding site for palmitic acid;²¹ we ruled out site 2 as a particular site for palmitic acid binding. This site is located among the C-terminal loop, β-strand C, and β-strand D. Since β-strand D is thermally unstable,⁷ it is not consistent to

the proposed secondary vitamin D binding site that is supposed to be thermally stable in nature (Table 1). Site 2 is thus unlikely for vitamin D interaction.

In order to search for a potential secondary vitamin D3 binding site location, we blocked sites 1 and 2 with the reason mentioned above and then performed GEMDOCK, while using retinol and palmitic acid as a “negative control”. Table 3 shows that, under this condition, sites 3 and 4 possessed high probability for vitamin D3 interaction (40 and 20%) relative to retinol and palmitic acid. The other domains (sites 5 and 6) were considered nonsignificant due to the low probability (<3%). Notably, the docked orientation of the vitamin D3 in sites 3 and 4 was distinctly different from that of other ligands (Figure 2C). Figure 2D tempted to suggest that the docked orientation of vitamin D3 in site 3 is more stable (with a low rmsd) than that in site 4 due to its large pocket size that could stabilize ligand docking. On the other hand, the hydrophobicity of site 4 is high relative to site 3 based on the electrostatic potential surface model between LG and docked vitamin D3 conformation (Figure 2E). The prediction of site 4 being involved in vitamin D3 interaction is consistent with ranking predicted by Q-SiteFinder which is based on van der Waals interaction (Table 2). Taken together, sites 3 and 4 are the potential candidates for the secondary vitamin D binding site

and thus led us to localize it via a high-quality LG-vitamin D3 crystal. We searched for an extra density around C-terminal α-helix (sites 3 and 4) of LG-vitamin D3 complex which was

prepared at pH 7 with a vitamin D3/LG ratio of 2 according to the previous crystallographic study.²⁰ We found a weak extra electron density located near the C-terminal α-helical region. Later, we used a rationally designed crystallography experiment for improving electron density of vitamin D3 of the secondary binding site. In the bioinformatic prediction of the secondary vitamin D3 binding site, we excluded false-positive site 2 on the previous biochemical findings. It was confirmed that there is no extra density around site 2 in the crystal structure of LG-vitamin D complex that was prepared using the previous condition.

For this reason, bioinformatic prediction excluded site 2 as a potential candidate for the secondary vitamin D binding site based on the previous biochemical findings that is proper for this study.

Crystallization and diffraction of LG-vitamin D3 complex prepared at various pH and

vitamin D3/LG ratios

In any event, starting the cocrystallization experiment with maximum ligand occupancy should provide a better opportunity in growing high-quality ligand-protein crystals. The affinity and concentrations of added ligands, as well as ligand solubility, would influence the occupancy of ligands at equilibrium. For 90% occupancy, the amount of added ligands must

than about 10 × Kd.³⁰ In practice, ratios of ligands to a given protein up to 10:1 or more are commonly used, but large excesses should be avoided owing to the possibility of ligand binding to nonspecific sites. In the case of a weak binding affinity, the concentrations of ligands (or ligand solubility) may have to be 10 mM or higher in order to observe crystallographic occupancy.^{37, 38} We suspect that the lower occupancy of vitamin D2 might

than about 10 × Kd.³⁰ In practice, ratios of ligands to a given protein up to 10:1 or more are commonly used, but large excesses should be avoided owing to the possibility of ligand binding to nonspecific sites. In the case of a weak binding affinity, the concentrations of ligands (or ligand solubility) may have to be 10 mM or higher in order to observe crystallographic occupancy.^{37, 38} We suspect that the lower occupancy of vitamin D2 might