Gene Editing Weekly — September 19–25, 2026: an AI find re-rates the sector, a prime-editing IND, and an epigenetic silencer for hepatitis B

The coverage window for this issue is September 19–25, 2026, Eastern Time. The seven days hold one sector-wide re-rating, one IND clearance, one preclinical result that opens a new modality, and no new patient data on a gene-editing therapy. The re-rating is the item most likely to be misread, so this issue starts there.

Track	Primary change	Evidence boundary	Next milestone
Market	Gene-editing shares fell on September 23 after Anthropic said its Claude model had flagged a CRISPR-like enzyme system in bacterial-virus DNA. 1	One day of price action around a finding whose function Anthropic describes as unknown, released as a preprint. 2	Whether the system is characterised, and whether the market's read survives it.
Regulatory	The FDA cleared Prime Medicine's IND for PM647, an in vivo prime editor for alpha-1 antitrypsin deficiency, on September 24. 3	An IND clearance. The company reported no patient data, and guides first clinical data to 2027. 3	First-in-human dosing, and the trial's lung-first, then liver, cohort sequence.
Clinical	Allotera opened a minimal-residual-disease cohort inside the pivotal T-RRex study of its CRISPR-edited CAR-T, sofi-cel. 4	A trial-design event. The company reported no efficacy data from the new cohort.	Whether sofi-cel converts MRD-positive remission into MRD-negative remission.
Research	CRMA-1001, a CRISPR-based epigenetic silencer, cut hepatitis B biomarkers by more than 3 log₁₀ in mouse models and left up to 90% of animals free of detectable surface antigen and viral DNA at six months. 5	Preclinical. The non-human primate arm covers tolerability, not efficacy. 5	An IND and a clinical dosing regimen.
Company	Metagenomi put its hemophilia A program, MGX-001, on an IND submission for the fourth quarter of 2026, with $120.7 million in cash and a runway it projects through the fourth quarter of 2027. 6	A corporate presentation furnished to the SEC, carrying preclinical durability of about 19 months in non-human primates. 6	The IND filing, and first patient dosing targeted for 2027.
Company	ERS Genomics licensed its foundational CRISPR-Cas9 patent estate non-exclusively to the contract research organisation Sai Life Sciences. 7	A licensing agreement. It carries no clinical or product claim.	Whether more discovery-service providers take the same licence.
Governance	Nature Medicine published a commentary titled "Embryo editing and embryo selection need joint governance." 8	A journal commentary by Hervé Chneiweiss, François Hirsch, Catherine Bourgain and colleagues, and a statement of position rather than a decision.	Whether a regulator or professional body takes the argument up.
Cost	Endpoints News reported that manufacturing the custom CRISPR medicine built for one patient cost $1.65 million, roughly double earlier estimates. 9	A reported figure attributed to Fyodor Urnov, with no journal publication behind it yet.	Whether the manufacturing blueprint now in development lowers the per-patient figure.
FDA ledger	CBER's "What's New for Biologics" carried two entries dated inside the window, one an administrative procedure and one an approval letter for a product outside genome editing. 10	A weekly posting ledger, which records what the agency posted and not everything it did. 10	A vaccines advisory committee on October 1, and the Expedited IND pilot's October 30 deadline.

Market

The sector re-rated on an AI find

On September 23, Anthropic said its Claude model had identified an enzyme system in bacterial-virus DNA whose architecture recalls CRISPR's: a reverse transcriptase, a neighbouring accessory protein, and a long array of evenly spaced DNA repeats. 1 Anthropic named the three-part system array-associated reverse transcriptases, or ART. The underlying reverse transcriptase already sat in the literature. Anthropic says Claude was the first to notice the repeat array and the accessory protein beside it, and the company released a preprint rather than a peer-reviewed paper. 2

Anthropic described how the search ran: about 950 agents working for 21 hours and spending 210 million tokens against a large DNA sequence database, launched from a single prompt asking for unfamiliar reverse transcriptases. 1 Those agents gathered more than 200,000 reverse transcriptases and narrowed them to 20 candidates for human review. 1 Anthropic states that ART's function is unknown, and that its first experiments show the repeat array is transcribed into a set of short RNAs, which is how a CRISPR array stores its targeting sequences. 1 Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, called the RNA-repeat finding "genuinely intriguing" and said it "merits further investigation." 1

The same day, gene-editing shares fell. CRISPR Therapeutics closed about 6% below Tuesday's $59.03, Intellia fell 3%, Beam 6%, Editas 8%. Prime Medicine fell about 12%, one session after a double-digit gain. 11

Loading chart…

The move repriced the sector's future tool chest. For ART to matter clinically, someone would have to work out what the system does, show it can be programmed to make a chosen change, find a way to deliver it into human tissue, and carry it through development. Anthropic's preprint stops after the first step. 1 Every editing medicine now in trials runs on a nuclease, a base editor or a prime editor, and each program's timeline is set by its own protocol and its own regulators. The record supports one reading of the selloff: traders marked down what the field's tools might be worth in a decade, while the trials below kept their schedules.

The question to carry forward is what ART turns out to be. Its function is the open variable, and the second AI-discovered system will show whether this is a repeatable way to find tools or a single result.

Clinical and regulatory

A second in vivo prime editor reaches the clinic

Prime Medicine said on September 24 that the FDA had cleared its investigational new drug application for PM647, a one-time in vivo prime editor for alpha-1 antitrypsin deficiency. 3 PM647 corrects the E342K, or Pi*Z, mutation in the SERPINA1 gene, and it uses the same liver-directed lipid nanoparticle as PM577a, the company's Wilson's disease program. 3 The Phase 1/2 study is global, single-arm, open-label and first-in-human. It starts with adults who have lung-only disease and, once tolerability is established there, adds a separate cohort of adults with significant liver disease. 3 The company guides initial clinical data to 2027. 3

PM647 arrives in a target that already carries several in vivo editing programs, and the three reports below all fall before this issue's window. Beam Therapeutics, whose BEAM-302 began dosing a pivotal cohort in July, reported on September 8 that a single 60 mg dose held total AAT above the protective threshold in 29 patients and reduced mutant Z-AAT by 84%. 12 YolTech reported on September 7 that a single 45 mg dose lifted mean AAT above the protective threshold in four patients, with up to 57% editing at the target site on liver biopsy and a 55 mg cohort still pending. 13 CRISPR Therapeutics lists CTX460, a SERPINA1 program built on its SyNTase editor, among its disclosed development candidates, with data guided to early 2027. 14

Three of those programs correct the same Pi*Z mutation, and a fourth sits behind them in preclinical development. Correction of the same base in the same gene means the programs will separate on dose, durability and immune profile rather than on target choice. Prime Medicine notes that roughly 100,000 people in the United States carry the PiZZ genotype, and about 200,000 across the United States and Europe. 3

Allotera opens an MRD cohort in a pivotal CAR-T study

Allotera Therapeutics said on September 23 that it had opened a minimal-residual-disease cohort in the global pivotal T-RRex study of sofi-cel, its off-the-shelf CAR-T for relapsed or refractory T-cell acute lymphoblastic leukaemia and T-cell lymphoblastic lymphoma. 4 The new cohort enrols patients who are in remission after standard therapy and remain MRD-positive, a state the company's chief medical officer, Cherry Thomas, called one of the strongest predictors of relapse. 4

Sofi-cel is made from healthy-donor T cells that CRISPR-Cas9 edits to delete CD7 and the T-cell receptor alpha constant gene, TRAC. Deleting CD7 keeps the CAR-T cells from attacking each other, and deleting TRAC addresses graft-versus-host disease. The study runs at 18 sites in the United States and Australia and is registered as NCT06514794. 4 The FDA has granted sofi-cel breakthrough therapy, regenerative medicine advanced therapy, fast track, orphan drug and rare paediatric disease designations, and the European Union has granted it PRIME designation. 4

The cohort tests a proposition that is easy to state and hard to demonstrate: an allogeneic product that clears residual disease before relapse can shift when patients are treated, and therefore how many reach a durable response. Enrolment and conversion rates are the numbers to follow.

The week's CBER ledger

The FDA's biologics ledger carried two entries dated inside the window: a standard operating procedure on regulatory meetings with sponsors, posted September 21, and a September 23 approval letter for ADZYNMA. 10 Both sit outside genome editing, so the ledger records no editing approval, complete response letter, advisory action, IND clearance or BLA acceptance for the week. 10

Two dates fall next. The vaccines advisory committee meets on October 1 to recommend strains for the 2027 Southern Hemisphere influenza season, 15 and applications for the Expedited IND Pilot Program close at 11:59 p.m. Eastern Time on October 30. 16

Research

An epigenetic silencer quiets hepatitis B

The week's research result that opens a new modality path came from Chroma Medicine's CRMA-1001, described in Nature Biomedical Engineering on September 21. 5 CRMA-1001 is an epigenetic silencer: it represses viral transcription by depositing DNA methylation on hepatitis B DNA, both the episomal form and copies integrated into the host genome. A single dose produced reductions of more than 3 log₁₀ in viral biomarkers in mouse models, and those reductions tracked with new methylation of viral DNA. Three doses left up to 90% of animals with no detectable hepatitis B surface antigen or viral DNA six months after treatment. 5

In non-human primates, liver transaminases rose only transiently at the highest dose tested. Profiling of expression and DNA methylation turned up no unintended targets in the human genome. 5 Nature's news desk covered the paper the same day under the headline "'Epigenetic' editing is here," writing that the technique banishes the virus. 17

The result changes the question a chronic hepatitis B programme has to answer. Approved therapies rarely produce a functional cure, since they leave transcriptionally active viral DNA in place. A silencer aims to switch that DNA off, which keeps the viral genome intact and makes the durability of the methylation, rather than the accuracy of a cut, the central risk. 5

The evidence stops at mice and non-human primates, and the authors' own framing is that the data support starting clinical development. 5 How long methylation holds in human liver, and whether it holds at a dose patients can take, are the two questions a first-in-human study would settle.

Company and capital

Metagenomi's hemophilia A timetable and its arithmetic

Metagenomi furnished a corporate presentation to the SEC for an event dated September 21, setting out the path for MGX-001, its in vivo CRISPR program for hemophilia A. 6 MGX-001 integrates a Factor VIII gene at a chosen site under an albumin promoter. The presentation reports stable Factor VIII expression for about 19 months in a non-human primate study and no off-target edits in the human hepatocyte studies it ran. 6 The company completed a pre-IND interaction in December 2025, plans to file the IND in the fourth quarter of 2026, and targets first patient dosing in 2027. 6

The presentation also carries the commercial case: $120.7 million in cash and marketable securities at the end of June 2026, a runway projected through the fourth quarter of 2027, roughly 25,000 haemophilia A patients in the United States and about 500,000 worldwide, and current Factor VIII replacement costing an estimated $565,000 to $750,000 per patient per year, or $18 million to $24 million over a lifetime. 6 Those figures set the price a one-time treatment can defend. A presentation furnished under Regulation FD sets out the company's own plans, which makes the runway and the IND date checkpoints rather than commitments.

A CRISPR patent estate reaches a service provider

ERS Genomics signed a non-exclusive licence with Sai Life Sciences on September 22, giving the Indian contract research, development and manufacturing organisation access to the foundational CRISPR-Cas9 patent estate developed by Emmanuelle Charpentier and her collaborators. 7 That portfolio covers more than 130 issued patents, over 50 of them in the United States. 7 Sai plans to use it for target validation, compound screening and engineered cell models offered to its pharmaceutical clients. 7

The licensing pattern is the point. CRISPR-Cas9 is now routine in discovery workflows, and the organisations that need freedom to operate include the service providers running those workflows for everyone else. ERS already counts GSK, Merck, Bayer, Thermo Fisher Scientific, Danaher, Charles River Laboratories and Cargill among its licensees. 7

Governance and cost

Embryo editing and embryo selection inside one frame

Nature Medicine published a commentary on September 23 titled "Embryo editing and embryo selection need joint governance," by Hervé Chneiweiss, François Hirsch and Catherine Bourgain with colleagues. 8 The title carries the argument: two practices that regulators and professional bodies treat through separate routes, changing an embryo's genome and choosing among embryos, belong to one governance question. Editors work at the second already, since preimplantation genetic testing and polygenic risk scoring both narrow which embryos are transferred, and both run ahead of any framework written for the first.

The commentary's practical weight depends on who takes it up. A joint framework would need a regulator or an international professional body to adopt it, and neither has announced one. 8

The cost of a medicine built for one person

Endpoints News reported on September 23 that manufacturing the custom CRISPR medicine built for a single patient, publicly known as Baby KJ, cost $1.65 million, roughly double earlier estimates, according to Fyodor Urnov. 9 The Innovative Genomics Institute is working with Danaher on a blueprint intended to make the process repeatable. 9

The figure matters because individualised editing has cleared its scientific hurdle and now meets an economic one. A treatment designed for one patient has no population across which to spread development cost, so the cost of making it, rather than the cost of discovering it, sets how many patients the approach can reach. The reported number and the blueprint's progress are the two data points to follow.

What to watch

ART's function. Anthropic's own description leaves the system uncharacterised. 1 A follow-up paper that names a reaction would settle whether the preprint opened a tool line or a curiosity.
PM647's first patient. Prime Medicine guides first clinical data to 2027, and the liver-disease cohort opens only after tolerability is established in lung-only patients. 3
The alpha-1 antitrypsin queue. Beam's pivotal cohort is dosing, YolTech's 55 mg cohort is pending, and CTX460 data are guided to early 2027. 12 13 14 Four programs aimed at one mutation will diverge on dose, durability and safety, and the first useful comparison will come from those readouts.
MGX-001's IND. Metagenomi's filing window closes with the year, and its cash is projected to last a year beyond it. 6
Regulatory dates. The vaccines advisory committee meets on October 1, and Expedited IND Pilot applications close on October 30. 15 16
Whether the embryo-governance argument finds an owner. Adoption would have to come from a regulator or an international professional body. 8