A CRISPR Editor One-Third the Size of Cas9 Just Hit 90% Efficiency. The Delivery Bottleneck May Be Over.

4,700 nucleotides. That is the cargo limit of an adeno-associated virus vector, the safest and most clinically validated delivery vehicle for getting gene-editing tools inside a living human body. SpCas9, the workhorse CRISPR enzyme behind Casgevy and a dozen clinical trials, requires roughly 4,200 nucleotides just for its coding sequence. Add a promoter, a guide RNA scaffold, and a polyadenylation signal, and the total payload hits 5,500 to 6,000 nucleotides. It does not fit. For a decade, that arithmetic gap between what Cas9 needs and what AAV can carry has been the single biggest bottleneck in gene therapy.

In April 2026, two independent research teams published results that may close it.

At the University of Texas at Austin, a team led by structural biologists identified a compact CRISPR enzyme called Al3Cas12f, isolated from a gut bacterium in the genus Alistipes. Published in Nature Structural & Molecular Biology, their study shows Al3Cas12f achieves editing efficiencies exceeding 50% at most genomic targets in human cells, with some sites reaching 90%. An engineered variant, Al3Cas12f RKK, pushes rates from below 10% to above 80% across a broad range of loci. At roughly one-third the size of SpCas9, the entire editor plus its guide RNA fits comfortably inside an AAV capsid.

Simultaneously, a team at Sun Yat-sen University in Guangzhou, China, published a complementary system in Nature Communications. Their engineered TnpB protein, derived from an IS200/IS605 transposon, serves as a Swiss-army knife for genome regulation: gene activation (2,889-fold increase with a minimized 93-nucleotide guide RNA), genome editing, and base editing, all packable into a single AAV. They demonstrated a proof-of-concept cancer immunotherapy called AAV-ImmunAct that activates the immune genes CXCL9, IL-15, and IFN-γ inside tumor cells, enhancing T cell killing of cancer cell lines and patient-derived organoids, and synergizing with anti-PD-1 therapy in humanized mice.

Neither team cited the other. Both arrived at the same conclusion from different starting points: the AAV size ceiling is no longer an engineering impossibility. It is a solved problem, at least in the lab.

Why Size Matters: A Payload-Gap Analysis

To understand why these results matter, consider the arithmetic that has constrained in vivo CRISPR therapy since 2012.

Component	SpCas9 System	Al3Cas12f System	TnpB System
Editor coding sequence	~4,200 nt	~1,500 nt	~1,200 nt
Guide RNA scaffold	~100 nt	~80 nt	93 nt
Promoter + poly(A)	~800 nt	~800 nt	~800 nt
ITRs (AAV packaging)	~300 nt	~300 nt	~300 nt
Total payload	~5,400 nt	~2,680 nt	~2,393 nt
AAV capacity	~4,700 nt	~4,700 nt	~4,700 nt
Remaining capacity	-700 nt (does not fit)	+2,020 nt	+2,307 nt

With SpCas9, the payload exceeds AAV capacity by roughly 700 nucleotides. Researchers have spent years developing workarounds: split-intein dual-AAV strategies, smaller Cas orthologs like SaCas9, and lipid nanoparticle delivery. Split-intein approaches can work but typically sacrifice 30% to 50% of editing efficiency because two separate viral particles must both infect the same cell and reassemble a functional enzyme. Lipid nanoparticles, used successfully by Intellia Therapeutics for their NTLA-2001 liver-targeting program, avoid the size constraint but currently reach only the liver effectively.

Al3Cas12f and TnpB flip the equation. With over 2,000 nucleotides of spare capacity, a single AAV can carry the editor, its guide, and still have room for a homology-directed repair template (typically 800-1,500 nt), a second guide RNA for multiplex editing, or additional regulatory elements to target specific tissues. This is not an incremental improvement. It transforms single-AAV delivery from "barely possible with compromises" to "comfortable with room to spare."

What This Unlocks: Counting the Diseases

Approximately 7,000 monogenic diseases are catalogued by the World Health Organization. Fewer than 200 have any approved treatment. Only a handful have approved gene therapies: Luxturna for inherited retinal dystrophy (2017), Zolgensma for spinal muscular atrophy (2019), Hemgenix for hemophilia B (2022), and Casgevy for sickle cell disease (2023, but ex vivo, requiring bone marrow harvest and chemotherapy conditioning).

Most of these 7,000 diseases are caused by mutations in genes smaller than 3,000 nucleotides. A compact editor system using ~2,400 nt leaves ~2,300 nt of usable payload, enough for a guide RNA targeting the mutation plus a short HDR template encoding the correction. For gene disruption applications, where you simply need to knock out a toxic gain-of-function allele, the spare capacity is even more generous: a second or third guide RNA adds only ~120 nt each.

Not every monogenic disease is equally amenable. Some require edits in tissues that AAV serotypes cannot efficiently transduce (neurons, for example, are accessible via AAV9, while muscle requires AAVrh74 or engineered capsids). Some mutations span deletions too large for HDR templates. But the constraint that was previously binary for most diseases ("can the editor fit in AAV at all?") has now shifted to an analog optimization problem ("which capsid, which promoter, which dose?"). That is a fundamentally different engineering challenge, and one the field already knows how to iterate on.

How Al3Cas12f Works

Al3Cas12f belongs to the Cas12f subfamily, a class of miniature CRISPR nucleases discovered through metagenomic surveys. Earlier Cas12f variants suffered from poor editing efficiency in mammalian cells, typically below 10%, making them academic curiosities rather than therapeutic candidates.

What makes Al3Cas12f different is its dimer architecture. Most Cas12f enzymes struggle because they form unstable complexes with their guide RNAs. Al3Cas12f features extensive interlocking interactions between its two subunits, described by the UT Austin team as a "mortise-and-tenon" joint, similar to the woodworking technique. These extended helices stabilize the enzyme-guide complex and promote efficient R-loop formation, the critical step where the enzyme unwinds target DNA and positions the cutting domains.

Crucially, Al3Cas12f's guide RNA scaffold is naturally streamlined. Other Cas12f systems carry extraneous stem-loop structures that reduce packaging efficiency. Al3Cas12f eliminates those, yielding a smaller total footprint without sacrificing recognition accuracy. When the researchers introduced three targeted mutations (the RKK variant), they enhanced electrostatic interactions with the DNA backbone, boosting editing efficiency from single digits to above 80% at previously resistant loci.

What TnpB Adds

If Al3Cas12f is a precision scalpel, the TnpB toolkit is a modular platform. TnpB proteins are ancestral relatives of Cas12, predating the CRISPR-Cas immune system by billions of years. Because they evolved for transposon mobility rather than immune defense, they carry minimal overhead.

What the Sun Yat-sen team demonstrated is not just gene cutting but gene activation. Their engineered enTnpBa variant boosted endogenous gene expression by 2,889-fold with a guide RNA scaffold of only 93 nucleotides. This is significant because many diseases are caused not by mutations that need correcting but by insufficient expression of a functional gene. Upregulating existing genes with a single AAV injection would bypass the need for DNA cutting entirely, avoiding the genotoxicity risks associated with double-strand breaks.

For their cancer immunotherapy application, the team packaged three gene-activation guides into one AAV vector, simultaneously turning on CXCL9 (which recruits T cells), IL-15 (which activates them), and IFN-γ (which primes the tumor microenvironment). In humanized mouse models, AAV-ImmunAct combined with anti-PD-1 checkpoint therapy produced stronger tumor regression than either treatment alone. Patient-derived tumor organoids confirmed the killing effect in human-origin tissue.

Strongest Counterargument

Every generation of CRISPR miniaturization has produced a wave of "this changes everything" headlines, followed by years of grinding clinical translation. SaCas9, a smaller Cas9 ortholog, was celebrated in 2015 for fitting into AAV and has yet to produce an approved therapy. CjCas9, even smaller, showed poor PAM flexibility. Cas12a variants like AsCas12a showed efficiency in cell culture that did not fully translate to animal models.

Al3Cas12f has not been packaged into AAV yet. All the published efficiency data comes from plasmid transfection in a leukemia-derived cell line (K562 cells). Plasmid transfection delivers orders of magnitude more editor than AAV transduction. When Al3Cas12f is constrained to the copy numbers that AAV actually delivers per cell (typically 1-100 vector genomes), efficiency could drop substantially. Previous compact editors have lost 50-80% of their cell-culture performance when moved to AAV delivery in animal models. Until the UT Austin team publishes AAV-packaged results in vivo, the 90% headline number should be treated as an upper bound, not a clinical expectation.

Additionally, off-target characterization remains incomplete. Neither paper reports unbiased genome-wide off-target detection (GUIDE-seq, DISCOVER-Seq, or CIRCLE-seq). For therapeutic applications, comprehensive off-target profiling is a regulatory requirement, and compact editors with novel PAM specificities sometimes exhibit unpredictable off-target profiles.

Limitations of This Analysis

Our payload-gap calculation uses published coding sequence lengths and standard regulatory element sizes. Actual AAV packaging behavior depends on secondary structure, manufacturing conditions, and serotype-specific capsid tolerance. Some groups report functional packaging of genomes up to 5.0 kb in certain AAV serotypes, while others find degradation begins above 4.5 kb. We use the consensus 4.7 kb limit.

Our estimate that "most monogenic disease genes are under 3,000 nt" derives from median human coding sequence length data (~1,340 nt, Lander et al., 2001) but does not account for specific mutation types (large deletions, repeat expansions) that require different therapeutic strategies. We also cannot predict immunogenicity of these novel proteins in human patients, which has derailed other gene therapy programs (notably high-dose AAV trials that triggered fatal hepatotoxicity).

What You Can Do

If you are a gene therapy researcher or biotech investor: Watch for Al3Cas12f AAV packaging data, which the UT Austin team identified as their next milestone. Positive results would validate the cell-culture numbers in a delivery-relevant context. Monitor whether Biogen's $1.37 billion Capsigen deal pivots to incorporate compact editors. Tissue-specific AAV capsids combined with miniature CRISPR systems represent the convergence that could industrialize single-shot gene therapy.

If you or a family member has a monogenic disease: Ask your specialist whether your specific mutation is a candidate for in vivo gene editing (as opposed to ex vivo approaches like Casgevy, which require bone marrow conditioning). Organizations like the National Human Genome Research Institute maintain searchable databases linking diseases to active clinical trials. These miniature-editor results are pre-clinical, so direct therapies are years away, but the pipeline is widening.

If you work in regulatory science: Compact editors with novel protein architectures will need adapted safety frameworks. Off-target detection methods validated for SpCas9 may require recalibration for different PAM specificities. Pre-existing immunity to gut-derived bacterial proteins (Al3Cas12f comes from a human gut commensal) is an open question that could affect dosing strategies.

The Bottom Line

For a decade, CRISPR gene therapy has been caught in a packaging paradox: the best editing tool was too big for the best delivery vehicle. Split-enzyme workarounds sacrificed efficiency. Alternative delivery methods sacrificed tissue reach. Two papers in April 2026 present compact editors that resolve this tradeoff directly, fitting inside a single AAV with over 2,000 nucleotides to spare and achieving editing rates that approach the SpCas9 benchmark. A $57.4 billion cell and gene therapy market projected for 2028 has been built largely on ex vivo approaches that require extracting cells, editing them in a lab, and infusing them back. If compact in vivo editors deliver in animal models what they have shown in cell culture, the market shifts from factory-dependent cell manufacturing to injection-based gene correction. Roughly 7,000 monogenic diseases are waiting. For the first time, the delivery truck is big enough for the tools and the cargo.