The P5 nick in the plasmid: how AAV manufacturing DNA ends up in a patient's liver

In Short. In one child’s liver after Zolgensma for SMA, manufacturing REP/CAP DNA sat at about 0.5-1% of the therapeutic transgene. A medRxiv reanalysis argues those contaminants start at an intact AAV P5 promoter parked downstream of the capsid gene on the packaging plasmid, an avoidable design choice rather than random leftover DNA.

---

Recombinant AAV (rAAV) delivers a therapeutic gene inside a viral shell. The shell’s protein parts and the helper genes that build them are supplied during manufacturing from separate DNA plasmids. Those helper sequences are tools for production. They are not the medicine. When packaging goes wrong in a patterned way, pieces of that helper DNA can ride along inside particles and later turn up in a treated person’s cells.

A recent biopsy study found that pattern in a child treated with Zolgensma (onasemnogene abeparvovec-xioi) for spinal muscular atrophy who later developed hepatitis. Manufacturing-derived AAV REP/CAP sequences appeared in about 5% of hepatocytes and at levels hard to explain as random plasmid scraps (Buddle et al., as cited in Brimble). Brimble and colleagues at St. Jude and collaborators then reanalyzed those sequencing data to ask how the contaminants were made (medRxiv DOI 10.64898/2026.09.10.26362787). Their answer centers on a small DNA switch called the AAV P5 promoter.

Cargo versus REP/CAP

The therapeutic payload is an SMN1-class gene cassette flanked by inverted terminal repeats (ITRs), the bookends that tell the AAV machinery what to package as the vector genome. REP and CAP are different. REP encodes Rep proteins that nick and package DNA. CAP encodes the capsid shell. In manufacturing, REP/CAP usually live on a packaging plasmid that never should become the patient’s lasting genetic cargo.

In this liver, REP/CAP-derived sequence was still abundant relative to the therapeutic transgene, corresponding to 0.5-1% of that therapeutic DNA in the authors’ report. That fraction is a measurement of what persisted in tissue for this one patient. It is not a class-wide impurity rate for every AAV product.

The P5 placement that pulls upstream DNA in

P5 is a native AAV promoter that also contains a Rep-binding element and a nearby nicking site. When an intact P5 sits immediately downstream of the capsid gene on the packaging plasmid, Rep-mediated incorporation can pull the upstream REP and CAP coding sequences into particles as contaminants. The process runs independently of packaging the ITR-flanked therapeutic genome and can fill fragments up to the roughly 5 kb packaging limit.

The Zolgensma packaging plasmid sequences are proprietary and not public. Brimble et al. therefore inferred layout from the patient’s reads. De novo assembly produced a contiguous contaminant contig spanning REP and CAP and ending inside P5 at the Rep nicking site, just downstream of the Rep-binding element. That boundary matches the known nick and supports initiation at P5. A second contig looked like reverse-packaged plasmid backbone, about 3% of the vector genome in their accounting, a previously described undersized-backbone route.

Among short reads only partially aligned to the REP/CAP contig, 17% (9 of 46 soft-clipped reads) linked to ITR-derived sequence at heterogeneous junctions. That linkage is a recombination signature between contaminant DNA and ITR-containing vector DNA after packaging or after transduction. It is evidence of mixing. It does not prove that a complete wild-type virus formed in this child.

Long reads put the breakpoint on P5

Nanopore coverage was thin: only 16 reads mapped to the REP/CAP contig. Still, breakpoints along that contig landed most often in P5, in 5 of 16 reads (31%). One read carried two independent P5 stretches, consistent with a short P5-initiated fragment joining a larger one. The authors, like Buddle et al., did not recover a full-length replication-competent AAV genome from these data. Limited depth leaves that as a negative finding with a small sample of long reads, not a certificate that such genomes are impossible.

Scope that stays honest

This is a mechanistic reanalysis of one patient’s liver sequencing after one commercial product. Proprietary plasmid maps were reconstructed from tissue, not assayed from a vial. Tissue levels need not equal drug-product release assays. The RNAseq depth in the original biopsy was insufficient to tell whether these REP/CAP fragments were transcribed in this liver. Prior lab work has shown that P5-associated contaminants can be transcriptionally active and can elicit T-cell responses in experimental systems. Those experimental facts motivate attention. They do not convert this biopsy into a measured immune story for this child.

Zolgensma remains a therapy that changed survival and motor outcomes for many infants with SMA. Rare serious liver and inflammatory complications have been reported in the field and are part of why product composition is studied carefully. The manufacturing lesson here is narrower. Putting P5 downstream of CAP is a common historical plasmid layout chosen for yield. It is also, on this evidence, a modifiable origin for a specific class of REP/CAP contaminants. Moving or altering that nickable P5, or using production systems that drive Rep without native P5, are the engineering responses the paper points toward. Prevalence across the broader AAV commercial class is not measured in this study and should not be invented from it.

Brimble M.A. et al. P5 promoter-mediated incorporation explains REP/CAP manufacturing contaminants in patient liver after rAAV gene therapy. medRxiv (v1, 2026-09-14). Preprint, not peer-reviewed.