Welcome to Plenty of Room!

If you get value out of this newsletter, don’t forget to share with your friends and colleagues: sharing is caring after all!

Today I check out an exciting paper using AI protein design to solve a very real problem: snakebites.

Plenty of Room is your guide to the cutting-edge news related to molecular machines.

Already subscribed? Share with a friend that might find this interesting! It really helps.

New here? Just go ahead and subscribe!

Let’s get into it now.

Antivenoms with AI-designed proteins

AI-designed proteins can block toxins contained in the venom of cobras and other snakes. Image credit: Nature.

So, I love snakes: I think they are the coolest animals out there. That being said, snakebites are a devastating public health threat in areas like sub-Saharan Africa, South Asia and Latin America. Snakebite envenoming is classified as a highest-priority neglected tropical disease by the WHO, and each year 2 million people suffer from snakebite envenomation, leading to over 100,000 deaths and countless cases of lifelong disabilities. Despite this, very little resources are dedicated to improve the current antivenom treatments. Traditional antivenoms are produced in a similar way to the first vaccines. An animal, often an horse or a sheep, is inoculated with the venom to stimulate antibody production. At that point, the animal’s plasma is extracted and purified. This process has several drawbacks:

• It’s expensive, it’s slow (in horses it takes up to a year!) and it’s difficult

• Can cause a lot of side effects

• It’s not very effective for toxins that have little capacity to elicit an immune response

These issues make antivenoms unreliable and inaccessible, especially in the low-resource settings where snakebites are more common.

An AI-Powered Solution to Antivenom Design

Okay, it’s clear that there is an unmet need for new and better antivenoms. The authors of today’s paper propose a solution, using AI-driven protein design. In particular, they focused their computational efforts on 2 families of toxins, both from 3 fingers toxins (3FTxs) family:

• α-neurotoxins: an important subclass of 3FTxs, they have a small globular core, three loops (the “fingers”) stabilized by disulfide bonds and a C-terminal tail. They bind to receptors in the skeletal muscle and are responsible for paralysis and death.

• cytotoxins: as the name suggests, they have a cytotoxic effect, damaging cell membranes by inserting into them and leading to necrosis.

The idea of the authors is to create small, stable proteins that can bind and neutralize these toxins, leveraging computational tools to make the process faster and more cost-effective.

The Design Pipeline

Okay, so, how did they do it? They actually used a similar pipeline to other works we have seen:

1. Target analysis: The researchers focused on structural motifs of the toxins, using crystallography data to build consensus models to ensure their binders could neutralize a broad range of toxins

2. Binder generation with RFdiffusion: The authors then used RFdiffusion, a generative protein design algorithm, to create the initial binder. RFdiffusion starts with a “cloud” of amino acids and iteratively refines it into a protein that is complementary to the target.

3. Optimization and refinement: once the binder scaffolds were generated, the sequences were created and refined using ProteinMPNN, a computational tool able to take a protein structure and design a sequence optimized for stability and solubility. The authors then used AlphaFold2 to predict the 3D structure of these optimized binders.

From Design to Testing

Once this workflow was over, they expressed and screened the designed proteins. For each target toxin, only around 50 designs were tested, a lot less than in traditional protein design! The proteins which showed the most promise were then further optimized by using partial diffusion (where only parts of the protein is subject to RFdiffusion). Using this workflow, the researchers developed 3 main binders:

• SHRT: specifically targets short-chain α-neurotoxins, it has a nanomolar binding affinity

• LNG: this one was designed for long-chain α-neurotoxins and it also shows a binding affinity in the nanomolar range

• CYTX: Designed for cobra cytotoxins (probably the hardest target) it has a worse binding affinity, but it has good solubility and stability

Preclinical Results: A Solution to Envenoming?

The authors also tested the biophysical properties of these antivenoms, which showed high thermal stability and solubility, great for low-resource settings. But maybe more importantly, they tested how good they were at toxin neutralization. In vitro, all the proteins were able to neutralize the protein they were designed for, with SHRT and LNG neutralizing 100% of the α-neurotoxin activity! In vivo, SHRT and LNG protected 100% of mice from a lethal dose of α-neurotoxin, without adverse reactions! On the other hand, unfortunately CYTX showed limited in vivo efficacy, so it will require further optimization (it’s hard to make drugs, people).

Advantages of AI-Designed Antivenoms

This paper was not just cool, but it also focused on a real world problem. And AI-designed proteins in this case have some huge advantages compared to traditionally produced one:

• Designed proteins exhibited higher affinities and minimal cross-reactivity, surpassing traditional antivenoms.

• Proteins were expressed in E. coli, enabling cost-effective and scalable production compared to animal-derived antibodies.

• High thermal stability ensures usability in low-resource settings, reducing reliance on cold chains.

• The ability to target multiple toxin families simultaneously (e.g., α-neurotoxins and cytotoxins) enhances therapeutic versatility.

Final Thoughts

I really liked their use of an already established pipeline to solve a real problem: it’s not just AI for AI sake, but has very real impact. Of course, there is still more to do (bringing a drug to market is a long path, especially if resources are scarce), but this is a great start. So, I strongly recommend your read the paper here!

And thank you for reading as always!

In Other News:

• Assembling Supercharged Nanocages: If you feel like you didn’t have enough nanocages (after our two issues series), this could the paper for you. The authors designed binary nanoparticle superlattices using two differently sized, supercharged protein nanocages assembled via electrostatic interactions. This paper shows that protein cage matrices can incorporate various nanoparticles without influencing the assembly, highlighting their versatility for creating complex materials.

• DNA Origami in Whispering Gallery: if sensors are your things, you might find this interesting: this study combines DNA origami with opto-plasmonic whispering gallery mode sensors to achieve highly sensitive, label-free biosensing. DNA origami serves as a scaffold to precisely assemble gold nanorod dimers, creating strong electromagnetic fields in the nanogap for enhanced detection of DNA hybridization events. Their platform enables real-time single-molecule detection of DNA!

• Protein Gatekeepers: For something combining normal nanotech and more biological nanotech, check out this paper: the authors integrated protein crystals which transition from a closed to an open state in presence of HCN into a silicon photonic sensor. This combination allows HCN detection while blocking interferents, demonstrating the potential of engineered proteins in solid-state devices. Pretty cool!

• Not yet a subscriber to Plenty of Room? Sign up today — it’s free!

• You think a friend or a colleague might enjoy reading this? Don’t hesitate to share it with them!

• Have a tip or story idea you want to share? Email me — I’d love to hear from you!

• You have something you would love me to cover? Just reach out here or on my social!