Science1 distinct publisher3 min readPublished
Rather than evolving a hotter integrase, the authors designed 144 versions of the DNA site it lands on, a change you install yourself and can therefore keep away from the enzyme's recognition surface. The rice lines inherited the edit.
The Scientist · Science desk
Compiled by The ScientistSomething wrong?How this is made
The design logic here is worth walking through. Bxb1 assembles as a tetramer on its recognition sites, and those sites are built from repeated motifs that the authors label motif1, motif2 and motif3 [14]. Because at least four distinct native sequences can be bound, each with its own kinetics, a wild-type att site is a mixture of strong elements and weak ones [13]. The move was to keep the favourable elements, discard those that bind poorly or not at all, and assemble chimeric sites from what survived [21].
A saturated mutant library was never going to be the practical route. The attB site is 38 base pairs, so a saturated mutant library would run to 4^38 sequences, about 7.6 x 10^22 of them [12][18]. The authors built 144 and reported that all of them were recognised by Bxb1, with different activities [1]. Rational design gave them a way to sample that space at a scale that random screening could not touch.
Work backwards from the fold-changes and the baselines appear. Dividing 51.9% by 1.7 puts wild-type attB near 30.5% in HEK293T cells; dividing 35.6% by 4.4 puts it near 8.1% in rice protoplasts [16][17]. So the human system was already reasonably good and gained a little; the plant system was poor and gained a lot. The single site variant did both, which is the more interesting result.
The safety argument turns on where the change sits. Activity-boosting mutations in engineered Bxb1 proteins reportedly cluster at the DNA-binding interface, which is also the surface that decides what the enzyme mistakes for a target [9]. Earlier work went down that road with directed evolution and AI-assisted prediction, producing variants such as eeBxb1, epBxb1 and reBxb1 [8]. An engineered att site changes a sequence you deliberately install by prime editing at a safe harbour [11], leaving the recognition surface untouched. The authors did add a protein mutation to reach their therapeutic-cassette numbers, but placed it in the catalytic domain rather than the DNA-binding interface [4][23].
The thing this doesn't tell you is what the assay was. The 51.9% figure and the 31% CD19 CAR figure are not the same experiment: the cassette work carries a real therapeutic payload and the protein mutant, and lands about 21 points lower [19]. HEK293T is also the human system on offer here, not a primary T cell. In rice, the 24% denominator is regenerated plants, and regeneration is itself a filter [5]. Long-read nanopore sequencing showed the 5.8 kb insertion was complete and precise [6], which is structure, not expression durability.
Still, this is a clean piece of reasoning rewarded with a clean result, and the heritability data in T1 lines is the strongest part of it [7]. My view, conditional on the genome-wide specificity profiling [15] holding up at loci other than the ones chosen here: if you are building a Bxb1 pipeline, the att sequence is the first knob to turn, because it is cheap to change and does not require you to characterise a new protein.
Ranked by verification strength, evidence, and original report placement.
The authors rationally designed 144 attB variants based on the canonical att consensus, all of which were recognised by Bxb1 recombinase with distinct recombination activities.
The engineered attB(V111) site achieved 51.9% integration efficiency in HEK293T cells, a 1.7-fold increase versus wild-type attB.
The same attB(V111) site achieved 35.6% integration efficiency in rice protoplasts, a 4.4-fold increase versus wild-type attB.
Paired with an engineered single protein mutant in the Bxb1 catalytic domain, the optimised system achieved targeted integration efficiencies of 31% for a CD19 chimeric antigen receptor cassette and 25% for an ornithine transcarbamylase expression cassette in human cells.
In rice, the engineered variants enabled integration of a 5.8 kb herbicide-resistance cassette at a targeted genomic locus, with stable integration detected in 24% of regenerated plants.
Oxford Nanopore long-read sequencing of edited plants showed complete and precise insertion with high specificity.
Distinct publishers with included, body-backed reporting in this cluster.
Follow any of these and your For You feed starts watching them — no settings page required.
science
Prime editing swaps a whole gene by annealing matched flaps on genome and donor1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One paper, but a well-instrumented one
Every figure in this story - 51.9%, 35.6%, 31%, 25%, 24% - comes from a single peer-reviewed paper, and nobody outside the originating lab has repeated a line of it. What raises it above an announcement is what the authors built into their own work: full-length donor recovery, genome-wide specificity profiling by sequencing, nanopore reads across the insertion, and a T1 generation carrying the edit. The text available to us stops early in the results, so the headline efficiencies rest on the abstract's summary rather than on figures a reader can inspect.
Still only in the inventors' hands
The engineered sites have been used in HEK293T cultures, rice protoplasts and the authors' regenerated seedlings - and nowhere else that this reporting shows. No second lab, no therapy programme, no breeding pipeline, no plasmid deposit or licence appears anywhere in the paper, so there is nothing to measure here beyond the publication event itself.
The friendliest number leads
51.9% is exactly as reported, but it is the site-level reading on a bare attB variant. The cargo anyone actually wants moved lands at 31% and 25%, and only with an extra Bxb1 mutation added; in plants, 35.6% in protoplasts thins to 24% of regenerated plants. The subtler stretch is the word safe: the safety case rests on where a mutation sits - catalytic domain rather than DNA-binding surface - which is sound design reasoning, and the genome-wide profiling supports the insertion side, but neither is an off-target measurement of the engineered sites across the variant set. Inflation of emphasis, not of fact.
The inventors are the only narrator
A team is publishing its own method into a race it names on the page: eeBxb1, epBxb1 and reBxb1 are other groups' hyperactive protein variants, and the paper's argument is that going after the DNA instead is the safer route. That is a legitimate scientific claim and simultaneously a positioning one. Nothing reaches us about funding, competing interests or patent filings, and no outside researcher is quoted anywhere, so the framing runs entirely one way.
Direction solid, magnitudes provisional
Two things are hard to argue with: chimeric att sites work, and rice lines passed the insert to the next generation under peer review. What we cannot hold tightly are the specific percentages - one lab, one paper, partial text, and integration efficiencies that are notoriously sensitive to cell type, target locus and readout. Ask again when someone else's protoplasts and someone else's CAR construct produce numbers.