Nature Biotechnology
(2026) Cite this article
Here we developed a DNA-centric strategy for optimizing site-specific recombination by rationally engineering chimeric attachment sites. The high-activity att variants enhance Bxb1-mediated integration efficiency in human cells and plants. Among these att variants, the engineered attB(V111) site achieved 51.9% integration efficiency in HEK293T cells (1.7-fold versus wild-type attB) and 35.6% in rice protoplasts (4.4-fold versus wild-type attB). When paired with an engineered single protein mutant in the Bxb1 catalytic domain, the optimized 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, these 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-based long-read sequencing of edited plants reveals complete and precise insertion with high specificity. Propagation of edited seedlings to T1 plants confirms heritable editing to future generations. This approach provides a safe, broadly applicable approach for recombinase-based genome editing.
Site-specific recombinases, such as the large serine integrase Bxb1, catalyze precise DNA recombination between a pair of short specific sequences (attP and attB) without requiring exogenous DNA repair factors1,2. This makes them exceptionally valuable for applications in cell and gene therapy, trait introgression in plant breeding and synthetic biology. A common strategy involves using prime editing to install an att site at a genomic safe harbor, followed by Bxb1-mediated integration of a donor cassette containing the other att site3,4. To improve the utility of this system, enhancing the integration activity of Bxb1 is crucial. Previous endeavors have focused on engineering the recombinase protein itself through directed evolution or artificial intelligence-assisted predictions, leading to hyperactive variants such as eeBxb1, epBxb1, reBxb1 and others5,6,7,8. However, mutations that confer high activity in protein variants often cluster at the DNA-binding interface, which increases the risk of unintended off-target effects with potentially unpredictable consequences5.
An alternative strategy to enhance recombinase activity is to engineer its att recognition sites. However, screening individual mutants for this purpose is time-consuming and laborious, especially since even the shorter attB sequence is still 38 bp long, making it theoretically impractical to construct a saturated DNA mutant library. Nevertheless, we reasoned that a more efficient strategy could be derived from learning the binding properties of recombinases. In nature, a recombinase often recognizes a pair of short DNA sequences to form a tetramer and thereby catalyze recombination. Consequently, at least four types of native sequences could be recognized by the recombinase with each exhibiting distinct binding kinetics9. We hypothesized that one could rationally combine the most favorable sequence elements from the known natural recognition sites to bypass extensive random screening efforts, thereby enabling directly engineering high-activity chimeric attachment sites. This approach strategically excludes suboptimal sequences that contribute to weak binding sites, or even those without any recognition activity.
Here we rationally designed 144 attB variants based on the canonical att consensus, all of which were recognized by Bxb1 recombinase with distinct recombination activities. From this library, we isolated hyperactive attB variants capable of driving efficient targeted integration across both human and plant systems. In human cells, these variants enable robust integration of therapeutic cargos, including a CD19-targeted chimeric antigen receptor (CAR) and an ornithine transcarbamylase (OTC) expression cassette. The platform is further validated by full-length donor recovery, genome-wide specificity profiling by next-generation sequencing (NGS) and Oxford Nanopore Technologies (ONT)-based long-read genome sequencing, and stable Mendelian inheritance of edits in T1 rice lines.
The Bxb1 attB and attP sites have been shown to follow binding based on distinct motifs (motif1, motif2, motif3) (Fig. 1a,b), which are proposed to correspond to specific protein-interaction sub-structures (hairpin, helix, loop)10. We therefore hypothesized that recombining these motifs from different halves of the natural attachment sites could generate additional, functional att sites recognized by Bxb1.
a, Sequences of m1, m2, m3, m4, m5 and m6. Each sequence comprises combinations of motif1, motif2 and motif3 derived from attB-L, attB-R, attP-L and attP-R, respectively. b, Sequences of Bxb1 att sites. The WT sequence arrangement ‘m1, m2, m3, m4’ is shown on the left, contrasted with the rationally designed variant ‘m4, m4, m4, m4’ on the right. The core GA dinucleotide is highlighted in red, and motif1, motif2 and motif3 are highlighted in distinct colors. c, Schematic of the dual-luciferase reporter system used for quantifying recombination activity in tobacco leaves. RB, right T-DNA border; LB, left T-DNA border. d, Fold-change in inversion activity of rationally designed att sites compared with the WT att sites. e, Fold-change in inversion activity of ‘m5, m5, m5, m5’ and ‘m6, m6, m6, m6’ att sites compared with ‘m4, m4, m4, m4’ in tobacco leaves. f, Schematic of strategy to characterize the integration activity of the rationally designed attB sites in human cells. g, Insertion efficiency of the attP into the TRAC site (left) and the integration efficiency of the WT attB and its variants in HEK293T cells. h, Sequence resource of 144 attB variants. i, Fold-change in integration activity of 144 attB variants relative to wt(attB) at the TRAC site in HEK293T cells. j, Integration activity of selected attB variants in HEK293T cells. k, Integration activity of selected attB variants at the GSH1 site in rice protoplasts. PE, prime editing (att insertion); IN, donor integration among all successfully prime-edited alleles. Data are presented as mean ± s.d. of three biological replicates in d, e, g, j and k. Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test, with ‘m1, m2, m3, m4’ designated as the control group in g.
To test this, we first defined the motif1, 2, 3 combinations on the attB-L as ‘m1’ (Fig. 1a,b). Similarly, we defined the corresponding motifs from attB-R as ‘m2’, from attP-L as ‘m3’ and from attP-R as ‘m4’ (Fig. 1a,b). The wild-type (WT) att sites are thus designated ‘m1, m2, m3, m4’ (Fig. 1b). We then synthesized a pair of de novo att sites denoted as ‘m4, m4, m4, m4’, in which all three motifs on the left and right of both attB and attP were replaced using the ‘m4’ sequence (Fig. 1b). We developed a quantitative recombination assay in tobacco leaves based on successful recombinase inversions correlating with firefly luciferase (F-LUC) activity (Fig. 1c). Using this assay, we confirmed that Bxb1 not only recognized the ‘m4, m4, m4, m4’ att site variants but also mediated inversions at higher efficiency than the WT ‘m1, m2, m3, m4’ sequence (Fig. 1d). This prompted us to create and test the homotypic variants ‘m1, m1, m1, m1’, ‘m2, m2, m2, m2’ and ‘m3, m3, m3, m3’. Bxb1 recognized all three sites, with ‘m1, m1, m1, m1’ and ‘m3, m3, m3, m3’ both demonstrating higher activity than the WT ‘m1, m2, m3, m4’ sequence (Fig. 1d).
Encouraged by these results, we designed and screened a broader set of chimeric att sites containing mixtures of motifs. From this screen, the combinations ‘m3, m4, m3, m4’ and ‘m1, m4, m1, m4’ emerged as the most active, showing the highest inversion efficiencies in tobacco leaf (Fig. 1d). Furthermore, to investigate the independent nature of these motifs, we modified the ‘m4, m4, m4, m4’ sequence by replacing its motif2 with either a GAC sequence from ‘m1’ or an AAC sequence from ‘m3’, generating the variants ‘m5, m5, m5, m5’ and ‘m6, m6, m6, m6’, respectively (Fig. 1a,e). Recombination was successfully detected for both variants, demonstrating that motif1 and motif2 are functionally modular and can be productively recombined. In addition to using a tobacco reporter assay, we evaluated the integration efficiency of our rationally designed attB variants for editing precise genomic insertions of a 5.6 kb plasmid donor in human cells using a prime-editing-assisted site-specific integrase gene editing (PASSIGE) strategy, which uses a single or dual pegRNA to install an att site in the human genome and then utilizes Bxb1-mediated insertion to integrate a donor sequence containing a matched att site, enabling precise, programmable genomic integration3,5. Consistent with our initial findings, the engineered attB variants mediated successful recombination activity and showed significantly higher integration efficiency compared with the WT attB and attP sequences (Fig. 1f,g).
From the homotypic evaluations, we identified ‘m2’ from attB-R as a potential limiting constraint to Bxb1 recombination activity (Fig. 1d). By analyzing the natural Bxb1 attachment sites, we noticed four distinct sequences for motif1, three for motif2 and one for motif3 (Fig. 1h). A systematic combinatorial design of all potential permissible sequences allows for a theoretical design of 144 distinct attB sequences (Fig. 1h). To systematically characterize these sequences, we constructed 144 individual donor plasmids, with each carrying a unique attB variant comprised of a combinatorial mixture of motifs (attB(V1) to (V144)). We individually analyzed the integration activity of each de novo attB sequence at the endogenous TRAC site (T cell receptor alpha constant, located at 14q11.2) in HEK293T cells using the same PASSIGE editing approach. All 144 attB(V) donor plasmids supported successful integration, demonstrating that Bxb1 retains recognition capability across a diverse set of chimeric sequences (Fig. 1i). The integration activity exhibited a broad dynamic range, with fold-change values relative to wt(attB) spanning from 0.23 to 4.75 (Fig. 1i). This result confirms the functional modularity of the att site motifs and highlights the potential for enhancing Bxb1-mediated recombination activity through engineering its DNA recognition sequences using modular assembly approaches.
We characterized the integration efficiency of WT Bxb1 (WT_Bxb1) with the top-performing attB variants—attB(V115), attB(V91) and attB(V111)—and found integration efficiencies among genomes with successful attP insertions to be 49%, 50% and 52%, respectively, which are, on average, 1.7-fold higher than that of wt(attB) (Fig. 1j). To evaluate the cross-kingdom applicability of this strategy, we tested the integration activity of top-performing variants at the GSH1 site in rice protoplasts using a PrimeRoot-mediated insertion method, which is using a prime editor to first install a lox66 sequence into a targeted genomic locus, followed by Cre-mediated integration of a donor carrying a lox71 sequence11. The rationally designed attB variants alone mediated a substantial enhancement in recombination integration efficiency, with increases from a baseline integration of 8% using the wt(attB) to 36% with attB(V111) (Fig. 1k). The robust performance of att variants in phylogenetically distant monocot crops demonstrate the broad utility and potential for large gene insertions using engineered att sequences across diverse eukaryotic systems.
Although many studies have already engineered and evolved Bxb1 recombinases, we sought to develop an in planta high-throughput screening approach to evolve Bxb1 integration activity directly in a plant-based cellular environment. Site-saturation mutagenesis was focused on the N-terminal catalytic domain (using NNK primers) to minimize potential off-target effects associated with DNA-binding alterations (Supplementary Fig. 1). We generated a recombinase selection circuit based on dual transfer DNAs (T-DNAs) co-delivered into a Nicotiana benthamiana leaf by Agrobacterium tumefaciens. We designed one T-DNA to harbor an attB site while the other an attP site along with a Bxb1 protein variant expressed using a CaMV 35S promoter. Successful recombination leads to fusion of the two T-DNAs, which we validated using a dual-luciferase reporter system and splitting the F-LUC across both T-DNAs (Extended Data Fig. 1). For selecting enriched recombinase variants, we designed a single reverse primer (R) on the recombinase which can amplify with one of two forward primers. Forward 1 (F1) is located on the attB-containing T-DNA, while Forward 2 (F2) is located on the recombinase T-DNA. Only successful attB/attP recombination permits amplification using F1/R while all variants can be amplified using F2/R (the amplified region stretches across the entire N-terminal domain of Bxb1 containing the mutated region). The NNK variant library was first generated in Escherichia coli, then delivered to leaves via Agrobacterium; following variant expression and a 2-day incubation, genomic DNA from tissues was extracted and purified. Two sets of PCR were performed using either F1/R or F2/R, and NGS was used to identify enriched, high-activity variants (Fig. 2a,b).
a, Schematic of the screening circuit in tobacco leaves and the strategy for identifying enriched variants. Primers used are listed in Supplementary Table. b, Workflow diagram of the Bxb1 variants screen in tobacco leaves. Credit: Illustration from NIAID NIH BIOART Source (bioart.niaid.nih.gov/bioart/685). c, Fold-change in total integration efficiency of Bxb1 variants at GSH1 site using PrimeRoot strategy in rice protoplasts. Each dot represents an individual biological replicate. d, Integration activity in rice protoplasts. Data are presented as mean ± s.d. of three biological replicates. Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparisons test. e, Targeted integration of the OsEPSPS gene cassette into upstream of OsEPSPS site in rice plants using an all-in-one plasmid. GOI, gene of interest. f, Statistics of OsEPSPS gene cassette targeted insertion events in rice plants. g, The T0 seedlings phenotype after 3 weeks of 1,350 g a.i. ha−1 Roundup spraying (30% glyphosate, 2× the minimum recommended dose in maize fields). Left is an edited rice plant, and the right is a WT rice plant.
Following library selection, individual candidate variants were first evaluated for integration activity in rice protoplasts using a PrimeRoot strategy. Several single mutants exhibited modestly increased activity compared with WT_Bxb1 (Fig. 2c). The combination of an enhanced protein variant Bxb1(R2Q) and the designed attB(V111) site resulted in an integration efficiency of 32% (Fig. 2d). Parallel comparisons among Cre (used in PrimeRoot), Cre-cm24 (utilized in Re-PCE), WT_Bxb1 and eeBxb1 (hyperactive triple mutants implemented in eePASSIGE) revealed that the attB(V111) and R2Q exhibited significantly enhanced integration activity (Supplementary Fig. 2).
Next, we sought to use the optimized attB(V111) in combination with Bxb1(R2Q) to generate herbicide-resistant rice plants. We designed an all‑in‑one vector for inserting a 5.8 kb cisgenic OsEPSPS pentamutant gene12 (driven by its native 2 kb promoter and followed by a 0.5 kb terminator) upstream of the endogenous OsEPSPS gene using a PrimeRoot-style, one-step plant stable transformation (Fig. 2e). First, we screened highly active dual-epegRNAs that efficiently installed the attB(V111) with a core GA dinucleotide at the target site13 (Supplementary Fig. 3). The donor construct was designed to undergo recombination through two sequential steps in vivo: first, a circularization step mediated by attP and attB(V111) flanked at the 5′ and 3′ of the donor whose core dinucleotides are GT, then this circularized donor undergoes integration mediated by a second orthogonal attP with a GA core dinucleotide that specifically recombines with the genomically installed attB(V111)-GA (Supplementary Fig. 4), thereby precisely inserting the cisgenic herbicide‑resistant OsEPSPS expression cassette into the rice genome. We successfully regenerated 129 transgenic-positive rice plants and obtained 31 rice plants with precise integration of the herbicide-resistant OsEPSPS cassette at the target site, yielding an integration efficiency of 24% for this large-fragment targeted insertion (Fig. 2f and Extended Data Fig. 2a,b). This stable plant transformation editing efficiency was much higher than those previously reported for other systems in rice, such as RePCE-mediated GFP insertion (9.3% for 720 bp)14,15 and Kp03-mediated insertion of a 3.4 kb donor (10.8%)16. Sequencing analysis confirmed that both the 5′ and 3′ integration junctions were precisely edited as predicted (Supplementary Fig. 5). We simultaneously analyzed for T-DNA excision using both PCR and nanopore-based long-read whole-genome sequencing and found donor excision efficiencies achieving 74.4% (Fig. 2f and Extended Data Fig. 2c,d). We next assessed the glyphosate resistance of the T0 plants regenerated following this insertion event. We sprayed 1,350 g a.i. ha−1 Roundup (30% glyphosate, 2× the minimum recommended dose in maize fields); 3 weeks later, we found that the edited plants grew well and demonstrated herbicide-resistant properties while the WT plant was dead (Fig. 2g).
