Membrane Protein Energy Functions for De Novo Design of Transmembrane β-Barrels

Abstract

As the field of proteomics rapidly, the ability to accurately model and design membrane proteins lags. Modeling and designing membrane proteins is notoriously more difficult than modeling and designing soluble proteins because of the heterogenous membrane environment, which is difficult to model computationally and to replicate experimentally.1 The ability to accurately design membrane proteins de novo – without homologies to naturally occurring proteins – is important to elucidate processes that govern membrane protein expression, folding, and insertion in vivo and to increase the utility of membrane proteins in novel therapeutics and biotechnologies. This research focused on the de novo design of transmembrane β-barrels (TMBs) using the macromolecular modeling and design Rosetta software suite.2 This project sought to answer the following questions: (1) Are current methods for TMB design successful? and (2) What design principles and methods are necessary for accurate TMB design? In order to address these questions, two recently developed methods, the Vorobieva design methods and the franklin2019 energy function, were utilized to design 8-strand and 12-strand TMBs.3,4 These methods are both physics based, developed using physical characteristics and knowledge of native membrane proteins and the membrane protein environment. However, the two methods differ in the fact that franklin2019 was developed to fit previously determined experimental data while the Vorobieva design methods were developed through an iterative hypothesis-design-test process.3,4 Our data indicate that the design methods tested in this study, which are state of- the-art physics-based methods, are not yet successful or generalizable for accurate and efficient TMB design. However, we verified several principles that are necessary for successful computational design of TMBs, including the importance of suboptimal β-turns and the importance of β-sheet propensity reduction. This study brings to the forefront the possibility that physics-based models, functions, and methods may not be the most efficient way to model and design membrane proteins and highlights the need to investigate novel artificial intelligence-based methods for membrane protein design.