Spun out of George Church's lab, this biotech upstart is mapping the AAV universe for Novartis, Sarepta to gaze
In a few days, through a series of video conferences, gene therapy researchers around the world will be presenting their latest findings at the virtual annual meeting of the American Society of Gene & Cell Therapy. Almost every discussion will feature a topic that has been central to the existence of the field but continues to perplex experts as they seek to refine the modality: the delivery of a gene to the tissue where it’s needed to fix disease.
For the first time, a biotech upstart will be publicly outlining their take on the problem — with an artificial intelligence flavor that Novartis and Sarepta are gobbling up.
Seasoned attendees of ASGCT would recognize the team behind Dyno Therapeutics. Since Eric Kelsic began building the platform in 2015 as a postdoc at George Church’s illustrious lab at Wyss Institute, he’s been making the rounds at scientific meetings. At Harvard, his group had demonstrated how — by doing high throughput screening on all capsid variants of one particular AAV serotype, modeling the space with machine learning, and finally building a profile of each capsid that can be ranked by different attributes — they could point to synthetic AAV capsid candidates that are superior to the handful of natural variants currently in use.
“This was by far the best application that I’d ever seen of AI in biology,” Alan Crane, an entrepreneur partner at Polaris and Dyno’s executive chairman, told Endpoints News. “It turns out potential partners were seeing it the same way, because when Eric came to me back in mid-2018, he already had this list of literally dozens of companies that have proactively approached him.”
Out of that pool Dyno had picked Novartis for a collaboration on eye disorders and Sarepta to team up on muscle diseases. Upfront payments, support, option fees and milestones from these two deals could add up to $2 billion, including $40 million from the research phase of the Sarepta deal.
“We always constantly try to make sure that we are ahead of the curve in terms of our technology and looking at next-generation treatments,” Louise Rodino-Klapac, Sarepta’s head of gene therapy, said. “So although we’re very happy with our current approach and our current vector, we’re thinking about the future potential technologies for other muscular dystrophies.”
Capsids — the protein shells that enclose genetic material of a virus — is one of three core components needed to form a gene therapy, she explained, alongside the transgene that’s missing or defective in a patient, and a promoter that turns the gene on in the cell. And small tweaks to the capsid can translate to profound changes in the final product’s immunogenicity, manufacturability, efficiency of delivery, specificity to target cells and package size.
All of these metrics are taken into consideration on Dyno’s CapsidMap platform, which takes “the most comprehensive approach to mapping out the AAV universe,” filling the gaps in each galaxy and telling stars from pure void, Kelsic said.
“We don’t want to improve one property but have other things get worse,” a challenge that others who have attempted to solve the problem have faced, he added.
With a new technology that promises to optimize viral vectors for individual applications like that, Crane predicted the company — which Polaris seeded with a modest $9 million — might never need additional venture funds.
While Dyno retains the option to create its own therapies, expect partnerships (and there are more coming) to remain at the center for a while.
“What I’ve observed in the industry — not only in gene therapy but in all areas — is as companies start to move into pipelines, they usually have to leave the platform behind,” Crane said.
Much work is to be done. Dyno currently has capacity to screen hundreds of thousands to millions of capsids and test them both in vitro and in vivo, but the plan is to scale up the infrastructure even further — both on the experimental and the computational fronts. The headcount is doubling from roughly 20 while all the machine learning gets moved onto the cloud.
A candidate won’t emerge any time soon, and even when it does materialize it would have to go through rigorous safety testing at the partners’ own R&D operations — a process that could take another one or two years. Still, Kelsic sees it as the quickest way to bring their work to patients even while they figure new things out.
“Especially when we’re thinking about technology, something George and I talked a lot about when we started this project, it still feels really early days for gene therapy,” he said. “There’s so much potential.”