How AI-designed proteins can awaken ‘silenced’ cells and help fight cancer
- Hope Reese
A Q+A with Shiri Levy about her groundbreaking biomedical research.
What if you could use AI to help the body help itself? Researchers are uncovering ways that it can. A new study from researchers at the University of Washington Medicine Institute for Protein Design shows how merging an AI-built protein with CRISPR technology can work to activate certain genes that are “turned off” — like those that stop the proliferation of cancer cells — thereby allowing the cell to function as normal.
The process works by bringing the protein to specific genes — one by one — that have epigenetic markers and removing them. Activating the natural gene expression this way could also replace traditional treatments, such as chemotherapy.
Right now, there’s nothing else like it. While there are natural elements that can express genes, this fusion of the AI-protein and CRISPR has created the first artificial gene activator.
“The vision is to really create gene therapies using precision,” said Shiri Levy, a postdoctoral fellow at the Hannele Ruohola-Baker Lab, and the author of the paper. I spoke to Levy about how the process works, how they were able to do all this without modifying DNA itself, and what’s next in AI-abetted gene therapy.
The conversation has been edited and condensed for clarity.
What was the big takeaway from your study?
Every cell of our body has the same DNA — however, many cells function very differently, and every cell that functions differently expresses different sets of genes. In order to express some genes, those genes need to be turned off.
This study looks at [a family of proteins] called Polycomb Repressive Complex 2 (PRC2) that represses genes through methyl marks. To study how necessary PRC2 is in this stem cell stage, we created a protein binder. This AI-designed protein is essentially a biological epigenetic inhibitor.
The capacity to use this AI to erase the specific epigenomic mistakes in aging and other diseases is now really pushing us closer to a great, safe therapy.
What is an epigenomic mistake?
An example for that would be the p16 gene that is supposed to block cell proliferation [as in cancer] and now it’s completely turned off. Or other genes that are responsible for muscle regeneration or immune response. Or any of these genes that are supposed to be turned on but instead are heavily decorated with epigenetic marks — what I call “mistakes.”
Those mistakes are acquired over time — they were not there before. To remove that mistake we’ll be able to use that tool and remove those marks.
Essentially, the genes that you are awakening should be on, but for some reason they’re not — so it’s not as if you’re messing with nature, turning something on that should be off?
Correct! The genes that are supposed to be on are now off — and because of that, they are causing diseases. To eliminate the disease, we need to turn them on in a very controlled way. Until now, the only way to do it was through chemical drugs. Now, we don’t have to apply chemical drugs to 30,000 genes — we can only do it on one that is causing this disease.
The genes in your study themselves are merely being turned on, not modified, right?
We don’t touch the genes at all — we do not manipulate it or add anything to it. All we do is act on it. That’s what we call epigenetics. In order to act on it, we remove a chemical molecule called methyl marks in order for the gene to be turned on. So instead of it being silenced, now it’s being awakened.
How does the AI-based protein work, exactly?
In 2017, using a computer, we [created] a protein from scratch that is better — much more specific and stable and smaller — than what nature has given us. This is basically nature in fast-forward.
We were able to take the AI-designed protein and fuse it to CRISPR/CaS9. We fused this protein to this CRISPR/CaS9 so we could manipulate single genes only. And this CRISPR/CaS9 [dCas9] essentially served as an UBER, and the AI-designed protein was like the passenger, telling it where to go – we were able to go to specific sites on the gene with that UBER, and now it turns the gene on.
What was the point of the inhibitor?
[We wanted to] find which repressive mark is responsible for silencing genes. You can imagine mountains of those marks — until now, in order to upregulate a gene, we would use a chemical drug to erase all those marks.
Now, we don’t have to remove all these marks — all we need to do is tile the region and find that one spot that is responsible for repressing and silencing genes. The moment we found that hotspot, the gene was awakened, and now it’s expressing.
You selectively turned on four different genes — what do those genes do?
TBX18 is a marker to the sinoatrial node — very important in development. Those cells are responsible for pacing the beating of the heart — they’re like pacemakers. T16 is a suppressor and a cell cycle regulator. In many cancers, particularly in brain cancer in kids, it stops the proliferation of cancerous cells. In those situations, T16 is heavily silenced. So wouldn’t it be amazing to turn that on, and block and [stop that cancer]?
The other two genes are CDX2 and GATA3. They’re responsible for placenta-making. It was really cool to use this tool to differentiate one cell type from another, and take induced pluripotent stem cells and create placenta progenitor cells.
Is there a downside to awakening a dormant gene?
We want to do that in a controlled way — we turn it on, then carefully monitor it. If it is misbehaving, we can manipulate it and restore its healthy function.
What are the remaining questions you have now?
The biggest mission for us now is to address it in many different applications. To study how this tool works in different cells, genes, organs, [and apply it to] the therapeutic world.