Retinoids are compounds consisting of three regions: a hydrophobic trimethylated cyclohexene ring, a conjugated polyene linker, and a polar carbon-oxygen functional group (a hydroxyl for retinol, and a carboxyl for retinoid acid). Retinoids, first described by Paul Karrer in the early 1900s [1], have been used as drugs for treating skin diseases and cancers [2,3]. They regulate cellular functions such as cell development, proliferation, differentiation, and apoptosis and play important biological roles in the skin, brain, stem cells, immune system, cancer cells, and so on [4]. Due to their biological significance and broad distribution in various tissues, they are actively utilized in the field of protein engineering [5] and drug design [6]. Many biological processes including retinoids have enabled its various pharmacological and biomedical applications in evaluating the effect of retinoids in hematological cancers [3,7], brain tumors [8], skin disease [2,9], and inflammation problems [9,10]. Retinoids require intracellular carrier proteins that bind themselves with high specificity and affinity for transport, bioavailability, and stability in an aqueous solution [11]. Cellular retinol-binding proteins (CRBPs) and cellular retinoic acid-binding proteins (CRABPs) are the carriers of retinoids and belong to the same family of intracellular lipid-binding proteins (i-LBP) [12]. They have similar tertiary structures and amino acid sequences but show clearly different substrate specificity. CRBPs bind retinol with high specificity and affinity whereas CRABPs are highly specific to retinoic acid.

CRBP(I) is a carrier protein of retinol and delivers retinol to enzymes that catalyze the conversion of retinol to retinaldehyde and retinyl ester [13,14]. The protein is a potential target of pharmacological and immunodetection strategies for some diseases [15,16] and a promising protein scaffold that can be engineered to the receptors with various molecular recognition functionality [17]. Extensive and intensive structural studies on the ligand binding mode of CRBP(I) and its variants have been performed, which allowed us to understand their important ligand binding features at the molecular level [[12], [13], [14], [15], [16], [17]]. CRBP(I) consists of a single polypeptide chain with approximately 130 amino acids and contains one binding site for retinol [18]. In the crystal structure of holo-CRBP(I), all-trans-retinol(atROL) is surrounded by three α-helices(αI-αIII) and ten stranded anti-parallel β-barrels(βA-βJ), where the binding cavity is completely shielded from the external solvent except eight buried water molecules for a structural network with the complex. The ligand-binding specificity of CRBP(I) is mainly attributed to the interactions with the binding pocket residues. The ligand-binding specificity of CRBP(I) is mainly attributed to the interactions with the binding pocket residues. Of the approximately 20 binding pocket residues, Lys-40, Trp-106, and Gln-108 surrounding the functional group of the ligand are known as key ligand-binding motifs [19,20]. Leu-29, Ala-33, Leu-36, Phe-57, Arg-58, and Ile-77, which form a portal region [[19], [20], [21]] for the ligand and solvent entry and exit, are located near the hydrophobic β-ionone ring of atROL. Among these major residues, Gln-108 interacts with the terminus of the ligand through a strong hydrogen bond, known to be a key interaction for the high specificity and affinity of CRBP(I) to atROL. The decisive impact of 108th residue on the substrate specificity was confirmed through various comparative and mutation studies for CRBP(I) and its homologous proteins [[22], [23], [24]]. These identified ligand binding features facilitate the drug design based on retinoids and receptor design based on CRBP(I).

Another important aspect of the protein-ligand interaction study is to understand that there is no binding between protein and ligand because the characteristics of non-binding can be applied in various fields such as prediction in drug discovery and suppressing non-specific binding in biosensor design [25]. However, most studies generally have focused on the binding characteristic and the features of non-binding modes have not been studied well. For CRBP(I), most studies have also targeted the binding mode and the study on the non-binding mode is very rare. A major reason is that non-binding complex does not exist and there is no way to study the features of proteins and ligands that do not bind with x-ray crystallography or any other experiments. Only computational modeling and simulation may be possible tools to investigate the features of a non-binding mode of protein-ligand interaction by modeling the virtual complex candidates.

It is known that atROL can bind to CRBP(I) but all-trans-retinoic acid (atREA) can't associate with CRBP(I). It was demonstrated that site-directed mutagenesis of Gln-108 to Arg-108 (Q108R) resulted in a major change in ligand binding specificity [24]. The Q108R mutant exhibited almost equivalent affinities for atROL and atREA, while wild-type CRBP(I) bound only atROL in the competitive binding experiments. Recently, we performed a molecular docking study on the binding modes of atREA in the binding pockets of wild-type CRBP(I) or its Q108R variant and compared them [26]. The docking simulation results of binding and non-binding complexes showed significantly different interaction patterns. In the ligand-binding complex (Q108R CRBP(I)-atREA), strong hydrogen bonds and salt-bridges were maintained between the 108th residue and the end of the ligand, and overall ligand-binding pose and interactions were similar to the crystal CRBP(I)-atROL structure. On the other hand, in the non-binding wild-type CRBP(I)-atREA complex, the 108th residue lost hydrogen bond, and ligand binding pose and pattern were substantially different from the crystal CRBP(I)-atROL structure. From these results, it was presumed that the specific interaction between 108 Q or R residue of CRBP(I) and the carboxyl or hydroxyl group of retinoid may be crucial in the forming of stable ligand conformation in the binding site.

Although our molecular docking simulation study mentioned above suggested some important features of binding modes of retinoids in CRBP(I), there was a fundamental limitation in the understanding of the non-binding mode of atREA in CRBP(I). In general, molecular docking simulation models the most probable and stable static structure of the protein-ligand complex based on a designed scoring function. However, there might not be the most probable and stable structure in the non-binding complex, and therefore the structure modeled by docking simulation may be just one of the ligand poses in the binding site which may not account for the behavior of atREA in CRBP(I). In this study, we investigated the dynamic behavior of wild-type and Q108R CRBP(I)-atREA complexes through molecular dynamics (MD) simulation. The vast snapshot information of the protein-ligand structures provided by molecular dynamics enables more stereoscopic and statistical analyses than docking simulation. We analyzed the RMSD of proteins and ligands in detail from the molecular dynamics simulation results, identified the fluctuations in the positional relationship of binding pocket amino acids that contribute to the binding, and examined how non-covalent interactions such as hydrogen bonds and salt-bridges dynamically change in the complexes. Finally, we tried to compare the differences between binding and non-binding complexes throughout all the aforementioned analyses. This study may provide not only a microscopic understanding of the ligand binding specificity of CRBP(I) but also more structural insights into protein-ligand interactions.