Molecular Dynamics Simulation Combined With Deep Learning To Explore The Effect Of Amino Acid Mutation Or Ligand Binding On Enzyme Activity

Protein(enzyme)function is highly dependent on its sophisticated molecular structure,the interplay of residues in protein can give rise to complex behavior to perform function.And the amino acid mutation or the combination of different ligands may cause changes in protein function,which are often accompanied by shifts in protein conformation and atomic/residue interactions.The development of molecular dynamics(MD)simulation has now become an important theoretical tool for studying protein structure information.The statistics and analysis of simulated trajectories are essential to understand the relationship between protein structure and function.However,no analytical method has been developed to infer the interactions that drive the complex movement of proteins.In recent years,graph neural networks(GNN)have been widely used in dynamic scenarios,such as simple physical movement and skeleton action recognition.Therefore,the application of graph neural networks to process data related to biomolecular dynamic systems is the future development trend.In this study,molecular dynamics simulation combined with graph neural network is used to in-depth study of enzymes with different regulatory mechanisms related to diseases.The specific research contents are as follows:1.The effect of mutation Y68I/G109 P on the catalytic activity of CbDPEaseThe anti-diabetic and anti-obesity physiological functions of D-psicose make it crucial to increase the yield of D-psicose.Since it is extremely scarce in nature and difficult to synthesize,it is important to increase the production of D-psicose 3-epimerase(CbDPEase)by increasing its catalytic efficiency.To study the theoretical mechanism of the mutation Y68I/G109 P to improve the catalytic activity of CbDPEase,MD simulation and adaptive steered molecular dynamics(ASMD)simulation were employed to conduct a detailed study on the complex of the D-fructose combined with wild-type and Y68I/G109 P CbDPEase.The results show that Y68I/G109 P enhanced the betweenness of functional residues in CbDPEase,leading to the increased interactions between D-fructose and CbDPEase.During the dissociation of the Dfructose from the Y68I/G109 P mutant,planes of benzene rings of F248 and W114 could be continuously parallel to the stretching direction of D-fructose.It made the tunnel have an open state and resulted in the stable interactions between D-fructose and the benzene rings.2.The effect of activating mutations(E203K and P124S)on the inhibition of MEK1 by trametinibActivation of the mitogen-activated protein kinase(MAPK)signaling pathway regulated by human MAP kinase 1(MEK1)is associated with the carcinogenesis and progression of numerous cancers.In addition,two active mutations(E203K and P124S)have been reported to enhance the activity of MEK1,thereby eventually leading to the tumorigenesis of cancer.Therefore,MD simulation and steered molecular dynamic(SMD)simulation were performed to explore the effects of active mutations(E203K and P124S)on the changes in the interaction of MEK1 with trametinib.The simulation results show that the active mutations E203 K and P124 S changed the secondary structure of the activation segment in MEK1 from a helix to a loop,and its conformation changed from a closed state to an open state.Especially for P124 S MEK1,the reverse movement between core kinase 1 and the activation segment reduced the interactions between MEK1 and trametinib,thereby weakening the inhibitory effect of trametinib on MEK1.Moreover,two active mutants made the allosteric tunnel of MEK1 wider and shorter than that of the non-active types(WT and A52 V MEK1).Hence,trametinib could dissociate from the active MEK1(E203K and P124 S MEK1)more easily compared with the wild-type MEK1.3.The effect of ligands on structural stability and selectivity of SULT2A1Cytosolic sulfotransferases(SULTs)acting as phase Ⅱ metabolic enzymes can be used in the sulfonation of small molecules by transferring a sulfonate group from the unique co-factor 3’-phosphoadenosine 5’-phosphosulfate(PAPS)to the substrates.In the present study,MD simulation and ensemble docking study were employed to theoretically characterize the mechanism for the effect of co-factor(PAP)and substrates/inhibitors(LCA,raloxifene,α-hydroxytamoxifen,ouabain,and 3’-phosphoadenylyl sulfate)on structural stability and selectivity of SULT2A1.During the MD simulation,the obviously rigid region and inward displacement were detected in the active-site cap(loop16)of the conformation containing PAP,which may be responsible for the significant changes in substrate accessibility and catalytic activity.The smaller substrates,such as LCA,could bind stably to the active pocket in the presence of PAP.However,the substrates or inhibitors with a large spatial structure,such as 3’-phosphoadenylyl sulfate and raloxifene,needed to bind to the open conformation(without PAP)prior to PAPS binding.4.Neural relational inference to learn interaction networks in proteins from molecular dynamics simulationTo develop MD simulation analysis methods to capture the interaction changes that drive the complex motion of proteins,we first attempted to apply a neural relational inference(NRI)model based on a GNN to analyze MD simulation in biological systems.In the allosteric regulation case study,protein Pin1 utilizes the allostery induced by ligand binding to regulate its function.NRI model successfully learned that the ligand binding made the WW domain and the peptidyl-prolyl cis/trans isomerase(PPIase)domain more coordinated and compact to strengthen the allosteric communication between the binding site and catalytic site.In addition,we used the NRI model to learn the simulated trajectories of the previous three systems(CbDPEase,SULT2A1,and MEK1).In the CbDPEase system,the Y68I/G109 P mutation enhanced the interaction between the domains surrounding the active site,so that the substrate D-fructose interacted strongly with the active site residues.In the MEK1 system,we found that there was almost no interaction between the residues/domains in non-active MEK1.On the contrary,in the active MEK1(E203K and P124 S MEK1),αA-helix,αC-helix,core kinase domain-1,activation segment,and proline-rich loop interacted strongly with the whole structure.Among them,the activation segment acted as a "messenger" to mediate signal propagation in MEK1.In the SULT2A1 system,we found that the binding of PAP enhanced the interaction between loop12 and loop16,which promoted the inward displacement of the active site cap(loop16)and lead to the closure of the active pocket.According to mutation and ligand binding sites,the enzymes involved in this study can be divided into two modes of regulation.One is that mutations and ligand binding sites are located near the active site,which can directly regulate the active site conformation(such as the regulation in the CbDPEase and SULT2A1);the second is that the mutation and ligand binding sites are far away from the active site,and the active state of the enzyme are changed through allosteric regulation(such as the regulation in the MEK1 and Pin1).This study shows that the NRI model is capable of inferring the interaction of residues that drive protein movement under different regulatory mechanisms,providing a new perspective for trajectory analysis.