Computational and artificial intelligence-based approaches for drug metabolism and transport prediction.

Drug metabolism and transport, orchestrated by drug-metabolizing enzymes (DMEs) and drug transporters (DTs), are implicated in drug-drug interactions (DDIs) and adverse drug reactions (ADRs). Reliable and precise predictions of DDIs and ADRs are critical in the early stages of drug development to reduce the rate of drug candidate failure. A variety of experimental and computational technologies have been developed to predict DDIs and ADRs. Recent artificial intelligence (AI) approaches offer new opportunities for better predicting and understanding the complex processes related to drug metabolism and transport. We summarize the role of major DMEs and DTs, and provide an overview of current progress in computational approaches for the prediction of drug metabolism, transport, and DDIs, with an emphasis on AI including machine learning (ML) and deep learning (DL) modeling.

Role of Conformational Dynamics of Sulfotransferases SULT1A1 and SULT1A3 in Substrate Specificity.

Sulfotransferases (SULTs) are phase II metabolizing enzymes catalyzing the sulfoconjugation from the co-factor 3'-Phosphoadenosine 5'-Phosphosulfate (PAPS) to a wide variety of endogenous compounds, drugs and natural products. Although SULT1A1 and SULT1A3 share 93% identity, SULT1A1, the most abundant SULT isoform in humans, exhibits a broad substrate range with specificity for small phenolic compounds, while SULT1A3 displays a high affinity toward monoamine neurotransmitters like dopamine. To elucidate the factors determining the substrate specificity of the SULT1 isoenzymes, we studied the dynamic behavior and structural specificities of SULT1A1 and SULT1A3 by using molecular dynamics (MD) simulations and ensemble docking of common and specific substrates of the two isoforms. Our results demonstrated that while SULT1A1 exhibits a relatively rigid structure by showing lower conformational flexibility except for the lip (loop L1), the loop L2 and the cap (L3) of SULT1A3 are extremely flexible. We identified protein residues strongly involved in the recognition of different substrates for the two isoforms. Our analyses indicated that being more specific and highly flexible, the structure of SULT1A3 has particularities in the binding site, which are crucial for its substrate selectivity.

Investigating the Unbinding of Muscarinic Antagonists from the Muscarinic 3 Receptor.

Patient symptom relief is often heavily influenced by the residence time of the inhibitor-target complex. For the human muscarinic receptor 3 (hMR3), tiotropium is a long-acting bronchodilator used in conditions such as asthma or chronic obstructive pulmonary disease (COPD). The mechanistic insights into this inhibitor remain unclear; specifically, the elucidation of the main factors determining the unbinding rates could help develop the next generation of antimuscarinic agents. Using our novel unbinding algorithm, we were able to investigate ligand dissociation from hMR3. The unbinding paths of tiotropium and two of its analogues, N-methylscopolamin and homatropine methylbromide, show a consistent qualitative mechanism and allow us to identify the structural bottleneck of the process. Furthermore, our machine learning-based analysis identified key roles of the ECL2/TM5 junction involved in the transition state. Additionally, our results point to relevant changes at the intracellular end of the TM6 helix leading to the ICL3 kinase domain, highlighting the closest residue L482. This residue is located right between two main protein binding sites involved in signal transduction for hMR3's activation and regulation. We also highlight key pharmacophores of tiotropium that play determining roles in the unbinding kinetics and could aid toward drug design and lead optimization.

Small GTPase Ran: Depicting the nucleotide-specific conformational landscape of the functionally important C-terminus.

The small GTPase Ran is the main regulator of the nucleo-cytoplasmic import and export through the nuclear pore complex. It functions as a molecular switch cycling between the GDP-bound inactive and GTP-bound active state. It consists of a globular (G) domain and a C-terminal region, which is bound to the G-domain in the inactive, GDP-bound states. Crystal structures of the GTP-bound active form complexed with Ran binding proteins (RanBP) show that the C-terminus undergoes a large conformational change, embracing Ran binding domains (RanBD). Whereas in the crystal structures of macromolecular complexes not containing RanBDs the structure of the C-terminal segment remains unresolved, indicating its large conformational flexibility. This movement could not have been followed either by experimental or simulation methods. Here, starting from the crystal structure of Ran in both GDP- and GTP-bound forms we show how rigid the C-terminal region in the inactive structure is during molecular dynamics (MD) simulations. Furthermore, we show how MD simulations of the active form are incapable of mapping the open conformations of the C-terminus. By using the MDeNM (Molecular Dynamics with excited Normal Modes) method, we were able to widely map the conformational surface of the C-terminus of Ran in the active GTP-bound form, which allows us to envisage how it can embrace RanBDs.

Machine learning and structure-based modeling for the prediction of UDP-glucuronosyltransferase inhibition.

UDP-glucuronosyltransferases (UGTs) are responsible for 35% of the phase II drug metabolism. In this study, we focused on UGT1A1, which is a key UGT isoform. Strong inhibition of UGT1A1 may trigger adverse drug/herb-drug interactions, or result in disorders of endobiotic metabolism. Most of the current machine learning methods predicting the inhibition of drug metabolizing enzymes neglect protein structure and dynamics, both being essential for the recognition of various substrates and inhibitors. We performed molecular dynamics simulations on a homology model of the human UGT1A1 structure containing both the cofactor- (UDP-glucuronic acid) and substrate-binding domains to explore UGT conformational changes. Then, we created models for the prediction of UGT1A1 inhibitors by integrating information on UGT1A1 structure and dynamics, interactions with diverse ligands, and machine learning. These models can be helpful for further prediction of drug-drug interactions of drug candidates and safety treatments.

Sampling of Protein Conformational Space Using Hybrid Simulations: A Critical Assessment of Recent Methods.

Recent years have seen several hybrid simulation methods for exploring the conformational space of proteins and their complexes or assemblies. These methods often combine fast analytical approaches with computationally expensive full atomic molecular dynamics (MD) simulations with the goal of rapidly sampling large and cooperative conformational changes at full atomic resolution. We present here a systematic comparison of the utility and limits of four such hybrid methods that have been introduced in recent years: MD with excited normal modes (MDeNM), collective modes-driven MD (CoMD), and elastic network model (ENM)-based generation, clustering, and relaxation of conformations (ClustENM) as well as its updated version integrated with MD simulations (ClustENMD). We analyzed the predicted conformational spaces using each of these four hybrid methods, applied to four well-studied proteins, triosephosphate isomerase (TIM), 3-phosphoglycerate kinase (PGK), HIV-1 protease (PR) and HIV-1 reverse transcriptase (RT), which provide extensive ensembles of experimental structures for benchmarking and comparing the methods. We show that a rigorous multi-faceted comparison and multiple metrics are necessary to properly assess the differences between conformational ensembles and provide an optimal protocol for achieving good agreement with experimental data. While all four hybrid methods perform well in general, being especially useful as computationally efficient methods that retain atomic resolution, the systematic analysis of the same systems by these four hybrid methods highlights the strengths and limitations of the methods and provides guidance for parameters and protocols to be adopted in future studies.

Insights into the substrate binding mechanism of SULT1A1 through molecular dynamics with excited normal modes simulations.

Sulfotransferases (SULTs) are phase II drug-metabolizing enzymes catalyzing the sulfoconjugation from the co-factor 3'-phosphoadenosine 5'-phosphosulfate (PAPS) to a substrate. It has been previously suggested that a considerable shift of SULT structure caused by PAPS binding could control the capability of SULT to bind large substrates. We employed molecular dynamics (MD) simulations and the recently developed approach of MD with excited normal modes (MDeNM) to elucidate molecular mechanisms guiding the recognition of diverse substrates and inhibitors by SULT1A1. MDeNM allowed exploring an extended conformational space of PAPS-bound SULT1A1, which has not been achieved up to now by using classical MD. The generated ensembles combined with docking of 132 SULT1A1 ligands shed new light on substrate and inhibitor binding mechanisms. Unexpectedly, our simulations and analyses on binding of the substrates estradiol and fulvestrant demonstrated that large conformational changes of the PAPS-bound SULT1A1 could occur independently of the co-factor movements that could be sufficient to accommodate large substrates as fulvestrant. Such structural displacements detected by the MDeNM simulations in the presence of the co-factor suggest that a wider range of drugs could be recognized by PAPS-bound SULT1A1 and highlight the utility of including MDeNM in protein-ligand interactions studies where major rearrangements are expected.

Revealing the activation mechanism of autoinhibited RalF by integrated simulation and experimental approaches.

RalF is an Arf GEF from Legionella pneumophilia, the bacterium that causes severe pneumonia. In its crystal structure, RalF is in the autoinhibited form. A large-scale domain motion is expected to lift the autoinhibition, the mechanism of which is still unknown. Since RalF is activated in the presence of the membrane, its active structure and the structure of the RalF-Arf1 complex could not have been determined experimentally. On the simulation side, it has been proven that classical Molecular Dynamics (MD) alone is not efficient enough to map motions of such amplitude and determine the active conformation of RalF. In this article, using Molecular Dynamics with excited Normal Modes (MDeNM) combined with previous experimental findings we were able to determine the active RalF structure and the structure of the RalF-Arf1 complex in the presence of the membrane, bridging the gap between experiments and simulation.

Nucleotide-Specific Autoinhibition of Full-Length K-Ras4B Identified by Extensive Conformational Sampling.

K-Ras is one of the most frequently mutated oncogenes in human tumor cells. It consists of a well-conserved globular catalytic domain and a flexible tail-like hypervariable region (HVR) at its C-terminal end. It plays a key role in signaling networks in proliferation, differentiation, and survival, undergoing a conformational switch between the active and inactive states. It is regulated through the GDP-GTP cycle of the inactive GDP-bound and active GTP-bound states. Here, without imposing any prior constraints, we mapped the interaction pattern between the catalytic domain and the HVR using Molecular Dynamics with excited Normal Modes (MDeNM) starting from an initially extended HVR conformation for both states. Our sampling captured similar interaction patterns in both GDP- and GTP-bound states with shifted populations depending on the bound nucleotide. In the GDP-bound state, the conformations where the HVR interacts with the effector lobe are more populated than in the GTP-bound state, forming a buried thus autoinhibited catalytic site; in the GTP-bound state conformations where the HVR interacts with the allosteric lobe are more populated, overlapping the α3/α4 dimerization interface. The interaction of the GTP with Switch I and Switch II is stronger than that of the GDP in line with a decrease in the fluctuation upon GTP binding.