Yunyun Gao

Nonstructural protein 3 (nsp3) is crucial for SARS-CoV-2 infection. It is the largest protein of the virus with roughly 2000 residues, and a major drug target. However, because of its size, disordered regions, and transmembrane domains, the atomic structure of the whole protein has not yet been established. Only 10 out of its 16 domains were individually determined in experiments. Here, we demonstrate how structural bioinformatics, AI-based fold prediction, and traditional experiments complement each other and can shed light on the makeup of this important protein, both in SARS-CoV-2 and in related viruses. Our method can be generalized for other multidomain proteins. Our prediction-based approach reveals a previously undescribed folded domain, which we could confirm experimentally. Our research also suggests a potential function of the domain Y1: this domain may be involved in the assembly of nsp3, nsp4, and nsp6 into the hexameric pore, which was discovered by electron tomography and exports RNA into the cytosol. The Y1 hexamer, however, could not be expressed on its own. We revise domain segmentation and nomenclature of nsp3 domains.

Non-structural protein 3 (nsp3) is crucial for the SARS-CoV-2 infection cycle. It is the largest protein of the virus, consisting of roughly 2000 residues, and a major drug target. However, due to its size, disordered regions, and transmembrane domains, the atomic structure of the whole protein has not yet been established. Only 10 out of its 16 domains were individually determined in experiments. Here, we demonstrate how structural bioinformatics, AI-based fold prediction, and traditional experiments complement each other and can shed light on the makeup of this important protein, both in SARS-CoV-2 and related viruses. Our method can be generalized for other multi-domain proteins, so we describe it in detail. Our prediction-based approach reveals a previously undescribed folded domain, which we could confirm experimentally. Our research also suggests a potential function of the nidovirus-wide conserved domain Y1: This domain may be involved in the assembly of nsp3, nsp4, and nsp6 into the hexameric pore, which was discovered by electron tomography and exports RNA into the cytosol. The Y1-hexamer, however, could not be expressed and purified on its own. We also provide a revised domain segmentation and nomenclature of nsp3 domains based on a compilation of previous research and our own findings.

During the automatic processing of crystallographic diffraction experiments, beamstop shadows are often unaccounted for or only partially masked. As a result of this, outlier reflection intensities are integrated, which is a known issue. Traditional statistical diagnostics have only limited effectiveness in identifying these outliers, here termed Not-Excluded-unMasked-Outliers (NEMOs). The diagnostic tool AUSPEX allows visual inspection of NEMOs, where they form a typical pattern: clusters at the low-resolution end of the AUSPEX plots of intensities or amplitudes versus resolution. To automate NEMO detection, a new algorithm was developed by combining data statistics with a density-based clustering method. This approach demonstrates a promising performance in detecting NEMOs in merged data sets without disrupting existing data-reduction pipelines. Re-refinement results indicate that excluding the identified NEMOs can effectively enhance the quality of subsequent structure-determination steps. This method offers a prospective automated means to assess the efficacy of a beamstop mask, as well as highlighting the potential of modern pattern-recognition techniques for automating outlier exclusion during data processing, facilitating future adaptation to evolving experimental strategies.

During the COVID-19 pandemic, the structural biology community swung into action quickly and efficiently, and many urgent questions were solved by macromolecular structure determination. The Coronavirus Structural Task Force evaluated all structures from SARS-CoV-1 and SARS-CoV-2, but errors in measurement, data processing and modelling are present beyond these structures and throughout the structures deposited in the Protein Data Bank. Identifying them is only the first step; in order to minimize the impact that errors have in structural biology, error culture needs to change. It should be emphasized that the atomic model which is published is an interpretation of the measurement. Furthermore, risks should be minimized by addressing issues early and by investigating the source of a given problem, so that it may be avoided in the future. If we as a community can do this, it will greatly benefit experimental structural biologists as well as downstream users who are using structural models to deduce new biological and medical answers in the future.

Antibodies protect from infection, underpin successful vaccines and elicit therapeutic responses in otherwise untreatable cancers and autoimmune conditions. The human IgG2 isotype displays a unique capacity to undergo disulfide shuffling in the hinge region, leading to modulation of its ability to drive target receptor signaling (agonism) in a variety of important immune receptors, through hitherto unexplained molecular mechanisms. To address the underlying process and reveal how hinge disulfide orientation affects agonistic activity, we generated a series of cysteine to serine exchange variants in the hinge region of the clinically relevant monoclonal antibody ChiLob7/4, directed against the key immune receptor CD40. We report how agonistic activity varies with disulfide pattern and is afforded by the presence of a disulfide crossover between F(ab) arms in the agonistic forms, independently of epitope, as observed in the determined crystallographic structures. This structural “switch” affects directly on antibody conformation and flexibility. Small-angle x-ray scattering and ensemble modeling demonstrated that the least flexible variants adopt the fewest conformations and evoke the highest levels of receptor agonism. This covalent change may be amenable for broad implementation to modulate receptor signaling in an epitope-independent manner in future therapeutics.

AlphaFold2 is a machine-learning based program that predicts a protein structure based on the amino acid sequence. In this article, we report on the current usages of this new tool and give examples from our work in the Coronavirus Structural Task Force. With its unprecedented accuracy, it can be utilized for the design of expression constructs, de novo protein design and the interpretation of Cryo-EM data with an atomic model. However, these methods are limited by their training data and are of limited use to predict conformational variability and fold flexibility; they also lack co-factors, posttranslational modifications and multimeric complexes with oligonucleotides. They also are not always perfect in terms of chemical geometry. Nevertheless, machine learning based fold prediction are a game changer for structural bioinformatics and experimentalists alike, with exciting developments ahead.

Contamination with diffraction from ice crystals can negatively affect, or even impede, macromolecular structure determination, and therefore detecting the resulting artefacts in diffraction data is crucial. However, once the data have been processed it can be very difficult to automatically recognize this problem. To address this, a set of convolutional neural networks named Helcaraxe has been developed which can detect ice-diffraction artefacts in processed diffraction data from macromolecular crystals. The networks outperform previous algorithms and will be available as part of the AUSPEX web server and the CCP4-distributed software.

AlphaFold2 predicts protein folds from sequence, which can be used for experimental structural biology, in construction and de novo protein design, prediction of complexes and perhaps even effects of mutations and conformational space exploration.

Contamination with diffraction from ice crystals can negatively affect, or even impede macromolecular structure determination and therefore, detecting the resulting artefacts in diffraction data is crucial. However, once the data have been processed, it can be very difficult to automatically recognize this problem. To address this, a set of convolutional neural networks named Helcaraxe has been developed which can detect ice diffraction artefacts in processed diffraction data from macromolecular crystals. The networks outperform previous algorithms and will be available as part of the AUSPEX webserver and CCP4-distributed software. Synopsis A program utilizing artificial learning and convolutional neural networks, named Helcaraxe, has been developed which can detect ice crystal artefacts in processed macromolecular diffraction data with unprecedented accuracy.