G-protein-coupled receptors control many processes, from pain and metabolism to blood-cell movement, but designing biological drugs that control their signalling precisely has been notoriously difficult. Researchers have now computationally designed tiny proteins that activate or block 11 receptors. In mice, one matched the effect of an approved drug while causing fewer unwanted effects.
Nearly every cell in your body carries molecular switches that respond to hormones, neurotransmitters and other signals. These G-protein-coupled receptors (GPCRs) are among medicine’s most important drug targets – but controlling them is unusually difficult because simply binding to the receptor is not enough.
“For many proteins, the lock-and-key analogy works quite well: if your protein binds and blocks the target, you are done. GPCRs are different. They are more like dimmer switches. You may want to turn the signal all the way up, turn it down or leave it at a particular level, depending on the receptor and the disease you want to treat,” explains Christoffer Norn, co-founder and CEO of Skape Bio and formerly a senior researcher at BioInnovation Institute in Copenhagen, Denmark.
Norn and an international team have now designed miniproteins that can activate or inhibit 11 GPCRs involved in itching, pain, migraine, metabolism, cancer and other processes. And the designs did not stop at cells: in mice, one miniprotein targeting C-X-C chemokine receptor type 4 (CXCR4) mobilised blood-forming stem cells at a level comparable to the approved drug plerixafor while causing fewer unwanted changes in circulating blood cells.
Why some of medicine’s most important drug targets are so hard to control
GPCRs form the largest and most diverse family of membrane receptors in the human genome. Sitting across the cell membrane, they translate signals from outside the cell into changes within it and regulate processes involved in everything from cardiovascular disease and metabolism to cancer and the nervous system.
“Around a third of the medicines on the market act through GPCRs. That tells you something about how good these receptors are as drug targets: when you influence them, you can often have a very large physiological effect,” says Christoffer Norn.
Researchers have spent decades developing small molecules, peptides, antibodies and other biological drugs that influence individual GPCRs.
“With de novo design, we are doing something fundamentally different. Instead of taking a molecule that nature has already made and trying to modify it, we can start with the interaction we want and ask: what protein would we need to build to make that happen?” says Christoffer Norn.
Norn experienced the rapid transformation of computational protein design first-hand during three years in the laboratory of David Baker at the University of Washington – the researcher who would later receive half of the 2024 Nobel Prize in Chemistry for computational protein design.
“A decade ago in protein design, you might ask what would happen if you exchanged one amino acid for another. By the end of my time with David, we could design completely new proteins from scratch, and the structures we measured afterwards could sometimes come back almost within the diameter of an atom of the computer model. The precision had become completely different,” says Christoffer Norn.
This leap in precision planted the seed for the GPCR project. With these receptors, a designed protein has to do more than fit its target: by stabilising specific receptor shapes, it also has to influence whether and how the signal is passed through the membrane.
Tiny proteins designed to do more than bind
For the study, the researchers focused on miniproteins. Small proteins and peptides already occur throughout biology – insulin and GLP-1 are in roughly the same size range – but here the entire molecule is built from the ground up for a specific receptor.
“A miniprotein might be only 40–70 amino acids long. The difference is that we are not taking something nature already made and adapting it. We design the entire protein de novo for the receptor we want to control,” says Christoffer Norn.
Their small size also leaves considerable room for engineering.
“That modularity is one of the attractive things. You can have one part that binds the receptor and gives you the function you want and another that extends the half-life or helps to transport it somewhere specific. Because we design the molecule ourselves, we have many places where we can build those properties in,” says Christoffer Norn.
But designing the proteins was only half the challenge. The researchers also needed a way to test very large numbers of candidates experimentally without taking these difficult receptors out of their natural membrane environment.
Building proteins to reach deep receptor pockets
The researchers used two complementary computational strategies. With RFdiffusion, they specified important contact points on the receptor and computationally built an entirely new protein around them. For especially recessed binding pockets, they first designed a five-amino-acid motif small enough to reach deep into the receptor and then built the rest of the miniprotein around that interaction.
“With some of these receptors, there is a particular point deep inside the binding pocket that you need to reach. We can first design a very small piece that makes that contact and then build the rest of the protein around it,” explains Christoffer Norn.
The second approach, MetaGen, worked differently: instead of generating a shape from scratch, it searched nearly 9,000 existing protein scaffolds selected from hundreds of millions of AlphaFold-predicted structures for geometries that could be adapted to the receptor.
“What is useful is the enormous diversity of shapes that is already out there in the predicted protein universe. You can find kinked helices, loops sticking out in unusual ways and many other geometries that you would rarely see in a conventional protein library. These can become completely new starting-points for design,” says Christoffer Norn.
Testing receptors without taking them out of the cell
GPCRs are difficult to screen experimentally because they normally sit embedded in cell membranes. Traditional high-throughput screening often requires extracting a target protein and stabilising it in detergent, liposomes or artificial membrane discs. But moving a receptor out of its natural environment can change how it acts.
“Once you take these receptors out of their natural membrane environment, you risk finding binders that look good in the screen and then do not work when you put the receptor back into a cell,” explains Christoffer Norn.
The researchers therefore developed receptor diversion, which allows them to screen the receptor while it remains inside a human cell. Each candidate miniprotein carries a tag that holds it inside the cell’s secretory pathway. A receptor that does not bind continues towards the cell surface; one that does bind is trapped with the miniprotein.
Both carry fluorescent markers.
“It is almost like a traffic light. If the red miniprotein and the green receptor stay apart, the design is not binding. If they overlap, you get yellow – and this tells us that the receptor has been trapped by the miniprotein,” says Christoffer Norn.
From a colour signal to a screen of 100,000 designs
Unexpectedly, the amount of red-green overlap also tracked how strongly the miniprotein bound the receptor. The same microscope image could therefore indicate not only whether a design worked but approximately how well it bound. When the researchers compared the method with independently measured binders, the two approaches agreed closely.
“We did not fully anticipate that at the beginning,” says Christoffer Norn.
To scale the method up, each candidate was linked to a unique DNA barcode. The researchers distributed as many as 100,000 designs across around 10 million cells, imaged the cells and then read the barcodes directly in the same microscope plates to determine which design was present in each cell.
“We can look at around 10 million cells and then read the barcode essentially one base at a time, so we know which protein design produced the signal in each cell. That lets us screen very large libraries without ever taking the receptor out of the membrane,” explains Christoffer Norn.
The most promising candidates were then purified for functional and structural testing.
“First, we need to know that the protein binds. Then we ask the much more important question: does it actually change the signalling of the receptor? And finally, for some of the designs, cryo-electron microscopy lets us go back and see whether the structure we made in the computer is really the structure we ended up with experimentally,” adds Christoffer Norn.
Designed proteins could turn signalling up or down across 11 receptors
The approach produced functional miniproteins against 11 GPCRs involved in itching, pain, migraine, metabolism, cancer and reproduction. Some were agonists that activated signalling; others were antagonists that blocked it.
“What was important for us was not just getting one receptor to work. We wanted to see whether the approach could work across very different GPCRs and whether we could make both agonists and antagonists. That begins to tell you that this could be a platform rather than a solution for one particular receptor,” explains Christoffer Norn.
For the itching- and pain-related receptor Mas-related G-protein coupled receptor member X1 (MRGPRX1), the researchers identified several agonists and computationally optimised one to increase its potency around 12-fold. It activated the receptor at nanomolar concentrations and showed little off-target activity when tested across 318 GPCRs.
The researchers also obtained 20 agonists against neurokinin 1 receptor (NK1R), with the strongest acting at around one nanomolar, and antagonists against CXCR4, C-C chemokine receptor type 5 (CCR5), the oxytocin receptor and several receptors involved in metabolism and migraine. One parathyroid hormone 1 receptor (PTH1R) antagonist inhibited signalling at around 450 picomolar.
“Directly out of the computer, we can already get very strong binding – in some cases down into the picomolar range. That is one of the things that makes these miniproteins attractive: you can start with a very compact molecule and then optimise the properties you need from there,” adds Christoffer Norn.
Cryo-electron microscopy showed how closely the designs matched the structures the researchers actually produced. For two MRGPRX1 agonists, the experimentally determined protein backbones differed from the computer models by only about 0.7 ångström – less than the width of a typical atom.
“With GPCRs, that level of precision really matters, because a small structural difference can change what the receptor does,” says Christoffer Norn.
From computer model to real structure. The turquoise miniprotein was designed to bind the gold-coloured MRGPRX1 receptor. When the researchers determined the structure experimentally, it closely matched the structure predicted by the computer model.
A designed protein produces the intended effect in mice
The strongest test outside cells involved CXCR4, which normally helps retain blood-forming stem and progenitor cells in the bone marrow. Blocking it releases them into the bloodstream so they can be collected for procedures such as stem-cell transplantation.
In mice, the designed miniprotein dCX1_001 mobilised these stem and progenitor cells at a level comparable to plerixafor, an approved CXCR4-blocking drug used in certain types of blood cancer. It also caused smaller increases in other circulating white blood cells and did not trigger the same adverse changes in blood-cell counts or systemic cytokine responses measured with the comparison treatment.
The researchers then used a gene-delivery system and later found genetically modified cells and vector DNA in blood-forming tissues, showing that the mobilised stem and progenitor cells could return to blood-forming tissues and re-engraft.
“This does not mean that we have a medicine. This is still preclinical. But it is an important step because we are no longer only showing that a designed protein binds a receptor in a cell assay. We can give it to an animal and see the biological effect we designed it to produce,” says Christoffer Norn.
From drug targets to biological unknowns
The immediate challenge is turning promising binders into molecules that can survive the much longer journey towards a medicine. Small miniproteins can be rapidly cleared from the bloodstream, so their lifetime in the body may need to be extended. In the study, a miniprotein targeting the migraine-related CGRP receptor had a half-life of only about 0.4 hours in mice; attaching it to an Fc domain, a larger antibody-derived component, extended this to about 25 hours.
“So you have to engineer the half-life for the particular application – for example by adding another domain or a chemical modification,” explains Christoffer Norn.
Because the proteins are designed de novo, researchers can also increase their stability, introduce sites for chemical modifications and potentially remove sequence features predicted to provoke unwanted immune responses.
“This is one of the freedoms you get when you design the whole protein yourself. Then the question becomes: where should the molecule go, how long should it stay there and how should the body get rid of it?” says Christoffer Norn.
These possibilities remain engineering goals rather than outcomes demonstrated by the present study.
The Skape Bio team is continuing to develop the designed miniproteins and explore whether the approach can be used both as a starting point for new medicines and as a tool for understanding receptors whose biology is still poorly understood.
What the designs still cannot control
A more fundamental limitation remains. Researchers can increasingly design where and how a miniprotein binds a receptor, but they cannot yet reliably predict what signal that interaction will produce. Designing against a receptor structure that looks active does not automatically create an agonist, just as targeting an apparently inactive structure does not guarantee an antagonist.
“We are very good at asking: can we make something that binds here? The harder question is what that binding does to the receptor. It is not simply on or off – there are different levels and different signalling outcomes. We have to learn much more about that part,” says Christoffer Norn.
The study also deliberately focused on GPCRs with relatively accessible binding pockets. Many receptors, especially those that normally bind small molecules, have sites buried deeper in the membrane and remain much harder for a protein to reach.
Yet some of the most intriguing GPCRs are orphan receptors whose biology remains poorly understood. Several have strong human genetic associations with obesity, immune-related conditions and other diseases, suggesting that they could be important drug targets even though researchers still lack good ways to control them experimentally.
“Sometimes human genetics gives us a very strong clue that changing the activity of a receptor could affect disease. But genetics tells us what happens when receptor function is altered over a lifetime. It does not necessarily tell us what happens if we activate or inhibit that receptor temporarily with a drug. For that, we need molecules that let us manipulate these receptors experimentally,” says Christoffer Norn.
Designing the tool before finding the natural ligand
This could reverse the usual sequence of discovery. If a receptor’s structure is known or can be predicted, researchers may be able to design a molecular tool before anyone has even identified its natural ligand.
“This is one of the things I find most exciting. We now have AlphaFold structures and algorithms that can generate binders from structure alone. So instead of waiting until somebody discovers the natural ligand, we may be able to make the tool ourselves and ask: what does this receptor actually do?” says Christoffer Norn.
The answer might reveal a new therapeutic target – or show that the genetic clue leads nowhere useful for treatment. Either result would resolve biology that today remains largely inaccessible.
The idea that began during Norn’s postdoctoral work with David Baker was later developed further through BioInnovation Institute and now forms the scientific basis of Skape Bio, which he co-founded.
“The first thing we have to prove is whether we can take molecules from this approach towards the clinic. But there is another direction as well. For receptors we barely understand, the designed protein can itself become the experimental tool – and learning what that receptor actually does may point us towards the next drug target,” concludes Christoffer Norn.
The route from a computer model to a medicine remains long, and none of the current miniproteins has yet been tested in humans. But the starting point has changed dramatically. Researchers no longer have to wait for nature or enormous screening libraries to provide a molecule capable of controlling a receptor. Increasingly, they can begin with the receptor itself – and try to design the molecule they need.
