Protein kinases catalyze the transfer of phosphate group from ATP to specific substrates, which is known as phosphorylation. The latter represents one of the most important regulatory mechanisms in human cells. Accordingly, protein kinases play an essential role in nearly all major signal transduction pathways that control intracellular communication. However, malfunctions of protein kinases may cause a loss or mix-up of intracellular communication, which is a major cause of cancers and many other diseases. It is for this reason that small molecules, protein kinase inhibitors (PKIs), are developed to competitively inhibit the ATP binding sites of protein kinases for the treatment of these diseases. Over the years, numerous PKIs with varying molecular frames have been developed. A large-scale data mining of the Protein Data Bank resulted in an in-house database of 2149 non-redundant high-resolution (2.5 Å or higher) x-ray crystal structures of PKIs bound to protein kinases. This provides a unique opportunity to investigate molecular recognition of PKIs, which is the subject of this dissertation.
Molecular recognition of PKIs in protein kinases is achieved, as usual, by non-bonded interactions. Traditionally, hydrogen binding and salt bridge interactions were regarded as the major intermolecular interactions for ligand-protein binding. However, our analysis showed that more than 99% of PKIs contained at least one aromatic ring, which led us to hypothesize that the non-bonded interactions involving π systems (CH-π interactions, π-π stacking interactions, cation-π interactions, SH-π interactions, OH-π interactions, NH-π interactions) may also play a crucial role in binding of PKIs to their target proteins. Supporting evidence for the hypothesis was sought in the following four projects.
Project #1: A chemoinformatic analysis of 2149 PKIs was performed. Traditional chemoinformatic tools were adopted, including analysis of molecular descriptors and classification of scaffolds. The first half of this project deals with analysis of molecular descriptors for 2149 PKIs. Molecular descriptors can function as a bridge between the molecular structures and chemical/biological properties, including druglikeness. Our focus was on molecular descriptors that are relevant for ligand binding, including number of hydrogen bond acceptors and donors, number of aromatic rings, aromatic ratio, polar surface area, total surface area, and the fraction of SP3 carbon. It was found that more than 99% of protein kinase inhibitors include at least one aromatic ring, pointing to a potential role of π systems in molecular recognition of kinase inhibitors. A careful analysis of the relationship between the hydrogen bond counts and the number of aromatic rings resulted in an important finding, i.e., the average weighted hydrogen bond count is inversely proportional to the number of aromatic rings. This finding led us to propose the exchange role for hydrogen bonding interactions and -system involved intermolecular interactions. That is, a loss of binding affinity caused by a decrease in hydrogen bonding interactions is compensated by a gain in binding affinity acquired by an increase in -system involved intermolecular interactions, and vise verse. This discovery represents a significant contribution to our knowledge of molecular recognition of PKIs in protein kinases. The second half of this project concerns the classification of scaffolds for 2149 PKIs. Two scaffolds schemes “plain ring systems” and “Murcko scaffolds” were employed for the analysis. The scaffold analysis resulted in a systematic classification of the core structures of protein kinase inhibitors.
Project #2: We have established a library of 3D binding motifs, sampling all major types of non-bonded interactions occurring in representative PKI bound protein kinases, including 13 CH-π interactions, 12 π-π stacking interactions, 8 cation-π interactions, 8 hydrogen bonding interactions, 8 salt bridge interactions and 2 OH-π interactions. This library of 3D motifs based on real life PKI bound complexes will provide medicinal chemists with proven working structural motifs that can guide future design of next generation PKIs. Furthermore, intermolecular interaction energies for all 51 motifs in the library were calculated at the highest level of electronic structure theory currently available, i.e., the coupled cluster methods with single, double, and perturbative triple excitations [CCSD(T)]. A comparison of the resulting energies for all nonbonded interactions indicated that the non-bonded interactions involving π systems (CH-π interactions, π-π stacking interactions, cation-π interactions, OH-π interactions) are significant contributors to PKI binding, with magnitudes comparable to that of hydrogen bonding. Moreover, the CCSD(T) level intermolecular interaction energies for various nonbonded intermolecular interactions will serve as a valuable reference for theoreticians to benchmark newly developed methods of intermolecular interaction energy calculations in future.
Project #3: Recently, a new generation of DFT methods have been developed with dispersion correction, making it possible to apply DFT methods for analyzing biological systems that are interacting mainly through dispersion interactions. However, the accuracies of dispersion corrected DFTs varied widely. In this project, a systematically benchmark study was carried out to evaluate the accuracy of nine widely applied DFT methods, along with the semiempirical method PM6, for calculations of non-bonded interactions. For the purpose, two datasets of non-bonded systems were chosen: a prototype dataset consisting of 8 small model complexes and the library of 51 3D motifs based on real life PKI bound protein kinases established in project 2. The former dataset consists of water dimer, formamide dimer, guanidinium…methyl acetate dimer, ammonium…benzene dimer, benzene…CH4 dimer, benzene dimer (parallel), benzene dimer (T shape) and indole…benzene dimer. The CCSD(T)/CBS method, widely accepted as a gold standard, was used to benchmark the performance of the following DFT methods: RI-B2PLYP, RIJK RI-B2PLYP, RI-PWPB95, RIJK RI-PWPB95, B3LYP, M062X, PW6B95, RIJCOSX PW6B95, ωB97X, RI TPSS, RI B97, RI BLYP. D3 level of dispersion correction was included for all DFT methods except M062X. It was determined that the double-hybrid RIJK RI-B2PLYP functional is the best DFT method for the treatment of non-bonded interactions in terms of both accuracy and computational efficiency. Another interesting finding from this project was the reasonably good performance of the semi-empirical PM6 method for calculation of ligand-protein interaction energies.
Project #4: In this project, we undertook the challenge of studying molecular recognition of FDA approved drugs that target protein kinases. Decades of intense development of PKIs resulted in 44 FDA approved kinase drugs. After a thorough search of the PDB, it was found that 30 out of 44 PKI drugs have available structures of drug-protein complexes. Those 30 FDA-approved kinase drugs are associated with 39 different protein kinases since some drugs can bind to more than one target of protein kinases. Firstly, non-bonded interactions, including hydrogen bonding, salt bridge interactions, π-π stacking interactions, CH-π interactions, cation-π interactions, NH-π interactions, OH-π interactions, and SH-π interactions, were identified and systematically analyzed for all 39 drug-protein complexes. Subsequently, the strengths of intermolecular interactions for those non-bonded interactions were quantified by means of the double hybrid RIJK RI-B2PLYP method. The latter was identified as the best performing DFT method in Project 3 above. This work resulted in the discovery that π-system involved non-bonded interactions, like CH-π interaction, π-π stacking interactions, cation-π interaction, XH-π (XH=HN, OH, SH) interactions, play an important role in binding of FDA approved kinase drugs with protein kinases. It is our expectation that knowledge of molecular recognition of FDA approved kinase drugs gained from this project will have a direct impact on structure-based drug design of next generation kinase drugs with higher potent and better selectivity.