Membrane Protein Synthesis: Robotics & ML Advancement!

50% of all drugs target membrane proteins, and 30% of genes code for one.

And yet, we struggle to synthetize them in their native, lipid-rich environment. But today we dive into a paper that mixes robotics, machine learning and experiments to perfect cell-free production of membrane proteins!

Let’s dive right in.

Protein Production, Simplified

Membrane Proteins: 30% Of All Proteins

Membrane proteins are easily some of the most important proteins out there. They are embedded in or interact with biological membranes, and around a third of human genes code for them!

They are essential for cellular functions, accessible on the cell surface, and involved in many diseases. This made them attractive for drug development: 50% of all drugs target membrane proteins!

But things are never easy in biology, aren’t they?

Why They Are Hard to Produce

While membrane proteins are important, producing them is hard.

Membrane proteins have hydrophobic regions, useful for interacting with the lipids in membranes. These regions don’t like being in water, and as a result the proteins aggregate, making them useless. There are some solutions, but they are inefficient or alter structure and function. This is why, out of the thousands of membrane proteins in the human genome, only a few hundreds have their experimental structure validated!

Usually, membrane proteins are produced using cells, which provide an amazing native environment for expression and folding. On the down side, the yield is often low, they don’t work for toxic proteins, and purification can be a nightmare!

A cool alternative is cell-free protein synthesis. This is the dream of anyone who’s too impatient to wait for cells to grow (like myself). Just mix all the components required, often using cells extract, add some DNA, and voilà, you can produce proteins! Cell-free systems have a few advantages:

• Rapid protein synthesis

• Flexibility in protein engineering

• Control on the synthesis environment

In addition, they allow the use of liposomes, which recreate the lipid-based environment that membrane proteins love! Liposomes can also help with purification, functional studies or use in therapeutics.

The big downside to cell-free systems? They are hard to design. Often there are a lot of parameters that should be optimized, and they are interdependent! Because of course they are.

MEMPLEX: Merging Computation and High-Throughput Assays

And here is where today’s paper enters. The authors created MEMPLEX, a platform that combines computational and high-throughput methods to rapidly design synthesis environments for membrane proteins!

MEMPLEX is split into two connected parts, one wet-lab focused and one computational. Let’s look at them!

Part 1: High-Throughput Assay Development

MEMPLEX’s experimental side is built on 3 main components:

1. Custom Droplet Printer: The team assembled a custom robot capable of printing droplets with nanoliter precision (!). Apart from being super cool, this enabled high-throughput, combinatorial screenings across many variables.

2. Cell-free protein synthesis: Based on E. coli extracts, with various chemicals and liposomes of around 100 nm diameter added.

3. Split-GFP Solubilization Reporter: Aggregation is a big problem for membrane proteins. To detect when a protein is soluble and inserted into liposomes, the authors created a reporter system. They split GFP in 2 parts: a small part is conjugated to the protein of interest, and the rest of the protein is just floating around. Only when the protein is properly folded and not aggregated do the two parts reassemble into fluorescent GFP. This binding cannot happen if the protein aggregates: so the fluorescence signal solubility. Such an elegant system!

Using these 3 components, the team screened over 10.000 reactions, with different chemical environments! They evaluated the platform on 28 membrane proteins, derived from bacteria, plants and humans. These proteins have therapeutics applications, biotech utility or medical relevance, with some of them having no experimentally validated structures.

But this was only the beginning.

Part 2: Machine Learning To Predict Synthesis Conditions

Did you think you could escape machine learning? Ah, of course not.

Once the researchers had screened all those reactions, they reached 3 conclusions:

• Smaller proteins had a higher yield than bigger proteins

• There was no universal reaction condition that worked for all proteins

• The variables (for example, magnesium and PEG concentration) were often interacting.

To make sense of these complicated interaction data, the researchers turned to machine learning. 

The team trained an ensemble of deep neural networks to predict the protein yield across various reaction conditions. The ensemble guided the next experimental conditions, and resulted in improved yield for 21 out of 28 of the proteins! In addition, they managed to produce 9 hard-to-synthetize proteins, some of which were never synthetized. And with a few tweaks, their model was able to also predict the ideal synthesis conditions for membrane proteins that the model had never seen.

Why It Matters and What’s Next?

MEMPLEX is a cool platform, and is a much needed step for membrane protein research! It offers:

• A scalable method to discover synthesis conditions

• A machine learning-guided framework for reducing experimental load (= less money needed)

• Insights into synthesis outcomes via embedded protein features

Of course, this is just the beginning! There is more work to be done:

• Authors here don’t check for the functionality of the synthetized proteins: creating functional assays could be the next step

• There are many more parameters that could be tested, when it comes to lipid types and chemical components

• Native-like membranes or hybrid vesicles could be used instead of liposomes, to mimic native conditions even better!

As protein science heads toward applications in medicine, manufacturing and bioremediation, producing proteins will be fundamental. And the success of these applications hinges on these tools, that make easier and faster to produce new proteins, even just at a prototype scale.

A bit like 3D printing is enabling faster prototyping in hardware!

This was a cool paper! If you want to dive deeper into it, go and read it here.

If you made it this far, thank you! Do you work with proteins? What do you think of cell-free systems? Do you use them for DNA or RNA? Reply and let me know!

More Room:

• BRAF Evading All The Stops: BRAF is one of those super important proteins we can’t stop learning new things about. This study shows that oncogenic BRAF mutations, including V600E, disrupt the protein’s autoinhibited state by displacing helix αC, pushing it into a preactivated form. Using cryo-EM, researchers found that the inhibitor PLX8394 restores autoinhibition by stabilizing helix αC in an inactive position, explaining its mechanism of action. Let’s hope it helps against cancer!

• Proteins and DNA: Best Friends. I love DNA nanotech, and I am learning more about protein assemblies. So, I was very excited to find this paper! It combines DNA nanostructures with protein assemblies to create programmable bio-nanomaterials. Using orthogonal conjugation methods, the team links coiled-coil protein origami to DNA, enabling the formation of patterned nanofibers with functional proteins. They also develop a reversible DNA-luciferase circuit, showcasing dynamic control. The work bridges DNA and protein nanotech, advancing the design of responsive, modular biomaterials. On my reading list pronto!

• Getting Inspired by PCR: I guess Kary Mullis would be proud. Or not, he was an interesting guy. This study presents a simple, enzyme-free DNA circuit called split-free autocatalytic amplification (SAA) that mimics PCR to achieve exponential signal amplification. With minimal components and no need for temperature control, SAA offers high efficiency, low background noise, and strong potential for sensitive biomarker detection in clinical and biochemical applications.

• 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.