Gene Therapy · asia
Fitting a Base Editor into a Single AAV: AI-Optimized Zinc-Finger Proteins Show Efficacy in Mice with Spinal Muscular Atrophy
A protein language model helped improve DNA-recognizing zinc-finger proteins, addressing the activity and targeting limitations of compact CRISPR editors. Delivered to mice in a single AAV, the new system corrected SMN2 splicing, but safety and durability questions remain before it can become a human therapy.
Gene-editing therapies must not only edit accurately; they must also be deliverable into cells. Many base editors exceed the packaging capacity of adeno-associated virus (AAV) and often have to be split between two vectors, increasing manufacturing and delivery complexity. A study published in *Nature Communications* proposes another approach: using zinc-finger proteins optimized with artificial intelligence to overcome the limited activity and targeting range of compact CRISPR systems despite their small size.
The research team’s zmBE platform links programmable zinc-finger proteins to compact base editors built around Un1Cas12f1 or OgeuIscB. The zinc-finger proteins can additionally bind the target DNA, helping position the editing complex. This increases editing efficiency at some sites and expands the range of targetable locations beyond those constrained by CRISPR recognition rules.
Designing zinc-finger proteins has traditionally required repeated screening; simply assembling modules does not guarantee effectiveness. The team therefore used a protein language model to evaluate and optimize candidate sequences, followed by experimental validation. Here, AI did not directly predict patient outcomes. Instead, it narrowed the protein-engineering search space to identify more suitable binding components for specific DNA sites. Sequencing data associated with the study have been deposited under PRJNA1250402 and GSE294498, providing a basis for external review and reanalysis.
The team further applied the platform to spinal muscular atrophy. The disease usually results from loss of SMN1 function. Humans also have the closely related SMN2 gene, but because exon 7 is often skipped during RNA splicing, it can produce only a limited amount of full-length SMN protein. The researchers used zmBE to alter a base near exon 7 of SMN2, increasing exon 7 inclusion and thereby shifting SMN2 toward a more functional transcript form.
According to the paper and PubMed record, the complete editing components can be packaged into a single AAV and improved disease features in a mouse model of spinal muscular atrophy. This validated two aspects simultaneously: AI-assisted zinc-finger protein design can support compact editors in vivo, and the compressed system retains enough activity to alter disease-related biology. However, the current evidence remains limited to mouse research and cannot be used to infer efficacy in humans.
Before moving toward the clinic, researchers must still clarify unintended DNA and RNA editing, bystander base changes, the additional off-target risks introduced by zinc-finger proteins, and AAV dosage, tissue distribution, and immune responses. Permanently rewriting DNA also makes long-term follow-up particularly important. A single vector lowers the delivery barrier but does not eliminate the core manufacturing, safety, and regulatory challenges of gene therapy. The near-term value of this work is primarily as an extensible engineering platform for compact editors, rather than as a therapy already approaching clinical use.