Del Zota The One Safety Question Made Exact

A Third of Normal Dystrophin — and the One Del-Zota Safety Question That Can Be Made Exact

A shear study on del-zota (AOC 1044, Avidity Biosciences) for Duchenne muscular dystrophy amenable to exon 44 skipping. Angle: SEAL. Lifecycle: standing charter, molecular court dark — the payload sequence is not public. The subject under grading is the safety instrument — never the medicine, the company, or a patient.

The disease, and the boys this drug is for

Duchenne begins in a single gene. A mutation in DMD breaks the reading frame, so little or no working dystrophin is made, and muscle rebuilds slower than it is spent. It is progressive: signs usually appear between two and five, walking gets harder through childhood, and the heart and the muscles of breathing are drawn in later. With today's care — corticosteroids, cardiac and respiratory support, a multidisciplinary team — many young men now live well into their thirties and beyond; the older survival numbers that still circulate belong to an earlier era of care, and the line is moving in the right direction.

One subgroup stands slightly apart, and it is exactly the one del-zota is built for: boys whose mutation is amenable to exon 44 skipping, roughly 6% of the Duchenne population. Natural-history cohorts show this group tends to progress more gently — independent walking is kept about two years longer on average, there are more "revertant" dystrophin-positive fibers, and a little exon 44 skips on its own, partly mending the frame with no drug at all. That is real good news for these families. It is also a confounder worth naming out loud, because it shapes how the early numbers should be read: a milder baseline makes a single-arm result look better against natural history than a randomized trial would.

The breakthrough

Del-zota (AOC 1044) is an antibody-oligonucleotide conjugate, and the idea behind it is the kind that changes what is reachable. A humanized antibody against transferrin-receptor-1 (TfR1) — abundant on muscle — carries multiple copies of an exon-44-skipping oligonucleotide into the fiber, a delivery gateway a naked or peptide-linked oligo cannot match; the peer-reviewed mechanism paper reports the conjugate delivering markedly more oligo to skeletal muscle than a peptide-conjugated comparator in preclinical models, and the Phase 1/2 program went on to measure real muscle delivery and skipping in people.

The payload's mechanism matters more than any single number here, because it governs the whole safety story. The oligonucleotide is a phosphorodiamidate morpholino (PMO) working by steric block: it base-pairs to the exon-44 region of the DMD pre-mRNA and hides that stretch from the spliceosome, so exon 44 is left out, the reading frame is restored, and an internally shortened but functional dystrophin is made. A PMO does not recruit RNase H and does not cleave RNA. So its off-target consequence is off-target splicing, not off-target cleavage — the distinction that decides which question below can be made exact.

The results are the strongest this subgroup has seen. On the biomarker (5 mg/kg, three doses), the sponsor reported exon-44 skipping rising from about 7% to about 44%, dystrophin rising from about 7% to about 32% of normal (a roughly 25-percentage-point increase, the sponsor's headline figure), and creatine kinase falling to near-normal — an order of magnitude past what the four approved exon-skippers reach for other exons. A September 2025 functional update (17 participants, ~12 months, against matched natural history) moved the right way across stair-climb, walk/run, time-to-rise, NSAA and PUL2. Those functional figures are sponsor-reported, open-label, small-N, measured against external natural history in a slower-progressing subgroup — genuinely promising, and not yet a randomized result. The confirmatory Phase 3, SAFARI44, is what converts them; del-zota carries Breakthrough Therapy, Fast Track, Orphan and Rare Pediatric designations, with a BLA submission planned for Q1 2026.

This is real hope, and it deserves to be read as hope. The case below is not about the medicine. It is about which safety question a computer can answer exactly, and which it cannot.

The one question that is genuinely discrete

Del-zota's most computable safety risk is the one every splice-switching oligo carries, and it is not hypothetical. An oligo of this type can bind and perturb splicing not only at its target but at other sites in the transcriptome that are only partly complementary. The anchoring measurement is Scharner et al. (NAR, 2020): of 108 predicted partially-complementary sites for one splice-switching oligo, 17 produced real mis-splicing — with effects appearing at as few as 12 of 18 nucleotides paired, tolerating G:U wobbles, terminal mismatches and short bulges. Their own verdict is the seam: these off-targets are hard to predict, and the standard in-silico ranker they used scores candidates by a floating-point minimum-free-energy value prone to both false positives and false negatives.

That is exactly where an exact instrument fits, and it fits well:

• The question is discrete. Watson–Crick pairing is a rule over an integer alphabet. "Where in the human transcriptome can this 20–25-mer bind within k mismatches and bulges?" is an exhaustive enumeration — finite, bounded, one right answer.
• The answer is re-derivable. An integer complementarity count returns the same off-target set on every machine, with no thermodynamic parameter set to choose and nothing to round. A regulator can reproduce it byte-for-byte, years later.
• The conventional score shears. A ΔG/MFE heuristic rides on a nearest-neighbor parameter set; published sets differ within their own stated uncertainty, and the candidate list reorders and reclassifies as parameters and rounding move. That is the same exact-versus-float shear measured out of domain on the fusion court — Study 34 on the observer/π-bracket axis, Study 35 on the time axis — and shown for antisense off-target on synthetic sequences locally (reproduce/aso-offtarget-exact-vs-float.swift, marker ASO_OFFTARGET_EXACT_IS_OBSERVER_INVARIANT) and in its CRISPR-guide sibling (reproduce/crispr-guide-offtarget-exact-vs-float.swift, marker CRISPR_OFFTARGET_EXACT_IS_OBSERVER_INVARIANT).

The honest verdict type is a design-stage SEAL: given the payload sequence, emit the exhaustive, re-derivable list of every near-complementary transcript site inside a stated mismatch/bulge budget — a certificate a reviewer reproduces exactly, in place of a float heuristic the field itself distrusts.

And here is the boundary, drawn as loudly as the claim: del-zota's PMO ("PMO44") sequence is not public. The peer-reviewed mechanism paper names the lead and its potency but does not print the bases, and reports no transcriptome-wide off-target analysis. So for the real molecule the molecular court is dark — this is a charter for a seal, not a completed one. The local demo runs on synthetic 20-mers; it shows the kind of exact-versus-float divergence, and no run on del-zota is claimed. The seal becomes a live verdict the moment the sponsor or a regulator puts PMO44 on the table.

Where the exact tool goes silent

The instrument answers a sequence question. Most of what remains of del-zota's safety is chemistry, immunology and pharmacokinetics — none of it a discrete enumeration, and the boundary belongs in the open. Hypersensitivity and infusion-reaction risk is real for this antibody-conjugate class and has been reported in the program; anti-drug-antibody and infusion-reaction risk is a wet-lab and clinical question, not a base search. TfR1 is also present at the blood-brain barrier and on red-cell precursors, so hematologic effects are a genuine class consideration — the sponsor specifically reported no symptomatic hemoglobin changes, which is reassuring, and which is a biology measurement, not a sequence one. Conjugate stability, drug-antibody ratio and payload release are PK and CMC. Tissue distribution, dosing interval, durability and delivery to the heart are PK/PD. And because a PMO does not cleave, the hepatotoxicity story that dominates gapmer antisense does not transfer directly; the off-target consequence here is splicing perturbation, which still begins at complementarity — where the instrument fits — but whose downstream biological harm is an assay question, where it does not.

(a) What we can honestly say about this drug's safety

One bounded thing, and it is worth saying precisely because it is bounded. The exact method can hand a reviewer a complete, re-derivable map of the payload's off-target complementarity — better and more reproducibly than the float ranker that is standard today — the moment the sequence exists to run it on. It cannot certify that del-zota is safe: it is silent on immunogenicity, the antibody and its conjugation, hematology, distribution and dosing. Underselling that would be as wrong as overselling it. The instrument covers one slice of safety, exactly, and only that slice.

(b) Whether better options exist

For the exon-44 subgroup there is no approved skipper today, so the honest comparators are these. Brogidirsen (NS-089/NCNP-02, Nippon Shinyaku) is the direct rival — a dual-targeting antisense oligo for the same population, in global Phase 2, with investigator-initiated long-term data (3.5-year at WMS 2025, 4.5-year at MDA 2026) reporting exon-44 skipping, dystrophin expression, maintained ambulatory function, and no serious or severe adverse events and no discontinuations over 3.5 years. It is the peer to watch; cross-trial efficacy comparisons are not valid without a head-to-head. The approved exon-skippers for other exons — eteplirsen, golodirsen, viltolarsen, casimersen — do not treat exon-44 patients, and their dystrophin restoration was low (~1–2% of normal) with clinical benefit still debated; they set a bar del-zota's ~32% appears to clear by a wide margin, with "appears" the honest word until SAFARI44. Elevidys, the AAV micro-dystrophin gene therapy, reaches some exon-44 patients and carries a real current signal: in November 2025 the FDA added a Boxed Warning for acute serious liver injury and restricted it to ambulatory patients after two deaths from acute liver failure — a fair comparison notes both its single-dose appeal and that hepatic risk. Beneath all of these, corticosteroids and cardiac-respiratory care remain the backbone any new drug is added to, not a rival it replaces.

The value

A re-derivable off-target-complementarity verdict costs one deterministic enumeration and can be produced before a first patient is dosed. What it is licensed against is a late-stage failure or a post-marketing safety signal — years, R&D, and, for a progressive childhood disease, the treatment window itself. For del-zota specifically, the instrument does not decide the drug; it hands a reviewer one bounded certificate they can reproduce exactly, and it is ready the moment PMO44 is on the table.

From day one — what the substrate would have found

Picture the exact instrument sitting beside the designers on the first day AOC 1044 was drawn — before an animal, before a patient. This is a counterfactual, and only that (ARGUMENT — nothing here was computed on the real PMO44 sequence, which is still dark; the number to run it on does not exist in public). But the shape of what it would have produced is concrete, and it is worth seeing.

At the design stage, the moment a candidate 20–25-mer was written down, the instrument would have returned a complete, re-derivable off-target complementarity map against the reference transcriptome: every site the payload can bind within a stated mismatch/bulge budget, enumerated exhaustively, with one right answer and nothing to round. Not a ranked shortlist that reorders when a parameter set is swapped — the full set, reproducible byte-for-byte by a reviewer years later. On the actual path, that seam was carried instead by a float MFE ranker the field itself distrusts (Scharner 2020) and by clinical readouts that arrive years downstream. The design-stage map is the same information, earlier, exact, and auditable — a candidate could be compared against its near-neighbors before a dollar of wet-lab was spent, and the site list handed forward to the assays that decide which of those sites actually mis-splice.

And the scale of this is not speculative. The demo already shows the exact-versus-float divergence on synthetic sequences (marker ASO_OFFTARGET_EXACT_IS_OBSERVER_INVARIANT, MEASURED on synthetic sequences — it screened no real drug). The library scale is already demonstrated too: Affine.Earth hosts and re-derivably grades molecular libraries at the 10⁵–10⁶ scale — 258,616 PDB structures and 125,302 protein/material aggregates on the live court (Study 14). An exact off-target enumeration against a reference transcriptome or genome is the same class of bounded, deterministic, re-derivable computation, at a scale the substrate has already shown it can carry. Ingesting a pinned reference transcriptome is a defined engineering step with a fixed answer, not a research problem. The one missing input is the sequence — and the day it is put on the table, the map is a single deterministic run away.

---

What we claim, and what we do not

We claim: Duchenne robs the exon-44 subgroup of dystrophin, and del-zota (AOC 1044) is a genuine first for them — the sponsor reports dystrophin restored to roughly a third of normal, an order of magnitude past the approved skippers for other exons. One specific safety sub-question — where else in the transcriptome the payload can bind within a stated mismatch/bulge budget — is a discrete, exact, re-derivable enumeration, categorically unlike an observer-dependent floating-point ranker, and a real risk for this class (Scharner 2020: mis-splicing at partial matches, effects at ≥12/18 nt). That exact-versus-float advantage is measured out of domain on the fusion court (Studies 34 and 35) and shown for antisense off-target on synthetic sequences locally; the substrate's contribution would be an auditable, design-stage SEAL.

We do not claim: that any run was performed on del-zota, or that any binding energy, off-target site, or safety verdict was computed for it or for the PMO44 sequence — that sequence is not public and the molecular court is dark. We do not claim the exact method answers whether del-zota is safe: it covers off-target complementarity only, it does not decide which sites actually mis-splice, and it is silent on the immunogenicity, antibody/conjugation, hematologic, distribution and dosing risks that dominate this drug's long-haul profile. We do not claim its open-label functional gains are randomized benefit — SAFARI44 is pending — nor that a better therapy exists; none is approved for exon 44. The honest edge word on anything unbound is: not known.

Sources

• Sept 2025 functional readout: Avidity / PR Newswire. July 2024 biomarker readout (dystrophin +25%, CK to near-normal): Avidity / PR Newswire.
• Mechanism / delivery (TfR1 AOC, steric-block PMO): Nucleic Acids Research 2025;53(6):gkaf241 — academic.oup.com · PubMed 40183632.
• Regulatory status (Breakthrough Therapy; pre-BLA / Q1 2026 submission): PPMD · PR Newswire pre-BLA. Trials: SAFARI44 NCT07587242 · managed access NCT07250737. Platform hold history (AOC 1001): NeurologyLive.
• Splice-switching off-target as discrete search vs float heuristic: Scharner et al., NAR 2020;48(2):802 — academic.oup.com.
• Comparators: brogidirsen PR Newswire · Cell Reports Medicine / MTNA; approved skippers & dystrophin controversy T&F review 2026 · Muscular Dystrophy News; Elevidys Boxed Warning Nov 2025 FDA · PPMD.
• Disease / natural history: Rare Disease Advisor · PMC6592545 · exon-44 milder LOA delay Neurology.
• Library scale (design-stage counterfactual): Affine.Earth Study 14 — Protein Lattice Manifold (258,616 PDB structures; 125,302 protein/material aggregates on the live court).
• Exact-vs-float principle only (out-of-domain): Affine.Earth Study 34 (observer axis), Study 35 (time axis); local reproduce/aso-offtarget-exact-vs-float.swift (marker ASO_OFFTARGET_EXACT_IS_OBSERVER_INVARIANT) and its CRISPR-guide sibling reproduce/crispr-guide-offtarget-exact-vs-float.swift (marker CRISPR_OFFTARGET_EXACT_IS_OBSERVER_INVARIANT) — synthetic sequences, no del-zota run.

Grades: VERIFIED / REPORTED / MEASURED / ARGUMENT / NOT_KNOWN. A reading that interprets a reported row is ARGUMENT, never a measurement. This study's molecular court is a charter — the PMO44 sequence is dark — and it says so; it extends, never contradicts, the fusion anchor (Studies 34 and 35).