Personalised Gene Editing: The Rise of Bespoke CRISPR

The question comes up at nearly every seminar I give, usually from someone in their second or third year, usually phrased some version of this way: if we can build a gene editor for one baby in six months, why can't we do it for everyone?

The standard answer is that the science is solved and the bottlenecks are money and regulation. I've given that answer myself. It's incomplete, and I think it's quietly misleading a cohort of students who are making career decisions based on it.

The science is not solved. What's solved is one narrow, extremely favourable case: correcting a point mutation in hepatocytes, in a patient whose disease is so rapidly lethal that a modest benefit is clearly worth an unquantified risk. Almost everything that makes that case tractable — liver tropism of lipid nanoparticles, a biochemical readout within days, a regulatory willingness to accept a single patient — falls away the moment you move to a different tissue or a slower disease.

So let me lay out what I think the evidence actually supports, in rough order of how much weight it can carry.

What "bespoke" means here, and what it doesn't

There's a distinction I keep having to redraw for people, including for an editor who once sent back a draft of mine with "CRISPR therapy approved for sickle cell — so this is already personalised?" in the margin.

No. The approved ex vivo CRISPR therapy for sickle cell disease and transfusion-dependent beta thalassemia uses one guide RNA, the same for every patient, disrupting an erythroid enhancer of BCL11A to raise fetal hemoglobin. It's personalised only in the sense that your own cells go out and come back. The construct is a product.

Bespoke editing means the therapeutic molecule itself is designed around a specific patient's variant. The precedent isn't CRISPR at all — it's the antisense oligonucleotide designed for a single child with a form of Batten disease, published in 2019, which established the template of a patient-specific drug developed and dosed under an individual investigational application.

The relevant distinction for anyone planning work in this area isn't gene editing versus not. It's platform versus product: how much of the safety, manufacturing and regulatory package can be reused when you swap the guide or the pegRNA.

The mechanisms, and why they suit n-of-1 work

Base editing

Cytosine and adenine base editors fuse a catalytically impaired Cas9 to a deaminase, converting C•G to T•A or A•T to G•C without a double-strand break and without a donor template. For a therapeutic programme aimed at one patient, that matters enormously: the editor protein is fixed, the mRNA is fixed, the lipid nanoparticle is fixed. What changes is a chemically synthesised guide RNA.

That's the engineering insight behind the whole bespoke model. You are not building a new drug. You are building a new targeting module for a drug you've already characterised.

Prime editing

Prime editing uses a nickase fused to a reverse transcriptase, with a pegRNA that both specifies the target and templates the edit. It covers insertions, deletions and all transition and transversion substitutions, which in principle addresses the large fraction of pathogenic variants base editors can't reach. The cost is a larger, more complex cargo and editing efficiencies that are more context-dependent and harder to predict from sequence alone.

If you want to see the current state of the primary literature rather than take my word for it, the PubMed record on in vivo prime editing moves fast enough that a review older than about eighteen months is a historical document.

Delivery is still the discipline

I've watched three separate groups, all strong, lose a year because they treated delivery as a downstream engineering detail. Ionisable lipid nanoparticles go to hepatocytes. That's most of what we can do systemically with transient editor expression right now. AAV reaches more tissues but gives you persistent expression of a nuclease or deaminase, which is exactly what you don't want. Ex vivo editing of hematopoietic stem cells sidesteps delivery entirely and inherits the toxicity of conditioning instead.

Pick your target tissue first. The editing chemistry is the easy part.

The strongest evidence: platform therapies in controlled or large trials

Here's where I'd push back on the field's own enthusiasm. There are, to my knowledge, no meta-analyses of therapeutic genome editing worth the name, because there aren't enough comparable randomised trials to pool. What we have:

• Approved ex vivo CRISPR disruption for sickle cell disease and beta thalassemia, licensed on the basis of single-arm pivotal studies with clinically meaningful endpoints — freedom from vaso-occlusive crises, transfusion independence — and now accumulating multi-year follow-up. Durable, as far as anyone can tell, which is what you'd expect from editing long-term repopulating stem cells.

• In vivo CRISPR knockout of hepatic TTR for transthyretin amyloidosis, delivered by LNP, which produced deep and sustained serum TTR reduction in early-phase work and has moved into randomised phase 3 testing against clinical endpoints. This is the single most informative programme in the field, because it will tell us whether durable in vivo knockout translates into outcomes under proper control.

• In vivo base editing of PCSK9 to lower LDL cholesterol, with dose-dependent reductions reported in early-phase studies. Mechanistically compelling; a long way from outcome data.

Everything above is platform, not bespoke. Each uses one fixed guide across all recipients. They matter for the bespoke question because they supply the safety and pharmacology dossier that individual programmes borrow from. You can track the registered studies through ClinicalTrials.gov and the published trial reports via a PubMed search on in vivo CRISPR clinical trials.

The n-of-1 evidence: real, important, and very thin

The case that changed the conversation was an infant with severe neonatal-onset carbamoyl phosphate synthetase 1 deficiency, treated in 2025 with a customised LNP-delivered base editor targeting his specific variant, designed, manufactured and dosed in roughly six months. The clinical course after dosing was encouraging. The primary report and the commentary around it are worth reading in full, including the supplementary material on off-target assessment, which is more candid than most.

What that case demonstrates is feasibility of timeline and process. What it cannot demonstrate, and what nobody involved claimed it demonstrated, is efficacy. One patient, no control, a disease with variable natural history, concurrent standard care, and an outcome measured partly in surrogate biochemistry. If you presented that design for any other intervention class you'd be shown the door.

I say this not to diminish it. I say it because I've now read three grant applications that cited that case as evidence that bespoke editing *works*, and reviewers noticed.

Alongside it sit a handful of first-in-human small cohorts: base editing to induce fetal hemoglobin, in vivo base correction of the Z allele in alpha-1 antitrypsin deficiency — notable as a genuine correction rather than a knockout — and the first prime editing programmes entering patients. Early, small, open-label. Treat them as pharmacology, not as outcomes.

Preclinical and animal work: where the honest uncertainty lives

The preclinical literature is where you find the problems the clinical reports are underpowered to detect. Deaminase-dependent guide-independent off-target editing of DNA and RNA. Bystander edits within the editing window. Large deletions and chromosomal rearrangements at nuclease-induced breaks. p53-mediated selection against edited cells. Immunogenicity against Cas proteins and against LNP components, which bears directly on whether you can ever redose.

Mosaic editing is the one I'd flag for anyone designing a programme. In a dividing or regenerating tissue, partial editing can be stable; in the neonatal liver it may not be. We have very little long-term human data on editing persistence after hepatocyte turnover.

Cost, and the accounting error I keep seeing

Approved cell and gene therapies carry list prices in the low millions of dollars per patient. People then multiply that by the number of patients with an ultra-rare variant, get an absurd number, and conclude bespoke editing is economically impossible.

That reasoning imports the wrong cost structure. The multi-million dollar price of a commercial product amortises a full development programme across a small population. A bespoke editor built on an established platform doesn't repeat that programme. The marginal costs are guide design and screening, GMP synthesis of one RNA, a tailored off-target panel, a bespoke toxicology package, regulatory submission and clinical delivery. Reported academic estimates have put individual cases in the range of a complex organ transplant — expensive, but inside the envelope health systems already fund.

The real economic barrier isn't manufacturing. It's that nobody owns the cost. There's no sponsor, no market, no reimbursement pathway, and no obvious mechanism for a hospital to recover what it spends. That's a policy problem wearing a technical disguise.

The regulatory debate I think we're getting wrong

Regulators have moved faster than the field gives them credit for. There are now dedicated genome editing guidances, platform technology designations intended to let a sponsor reuse a characterised component across products, and explicit interest in individualised therapies for ultra-rare disease. The FDA's posture has been more accommodating than most people assumed it would be five years ago.

The hard question isn't permission. It's evidence. If every bespoke therapy is an uncontrolled single case, we will accumulate hundreds of anecdotes and never learn whether the platform is safe at population scale. Positive cases get published; silent failures mostly don't.

My view: the field needs mandatory, prospective registration and long-term follow-up of every individualised editing case, successful or not, with pre-specified off-target reporting standards. Voluntary consortia won't get there. Neither will journals. This should be a condition of the regulatory authorisation itself.

What I'd watch over the next few years

1. The randomised in vivo knockout read-outs. Whether durable hepatic gene knockout changes hard clinical outcomes is the question the entire in vivo field rests on.

2. Extrahepatic delivery. Targeted LNPs, engineered capsids, delivery to hematopoietic stem cells in situ. This is the real rate limiter, and I'd steer a new postdoc here over editor engineering.

3. Prime editing efficiency in patients. Preclinical efficiencies have improved sharply; human data will tell us how much of that survives contact with a real tissue.

4. Redosing. Anti-LNP and anti-Cas immunity determines whether partial editing can ever be topped up. Almost nothing published addresses this in humans.

5. Who pays. Watch for the first national health system that builds a standing reimbursement route for individualised therapies rather than funding them case by case.

One practical note for students entering this area. The literature turns over fast enough that the usual workflow — search, download, lose the PDFs, rewrite the same background section every six months — actively costs you. I keep a living evidence map for this topic and update it as trials report, which is easier in a workspace where searching, reading, annotating and drafting all sit together. Exploring the literature in PACR keeps the context from a search attached to what you write from it, so last year's reading is still usable when the next trial reads out. If you're starting a review in this space, setting up a PACR workspace before you start collecting papers is worth the twenty minutes.

If you take one thing from this: the bespoke editing story is a genuine advance in process, not yet a demonstrated advance in outcomes. Those are different claims, and the gap between them is where the next decade of work actually sits. Design your experiments for the second one.