Seminars2017 –

Calcium ions play crucial roles in many important biological processes, such as muscle contraction, signal transduction in nervous system, cell signaling regulation and cell fate determination. However, the existing Ca²⁺ models in the classical non-polarizable force field are inaccurate in calculating the interaction energies between calcium ions and protein molecules, which has hindered the computational studies of Ca²⁺-interacting proteins. We developed a new multi-site Ca²⁺ model in the framework of the classical non-polarizable force field, which can not only reproduce the energetical and dynamical properties of calcium ions in solution, but also describe the interaction energies between calcium ions and proteins more accurately. We utilized this new Ca²⁺ model to simulate ion permeation through the RyR channels, and revealed the detailed Ca²⁺ permeation behavior. We hope this Ca²⁺ model can find more applications soon.

The significant progress of the cryo-EM poses a pressing need for software for structural interpretation of EM maps. Particularly, protein structure modeling tools are needed for EM maps determined at a resolution around 4 Å or lower, where finding main-chain structure and assigning the amino acid sequence into EM map is challenging. In this seminar, we discuss computational protein structure modeling tools we have been developing and future directions, opportunities, and challenges.
We have developed a de novo modeling tool named MAINMAST (MAINchain Model trAcing from Spanning Tree) for EM maps for this resolution range (Nature Communications, 2018). MAINMAST builds main-chain traces of a protein in an EM map from a tree structure constructed by connecting high-density points without referring to known protein structures or fragments. The method has substantial advantages over the existing methods. MAINMAST showed better modeling performance than existing methods. The method is further enhanced recently to be able to model symmetric protein complexes and ligand (drug) molecules that bind to a protein in a map. Moreover, to provide structure information for maps determined at a lower resolution (5~10 Å), we have recently developed a new tool, Emap2sec, which uses convolutional deep learning for detecting secondary structures of proteins (accepted and to appear, 2019). Emap2sec scans an EM map with a voxel and assigns a secondary structure, i.e. alpha helix, beta strand, or coil, from density patterns of the voxel and its neighbors.

I will describe methods we have developed through the years aiming at incorporating the effects of nuclear motions and states in theoretical computations, ranging from MD and QM/MM in the prediction of electronic spectra to full quantum treatment of many-mode nuclear motion. I will discuss why and how we have changed from a QM/MM approach for classical simulations of electronic spectra to so-called incremental type approaches for constructing quantum computed potential energy surfaces for many mode vibrational computations. In the context of many-mode dynamics progress in the vibrational coupled cluster approach developed in our group through the years will be discussed. Along the way I will also discuss the perspectives of boosting computations using techniques such as machine-learning and tensor decomposition. I will finally also discuss the computation of Franck-Condon factors using anharmonic wave functions with thiophenes as an example.

I will present recent advances in quantum chemistry with an emphasis on software development in the BAGEL program package. First, I will show that, with BAGEL, one can routinely perform first principles all-electron DFT simulations of 1000-2000 atoms in ∼15 min using a computer cluster with a few dozen of nodes, which (we hope) will allow the users to replace some of the computational protocols conventionally performed by classical force fields. Second, I will review several high-accuracy wave function methods in BAGEL, which would replace some of the simulations currently done by DFT. Finally, I will discuss how parallel software, like BAGEL, can accelerate the users’ workflow and change how quantum chemistry is used in academia and in industry.

In this talk, I would like to share our ongoing efforts toward computational (structural) glycobiology in terms of (1) glycan structure and dynamics in glycoproteins, (2) roles of glycans as ligands in protein-glycan and protein-protein interactions, (3) glycolipid structure and dynamics, and (4) bacterial outer membranes containing lipopolysaccharides and their interactions with membrane proteins. We have developed various tools available in GlycanStructure.ORG (http://www.glycanstructure.org): Glycan Reader for automatic detection and annotation of carbohydrates, their chemical modifications, and glycosidic linkages in PDB files, Glycan Fragment DB for finding carbohydrate fragment structures in the PDB and torsion angle distributions of specific glycosidic linkages of a query glycan structure, Glycan Modeler for modeling glycan structures from its sequence, and GS-align for glycan structure alignment and similarity measurement. In addition, we are in the process of building Glycan Binding Structure Database and Glycan Microarray Database. A PDB survey study of N-glycan structures and protein-glycan interactions is also presented for modeling glycan structures and protein-glycan interactions.

The 12−6 Lennard-Jones (LJ) nonbonded model is routinely used to represent metal ions in the simulation of the structure and thermodynamics of a range of coordination compounds. However this model often fails to simultaneously reproduce these properties, which limits its applicability. Our 12−6−4 LJ-type nonbonded model, that includes a 1/r ⁴ term to incorporate charge-induced dipole interactions, reproduces multiple experimental properties of highly charged metal ions [1-3]. We recently optimized 12−6−4 LJ parameters for Cd²⁺, Ni²⁺, Fe²⁺, and Zn²⁺ binding to ethylenediamine (en) in water in order to capture detailed mechanistic insights into the chelate effect [4]. This study highlighted the role of water molecules in the first solvation shell of the metal ion in facilitating chelate ring formation. In this talk, we’ll present recent efforts in furthering the application of our model to simulate the structural and thermodynamic properties of the self-assembly process involving metal ions with organic and biological ligands.

References:
[1] Li, P. and Merz, K. M., Jr. J. Chem. Theory Comput., 2014, 10, 289−297.
[2] Li, P.; Song, L. F. and Merz, K. M., Jr. J. Phys. Chem. B, 2014, 119, 883−895.
[3] Li, P.; Song, L. F. and Merz, K. M., Jr. J. Chem. Theory Comput., 2015, 11, 1645−1657.
[4] Sengupta, A. Seitz, A.; Merz, K. M., Jr. J. Am. Chem. Soc., 2018, 140, 15166–15169.

The basic outcome of an ordinary molecular simulation is a trajectory, or sequence of molecular configurations, and any equilibrium or non-equilibrium observable can be derived from a sufficiently long trajectory. Likewise, arbitrary observables can be derived from an appropriate ensemble of trajectories. The talk will explain how the basic physics of trajectory ensembles can be exploited for enhanced sampling and for improved analysis of ordinary trajectories via Markov models. In particular, the “weighted ensemble” enhanced sampling approach has been employed to estimate protein folding times up to the second timescale. Further, the related trajectory theory enables construction of Markov models of protein folding that remain unbiased at very short lag times, greatly enhancing their capability to describe fundamental aspects of mechanism (pathways).

In this talk I will present the results of our analyses of amyloid formations of peptide fragments by replica-exchange molecular dynamics simulations. Two peptide fragments were studied. For the former system, we found that there is a clear phase transition temperature in which the peptides aggregate with each other. Moreover, we found by the free energy analyses that there are two major stable states: One of them is like amyloid fibrils and the other is amorphous aggregates. For the latter system, we focused on the concentration dependency. We showed that high concentration environment of fibril-forming peptides is likely to cause the protein aggregation.

Considerable progress has been made, using experiments and computations, to decipher the general principles governing the mechanism of formation of oligomers and fibrils of amyloid proteins implicated in diseases, including the amyloid-β protein (Aβ) associated with Alzheimer’s disease (AD). However, the identification of the link between protein aggregation and the systems of disease at the molecular level has proved elusive. The biogenesis of Aβ starts with interaction of the Amyloid Precursor Protein (APP) with secretases in the presence of membrane. Subsequently, interactions with cholesterol and other proteins such as the cellular prion protein (PrPC) determine the route to oligomer formation and the extent of cytotoxicity. We report on theoretical and computational studies designed to systematically investigate the biogenesis of Aβ, its propensity toward aggregation, and putative mechanisms of cytotoxicity in order to highlight critical areas for future research.

Intrinsically disordered proteins (IDPs) are increasingly realized to play a wide range of functional as well as pathological roles in biology. However, biophysical characterization of these proteins is experimentally challenging due to the extremely heterogeneous ensemble of structures which they populate. Computational tools, in particular molecular simulations, can therefore play a role in elucidating structure, function and dynamics in IDPs. Here, Dr. Best will show how both detailed atomistic simulations, as well as simplified coarse-grained models can be used to assist in the interpretation of experiments on IDPs. In particular, Dr. Best will describe recent work characterizing an ultra-high affinity complex between two charged IDPs which, remarkably, remain completely disordered upon binding.

Computation has emerged as a powerful tool for accelerating the discovery of new materials and molecules: first through first-principles simulation in high throughput screening and very recently even further with machine learning (ML). In the first part of my talk, I will discuss the unique challenges that remain for the accelerated discovery and design of inorganic complexes, despite the highly tunable electronic structure properties that make these materials so compelling for applications in energy storage and catalysis. I will describe our recent efforts to overcome the high cost and low accuracy of electronic structure for inorganic chemistry through developing ML models (e.g., artificial neural networks and kernel ridge regression) that predict key first-principles energetic (e.g., spin-state ordering and redox potential) and geometric properties to within the accuracy of the underlying simulation method that provides the training data. I will describe our efforts to use these ML models for extrapolative applications (i.e., design) by incorporating heuristics for ML model uncertainty in a genetic algorithm to discover new inorganic compounds. In the second part of my talk, I will discuss how we leverage recent advances in stream processors to carry out large-scale electronic structure in QM and QM/MM simulation with over 1000 atoms treated quantum mechanically. I will describe how these simulations have given us new insights into enzyme mechanism, what potential challenges there are in applying conventional electronic structure methods on such large systems, and how we can understand when large scale electronic structure is needed in enzyme simulation.

Communication and transport across lipid membranes control a large variety of cellular processes but remain poorly understood, largely because membrane proteins are difficult to study experimentally. The scarcity of high-resolution membrane protein structures and mechanistic insights hinder the development of effective therapeutics, with membrane proteins estimated to represent around 60% of possible cellular drug targets. To address these challenges, we have recently developed an ensemble of integrated computational/experimental approaches to accurately model and design membrane protein structures and functions. With our methods, we can (1) predict the functional consequences of membrane protein sequence variations, (2) uncover new molecular determinants of membrane protein structure and function, and (3) rationally design membrane receptors with novel biophysical and signaling properties. We leverage these combinatorial approaches to engineer biosensors, as well as reprogram and create novel signaling pathways for applications in synthetic, systems biology and personalized medicine, including immunotherapeutic interventions.

Chromatin architecture plays a key role in embryonic stem cell programming, human embryo development, brain function and cancer. Specifically, epigenetic methylation and acetylation marks are thought to control gene expression by dramatically altering global chromatin architecture; however the exact mechanism by which a single methyl group can induce a large scale conformation change of chromatin is not well understood. By examining histones in a dense nucleosome context, we aim to gain insight into possible scenarios by which methylation can alter chromatin conformation. Using coarse grain models of chromatin as a basis, we construct all atom chromatin models and simulate these with the GENESIS molecular dynamics code on the large-scale high performance platforms at Los Alamos National Laboratory. The multi-disciplinary effort combined computer science, high performance computing, chip design, biophysics, structural biology, and cell biology, including researchers from RIKEN, LANL, NYU, Intel and Cray. Several performance optimizations for the KNL architecture enabled scaling to large numbers of cores.

The influence of lone-pair electrons on the directionality of hydrogen bonds that are formed by oxygen and nitrogen atoms in the side chains of nine hydrophilic was investigated using molecular dynamics simulations. The simulations were conducted using two types of force fields; one incorporated lone-pair electrons placed at off-atom sites and the other did not. The density distributions of the hydration water molecules around the oxygen and nitrogen atoms were calculated from the simulation trajectories, and were compared with the empirical hydration distribution functions, which were constructed from a large number of hydration water molecules found in the crystal structures of proteins. Only simulations using the force field explicitly incorporating lone-pair electrons reproduced the directionality of hydrogen bonds that is observed in the empirical distribution functions for the deprotonated oxygen and nitrogen atoms in the sp ²-hybridization. The amino acids that include such atoms are functionally important glutamate, aspartate, and histidine. Therefore, a set of force field that incorporates lone-pair electrons as off-atom charge sites would be effective for considering hydrogen bond formation by these amino acids in molecular dynamics simulation studies.

Docking (posing) calculations coupled with binding free energy estimates (scoring) are a mainstay of structure based drug design. Docking and scoring methods have steadily improved over the years, but remain challenging because of the extensive sampling that is required, the need for accurate scoring functions and challenges encountered in accurately estimating entropy effects. To address these issues we have been developing a number of novel strategies in our laboratory. In particular, in this presentation, we will describe the Movable Type sampling (MTS) method developed in our laboratory that estimates binding free energies, entropies and enthalpies. The utility of MTS will be explored through a series of examples that involve rigid or flexible protein-ligand docking, protein-protein docking and entropy estimation. We will show that MTS allows us to compute thermodynamic quantities associated with myriad biological processes rapidly, accurately and yields structural information at a minimal computational cost relative to currently available methods.

I will start with an overview of my research program, which includes studies on protein association, macromolecular crowding, and peptide self-assembly. The focus will then shift to our studies on ion channels. These membrane proteins, relative to water-soluble proteins, have less intrinsic stability and are more prone to influences of the solubilizing environments. Indeed, our recent assessment of helical membrane protein structures in the Protein Data Bank identified many cases of potential distortions in transmembrane domains, attributable to sample preparations used for X-ray crystallography and solution NMR spectroscopy [1]. To achieve native-like structures, we use solid-state NMR data for refinement through restrained molecular dynamics simulations in native-like environments, i.e., in lipid bilayers. For the Influenza M2 protein (an acid-activated proton-selective channel), our study further targeted its functional center, i.e., a histidine tetrad within the channel pore that acts as both the pH sensor and ion selectivity filter. Based on solid-state NMR data and quantum chemistry calculations, we developed a mechanism for acid activation and proton conductance [2]. We have also remodeled transmembrane domains from crystal structures, both for correcting distortions [3] and for generating structural models in different functional states [4]. Lastly we have used molecular dynamics simulations and structural modeling to develop mechanisms for ionotropic glutamate receptors on channel gating, partial agonism, and disease-associated mutations, and are integrating these results into electrophysiological studies [5, 6].

[references]
1. H.-X. Zhou and T. A. Cross (2013). Influences of membrane mimetic environments on membrane protein structures. Annu. Rev. Biophys. 42, 361-392.
2. M. Sharma, M. Yi, H. Dong, H. Qin, E. Peterson, D. D. Busath, H.-X. Zhou, and T. A. Cross (2010). Insight into the mechanism of the Influenza A proton channel from a structure in a lipid bilayer. Science 330, 509-512.
3. G. Heymann, J. Dai, M. Li, S. D. Silberberg, H.-X. Zhou, and K. J. Swartz (2013). Inter- and intrasubunit interactions between transmembrane helices in the open state of P2X receptor channels. Proc. Natl. Acad. Sci. USA 110, E4045-E4054.
4. J. Dai and H.-X. Zhou (2014). General rules for the arrangements and gating motions of pore-lining helices in homomeric ion channels. Nat. Commun. 5, 4641.
5. R. Kazi, J. Dai, C. Sweeney, H.-X. Zhou, and L. P. Wollmuth (2014). Mechanical coupling maintains the fidelity of NMDA receptor-mediated currents. Nat. Neurosci. 17, 914-922.
6. H.-X. Zhou and L. P. Wollmuth (2017). Advancing NMDA receptor physiology by integrating multiple approaches. Trends Neurosci. 40, 129-137.

Bacterial ribosomal RNA (rRNA) is a target for small molecule antibiotics whose binding inhibits protein synthesis. However, rRNA constitutes two-thirds of the ribosome by mass so it offers many other possible interaction sites. We explored bacterial rRNA as a target for complementary oligomers that would bind observing the Watson-Crick pairing rules. We analysed various properties of the rRNA regions such as accessibility, functionality, hydrogen bond patterns, easiness of opening for strand invasion and flexibility. To determine 16S rRNA flexibility in the ribosome context, we performed all-atom molecular dynamics simulations of the small ribosome subunit in explicit solvent. Based on these properties we selected rRNA targets for hybridization with complementary oligoribonucleotides. Next, we tested translation inhibition efficiencies of these ribosome-interfering oligomers in a cell-free translation system. Selected rRNA sites were targeted with peptide nucleic acid oligomers and tested for inhibition of bacterial growth.

Understanding the universal mechanism of glass transitions is a challenging problem for condensed phases, despite extensive efforts in theories, simulations, and experiments. A remarkable feature of glass-forming liquids is the drastic slowing down that accompanies non-exponentially and non-Gaussianity observed in various time correlation functions. On the contrary, the amorphous structures upon supercooling remain unchanged and are similar to those in liquid states. In this talk, I first provide the general review about glass transition problem and then introduce my recent simulation studies. The particular interest is related to temperature dependence of transport coefficients such as diffusivity, viscosity, and structural relaxation time in glasses. This temperature dependence is characterized by the degree of the Arrhenius property, which is referred to as fragility. It is well known that anisotropic tetrahedral network-forming liquids (SiO2) exhibit the Arrhenius behavior, while isotropic short-ranged potential liquids (metallic alloys) act as another type of glass former exhibiting super-Arrhenius temperature dependence. Here, it is demonstrated that the fragility can be controlled over a wide range by tuning the potential in a single simulation model. This model uses the short-ranged and isotropic pairwise potential. However, the reduction of the potential depth, eventually transforming from tetrahedral into isotropic structures, seamlessly changes the temperature dependence from Arrhenius to super-Arrhenius.

Protein folds in to a unique structure, but have some flexibility to function efficiently. The importance of flexibility, or protein dynamics such as configurational fluctuations and conformational transitions, have become evident in recent studies, yet understanding how it acts, especially at molecular level, is still a challenging task. In this talk I will discuss our recent studies on two topics, protein folding and enzymtic reactions. For the folding, we analyze multiple ∼μs long molecular dynamics trajectories from Anton to study how folding/unfolding proceed behind a seemingly two-state folding free energy profile. For the enzymatic reaction, the peptidyl-prolyl cis-trans isomerization reaction in Pin1 is studies, and the transition mechanism is discussed in detail. These results show that the heterogeneous dynamics of the proteins found at molecular level play a fundamental role in folding into the native structure and catalyzing the reaction efficiently.