AI-Designed Transmembrane Proteins: Glowing Gatekeepers of Cellular Membranes

Transmembrane proteins are the gatekeepers of the cell and an integral part of cellular membranes. But designing them is tough, especially if you want to have them do something. And there is no better way to check your proteins than to make them glow up!

Light Up the Membrane

Transmembrane proteins pierce the lipid bilayer, connecting the inside and the outside. They let signals pass, move nutrients in, and send waste out.

And they love interacting with small molecules. Hormones and neurotransmitters are involved in signal transmission, and transmembrane proteins can shuttle glucose. And the activity of these proteins is regulated by small molecules!

Combining small molecules and transmembrane proteins is promising. If we could design transmembrane proteins, we could create new probes, sensors, receptors, transporters! Even new enzymes, embedded in membranes.

Unfortunately, nature doesn’t offer a lot to work with. It’s hard to create new proteins with the limited number of existing structures and ligands. Plus, transmembrane proteins require balancing the binding and the membrane interactions, which have opposite requirements. Hard to get it just right!

But what if we could design new transmembrane binders from the ground up?

Dreaming of Fluorescence

Here is where today’s paper comes in. The authors developed a computational pipeline to create de novo transmembrane proteins that can bind small molecules. Bonus: they are fluorescent!

Fluorescence is everywhere in bio: even I work on fluorescent nanomaterials! It’s a useful and simple way to check if your system is working, and it gives deep insights into how molecules are interacting!

Is there a better way to check your proteins than seeing them glow?

Two Steps to Design

The team took a two-step design approach.

In the first step, they designed stable, water-soluble fluorescence-activating proteins (wFAP). In the second one, they changed the surface of the proteins to create a transmembrane form (tmFAP).

But before we dive into the protein design, let’s get clear: where is the fluorescence coming from?

There are different ways to get fluorescence. GFP is a classic in biology, with 3 amino acids coming together to form a chromophore. But this is very uncommon! Most fluorescent materials, like proteins, RNA, or even DNA, rely on small molecules.

In this case, the team took a page from the RNA world. They used different variants of HBC (it has a very long and chemical name, and I’ll avoid it). HBCs are non-fluorescent in solution, but they become fluorescent after binding to the RNA aptamer Pepper.

This fluorescence activation is linked to a conformation change in HBC, which the RNA constrains in a planar conformation. So, the authors thought, why don’t we just copy that?

First Step: Designing Soluble Proteins

So, the first step is to design stable four-helix bundles. These need to have a central pocket that can bind and activate HBC, while also protecting it from the environment.

The researchers generated compact, four-bundle-based protein scaffolds and explored the design space to create pockets for HBC. The optimal pocket would be able to tightly bind the ligand and change its conformation!

The team used Rosetta and AlphaFold2 to optimize and fine-tune the designs. They filtered the AlphaFold2-predicted structures to get only the good designs, with a predicted pocket geometry within <0.8 Å Cα RMSD to the design model! Very strict.

Out of these optimized designs, they expressed 5 in E. coli. 3 lit up! And one had strong fluorescence: wFAP0. The team used directed evolution (an iterative mutagenesis technique to improve protein functionality) to improve the brightness and the binding of the protein to HBC.

The optimized proteins, wFAP1.1 and wFAP1.2, showed nanomolar binding affinity and high brightness (higher than GFP, the standard fluorescent protein). The proteins boosted a quantum yield (a measure of the fluorescence efficiency, specifically the ratio of emitted photons to absorbed photons) of 0.88 and 0.7: impressive!

Second Step: Conversion to Transmembrane Proteins

So, now the team has soluble fluorescent proteins. Time to turn them into transmembrane proteins!

This meant turning the water-friendly outer surface into a hydrophobic one, without disturbing the ligand-binding pocket. They turned to the most annoying feature of generative AI models: hallucinations.

ChatGPT making up random facts is annoying. And yet, this “bug” can be turned into a feature for protein design. “Hallucinated” proteins are structures that AlphaFold2 sees as an example of “ideal” proteins, but they don’t actually exist. But they still work!

So, the team used AlphaFold2-guided hallucination via ColabDesign:

• They froze the pocket residues and the overall structure to preserve them.

• Then, they optimized the rest of the protein to produce a hydrophobic surface. This meant positive residues inside, which helps the protein localization, and a longer span (the proteins have to cross the whole membrane!).

• Lastly, they selected proteins based on:

  • AlphaFold2 confidence

  • RMSD to original pocket

Then, it was time to test the tmFAP designs experimentally!

They successfully obtained the fluorescent tmFAP1 and tmFAP2. They further optimized tmFAP1, obtaining tmFAP1.2, a bright protein with nanomolar affinity for HBC and a quantum yield of 0.84.

And tmFAPs are specific for only one HBC variant, opening the door for multiplexing!

Do the Designs Match Reality?

The team also asked how well the designs matched reality. And the answer is simple: great!

They obtained structural models for some of the FAPs:

• wFAP1.1: X-ray crystallography models with a resolution of 2.3 Å in the apo form and 2.1 Å in the dye-bound form. The pocket is close to the design, and they could see that the HBC ligand sits planar and it’s fully encapsulated.

• tmFAP1.1: Obtained cryo-EM model, stabilized by fusing it to BRIL and an antibody fragment. They achieved 2.79 Å resolution, and here as well, they could see the ligand density in the pocket!

So, they showed that the de novo structures matched design coordinates to high precision. Amazing!

Functional Tests in Living Cells

Finally, they tested the functionality of the FAPs in different live cells:

• E. coli: wFAPs had a strong cytosolic signal when HBC dye was present, while tmFAPs were clearly localized in the membrane.

• Chinese hamster ovary cells: wFAPs can be targeted to the cytosol or mitochondria via signal peptides, and tmFAPs localize to the plasma membrane or endomembranes depending on different signals.

• Xenopus oocytes: tmFAP3 targeted to the plasma membrane showed strong surface-localized fluorescence!

A Bright Future for Protein Design

Cool work! Amazing how they combined physics-based computational models, generative models, and traditional directed evolution to get new fluorescent proteins! And they also admitted that their first attempts didn’t work, which is refreshing to see in such a paper.

The new fluorescent proteins are brighter and smaller than the alternatives. They would make great sensors for pH, membrane gradients, or metabolites, merging the high sensitivity of small molecules and the specific localization of proteins.

And their two-step approach is not limited to fluorescence either! It could be applied to create transmembrane proteins of any kind:

• Receptors: A clear case for drug delivery!

• Ion channels and transporters: They could help with drug resistance in bacteria or cancer!

• Enzymes: New transmembrane enzymes could open the way for new cellular chemistry!

So, cool paper! I really enjoyed them documenting their troubles; you don’t see that often, and certainly not when you publish in Nature, let’s be honest here.

So, go and read it for yourself here!

If you made it this far, thank you! Are you interested in protein design? What do you think could be the next big target? Reply and let me know!

More Room:

• Organizing Bubbles: Using DNA origami. This study introduces a DNA-origami–based polymerization platform that organizes liposomes into programmed structures. Using square DNA origami with tunable sticky-end features, the team assembled liposomes of uniform size into 2D arrays, finite lattices, and rings, with control over arrangement, chirality, and flexibility. The method also supports stepwise assembly and targeted disassembly, offering dynamic control. Cool!

• Writing Shakespeare in DNA: DNA data storage is seeing more and more innovations. And I love it! This study presents a DNA-based platform for both data storage and computing. Using a combinatorial assembly method and a trie-like DNA data structure, the authors store data in key–value form and demonstrate it by encoding “The Complete Works of William Shakespeare” into DNA with error rates comparable to conventional media. They introduce two new DNA computing instructions enabling both exact and approximate text searches directly on the DNA-encoded data, achieving perfect recall and high precision. The method is scalable and cost-efficient. From the folks at Catalog DNA!

• Proteins, Clusters and Cryo-EM: An interesting structural biology paper. It uses cryo-EM to explore how membrane proteins form higher-order transient clusters that link components of signaling pathways. The authors find that some proteins cluster via structured extracellular or intracellular domains, while others use intrinsically disordered cytoplasmic regions. These different interaction modes produce clusters with varying compactness and organization, revealing that nature employs multiple strategies to connect membrane proteins in self-assembled assemblies.

• Share Plenty of Room with founders or builders

  I help biotech and deep tech companies transform complex technologies into engaging content that builds credibility with investors, partners, and potential hires. Let’s chat!

• Know someone who’d love this?
  Pass it on! Sharing is the easiest way to support the newsletter and spark new ideas in your circle.

• Got a tip, paper, or topic you want me to cover?
  I’d love to hear from you! Just reply to this email or reach out on social.