AI-Guided ASO Development for Ultra-Rare Diseases

David Dimmock, MBBS

Creyon Bio developed an antisense oligonucleotide (ASO) therapeutic that demonstrated the ability to knock down toxic Transportin-2 (TNPO2) protein in an allele-selective manner. The target of the ASO was informed using an artificial intelligence (AI) model, and it was used in a case study for a single patient with a de novo heterozygous pathogenic variant causing a missense change in the gene that codes for TNPO2.

Investigators found that the child had a dramatic reduction in seizure frequency subsequent to the second dose and regained developmental milestones that were lost and developed new skills following the third dose. Notably, these improvements regressed and were regained between and after doses. At the American Society of Gene & Cell Therapy (ASGCT) 27th Annual Meeting, held May 7 to 10, 2024, in Baltimore, MD, CGTLive® interviewed David Dimmock, MBBS, chief medical officer at Creyon, to learn more about the approach the company used to develop this ASO.

CGTLive: Can you tell us a bit about using AI for ASO design?

David Dimmock, MBBS: A fundamental problem with drug design is that most drugs fail because they're not safe. Often that failure happens when you get to the point where you're animal testing, or unfortunately, sometimes in humans. Oligos are a relatively simple medical modality in one sense. You can target them very specifically, but about 90% to 95% of oligos are just intrinsically unsafe out of the box, and would fail animal studies. So what Creyon did is they looked at a representative sample of many oligos, put those into animals, and were able to then take those features and predict whether or not this part or this part or this part of an oligo was going to be safe or not safe. So when you walk in and you know what sequences are going to be safe or unsafe, it then allows you when you walk in to engineer an oligo in the first place that has the pieces in that are going to work; and the pieces that won't work or cause injury—you either avoid those sequences or if you're forced, as we were in this case to a specific sequence, you can actually change the way the oligo is designed, the sugar, the base, or the linker.

Can you talk a little bit about the case study and the design process for that patient?

The family that we're talking about have been very generous in allowing us to share their story. They came to us through the TNPO2 Foundation just to the point where our platform was mature enough that we were ready to have a test case to take through and see how well it worked. We were at the fortunate point where we had the platform up and running in a sort of a beta testing phase when these parents were unfortunately presented with I think what is close to a parent's worst nightmare, their child developing intractable seizures—whereas the child had been previously interacting with them, rolling over—he'd lost the ability to do those things. So the question and the hope was: Could we actually design an oligo that could get rid of this bad protein that was building up in the brain whilst leaving the good copy of that gene alone? And would that actually make a difference? The challenge with this disease is we knew the genetic changes that were causing the disease [but] at this point we didn't have any long read sequencing, so we didn't know of any other parts of the genome that were on the bad copy of the gene. So, we were forced to design the oligo to actually hit the disease-causing variant. We were in a fortunate place with the platform where we could actually walk in and we could design oligos that we had a very high confidence were gonna be safe. And actually, when we did that, we designed 96 oligos. We knew we were in hurry, we didn't want to fail. More than 80% of those—in fact, close to 90% of those oligos—were safe when we put them in mice. We actually tested all of these oligos in mice, both in their brains and systemically, looking at their liver and their kidneys. The thing that made this case particularly tricky was we had to get rid of the bad copy of the gene whilst leaving the good copy alone. So we had to make sure that the ASO that was knocking down the bad copy of the gene just hit the bad copy and didn't hit the good copy. That made the place where we could target really quite small.

This transcript has been edited for clarity.

Click here to view more coverage of the 2024 ASGCT Annual Meeting.

REFERENCE
Dimmock DP, Bird L, Bouck A, et al. AI Enabled ASO Design Can Lead to Rapid Initiation of Treatment for an Ultra-Rare Disorder Leading to Allele Selective Knockdown of a Toxic Protein and Consequent Clinical Improvement. Presented at: ASGCT 27th Annual Meeting, May 7-10; Baltimore, Maryland. Abstract #309