Jennifer Doudna, Ph.D., the Nobel Laureate best known for co-inventing the CRISPR system that has become the foundation for gene editing, is dipping her toes into a new, buzzy field: AI-powered protein design.
In a new paper, published today in Science, Doudna’s team described a new platform for generating novel gene editing enzymes. To prove the platform’s worth, the researchers designed a never-before-seen protein that can bind to and cut DNA.
The new enzyme is a TnpB, a tiny type of RNA-guided nuclease that is thought to be the evolutionary ancestor of the CRISPR-Cas system made famous by Doudna and her fellow Nobel winner Emmanuelle Charpentier, Ph.D.
But unlike other methods that repurpose enzymes found in nature, this new platform is capable of crafting proteins that are far afield from those forged by millennia of evolution.
“We were able to develop non-natural nucleases that were active in human, plant and bacterial cells,” Doudna told Fierce. “This approach opens the door to the possibility of designing an enzyme on-demand for a particular problem, whether it’s for treating a genetic disease or helping crop plants adapt to a changing climate.”
The new design strategy was the brainchild of structural biologist Petr Skopintsev, Ph.D., a scientist in Doudna’s lab at the University of California, Berkeley’s Innovative Genomics Institute. He became curious about a new model (PDF) from tech giant Meta’s AI research division that is like an inverse version of Google DeepMind's AlphaFold. Rather than predicting a protein’s 3D shape from its amino acid sequence, the inverse model goes the other way around.
“You define a backbone structure of a protein and find sequences that fold into that,” Skopintsev told Fierce. When he entered TnpB into the model, he found results that “made a lot of sense” to his experienced eye. By then restricting the model’s output to include known DNA-binding sites, the team could generate reams of potential new enzymes.
Another Doudna lab scientist, Isabel Esain-Garcia, Ph.D., quickly got involved to validate the generated proteins in genome editing experiments, she told Fierce.
TnpBs are a flexible group of proteins, and as a result could be engineered using the new platform for a variety of purposes as well. In addition to potential uses in personalized medicine or crop altering, Esain-Garcia is excited by the fact that the little enzymes are much easier to deliver than others.
“One of the key advantages of these proteins is that they're tiny,” she said. “We think that this is going to be very beneficial [for] organ-specific delivering, just because it's going to be very easy to pack these editors.”
Delivering minuscule gene editing tools is already the domain of another Doudna endeavor, Mammoth Biosciences. But while patents have been filed for this new platform, Doudna stayed mum when asked if another startup is in the works.
“The novelty isn’t in the fact that they’ve created another small nuclease, but rather in the approach that they took,” Benjamin Kleinstiver, Ph.D., a gene editing enzyme engineer at Mass General Brigham and Harvard Medical School, who was not involved with the work, told Fierce. “The fact that they can make so many mutations in certain domains of TnpB while retaining nuclease activity is impressive. And that there can be different enzymes that achieve maximal activity on each target site in human cells is interesting.”
The Doudna approach is “somewhat analogous,” Kleinstiver added, to another recently developed by Profluent Bio, a Bay Area biotech founded by Berkeley alum Ali Madani, Ph.D.
While he doesn’t see any new capabilities for genome editing just yet, the Doudna lab’s method has interesting implications for intellectual property and patents, he said, a major area of contention in the gene editing field.
Many enzyme patents are filed to be broad enough to cover other enzymes that share a certain percentage of the same sequence, Kleinstiver explained. “These new methods, if they can sufficiently diversify the sequences past these thresholds, may be able to create IP-evading enzymes,” he said.
While Doudna’s lab has discovered proteins using AI before, and designed them with pre-AI methods, this is the first time the celebrated biologist has used the hyped tool to create new proteins. AI protein design is chiefly the domain of another more recent Nobel Laureate, David Baker, Ph.D.
Doudna’s team works closely with Baker’s, she told Fierce, “so hopefully it’s more of a ‘1 + 1 = 3’ situation where we can use our complementary strengths to make some new breakthroughs we couldn’t do alone.”
Kleinstiver, for his part, isn’t concerned about Doudna entering the mix.
“The more the merrier,” he said. “Great to see many labs in the field using some of the latest protein design models to engineer genome editing technologies.”