What makes a molecule a drug? i.e. drug-likeness is a question of great theoretical and practical importance in drug design. Pharmacokinetics and drug-target binding affinity are two equally important aspects of drug-likeness. The former was elegantly profiled by the Lipinski’s rule-of-five which had a major influence on both the selection of compounds for high-throughput screening and the design of lead generation libraries over the past two decades. The latter has the promise to profoundly impact lead optimization in drug discovery. However, currently, there exist no clear guidelines that systematically delineates the molecular determinants for drug-target binding. The latter forms the main subject of investigation of this thesis.
We have carried out a structural and chemoinformatic analysis of 1589 FDA-approved drugs. A wide variety of molecular descriptors have been analyzed. This analysis resulted in the finding that about 80% of the FDA-approved drugs are aromatic molecules. More importantly, an inverse correlation between the number of hydrogen bonding (donors + acceptors) and the number of aromatic rings of drugs has been discovered. Accordingly, we hypothesized that aromatic groups of drug molecules may play a pivotal role in drug-protein binding.
To test this hypothesis, a large-scale data mining of the Protein Data Bank was carried out to decipher molecular determinants for recognition of the FDA-approved drugs by proteins. Non-bonded intermolecular interactions (hydrogen bonding, ionic interactions, p-p stacking interactions, CH-p interactions, and cation-p interactions) between drug molecules and their surrounding protein residues in their binding pockets were systematically profiled for the 161 available drug-protein complexes. It was found that the aromatic moieties of drug molecules can form a large number of intermolecular interactions with their surrounding protein residues, via CH-p interactions and p-p stacking interactions, and to a lesser extent, via cation-p interactions.
Furthermore, high-level quantum chemical calculations were performed to quantify energetics of drug-protein binding. For this purpose, two parallel projects were carried out. In the first project, representative 3D motifs for each type of intermolecular interaction modes were selected to establish a library of intermolecular interactions in drug-protein binding. The library consists of a total of 56 representative 3D drug binding motifs, including 11 hydrogen bonding pairs, five salt-bridge pairs, 17 p-p interaction pairs, 16 CH-p interaction pairs and 7 cation-p interaction pairs. A high-level CCSD(T)/CBS quantum chemical calculation was performed to quantify energetics of binding for those representative modes of interactions, demonstrating the substantial strengths of CH-p interactions and p-p stacking interactions. In the second project, drug-protein interactions in 19 represntative drug-protein complexes were comprehensively analyzed in its entirety. To quantify the contribution of interactions involving aromatic rings of drugs for their binding in protein targets, we have calculated drug-protein binding energies in the 19 drug-protein complexes at the RI-MP2/cc-pVTZ level. In addition to confirming the importance of hydrogen bonding and electrostatic interactions for drug-protein binding, it was found that the contributions arising from interactions involving aromatic p moieties of drugs (e.g. p-p stacking, CH-p interactions and cation-p interactions) is dominantly significant.
Our molecular recognition study was further extended to a project aiming at understanding the underlying mechanism behind the robust resistance profiles of the second generation non-nucleoside transcriptase inhibitors etravirine and rilpivirine against HIV-1 RT mutants that are resistant to the first generation non-nucleoside inhibitors. A quantitative analysis of etravirine and rilpivirine binding pockets in the wild-type HIV-1 RT and several clinically important mutant strains at the molecular level was carried out using a high level quantum chemical calculation (MP2/cc-pVTZ). The contributions of individual amino acids of RTs that constitute their binding pockets have been evaluated in a pairwise manner. This work resulted in the discovery that non-bonded intermolecular interactions involving aromatic moieties of these drugs are critical for their unattenuated activities against mutated RTs. In particular, interactions between etravirine/rilpivirine and the highly conserved amino acids: Trp229, Leu234, Phe227, Tyr318 and Val106 persist despite mutations. It is expected that a detailed understanding of the intermolecular interactions responsible for the specific molecular recognition between etravirine/rilpivirine and RTs as reported here will have far-reaching implication for the rational design of new NNRTIs that are even more resilient to mutations.
Moreover, electronic structure calculations of non-bonded intermolecular interactions for large biomolecular systems represent a challenging undertaken due to the large system size and the necessity to include electron correlation correction. Traditional wave function based electron correlation methods, such as the configuration interaction (CI) method and the coupled cluster (CC) method, are accurate, but very demanding in computational resources. The newly emerged variants of the DFT method, named dispersion-corrected DFT methods, represent an alternative approach for the treatment of electron correlation in many-electron systems. Thus, as an integral part of our focus in this thesis, a benchmark study of nine popular dispersion-corrected DFT methods for noncovalent interactions in eight prototypic model systems was carried out. The gold standard CCSD(T) method was used to benchmark the performance of the following DFT methods: RI-B2PLYP, RIJK RI-B2PLYP, RI-PWPB95, RIJK RI-PWPB, B3LYP, M062X, PW6B95, RIJCOSX PW6B95, ¿B97X, RI TPSS, RI B97, RI BLYP. For each DFT method, three basis sets of Ahlrichs def2 were applied: QZVP, TZVP, and SVP, with the three-body D3 level of dispersion correction. It was found that dispersion-corrected DFT methods are highly accurate, with the best performers ranking as: (1) RIJK RI-B2PLYP, (2) RIJCOSX PW6B95 and (3) RIJK RI-PWP95. In addition, the multibody effect in drug-protein interactions was studied in a set of 19 drug-protein complexes by means of a low level dispersion-corrected DFT method using the TPSS density functional, employing both the pairwise additive approach and the whole-body approach. It was observed that, on average, the intermolecular interaction energies obtained using the pairwise additive approach yielded an overall error of 10.06%.