Membrane protein folding and design: from structural prediction to protein expression

The de novo protein design involves the computational generation of novel proteins that can fold into specific 3D structures by coupling the innovative technologies of artificial intelligence to computational protein design, attaining remarkable achievements and applications in proteins behavior. Nevertheless, comprehending how proteins achieve their natural configurations from their amino acid sequences remains an enigma and, since misfolding and protein aggregation leads to several diseases, a thorough understanding of protein folding and dynamics between sequences, structures, and functions is necessary. While the design of water-soluble proteins has been so far more successful, the generation of membrane proteins remains a very challenging topic and lags far behind due to their hydrophobic nature. As a result, improving the accuracy and reliability of computational design approaches is crucial. In this work we seek to contribute to the development of more effective and reliable strategies for membrane protein design, presenting the experimental validation of two novel membrane proteins, named TM2mec and memGFP, generated from scratch using two distinct fixed-backbone design strategies. The design of TM2mec relied on the use of the bacterial Small Conductance mechanosensitive ion channel (MscS) as a backbone, incorporating its features into a minimalist membrane protein that responds to mechanical cues. We exploited the state- of-the-art Rosetta Design algorithm and generated a set of candidate structures furtherly refined with molecular dynamics simulations. Conversely, the memGFP was designed with the innovative goal of generating a membrane protein functionally akin to the green fluorescent protein (GFP) with the ability to translocate within the membrane and emit fluorescence once properly folded and embedded in the lipid bilayer. This would enable a proof of principle for an innovative real-time visualization and tracking of membrane proteins folding and behavior in response to different stimuli using fluorescence as readout. We employed a combination of computational tools, rational- and innovative AI-design methods, to generate thousands of eligible versions of a b-barrel structure having the GFP chromophore. Parallelly, we pursued the optimization of an unconventional Voltage-gated Ion Channel, named Hv1 proton channel, notorious for its inherent instability. We seek to develop a hybrid approach that combines traditional molecular biology approached with deep neural network-based tools to design a more stable version, generating an Hv1 chimera more suitable for future structural purposes, eventually elucidating the mechanistic determinants behind the function of this peculiar VGIC. Overall, our studies demonstrate the effectiveness of deep learning approaches as resource for design and optimization of both de novo and naturally occurring membrane proteins. However, their inherent complexity, coupled with our limited understanding of folding mechanisms, pose significant difficulties for their accurate modeling. As such, it is critical to continue improving and refining these algorithms, as well as developing complementary experimental techniques for making progress toward the development of new and improved biomolecules, particularly those involving membrane proteins as key components.