Protein sheaths designed by a deep learning algorithm can be used to solubilise cell membrane proteins without significantly affecting their structure or function, researchers in the US have shown. The technology should make it much easier to study membrane proteins, and could potentially lead to the development of new drugs or vaccines.
Nearly a third of the human proteome comprises proteins that sit within or span the cell membrane (for example, receptors and ion channels) and they are the target of more than half of clinically approved drugs. Studying them, however, is fraught with difficulty, as they are naturally insoluble in water: ‘Membrane proteins are embedded in this greasy environment that is the lipid bilayer, so the way to take them out is by using detergent,’ explains computational biologist Ljubica Mihaljević at the University of Washington’s Institute of Protein Design in Seattle; ‘It’s like washing dishes that are oily: it breaks up the membrane, and those detergent micelles wrap around the protein, shielding it from the solvent but enabling it to be soluble in solution.’
This process is labour intensive, unreliable, and means further separation steps are needed to isolate the proteins before they can be used. In 2015, a new approach was developed by another US-based research group that involves encircling a membrane protein with a naturally-occurring apolipoprotein AI.1 This protein is amphipathic – it has a lipophilic interior that interacts with hydrophobic residues on the membrane protein’s surface, and a hydrophilic exterior that makes the whole complex soluble in aqueous solution. However, the interaction between apolipoprotein AI and proteins is non-specific.
Wrappers delight
Mihaljević and colleagues in the group of David Baker – who shared the 2024 Nobel Prize in chemistry for computational protein design – have now used their deep learning algorithm RF diffusion to custom design protein sheaths called Wraps (water-soluble RF-diffused amphipathic proteins) that are tailored to encapsulate the cell-bound regions of specific target proteins while leaving their functional ends free.2 The amino acid sequence for these Wraps are then genetically encoded and expressed along with the membrane protein in Escherichia coli bacteria. The team has used the approach to capture and analyse a variety of membrane proteins.
The researchers looked at a Mycobacterial porin protein whose structure had previously been solved using traditional methods. ‘What we are able to do right now is confirm a structure that is unchanged based on the previous structure,’ says Mihaljević; ‘Unfortunately other proteins that we were able to wrap were too small to be able to actually solve the structure.’ As methods such as cryo-electron microscopy improve, the researchers hope it will be possible to ascertain structures of smaller wrapped proteins directly.
The researchers then studied other proteins, including outer-membrane proteins of the Treponema pallidum bacterium, which causes syphilis. The researchers used the AlphaFold protein folding algorithm to predict the shapes of the folded proteins and then designed Wrap sheaths to fit those structures. They found that the resulting sheaths could bind the proteins and stabilise them in solution. The bound proteins showed reactivity towards serum extracted from syphilis-infected rabbits. Mihaljević says the current work is ‘just a proof of concept,’ but reveals that collaborating laboratories are studying the technology’s potential for monoclonal antibodies and vaccines.
Synthetic biologist Chang Liu at University of California, Irvine says that ‘the challenge of solubilising cell membrane proteins is a big one and this makes a really significant contribution’. He believes one of the first applications could involve biochemical assays of samples isolated from patients: ‘Even in those more common drug development and biological applications, you always run in to this challenge that you can’t get a reliable source of the membrane protein to work with, and I’m somewhat convinced that this method and others in this realm will make a step change in that direction.’ The technique is promising enough, he says, for researchers to start applying it: ‘You’ll find out the use cases in which it works really well and the use cases in which other developments are necessary,’ he concludes.
