Predicting AAV Delivery Across the Blood-Brain Barrier

Short answer

How do you predict whether an AAV capsid will cross the blood-brain barrier? You model receptor-mediated transcytosis, not diffusion. Capsids are large and charged, so small-molecule permeability rules score them at essentially zero while some capsids clearly do reach the brain. A usable prediction combines receptor engagement, published in vivo biodistribution and sequence similarity to known penetrant capsids — and it must carry a species flag, because the best-known CNS capsid works through a receptor primates do not have.

Why ADME heuristics do not transfer

Brain-penetration rules borrowed from small-molecule discovery — lipophilicity, polar surface area, molecular weight cut-offs — assume passive diffusion across the endothelium. An AAV capsid is a large, charged protein shell that does not diffuse, so those rules score every capsid at the floor and lose all discriminating power.

The actual mechanism is receptor-mediated transcytosis across brain microvascular endothelium. Atlas Bio's published model follows that biology directly, scoring engagement with transcytosis receptor pathways including LRP1, transferrin, GLUT1 and integrins, and reports a separation of 'CNS 0.641 vs non-CNS 0.244' where the prior ADME-style heuristic produced 0.000.

The species trap, in detail

The field's cardinal translation failure is well documented. Hordeaux et al. (Molecular Therapy 2018) showed that the remarkable CNS transduction of the AAV9 variant PHP.B is restricted to the model in which it was selected — a C57BL/6J background — and was not found in non-human primates or in the commonly used BALB/cJ strain. The same group reported serious acute toxicity potential in non-human primates after high-dose systemic administration.

The mechanism was then pinned down: Hordeaux et al. (Molecular Therapy 2019) mapped the phenotype to a haplotype of Ly6a, which encodes a GPI-anchored protein, and linked high blood-brain-barrier transport to binding between AAV-PHP.B and LY6A. Because that receptor is not available in primates, a mouse CNS result obtained through it does not transfer. Any model that ranks capsids for brain delivery must detect this dependency and attach the species flag to the prediction.

What a defensible BBB score is built from

Atlas Bio publishes the composition of its score: receptor engagement weighted 0.40, published CNS biodistribution similarity 0.25, ESM-2 sequence similarity to known penetrant capsids 0.20, and surface charge profile at physiological pH 0.15, with the weighted sum bounded between zero and one and calibrated so that the murine gold-standard CNS-penetrant sits near 0.71.

The validation claim is ordinal rather than absolute: the ranking reproduces published in vivo CNS biodistribution orderings without per-capsid tuning, anchored on literature including the finding that intravenous AAV9 bypasses the barrier and transduces central nervous system cells, with astrocytes predominating in adults and neurons in neonates (Foust et al., Nature Biotechnology 2009).

Receptor engagement across named transcytosis pathways
Similarity to reported in vivo CNS biodistribution from peer-reviewed studies
Sequence-embedding similarity to known penetrant capsids
Surface charge at physiological pH

Three limits to state before anyone uses the number

Atlas Bio publishes its own caveats, and they are the right ones to demand of any vendor. The species bridge is not modelled, so the score is a relative ranking rather than an absolute, species-transferable prediction. Route of administration changes which receptor pathways dominate, and the published score assumes systemic intravenous delivery. Pre-existing immunity overrides chemistry: a patient seropositive for the capsid family will not achieve brain delivery regardless of the score.

Used inside those limits, a BBB score is a sequencing tool for expensive work — it tells you which candidates deserve the first biodistribution study, and which carry a species dependency that will not survive the pivot to primates.

How Atlas Bio does this

Atlas Bio's BBB Transcytosis Model scores brain penetration from receptor biology rather than ADME heuristics, names the contributing factors behind each score, and surfaces species-specificity warnings on central-nervous-system predictions.

Common questions

Why did PHP.B fail in primates?

Its brain transport depends on binding to LY6A, a GPI-anchored protein whose relevant haplotype is a mouse-strain feature. Hordeaux et al. showed the enhanced CNS tropism did not appear in non-human primates or in BALB/cJ mice.

Can a BBB score be compared across routes of administration?

No. Intravenous, intrathecal and intracerebral delivery engage different pathways and bypass different barriers. A score calibrated on systemic intravenous data should not be reused to rank candidates for direct central nervous system administration.

Does a high BBB score mean a patient will respond?

No. Pre-existing neutralising immunity to the capsid family can prevent delivery entirely, independent of transcytosis biology, which is why seroprevalence-adjusted output and patient screening sit alongside the score.

Sources

Talk to Atlas Bio.

Sponsors, researchers and clinicians: tell us the decision you are trying to make and we will show what the platform can and cannot answer.