Part 2 - The Vector Debate

The AAV Question

Gene4Neuro Series 

Ask a group of people working in gene therapy for neurological disease which vector they would bet on, and a substantial fraction will still say adeno-associated virus. There are good reasons for this.  AAV's track record is comparatively strong: decades of clinical safety data, established manufacturing infrastructure, tissue tropism with some serotypes, such as AAV9, with preference for the nervous system, and a regulatory pathway that provides a useful precedent for prospective therapies. The approved CNS gene therapies that exist today — the ones that have actually reached patients — are built on AAV.

And yet, inside the conversations we have across our network — with platform technology innovators, clinical developers, investors, and manufacturing specialists — a more complex picture emerges. AAV is not standing still, but neither is the rest of the field. The vector debate has quietly become one of the most contested strategic questions in genetic medicines for the brain.

So we want to take a closer look

.

What AAV does well — and where it runs into walls

AAV's strengths are real. For a one-time intervention in a pediatric rare disease with a defined genetic target and a small patient population, the case remains compelling. The tissue tropism profiles of established serotypes are somewhat well-characterized. The regulatory packages, while challenging, represent a path that has been walked before. The manufacturing ecosystem, imperfect as it is, exists.

But the constraints are also real, and they matter more as the field expands. The capacity  ceiling of AAV  rules out genes larger than 5 kilobases without engineering workarounds that introduce their own complexity. Pre-existing neutralizing antibodies, present in a  substantial fraction of adult populations, can either block delivery entirely or drive immune responses at the doses required to penetrate the CNS. And the redosing problem — AAV is not designed for repeat administration — sits uncomfortably against a neurodegenerative disease field that is starting to think about therapies for conditions that evolve over decades rather than months.

Immunogenicity and hepatotoxicity signals at high systemic doses have added a safety dimension to this conversation that was not fully anticipated when many current programs were designed. The field is learning in real time what it means to deliver very large quantities of engineered viral particles into a human being — and some of those lessons have been difficult ones.

The engineering response

The most sophisticated response to AAV's limitations has been to engineer the vector itself rather than replace it. This work is happening at the frontier of biology and machine learning, and it is producing results that would have seemed implausible a few years ago.

Ben Deverman's lab at the Broad Institute of MIT and Harvard — one of the voices we heard from at Bio-Neuroscience in February — has been at the leading edge of this effort. His team's development of novel capsid engineering platforms, including ML-guided approaches that can screen vast sequence spaces for capsids with defined human-relevant properties, represents a genuine game changer  from the empirical selection methods that produced most current clinical vectors. The publication in Science in 2024 of an engineered AAV that targets human transferrin receptor 1 to cross the blood-brain barrier — delivering genes brain-wide in humanized mice after intravenous administration — is one of the most closely watched results in the field right now. Work like this suggests that the question is not whether AAV can be improved, but how far that improvement can go, and how quickly it translates from laboratory to clinic.

Sanofi, whose Global Head of Rare and Neurological Diseases Pablo Sardi opened Gene4Neuro 2025 with a keynote on the science and challenges ahead, is pursuing an analogous strategy at industrial scale — using AI to design novel AAV capsids alongside antibody-based brain shuttle approaches. The convergence of academic platform engineering and pharma-scale AI application is one of the more encouraging structural developments in the field.

The challengers are no longer hypothetical

Non-viral delivery platforms are advancing in parallel. Lipid nanoparticles brought a proof of concept to global attention  through mRNA vaccines; the translation to CNS applications is not straightforward, but the pace of CNS-targeted LNP engineering has accelerated considerably on the back of that investment. Engineered virus-like particles, exosome-based delivery, and polymer nanoparticles each carry distinct tissue tropism profiles and immunogenicity characteristics — and none requires the manufacturing infrastructure that AAV demands, with its associated cost consequences for pricing and access.

None of these alternatives has the  clinical track record of AAV yet. But track records are built from trials, and trials are being designed now.

What the choice actually depends on

From our vantage point — sitting across conversations with researchers, clinicians, investors, pharma executives, CEOs and company builders in this space — the honest answer is that AAV versus non-viral may not be a field-wide choice  waiting to be made. It is a disease-specific, patient-specific, and target-specific question. For a pediatric monogenic rare disease with a defined gene defect and a one-time intervention logic, the case for an engineered AAV is strong and getting stronger. For a late-onset neurodegenerative condition requiring broad neuronal  distribution in an adult population with variable anti-AAV antibody status, the calculation looks different.

What the field cannot afford is a platform monoculture — an over-investment in one approach that crowds out the diversity of solutions the hardest CNS targets will ultimately require. That is a question about capital allocation as much as science, and it is one we will be exploring directly in Pamplona.

We would love to hear where this debate stands in your work and your organization. The people shaping the answer are in this community — and many of them will be in the room at Gene4Neuro.