It's the last Friday of July, and this week's gene therapy story is a study in who leaves a room and who stays. The names walking out are the ones that spent the most to be there. The one arguing they left too early runs a company that has never sold a drug and doesn't plan to.
Pfizer has pulled Beqvez, its hemophilia B gene therapy, off the market after booking zero commercial sales and told the field it won't build another AAV-based gene therapy. Vertex has walked away from AAV research altogether. Roche's Spark Therapeutics still holds Luxturna, the closest thing the field has to a flagship, and still generated just $40m in the first half of 2025. Between them, some of the largest commitments ever made to AAV gene therapy are being unwound.
Eric Kelsic's timing is either terrible or perfect.
Kelsic founded and runs Dyno Therapeutics, a Watertown, Massachusetts outfit that has spent seven years on one narrow problem: the capsid, the protein shell that ferries a genetic payload into a cell. His argument is not that today's AAV therapies are good enough. It's that they are a rough draft, and Dyno has already written much of the next one.
"With that one capsid, you can deliver many payloads, so you can help a lot of patients across many diseases," he told In Vivo.
Selling The Shovels
Dyno has never tried to bring a drug to market, and that was the point. During a postdoc in George Church's lab at Harvard around 2015, Kelsic decided the field's real bottleneck was delivery, and that the most valuable place to stand was not next to the prospectors but behind them, handing out better tools. "Working to help our partners make better medicines, rather than competing with them," is how he puts it.
The platform, CapsidMap, trains machine learning on hundreds of thousands of capsid sequences from animal studies to design better variants. The old directed-evolution approach runs into a wall: every extra mutation you stack on a capsid raises the odds it stops working, so gains decay fast. Dyno's pitch is that it can predict which combinations will hold up before anyone commits to a wet lab.
Big pharma has bought in, at least on paper. Roche signed on in 2020 in a deal worth up to more than $1.8bn, then exercised an option to license a capsid for an undisclosed neurological disease, a move that triggered a $7m payment and opened a path to more than $220m in further milestones. Astellas took a muscle capsid for $15m upfront. Novartis and Sarepta are in the network too, on eye and muscle programs respectively, with a combined potential value north of $2bn.
The New Vectors
At the most recent American Society of Gene & Cell Therapy meeting, Dyno unveiled four new vectors across muscle and CNS and a wider partnering program. The muscle capsid, Dyno-bn8, reached about 80% of skeletal muscle fibers in non-human primates at a dose 25 times lower than Elevidys, Sarepta's approved Duchenne treatment, and did so while largely steering clear of the liver. That last part matters: off-target delivery to the liver has been tied to toxicity, including two patient deaths linked to Sarepta over the past two years.