Modeling Alternate Conformations with Alphafold2 via Modification of the Multiple Sequence Alignment

The unprecedented performance of Deepmind's Alphafold2 in predicting protein structure in CASP XIV and the creation of a database of structures for multiple proteomes is reshaping structural biology. Moreover, the availability of Alphafold2's architecture and code has stimulated a number of questions on how to harness the capabilities of this remarkable tool. A question of central importance is whether Alphafold2's architecture is amenable to predict the intrinsic conformational heterogeneity of proteins. A general approach presented here builds on a simple manipulation of the multiple sequence alignment, via in silico mutagenesis, and subsequent modeling by Alphafold2. The approach is based in the concept that the multiple sequence alignment encodes for the structural heterogeneity, thus its rational manipulation will enable Alphafold2 to sample alternate conformations and potentially structural alterations due to point mutations. This modeling pipeline is benchmarked against canonical examples of protein conformational flexibility and applied to interrogate the conformational landscape of membrane proteins. This work broadens the applicability of Alphafold2 by generating multiple protein conformations to be tested biologically, biochemically, biophysically, and for use in structure-based drug design.

Explore Protein Conformational Space With Variational Autoencoder

Molecular dynamics (MD) simulations have been actively used in the study of protein structure and function. However, extensive sampling in the protein conformational space requires large computational resources and takes a prohibitive amount of time. In this study, we demonstrated that variational autoencoders (VAEs), a type of deep learning model, can be employed to explore the conformational space of a protein through MD simulations. VAEs are shown to be superior to autoencoders (AEs) through a benchmark study, with low deviation between the training and decoded conformations. Moreover, we show that the learned latent space in the VAE can be used to generate unsampled protein conformations. Additional simulations starting from these generated conformations accelerated the sampling process and explored hidden spaces in the conformational landscape.

Explore protein conformational space with variational autoencoder

Molecular dynamic (MD) simulations have been actively used in the study of protein structure and function. However, extensive sampling in the protein conformational space requires large computational resources and takes a prohibitive amount of time. In this study, we demonstrated that variational autoencoders (VAEs), a type of deep learning model, can be employed to explore the conformational space of a protein through MD simulations. VAEs are shown to be superior to autoencoders (AEs) through a benchmark study, with low deviation between the training and decoded conformations. Moreover, we show that the learned latent space in the VAE can be used to generate unsampled protein conformations. Additional simulations starting from these generated conformations accelerated the sampling process and explored hidden spaces in the conformational landscape.

SARS-CoV-2 spike opening dynamics and energetics reveal the individual roles of glycans and their collective impact

The trimeric spike (S) glycoprotein, which protrudes from the SARS-CoV-2 viral envelope, is responsible for binding to human ACE2 receptors. The binding process is initiated when the receptor binding domain (RBD) of at least one protomer switches from a "down" (closed) to an "up" (open) state. Here, we used molecular dynamics simulations and two-dimensional replica exchange umbrella sampling calculations to investigate the transition between the two S-protein conformations with and without glycosylation. We show that the glycosylated spike has a higher barrier to opening than the non-glycosylated one with comparable populations of the down and up states. In contrast, we observed that the up conformation is favored without glycans. Analysis of the S-protein opening pathway reveals that glycans at N165 and N122 interfere with hydrogen bonds between the RBD and the N-terminal domain in the up state. We also identify roles for glycans at N165 and N343 in stabilizing the down and up states. Finally we estimate how epitope exposure for several known antibodies changes along the opening path. We find that the epitope of the BD-368-2 antibody remains exposed irrespective of the S-protein conformation, explaining the high efficacy of this antibody.

Attractive and repulsive residue fragments at the interface of SARS-CoV-2 and hACE2

The initial stages of SARS-CoV-2 coronavirus attachment to human cells are mediated by non-covalent interactions of viral spike (S) protein receptor binding domains (S-RBD) with human ACE2 receptors (hACE2). Structural characterization techniques, such as X-ray crystallography (XRC) and cryoelectron microscopy (cryo-EM), previously identified SARS-CoV-2 spike protein conformations and their surface residues in contact with hACE2. However, recent quantum-biochemical calculations on the structurally related S-RBD of SARS-CoV-1 identified some contact-residue fragments as intrinsically attractive and others as repulsive. This indicates that not all surface residues are equally important for hACE2 attachment. Here, using similar quantum-biochemical methods, we report some four-residue fragments (i.e quartets) of the SARS-CoV-2 S-RBD as intrinsically attractive towards hACE2 and, therefore, directly promoting host–virus non-covalent binding. Other fragments are found to be repulsive although involved in intermolecular recognition. By evaluation of their respective intermolecular interaction energies we found two hACE2 fragments that include contact residues (ASP30, LYS31, HIS34) and (ASP38, TYR41, GLN42), respectively, behaving as important SARS-CoV-2 attractors. LYS353 also promotes viral binding via several mechanisms including dispersion van der Waals forces. Similarly, among others, three SARS-CoV-2 S-RBD fragments that include residues (GLN498, THR500, ASN501), (GLU484, PHE486, ASN487) and (LYS417), respectively, were identified as hACE2 attractors. In addition, key hACE2 quartets identified as weakly-repulsive towards the S-RBD of SARS-CoV-1 were found strongly attractive towards SARS-CoV-2 explaining, in part, the stronger binding affinity of hACE2 towards the latter coronavirus. These findings may guide the development of synthetic antibodies or identify potential viral epitopes.

An Efficient Method to Quantify Structural Distributions in Heterogeneous cryo-EM Datasets

Cryogenic Electron Microscopy (cryo-EM) preserves the ensemble of protein conformations in solution and thus provide a promising way to characterize conformational changes underlying protein functions. However, it remains challenging for existing software to elucidate distributions of multiple conformations from a heterogeneous cryo-EM dataset. We developed a new algorithm: Linear Combinations of Template Conformations (LCTC) to obtain distributions of multiple conformations from cryo-EM datasets. LCTC assigns 2D images to the template 3D structures obtained by Multi-body Refinement of RELION via a novel two-stage matching algorithm. Specifically, an initial rapid assignment of experimental 2D images to template 2D images was applied based on auto-correlation functions of image contours that can efficiently remove the majority of irrelevant 2D images. This is followed by pixel-pixel matching of images with fewer number of 2D images, which can accurately assign the 2D images to the template images. We validate the LCTC method by demonstrating that it can accurately reproduce the distributions of 3 Thermus aquaticus (Taq) RNA polymerase (RNAP) structures with different degrees of clamp opening from a simulated cryo-EM dataset, in which the correct distributions are known. For this dataset, we also show that LCTC greatly outperforms clustering-based Manifold Embedding and Maximum Likelihood-based Multi-body Refinement algorithms in terms of reproducing the structural distributions. Lastly, we also successfully applied LCTC to reveal the populations of various clamp-opening conformations from an experimental Escherichia coli RNAP cryo-EM dataset. Source code is available at https://github.com/ghl1995/LCTC.