Three-dimensional Structures Of Biological Macromolecules

National Heart, Lung, And Blood Institute

Investigators

Linked publications, trials & patents

Abstract

Protein structure prediction via deep learning of protein folding Modern protein structure prediction (PSP) systems generally comprise four components: (i) an input module that takes a single protein sequence to generate additional input features, almost always including a multiple sequence alignment (MSA) of homologous proteins, (ii) a trunk, typically a neural network capable of sophisticated pattern recognition, which transforms features from the input module to spatial information that partially encode the 3D structure, (iii) an output module that converts this spatial information into an initial 3D structure, sometimes without explicit side-chain atoms, and (iv) a refinement module that improves the initial structure and produces all atomic coordinates. Traditionally, these modules relied on a mixture of physics-based energy functions, knowledge-based statistical reasoning, and heuristic algorithms. The last few years however have witnessed an infusion of machine learning, particularly neural networks, into every aspect of PSP. What started as a trickle of progress accelerated over the subsequent decade and, last year, reached a crescendo with DeepMind's AlphaFold2 14, a system that arguably solves single apo domain PSP. Currently ML for PSP use binary contact map (BCM) or discretized inter-residue distances as output. All information comes from existing structures. To improve ML for PSP, increase information source will be beneficial. We believe protein folding pathway will provide abundant information for PSP. Therefore, we employ ML to recognize the folding movement at every stage of folding pathway to produce the movement of proteins. We utilize the Nudged elastic band simulation to produce pathway from the extended state to the folded state. The movement of protein at each conformation are studied with ML. This work is still in progress and hopefully lead to more accurate PSP, as well as folding pathway identification. Laser Induced Alignment of Proteins for Single Particle Imaging Laser-induced alignment of particles and molecules was long envisioned to support three-dimensional structure determination using single-particle imaging with x-ray free-electron lasers PRL 92, 198102 (2004). However, geometric alignment of isolated macromolecules has not yet been demonstrated. We developed a simulation framework in which protein-laser interactions are simulated treating proteins as rigid bodies, which allows the modeling of large number of particles (hundred thousand) for a long-time scale. Using this simulation framework, we analyzed and demonstrated how the alignment of large nanorods and proteins is possible with standard laser technology, and performed a comprehensive analysis on the dependence of the degree of alignment on molecular properties and experimental details. Calculations of the polarizability anisotropy of about 150,000 proteins yielded a skew-normal distribution with a location of 1.2, which reveals that most of these proteins can be aligned using appropriate, realistic experimental parameters. Moreover, we explored the dependence of the degree of alignment on experimental parameters such as particle temperature and laser-pulse energy. Structure analysis and simulation study of microtubule dynamics Microtubules are a structurally and functionally important components of the eukaryotic cytoskeleton that play a crucial role in cell division and intracellular trafficking. Disrupting microtubule dynamics is a major strategy in cancer therapy. Through structure analysis of tubulin conformations in microtubules and in unassembled forms, we discover a rotation of the tubulin intermediate domain that switches tubulins from a closed state to an open state. This rotation is responsible for the conformational change of tubulins during microtubule growth. Based on the observation that a GDP shift coincident with the rotation; I propose a hypothesis that the GTP hydrolysis produces a GDP stroke that causes the rotation. Through self-guided Langevin dynamics simulations of tubulin monomers and heterodimers, with and without the GDP stroke, this work proves that the GDP stroke does cause the rotation. At the closed state, tubulins polymerize into a curved protofilament. In the open state tubulins can dock into the open pockets to form a straight protofilament. Lateral interactions between straight protofilaments stabilize microtubules. Based on these results, this work proposes a hydrolysis driven mechanism that can well describe microtubule dynamics. This new mechanism may facilitate new strategies in regulating microtubule dynamics. Identify the conformational states of glycine receptor alpha3 through umbrella sampling Ligand-gated ion channels allows cells to respond rapidly to changes in their external environment. The structure change from one state to other states is the key to understand how ion channels function. In many cases due to the limitation in experiment, only some state structures, a closed, open state, or desensitized state, are available. This work presents a method that utilizes the umbrella sampling to drive conformation changes from one state to the other state. The trans-membrane domain of glycine receptor is embedded in POPC lipids. A reaction coordinate describing relative orientation of lining substructures of ion channels is proposed and the free energy profile along the reaction coordinates is produced. For glycine receptor alpha-3 pentamer, we find that there are two free energy wells separated by a barrier in the free energy profile. The desensitized state corresponding to one well at a larger reaction coordinate and the closed state corresponding to the other well at a smaller reaction coordinate. The open state locates at the barrier region and the high free energy of the open state make it unstable, which is a main reason that the open state is difficult to be captured in experiment. This free energy profile also explains the observation that the open state structure quickly decays to the other states in some computational studies. Examining the conformations of these state shows that glycine binding produces an expansion movement at the ECD and TMD interface, which opens the ion gate at the middle of the ion channel. Continued opening will result in the structure decaying into the free energy well of the desensitized state. At the desensitized state, the proline residues at bottom of the ion channel close to lock up the channel. Characterization of T-cell immunoglobulin and mucin-domain containing protein (TIM) structural dynamics by molecular dynamics simulations Mucin-domain glycoproteins are characterized by extremely dense O-glycosylation that contributes to a unique, bottle-brush secondary structure which can extend away from the cell surface or form extracellular gel-like secretions. Mucin-type O-glycans are characterized by an initiating -N-acetylgalactosamine (-GalNAc) that is further elaborated into several core structures containing sialylation, fucosylation and/or ABO blood group antigens. As a result, mucin domains serve as highly heterogeneous swaths of glycosylation that exert both biophysical and biochemical influence on the cellular milieu. In particular, the T-cell immunoglobulin and mucin-domain containing family of proteins (TIM-1, 3, 4) decorate immune cells and act as key checkpoint inhibitors in cancer. However, their dense O-glycosylation remains enigmatic both in glycoproteomic landscape and structural dynamics, primarily due to the challenges associated with studying intrinsically disordered mucin domains. Through the discovery and use of a novel mucinase that selectively cleaves along the mucin glycoprotein backbone, we fully characterize the glycoproteomic landscape and construct the first all-atom model of TIM3 and TIM4.