Improved in-cell structure determination of proteins at near-physiological concentration.

Teppei Ikeya,Tomomi Hanashima,Shiro Ikeda,Masaki Mishima,Yutaka Ito,Peter Güntert,Saori Hosoya,Manato Shimazaki

doi:10.1038/srep38312

Abstract

Investigating three-dimensional (3D) structures of proteins in living cells by in-cell nuclear magnetic resonance (NMR) spectroscopy opens an avenue towards understanding the structural basis of their functions and physical properties under physiological conditions inside cells. In-cell NMR provides data at atomic resolution non-invasively, and has been used to detect protein-protein interactions, thermodynamics of protein stability, the behavior of intrinsically disordered proteins, etc. in cells. However, so far only a single de novo 3D protein structure could be determined based on data derived only from in-cell NMR. Here we introduce methods that enable in-cell NMR protein structure determination for a larger number of proteins at concentrations that approach physiological ones. The new methods comprise (1) advances in the processing of non-uniformly sampled NMR data, which reduces the measurement time for the intrinsically short-lived in-cell NMR samples, (2) automatic chemical shift assignment for obtaining an optimal resonance assignment, and (3) structure refinement with Bayesian inference, which makes it possible to calculate accurate 3D protein structures from sparse data sets of conformational restraints. As an example application we determined the structure of the B1 domain of protein G at about 250 μM concentration in living E. coli cells.

Highlights

Nonlinearly sampled data, automated chemical shift assignment, and robust structure calculation with Bayesian inference that can make optimal use of the limited experimental information (Fig. 1)
Structure calculations were performed employing the program CYANA19,20 with the newly developed CYBAY (CYANA Bayesian inference) module, which is able to extract a maximum of structural information from the limited and ambiguous experimental NOESY data with much broader line shapes and low sensitivity that is available for proteins in cells
Since expecting a similar or even more severe problem in HCCH- and NOESY-type spectra of GB1 in E. coli cells, in which strong self-correlated diagonal signals and plenty of much weaker correlation cross peaks are observed all together, we examined the reproducibility of the Quantitative Maximum Entropy (QME) processing on reconstructing these spectra

Summary

Introduction

Nonlinearly sampled data, automated chemical shift assignment, and robust structure calculation with Bayesian inference that can make optimal use of the limited experimental information (Fig. 1). It is usually not trivial to determine resonance assignments from NOESY spectra by conventional manual analysis, the automated chemical shift assignment algorithm FLYA permitted to comprehensively analyze all spectra and to objectively validate the obscure resonances from the manual approach. Structure calculations were performed employing the program CYANA19,20 with the newly developed CYBAY (CYANA Bayesian inference) module, which is able to extract a maximum of structural information from the limited and ambiguous experimental NOESY data with much broader line shapes and low sensitivity that is available for proteins in cells

Methods

Results

Conclusion