A Gene That Reads Correctly Either Way Round The Promoterless Bidirectional Insertion Template

This application covers a piece of DNA designed to work no matter which way round it lands in the genome. Claim 1 is a composition of matter claim on the construct itself: two segments, one carrying a coding sequence and the other carrying the reverse complement of a coding sequence, and no promoter anywhere in the construct to drive expression. It was filed jointly by Intellia Therapeutics, Inc. and Regeneron Pharmaceuticals, Inc., names Jonathan Douglas Finn and Hon-Ren Huang as inventors, and is a divisional of Australian application 2019360269. The specification claims priority from two United States provisional applications, 62/747,393 filed 18 October 2018 and 62/840,343 filed 29 April 2019.

The Problem

Putting a working copy of a gene into a patient’s liver with CRISPR sounds straightforward. Cut the chromosome at a chosen site, supply a DNA template, and let the cell’s repair machinery paste the template into the break. The catch is in the last step. Hepatocytes in an adult liver are mostly not dividing, and the repair pathway available to a non-dividing cell is non-homologous end joining, which ligates whatever blunt DNA end it finds to whatever blunt DNA end it finds next. It has no sense of direction.

That means a conventional one-way cassette, with the gene reading left to right, gets pasted in backwards roughly half the time. A backwards insertion is silent. It consumes an editing event, occupies the target site and contributes nothing. The usual answer is to add homology arms so that homology-directed repair controls the orientation, but homology-directed repair barely operates in a non-dividing cell, and the arms consume packaging capacity in the adeno-associated virus used to carry the template.

The specification also documents a trap for anyone screening these designs. In an immortalised mouse liver cell line, the vector with 200 base pair homology arms was the only one that produced detectable human Factor IX. The vectors without homology arms produced none. In live mice, the ranking reversed. Across the later screening examples the same disconnect recurs: constructs achieving high levels of editing, measured as insertion or deletion formation at the cut site, frequently did not produce more insertion or more protein, and the document states plainly that significant levels of editing did not necessarily result in more efficient insertion or expression.

What This Invention Does

Claim 1 is a bidirectional nucleic acid construct comprising a first segment with a coding sequence for an agent such as a polypeptide, and a second segment with a reverse complement of a coding sequence of the agent, with the closing requirement that the construct does not comprise a promoter that drives the expression of the agent. Claim 2 is the same structure where the two segments encode two different agents. Both are product claims on the DNA, not on any method of using it.

The two limitations do different jobs. The mirrored pair means the construct is readable from either strand, so whichever way non-homologous end joining ligates it into the break, one of the two segments is now pointing the right way and is transcribed. The absence of a promoter means the construct does not carry its own on switch and must borrow one from the insertion site. In the worked examples that site is intron 1 of the albumin gene, which the specification treats as a safe harbour because the albumin promoter is strong and liver specific, the albumin signal peptide handles secretion, and a modest reduction in circulating albumin is tolerated. Claims 13 and 14 add splice acceptor sites, including a first acceptor 5-prime of the first segment and a second acceptor 3-prime of the second segment, so that whichever segment ends up in the sense orientation can be spliced onto the albumin exon 1.

Claim 9 makes explicit what the examples demonstrate: the construct does not comprise a homology arm. Claims 4 to 6 cover a subtler design point. Because two copies of the same coding sequence pointing in opposite directions on a single-stranded template would fold back on themselves, the second segment is allowed to use different codons for the same amino acids, and the specification says this is done in order to reduce hairpin formation. Claim 5 sets the tolerated complementarity anywhere from about 30 per cent up to about 99 per cent, so the two halves can be as similar or as divergent as the folding problem requires.

The mouse data is the clearest demonstration. Mice received adeno-associated virus at 3e11 vector genomes plus a lipid nanoparticle carrying Cas9 messenger RNA and a guide RNA against intron 1 of mouse albumin, at 4 mg/kg of total RNA cargo. Editing was essentially identical across the three template designs, at 66.72, 68.10 and 70.16 per cent indels. Circulating human Factor IX was not. The self-complementary blunt template gave 0.75 ug/mL, the single-stranded template with homology arms gave 0.96 ug/mL, and the single-stranded bidirectional template gave 2.92 ug/mL. Same cut, same editing rate, three to four times the protein.

Durability held up over a one-year study. At 1 mg/kg of lipid nanoparticle, human Factor IX ran from 4.02 ug/mL at week 2 to 4.72 ug/mL at week 52, and at 0.25 mg/kg from 0.87 to 0.82 ug/mL across the same period. Serum albumin was measured at weeks 2 and 41 and was maintained. In cynomolgus monkeys the first reported study produced plasma Factor IX of about 135, 140, 150 and 110 ng/mL at days 7, 14, 28 and 56, spanning about 75 to 250 ng/mL. Because the variant used was the hyperfunctional R338L Factor IX with roughly eight-fold specific activity, the specification calculates this as about 20 to 40 per cent of wild type Factor IX activity, with a range spanning 12 to 67 per cent. A later example using a different guide reported more than three-fold higher protein again, about 3 to 4 ug/mL at day 14 and about 3 to 5 ug/mL sustained through day 28, which the document places at or above the cited normal human range of 3 to 5 ug/mL.

Key Features

• Two coding segments pointing opposite ways. The construct carries a coding sequence in one segment and the reverse complement of a coding sequence in the other, so an insertion event that lands the construct backwards still leaves a readable copy facing the promoter.
• No promoter in the construct. Claim 1 requires the absence of a promoter driving expression of the agent, which forces the inserted gene to run off the endogenous promoter at the target site and keeps expression tied to the tissue where that promoter is active.
• No homology arms. Claim 9 covers the homology-arm-free construct, and the in vivo data supports it: templates without arms outperformed templates with 200 to 2000 base pair arms in non-dividing hepatocytes, while freeing up packaging space in the viral vector.
• Codon divergence between the two halves. Dependent claims allow the second segment to encode the same polypeptide using alternative codons, described in the body as a way of reducing hairpin formation in what would otherwise be a long self-complementary single strand.
• Splice acceptors on both ends. The construct can carry a splice acceptor 5-prime of the first segment and another 3-prime of the second segment, so the transcript from either orientation can be joined to the host gene’s first exon and use the host signal peptide.
• Claimed across delivery formats. Beyond the construct itself, the claim set reaches vectors including adeno-associated virus serotypes, lipid nanoparticles containing the construct, host cells, methods of modifying a locus using an RNA-guided DNA-binding agent and a guide RNA, and the cells produced by those methods.

Who Is Behind It

Intellia Therapeutics is a Cambridge, Massachusetts gene editing company built around in vivo CRISPR delivery by lipid nanoparticle, and Regeneron Pharmaceuticals is the Tarrytown, New York biotechnology company best known for its antibody franchise and its genetics institute. The two have a long-running collaboration on in vivo CRISPR targets, and haemophilia A and B are among the programmes they took on jointly. Intellia’s own glossary entry for the Factor IX gene describes F9 as the target for the two companies’ haemophilia B targeted gene insertion programme, which is precisely the worked example running through this specification.

The joint filing is worth noting in its own right. Both companies are named as applicants on the title page, and the two inventors sit on the Intellia side of the work. The Australian agent of record is Pearce IP Pty Ltd.

The family position is stated on the face of the document. The body text opens by claiming the benefit of priority from United States provisional applications 62/747,393 filed 18 October 2018 and 62/840,343 filed 29 April 2019, and every page of the specification carries the international headers for WO 2020/082041 and PCT/US2019/057084. The title page records the application as a divisional of AU 2019360269, which is the Australian national phase of that international application. Priority country is the United States.

Why It Matters

Gene insertion is the harder half of gene editing. Knocking a gene out only requires a cut, and a messy repair is often good enough because the point is to break something. Putting a working gene in requires the cell to accept a piece of supplied DNA at a chosen address, in a chosen orientation, in a cell type that is not dividing and therefore cannot use the precise repair pathway. This construct is an engineering answer to that constraint rather than an attempt to change the biology: if you cannot control the orientation, make orientation stop mattering.

The commercial logic behind the promoterless design is equally practical. A gene delivered with its own strong promoter expresses wherever it lands, including places it should not. Borrowing the albumin promoter means the transgene is expressed at hepatocyte levels, in hepatocytes only, and the albumin locus supplies a secretion signal that would otherwise have to be engineered in. It also ties the therapy to one of the most highly expressed genes in the body, which is why 60 to 70 per cent editing translates into microgram per millilitre protein levels rather than trace amounts.

There is a cautionary lesson in the data that applies well beyond this application. Editing efficiency, the number everyone reports, repeatedly failed to predict insertion or expression across the screens in this specification, and the in vitro cell line ranking inverted when the same constructs were tested in animals. For a field that routinely publishes indel percentages as the headline result, a specification that documents the disconnect this explicitly is a useful thing to have on the public record.

Related Concepts

• Gene therapy – the broader field this construct is a delivery component of, and the framing the specification opens with.
• Haemophilia B – the Factor IX deficiency that the worked examples target through insertion at the albumin locus.
• RNA splicing – the mechanism the splice acceptor claims rely on to join the inserted transgene to the host gene’s first exon.
• Cas9 – the nuclease named in the method claims, including the nickase and S. pyogenes variants.
• Codon usage bias – the redundancy in the genetic code that lets the two mirrored segments encode one protein without being perfectly complementary.
• Indel – the insertion or deletion signature used throughout the examples to measure editing, and the number that turned out not to predict expression.

---

AU 2026202230 was published in the Australian Official Journal of Patents on 9 April 2026 and is open for public inspection. Patent applications represent inventions that are sought to be protected and do not necessarily reflect commercially available products.