Podcast recap: Jagesh Shah on why delivery is genetic medicine's real bottleneck

In genetic medicine, we increasingly know what to deliver. The unsolved problem is getting it to the right place in the body. That was the central thesis in the latest episode of The Genetics Podcast, featuring Dr. Jagesh Shah, Chief Scientific Officer at Mirai Bio, in conversation with host Patrick Short.

Jagesh, a former Harvard Medical School systems biology professor who co-founded Cobalt (which later merged into Sana Biotechnology), joined Flagship Pioneering's Mirai in 2024 to focus exclusively on the delivery challenge for nucleic acid medicines.

The conversation offers a clear-eyed look at why delivery has become the field's defining constraint, and how one company is engineering its way through it.

Why the body resists therapeutic nucleic acids

The challenge begins with evolution. Over hundreds of millions of years, biological systems have developed robust defenses against foreign nucleic acids. Those defenses were shaped to fend off viral infections. Now, as Jagesh describes it, the field is trying to "flip that around and try to make it a therapeutic."

This evolutionary barrier explains why delivery remains difficult despite decades of research. Lipid nanoparticles (LNPs) have been studied for roughly six decades and proved remarkably successful in COVID-19 vaccines. But the reactogenicity that helped stimulate immune responses in a vaccine context is undesirable for chronic therapeutic use. The COVID vaccine success demonstrated feasibility. It did not solve the broader delivery problem.

Today, the tissues that are reliably reachable include the liver, the central nervous system (via cerebrospinal fluid), and the eye. The next frontier, Jagesh explained, includes skeletal and cardiac muscle, as well as the brain via the bloodstream. This transendothelial delivery challenge represents one of the field's most difficult problems.

Modularity as the design principle

Mirai's approach rests on a modular LNP platform architecture. The concept separates the "outside" of the nanoparticle, which governs tropism and biodistribution (where the particle goes), from the "inside," which handles cargo protection and endosomal escape (how the payload gets into the cell once it arrives).¢

This separation enables components to be swapped to redirect delivery without rebuilding the entire system. As Jagesh put it: "You can't have modularity without abstraction. So I would like the outside to just be the directional vector... and then the inside be the one that when you get to your destination, then it takes over to deliver the nucleic acid."

This design philosophy underpins Mirai's ability to target multiple tissue types. At ASGCT, the company presented data on delivery to adipocytes and CD8 T cells. In one preclinical study, Mirai demonstrated sustained B cell depletion for approximately one week in a non-human primate via an in vivo CAR approach, a result that was also shown to be tolerable.

Machine learning as a collaborator

Mirai uses a machine learning feedback loop to navigate the vast chemical space of lipid formulations. Rather than exhaustive screening, the system learns from each in vivo experiment to guide the next round of design. The objective function can be tuned to optimize for specific goals, such as reaching the spleen while avoiding the liver.

"Within three or four rounds, we're at a very high performing lipid nanoparticle," Jagesh said. "And that in vivo work is done quite efficiently... four rounds is about six months."

Jagesh was candid about where ML still struggles. Encoding a chemical entity into numbers remains hard. So does modeling the multi-scale journey from lipid molecule to nanoparticle to bloodstream to target cell. The team's goal is not just to achieve performance but to "crack the box" and understand the underlying principles.

He also tempered expectations about automation: "Not to make it sound like it's an ML overlord or anything, but this is a collaboration with the machine." The vision of "chat LNP," where a user could simply type in the desired destination and potency, remains aspirational.

A platform model that spreads risk

Mirai does not develop its own therapeutic assets. Instead, it builds and licenses modular delivery systems to partners pursuing their own nucleic acid cargos. This platform model enables multiple programs to benefit from shared infrastructure while spreading risk across a broader portfolio.

"People really have good nucleic acid ideas out there," Jagesh explained, "and our job is to enable the delivery to the sites that they wanna be able to access."

This approach also offers a form of future-proofing. Whatever new therapeutic modality emerges next, whether it is the next CRISPR-Cas9 or something yet to be discovered, it will likely be encoded in RNA or DNA. A robust LNP delivery platform can carry it. Unlike AAV, which faces strict cargo size limits, LNPs have no clear ceiling. Self-amplifying mRNA cargos in the range of 12 to 16 kilobases have already been delivered successfully.

The regulatory pace gap

Jagesh raised a structural concern that extends beyond any single company. If the field has so many nucleic acid cargo ideas, the rate of approved therapeutics should be higher than it is. The pace has not matched expectations.

Part of the issue is regulatory. The field has two reference points at opposite extremes: the COVID vaccines, which were developed at pandemic speed and global scale, and the N-of-1 treatment for Baby KJ, a bespoke intervention for a single patient. Neither model translates directly to most rare disease programs.

Closing the gap between these poles is essential for genetic medicine to reach the patients who could benefit. Delivery infrastructure is one piece of that puzzle. Regulatory frameworks that can accommodate faster development without compromising safety are another.

Key takeaway

This conversation reinforces a point that has surfaced across recent episodes of The Genetics Podcast: the field is shifting from discovery to execution. We increasingly know what to target. Now the question is whether we can deliver it.

Dr. Jagesh's work at Mirai represents one approach to that challenge. Modular design, ML-guided iteration, and a platform model that enables partners may together help close the delivery gap that has held the field back.

Listen to the full episode below.