A study published in Nature on March 18, 2026 describes a method to generate CAR-T cells directly inside a living patient — without removing blood, engineering cells in a manufacturing facility, and reinfusing them. The approach, from researchers at UCSF and UC Berkeley, addresses one of the most persistent practical barriers to CAR-T cell therapy: the complexity and cost of cell manufacturing.
The paper is “In vivo site-specific engineering to reprogram T cells” by researchers in the Eyquem laboratory at UCSF and the Doudna laboratory at UC Berkeley (Nature, 2026).
The Problem With Current CAR-T Therapy
CAR-T cell therapy works by engineering a patient’s own T cells to express a chimeric antigen receptor (CAR) — a synthetic protein that directs the T cell to recognize and kill cancer cells expressing a specific surface marker. The therapy has been transformative for certain blood cancers: CD19-targeted CAR-T cells achieve complete remission in patients with B cell lymphoma who have exhausted other options.
But the manufacturing process is a serious bottleneck. Currently, producing CAR-T cells requires:
- Leukapheresis to collect T cells from the patient’s blood
- Shipping those cells to a specialized manufacturing facility
- Engineering the cells with a viral vector to insert the CAR gene
- Quality control testing
- Shipping the cells back and reinfusing them
This process takes three to five weeks and costs several hundred thousand dollars per patient. During that waiting period, some patients progress on disease. Patients in poor health may not be able to undergo leukapheresis at all. And the manufacturing infrastructure doesn’t scale easily to broad patient populations.
The promise of in vivo CAR-T generation is to replace this process with something closer to a vaccine injection — delivering the CAR gene directly to T cells inside the body.
What the New Study Did
The UCSF/Berkeley team developed a two-vector system to achieve site-specific integration of the CAR gene into human T cells without removing them from the body.
The approach uses enveloped delivery vehicles and adeno-associated viruses (AAVs) to deliver a large DNA payload to circulating T cells. Critically, the system targets integration to a specific location in the genome: the T cell receptor alpha chain (TRAC) locus. This is the same locus that has been the focus of engineered CAR-T cell manufacturing for years because inserting the CAR at this site does two things at once: it disables the endogenous T cell receptor (reducing the risk of graft-versus-host disease in allogeneic applications) and places CAR expression under the control of the natural TRAC promoter, producing regulated expression rather than constitutive high expression from an exogenous promoter.
The regulated expression matters because constitutive high-level CAR expression is a driver of T cell exhaustion, one of the main reasons CAR-T cell responses can wane over time.
Previous in vivo delivery approaches relied on random integration of the CAR gene using viral vectors. Random integration is less precise, produces inconsistent expression levels, and carries a theoretical risk of insertional mutagenesis — the chance that the gene inserts near an oncogene and inadvertently drives cell transformation. Site-specific integration addresses all three problems.
Results in Animal Models
In humanized mouse models of B cell aplasia, aggressive leukemia, and multiple myeloma, the in vivo-generated TRAC-CAR T cells achieved therapeutic levels of CAR-positive T cells and sustained tumor control comparable to CAR-T cells made using conventional methods.
The targeted integration approach outperformed standard methods using random viral integration, producing T cells with more consistent expression and better functional persistence. In some models, the in vivo-generated cells outperformed ex vivo-manufactured cells.
The researchers also tested the approach against a solid tumor model, where CAR-T therapies have historically performed poorly. Results were more preliminary in that context, though the authors note this as a direction for continued development.
Why This Matters
Two main barriers have prevented in vivo CAR-T generation from working until now: the difficulty of delivering large DNA payloads to specific cell types in a living organism, and the challenge of achieving precise, site-specific integration rather than random insertion.
The two-vector system addresses both. The delivery component — enveloped delivery vehicles combined with AAV — can carry the relatively large CAR gene construct and achieve sufficient transduction of circulating T cells. The site-specific integration component — targeting the TRAC locus — addresses the precision problem.
If this approach translates to humans, the practical implications are substantial. Patients could potentially receive an injection (or infusions) rather than undergoing leukapheresis and waiting weeks for cell manufacturing. Manufacturing costs could drop dramatically. Patients who are currently too ill for conventional CAR-T therapy might become eligible. Geographic access — currently limited by the handful of manufacturing centers worldwide — could improve.
Three of the study’s authors have co-founded Azalea Therapeutics to advance this technology toward clinical development.
Limitations and What Comes Next
The current results are in mouse models, including humanized mice — a useful but imperfect proxy for human immune biology. Several questions remain before clinical translation:
Delivery efficiency in humans. Achieving sufficient transduction of circulating T cells in a patient with a fully intact immune system, with competing cell types and natural barriers, may be harder than in a mouse model. The delivery vehicles will need to demonstrate adequate in vivo transduction rates in human trials.
Durability of the response. The mouse experiments showed sustained tumor control, but the follow-up periods are limited. Long-term durability of in vivo-generated CAR-T cells in patients — particularly compared to conventionally manufactured cells with decades of clinical follow-up — is unknown.
Safety of in vivo gene delivery. Systemic delivery of gene-editing machinery to T cells in a living patient introduces immunogenicity concerns (the body may mount immune responses against the delivery vehicles or the CAR protein itself), off-target effects in non-T cell populations, and the broader safety profile that will need to be characterized in Phase 1 clinical trials.
Solid tumors. The approach worked best in hematologic malignancies (blood cancers). Solid tumors, which have suppressive tumor microenvironments that disable T cell function, remain the harder problem — and the problem that affects far more patients.
What to Watch For
A Phase 1 clinical trial for in vivo CAR-T generation is the obvious near-term milestone. Azalea Therapeutics was formed specifically to move this forward. Based on typical timelines in gene therapy, a first-in-human study could plausibly initiate within two to three years if preclinical development proceeds without major setbacks.
There are also other groups working on in vivo CAR-T generation using different delivery modalities, particularly lipid nanoparticle (LNP) systems — the same delivery technology that proved essential for mRNA COVID vaccines. A companion editorial in Nature Medicine by independent researchers notes that in vivo generation of CAR-T cells is now transitioning from theoretical to achievable, with multiple platforms at various stages of development.
The field is moving. This paper represents one of the more technically sophisticated approaches to date, with in vivo site-specific integration at a functionally relevant locus. Whether it translates to patients at the efficacy levels seen in mouse models remains to be shown — but the proof of concept is real, and it’s built on solid mechanistic ground.
Source
Eyquem laboratory and Doudna laboratory. “In vivo site-specific engineering to reprogram T cells.” Nature, March 18, 2026.
See also: “Engineering in vivo CAR-T cells” (editorial commentary, Nature Medicine, 2026).