Carbohydrate-Protein Interactions: Advances and Challenges

Abstract

A carbohydrate, also called saccharide in biochemistry, is a biomolecule consisting of carbon (C), hydrogen (H) and oxygen (O) atoms. For example, sugars are low molecular-weight carbohydrates, and starches are high molecular-weight carbohydrates. Carbohydrates are the most abundant organic substances in nature and essential constituents of all living things. Protein-carbohydrate interactions play important roles in many biological processes, such as cell growth, differentiation, and aggregation. They also have broad applications in pharmaceutical drug design. In this review, we will summarize the characteristic features of protein-carbohydrate interactions and review the computational methods for structure prediction, energy calculations, and kinetic studies of protein-carbohydrate complexes. Finally, we will discuss the challenges in this field.

1. Introduction

Protein-carbohydrate interactions, also called protein-saccharide interactions, are gaining increasing popularity due to their fundamental roles in numerous aspects of biology and food industry. Saccharide is a synonym of carbohydrate, which is a biomolecule consisting of carbon, hydrogen, and oxygen atoms. Carbohydrates are one of the most abundant materials on earth. Based on the different lengths of repeating units, carbohydrates (i.e., saccharides) are divided into four chemical groups: monosaccharides, disaccharides, oligosaccharides and polysaccharides, as shown in Figure 1.

Figure 1.

Four types of saccharides. (a) Two typical monosaccharides, β-d-Fructofuranose and α-d-Glucose. Monosaccharides with five-membered rings are called furanose, and monosaccharides six-membered rings are called pyranose. (b) Disaccharide is formed when two monosaccharides are joined by a carbon-oxygen-carbon linkage known as glycosidic linkage. Sucrose is a common disaccharide composed of glucose and fructose. (c) An oligosaccharide is a saccharide polymer having three to ten monosaccharides. (d) Polysaccharides consist of (usually many) monosaccharides. Amylose is a polysaccharide made of many α-d-Glucose units.

Carbohydrates have a wide variety of functions, one of which is to serve as energy storage molecules¹. For example, starch functions as a storage for the energy from the photosynthetic process in plants. In animals, glycogen is the counterpart form of energy storage². Carbohydrates also function as structural components. Cellulose, considered the most common organic compound in the world, provides the cell with structural support. The second most common polysaccharide in the world after cellulose, chitin, also provides animals with similar functions. Carbohydrates serve as a recognition element as well. Glyconjugates, which are carbohydrates on cell surfaces that are bonded to other chemical species such as proteins, peptides and lipids, play a role in cell-cell interactions. One class of proteins called lectins, which are ubiquitous in nature and are found in animals, plants and microorganisms, can bind carbohydrates with high specificity. Such an interaction constitutes the basis for lectin-mediated drug targeting^{3, 4} and drug delivery⁵. In addition, carbohydrates mediate biological functions by interacting with enzymes and antibodies. Therefore, protein-carbohydrate interactions are essential to immune systems, fertilization, and blood clotting⁶.

This review will present the features of protein-carbohydrate interactions and summarize the computational methods for predicting protein-carbohydrate interactions. Challenges in this field will be discussed at the end.

2. Characterization of protein-carbohydrate interactions

Experimentally, X-ray diffraction and nuclear magnetic resonance (NMR) spectroscopy have been employed to determine the geometry of protein-carbohydrate complexes at the atomic level. X-ray crystallography provides static pictures of stabilized complexes, whereas NMR in combination with computational studies reveals dynamic structural behaviors of the carbohydrate binding process⁷. Unlike a protein, whose primary structure is linear, carbohydrates vary from linear to highly branched structures, as illustrated in Figure 1c. High flexibility is thus the major obstacle to obtain high-resolution crystal structures of carbohydrates. Therefore, despite the huge successes of X-ray crystallography in solving protein structures, it is difficult to determine the structures of carbohydrates with X-ray crystallography directly. NMR techniques are often used to study carbohydrates in solutions. Particularly, solid-state NMR (ssNMR) is a useful tool to observe the behaviors of different saccharides in their intact chemical forms bound to receptors. For instance, the structure and dynamics of Escherichia coli peptidoglycan, an essential component of the bacterial cell wall, was investigated by using ssNMR⁸. Because NMR data are the average of a conformational ensemble, which may lead to a virtual conformation, NMR should be used with caution when determining 3D carbohydrate structures⁹. Recently, cryo-EM has been used to solve the structures of carbohydrate-binding proteins^10–13. However, it remains a challenge for cryo-EM to determine carbohydrates structures. Although it is possible to determine the structure of a monosaccharide and its ring conformation at a high resolution, neither the monosaccharide nor its ring conformation can be identified at a resolution lower than 2.4 Å.¹³

To understand the formation of protein-carbohydrate structures, much attention has been drawn to studying protein-carbohydrate interactions. Similar to other protein-ligand interactions, protein-carbohydrate interactions are characterized by several biophysical properties such as thermodynamic parameters, dissociation constants and kinetics (e.g., k_on and k_off rate constants). A number of techniques including isothermal titration calorimetry (ITC)¹⁴, frontal affinity chromatography (FAC)¹⁵ and surface plasmon resonance (SPR)¹⁶ have been used to measure the dissociation constants for protein–sugar recognition. SPR is one of the most widely used techniques for the determination of kinetic parameters such as association/dissociation constants and on/off rates (k_on/k_off). This method has been successfully applied to exploring the kinetics and mechanisms of lectin-carbohydrate recognition¹⁷. It should be noted that SPR analysis requires immobilization of the protein and the carbohydrates. Recently, other approaches have been developed to investigate carbohydrate-protein interactions. For example, lectin-carbohydrate interactions were measured using a glycan microarray¹⁸. Kinetic analysis of the conformational changes in the mutants of maltose-binding proteins (MBPs) was performed using single-molecule fluorescence resonance energy transfer (smFRET)¹⁹. The binding kinetics of carbohydrate-protein interactions on unfixed cancer cell surfaces was studied using a quartz crystal microbalance (QCM) biosensor²⁰.

It has been reported that protein-carbohydrate interactions are relatively weak^{7, 21–23}. For example, the dissociation constants for most lectin-monosaccharide interactions are in the mM range²⁴. The protein-carbohydrate binding affinity increases with increasing sizes of oligosaccharide²⁵, except in the assessment of oligomannosyls (consisting 1 to 6 monosaccharide units) for their abilities to inhibit yeast cell adherence²⁶. However, the dissociation constants are usually greater than μM^{27, 28}. Such weak interactions facilitate the formation of transient states and therefore help partner proteins to reach their appropriate locations²².

Generally, the carbohydrate-binding sites on the hydrophilic surfaces of lectins and other proteins are shallow²⁹. Therefore, the binding affinities between lectins and their carbohydrate ligands are lower than many other protein-ligand binding affinities. However, like other protein-ligand interactions, the binding specificity of protein-carbohydrate interactions originates from the combination of a variety of interactions, including hydrogen bonding, hydrophobic interactions and electrostatic interactions. Due to the existence of numerous hydroxyl groups in every carbohydrate, hydrogen bonds probably contribute the most to the protein-carbohydrates interactions, especially in lectin-carbohydrate complexes. Hydroxyl groups may act as donors or acceptors. Sometimes, the hydroxyl groups participate in hydrogen bonding as both donors and as acceptors, known as “cooperative hydrogen bonding”. The formation of the hydrogen bonds often involves carbohydrate hydroxyl groups contacting with aspartates, glutamates, asparagines, glutamines, arginines, and serines, as well as with backbone amines and carbonyl groups. Bidentate hydrogen bonds are formed when two adjacent hydroxyl groups of a carbohydrate molecule establish hydrogen bonds with both carboxylate oxygens of either aspartates or glutamates. This feature is observed in enzymes, as well as in lectins³⁰. Hydrogen bonding interactions are found to be important for selectivity, by using engineered sugars³¹ and fluorodeoxy derivatives of natural lectin ligands^{32, 33}.

In some carbohydrates, the clustering of several adjacent C-H groups arising from the steric disposition of sugar hydroxyl groups creates a hydrophobic surface that can form nonpolar interactions with the aromatic rings of Trp, Tyr, and Phe residues in proteins. Such CH-π stackings between saccharides and aromatic rings in carbohydrate-binding sites have been demonstrated experimentally³⁴. Efforts have been made to better understand CH-π interactions^35–38. The attractive forces are thought to result from 1) an entropic contribution arising from the mutual shielding of both nonpolar surfaces from bulk water, and 2) the enthalpic contribution of non-conventional hydrogen bonds between the partially positively charged C-H groups and the quadrupole of the π-system of the aromatic ring³⁹.

Electrostatic interaction is also crucial for carbohydrate-protein recognition. Charged residues and ions are often found in active sites of carbohydrate-binding proteins^{40, 41}. Salt bridges are formed between the charged residues of saccharides and protein residues with opposite charges. In some instances, divalent cations, such as calcium and magnesium, bridge the hydroxyls of carbohydrates and negatively charged aspartates/glutamates, facilitating proteins to recognize their carbohydrate ligands⁷.

Apart from noncovalent interactions, glycosidic linkages also play a vital role in protein-carbohydrate complexes, as shown in Figure 1 and Figure 2. Although glycosidic torsion angles govern the flexibility of carbohydrates, it has been reported that torsion angles in glycosidic linkages have angular preferences instead of random distributions⁴². Hence, taking into account glycosidic torsion angles provides further insight into the assessment of protein-carbohydrate complex structures^{42, 43}.

Figure 2.

Glycosidic torsion angles of disaccharides. ϕ (O₅- C₁- O_x^’- C_x^’), ψ (C₁- O_x^’- C_x^’- C_x-1^’) and ω (O₆^’- C₆^’- C₅^’- O₅^’). x is denoted as the index of a carbon atom of the neighboring monosaccharide. The C₁ atom is the anomeric carbon. In a cyclic carbohydrate, the anomeric carbon is the carbonyl carbon in acyclic form. The orientation of the connecting oxygen determines whether the anomer is α or β⁴³. If the oxygen is below the plane of ring, the anomer is α. Otherwise, the anomer is β.

3. Predicting protein-carbohydrate interactions

Due to the limited number of experimentally solved protein-carbohydrate structures, computational methods provide valuable complementary tools to investigate protein-carbohydrate interactions. One of these methods is molecular docking. Molecular docking is a powerful computational approach that is widely used to predict receptor-ligand complexes^{44, 45}. Here, a receptor refers to a protein or a protein oligomer, and a ligand is either a small molecule or another protein. Scoring functions for docking are designed to rank different ligand poses through the evaluation of the strength of the interactions. Over the last two decades, more than 60 different docking programs have been developed and successfully used in predicting protein-ligand structures⁴⁶, such as DOCK⁴⁷, AutoDock⁴⁸, ZDock⁴⁹, HADDOCK⁵⁰ and MDock^{51, 52}. However, most docking programs often lead to unnatural glycosidic angles when applied to carbohydrates⁴². Due to the peculiarity of protein-carbohydrate interactions, a few programs have been developed for protein-carbohydrate docking^{43, 53} to address this challenge.

As aforementioned, the CH-π interaction is important for protein-carbohydrate complex formation. By adding an energy term for the CH-π stacking interactions to the SLICK scoring function⁵⁴, BALLDock⁵³ has been successfully applied to a set of carbohydrate-lectin complexes. The SLICK scoring function is composed of four terms, including a CH-π interaction term, a hydrogen bonding term, a softened van der Waals term, and a Coulombic electrostatic term. The score S is given by

$$S=s_0+s_{CH\pi }{S}_{CH\pi }+s_{hb}{S}_{hb}+s_{vdw}\Delta E_{vdw}+s_{es}\Delta E_{es}$$ (1)

while the weighting coefficients s are manually optimized. BALLDock and the SLICK scoring function successfully redocked 17 out of 18 plant lectin complex structures in a calibration set with high accuracy, much better than another docking program FlexX⁵⁵. In a larger test set, 14 out of 22 complexes consisting of plant lectins and non-plant lectins were obtained, with the root-mean-square deviations (rmsd) of the best ranked ligand orientations less than 1.5 Å.

Inclusion of conformational preference provides another opportunity to improve the efficiency and accuracy of carbohydrate docking. Recently, carbohydrate Intrinsic (CHI) energy functions⁴² were developed to estimate the energy arising from distortion of the glycosidic linkages. The CHI energy functions were derived from fitting the Gaussian expansions to average energy curves obtained from quantum mechanical calculations:

$$f\left(x\right)={\sum }_{i=1}^N{a}_i{e}^{-\frac{{\left(x-b_i\right)}^2}{c_i}}$$ (2)

where f is the CHI energy function, N is the number of individual Gaussian functions, x is the torsion angle (i.e., ϕ, ψ or ω shown in Figure 2), and a_i, b_i and c_i are the magnitude, width and midpoint of the distribution, respectively. By implementing the CHI-energy functions into the AutoDock Vina scoring function⁵⁶, Vina-Carb⁴³ increased the docking accuracy compared to AutoDock Vina. When Vina-Carb was applied to antibodies, lectins and carbohydrate binding modules (CBM), the success rates of predicting accurate binding modes reached 86%, 50%, 42%, respectively, in comparison with 70%, 50%, and 0% for AutoDock Vina. In addition, inclusion of explicit water during carbohydrate docking can improve docking accuracy^{57, 58}.

Molecular dynamics (MD) simulation⁵⁹ and hybrid method (quantum mechanics/molecular mechanics or QM/MM) are also important computational tools to explore protein-carbohydrate interactions. For example, QM/MM combined with MD simulation was conducted on a norovirus capsid protein bound with a Lewis antigen⁶⁰. In a different report⁶¹, comparative studies on MM/Poisson–Boltzmann surface area (PBSA), MM/ generalized Born surface area (GBSA) and QM-MM/GBSA were performed on monosaccharides bound to Ralstonia solanacearum lectin (RSL), a six-bladed β-propeller trimeric lectin with 90 amino acid residues in each monomeric chain. The limitations to these approaches are short timescale and inaccuracies in force fields.

It should be noted that PBSA and GBSA are two typical implicit solvent models for computing theoretical binding free energies of carbohydrate-protein complexes. Accurate prediction of binding free energies is essential to understanding biomolecular recognition. Generally, the binding free energy consists of enthalpy-related and entropy-dependent components⁶². Some of these terms can be directly obtained from MD simulations, such as the values for van der Waals and electrostatic energies. Desolvation free energy is usually estimated by using implicit solvent approximation, such as PBSA and GBSA. Comparing to explicit solvent models, implicit models represent solvent molecules as a continuum dielectric media instead of individual “explicit” solvent molecules. This method improves the computational speed significantly. PBSA is based on the Poisson continuum dielectric model and GBSA is an approximation to the PBSA with more computational efficiency but less precision. These two methods have been widely used in the study of lectin binding^63–65. Although implicit solvent models simplify the calculation and reproduce the behavior of bulk water, not all water molecules behave like bulk water. Some tightly bound waters play as a part of the protein or ligand^{62, 66}, and hydrogen-bond networks are important to stabilize protein-carbohydrate complexes, such as Polyporus squamosus lectin (PSL) bound with a trisaccharide⁶⁷. These additional effects are omitted when applying implicit solvent models to binding free energy calculations.

Finally, kinetics of carbohydrate-protein binding has also been studied with MD simulations. Recently, Negami et al.⁶⁸ employed coarse-grained molecular dynamics (CGMD) simulations with the MARTINI force field⁶⁹ to study the levansucrase–carbohydrate system. The binding rate constants were estimated using a Bayesian approach. Solanke et al. studied the hevein domain (HEV32) bound with N-acetylglucosamine mono-, di- or trisaccharides by using unbiased MD simulation⁷⁰. Three of the six simulations successfully reproduced the correct binding modes in agreement with experimental results. Unbinding events were observed in one simulation (monosaccharide) as well. The dissociation constant was estimated based on these binding/unbinding events.

4. Discussion

In this review, we have briefly summarized the recent advances in the field of protein-carbohydrate interactions. Experimentally, it remains a challenge to determine the structure of a protein-carbohydrate complex, despite the achievements of X-ray diffraction, NMR and cryo-EM. New refinement protocols need to be developed for X-ray and cryo-EM methods⁷¹. The main drawback of NMR-based studies is its relatively intrinsic low sensitivity. Efforts have been made to overcome this problem, such as using higher magnetic fields⁷². Pertaining to computational methods, molecular docking is most efficiently used to predict the structures of protein-carbohydrate complexes. Vina-Carb is the most successful carbohydrate docking program to date. However, due to missing the CH-π interaction, it fails when predicting the binding modes of carbohydrates that stack against aromatic amino acids. Additionally, pyranose (i.e., saccharides that have a six-membered ring) and furanose (i.e., saccharides with a five-membered ring) commonly have distortions in presence of carbohydrate-active enzymes⁷³. Unlike pyranose, furanose rings can adopt different conformations with little difference in their free energies⁷⁴, resulting in more ring flexibility⁷⁵. Thus, improvement in the scoring functions for carbohydrate-protein docking is urgently needed.