Structural mechanism governing the directionality of bridge recombination

Insertion sequence (IS) elements are transposable DNA sequences found in prokaryotic genomes and are categorized into approximately 30 families, including IS110 (ref. ³). The transposition cycle of IS110 family elements involves two sequential recombination events: excision of the element from the host genome to form a circular double-stranded DNA intermediate and insertion of this circular form into a new target site^{4,5,6,7,8,9,10} (Fig. 1a,b and Extended Data Fig. 1a). Although insertion has been the focus of recent structural and functional studies^1,2, the mechanism by which IS110 elements are excised under native conditions has remained poorly understood.

Unlike typical transposable elements¹¹, IS110 elements, such as IS621 from Escherichia coli⁶, comprise a left end (LE), a recombinase gene and a right end (RE), with conserved CT (cytidine–thymidine) core dinucleotide sequences at both boundaries (Fig. 1a,b and Extended Data Fig. 1a). The IS110 recombinase has a distinctive domain architecture: a RuvC-like nuclease domain with a non-canonical DEDD catalytic motif (hereafter, RuvC for simplicity), a coiled-coil domain that mediates dimerization and a Tnp domain harbouring a conserved catalytic serine residue^1,6 (Extended Data Fig. 1b). During transposition, the IS110 recombinase catalyses recombination between the circular intermediate (donor DNA (dDNA), containing left donor (LD) and right donor (RD) sequences) and the genomic target site (target DNA (tDNA), containing left target (LT) and right target (RT) sequences), generating the LT–RD and LD–RT junctions at the insertion site, referred to as the left half (LH) and right half (RH), respectively (Fig. 1b and Extended Data Fig. 1a). Excision is the reverse process: recombination of the LH and RH regenerates the circular donor and the original target site.

We recently established that the IS621 circular intermediate expresses a non-coding bridge RNA (bRNA) from a promoter reconstituted across the RE–LE junction¹. The bRNA comprises two programmable DNA-binding loops: a target-binding loop (TBL) and a donor-binding loop (DBL), each containing guide segments that hybridize with tDNA and dDNA, respectively, enabling the IS621 recombinase to catalyse sequence-specific recombination between diverse DNA substrates¹ (Extended Data Fig. 1c). Cryo-electron microscopy (cryo-EM) structures of the IS621 recombinase–bRNA complex bound to insertion substrates (referred to as the insertion complex) showed a tetrameric synaptic complex in which composite RuvC–Tnp active sites catalyse stepwise strand cleavage and exchange² (Extended Data Fig. 1d). Furthermore, we and others demonstrated that the bridge recombinase orthologue ISCro4 can mediate programmable rearrangement of the human genome^12,13.

Studies on the IS110 family member ISEc21 showed that a non-coding RNA derived from the upstream non-coding region co-purifies with the recombinase and is essential for circular intermediate formation, with a minimum number of target-derived flanking bases required on each side of the element for excision to proceed¹⁴. In a related IS110 family member, ISPpu9, multiple distinct circular intermediates are generated through recombination between different combinations of terminal sequences, in a process requiring an active recombinase and specific nucleotides at the element ends¹⁵. However, it remains unclear how the bridge recombinase–bRNA complex mediates excision under native conditions in bacterial cells, how the left and right element junctions are recognized, and how the bRNA is produced before circular intermediate generation, given that bRNA expression is itself driven by circularization. Here, we biochemically and structurally characterized the IS621 excision reaction, providing a resolution to this chicken-and-egg problem.

To investigate how the IS621 elements are excised by the IS621 recombinase–bRNA complex in bacterial cells, we transformed E. coli strain BL21(DE3), which naturally lacks IS621 elements, with plasmids encoding IS621 in one of two configurations: either mimicking the excised circular intermediate, in which the RE–LE junction is reconstituted to generate a promoter upstream of the bRNA (Extended Data Fig. 2a), or mimicking the genomic (linear) arrangement, in which the element is in its LE–recombinase–RE configuration and lacks the junction promoter (Extended Data Fig. 2b). We then profiled transcription from both constructs by RNA sequencing (RNA-seq). Consistent with our previous study¹, we observed transcription of the bRNA from the reconstituted RE–LE junction promoter (Extended Data Fig. 2a). We also detected low-level transcription across the full-length IS621 element in the LE–recombinase–RE configuration, which lacks the junction promoter (Extended Data Fig. 2b), suggesting that both the recombinase and bRNA can be expressed at low levels even without promoter reconstitution. To assess the expression of the IS621 elements encoded in the E. coli genome, we sequenced the whole transcriptome of the E. coli strain Mach1, which contains three copies of the IS621 element (Fig. 1c and Extended Data Fig. 2c). We observed transcription originating from each genomic locus, with some transcripts extending into the region encoding the bRNA (Fig. 1d).

To determine whether excision is supported by low-level transcription from the LE–recombinase–RE configuration, we performed polymerase chain reaction (PCR) targeting both the expected circular intermediate (RE–LE) and the post-excision junction in E. coli BL21(DE3) cells harbouring a plasmid encoding IS621 in the linear configuration (Extended Data Fig. 3a). Both products were detected from the plasmid lacking the junction promoter (Extended Data Fig. 3b,c), indicating that read-through transcription can produce sufficient amounts of the recombinase and bRNA to mediate excision. When a transcriptional terminator was inserted upstream of the LE on the plasmid (Extended Data Fig. 3a), neither the expected circular intermediate nor the post-excision junction was detected by PCR (Extended Data Fig. 3b,c), indicating that transcription through the element is required for excision. Deletion of the bRNA-encoding region from the LE yielded similar results (Extended Data Fig. 3b,c), confirming that the bRNA itself, rather than the process of transcription, is essential for excision.

We next asked whether excision also occurs from native IS621 loci in the E. coli genome. PCR analysis of total DNA from E. coli Mach1 detected the expected circular intermediate (RE–LE), whereas no product was observed in E. coli BL21(DE3), which lacks IS621 elements (Fig. 1c,e). Furthermore, PCR analysis of total DNA from E. coli Mach1 detected post-excision junctions at the expected genomic sites of all three elements (Fig. 1c,f), suggesting that the circular intermediates originate from any of the three loci across the population under basal conditions. As the three IS621 copies are highly homologous, we cannot exclude the possibility that the bRNA or recombinase expressed from one locus acts in trans to facilitate excision at another. Collectively, these results indicate that the bRNA and recombinase-coding transcripts are expressed from the IS621 elements in the E. coli genome and mediate the excision of these elements.

To investigate whether the IS621 recombinase–bRNA complex catalyses the excision reaction, we next measured the efficiencies of the excision and insertion reactions in vitro, using quantitative PCR (qPCR). As we previously found that IS621-mediated insertion is enhanced by extending the right-target guide (RTG; positions 74–76) and right-donor guide (RDG; position 161) segments of the bRNA^1,2 (Extended Data Fig. 4), we assessed the impact of bRNA modifications on the efficiencies of both processes. We used the purified IS621 recombinase and the 177-nt bRNA with DNA substrates for excision (51-bp LH and 121-bp RH) or insertion (86-bp tDNA and dDNA) (Fig. 1g and Extended Data Fig. 5a). Consistent with our previous findings^1,2, we observed substantial enhancements in insertion efficiency with the RTG extension, with further additive improvement in combination with the RDG extension (Extended Data Fig. 5b). Notably, excision was roughly 2,000-fold less efficient than insertion (0.03% compared with 63%) (Fig. 1h and Extended Data Fig. 5b), with efficient insertion requiring both the TBL and DBL (Extended Data Fig. 6a). Unexpectedly, DBL alone mediated both excision and insertion (Extended Data Fig. 6a), suggesting that DBL–DNA base pairing alone is sufficient for substrate recognition and that two DBL-containing recombinase dimers can assemble into a productive tetramer even without TBL, consistent with our previous structural observations². Every bRNA modification that increased the insertion efficiency caused a corresponding decrease in the excision efficiency (Extended Data Fig. 5b).

We next explored whether extended bRNA transcripts, potentially resulting from read-through transcription originating at the promoter of an upstream gene, could support recombination in vitro. Using several transcripts (up to 775 nt) extended at the 5′ end, the 3′ end or both ends, designed based on a transcription start site mapped to an upstream coding sequence (Extended Data Fig. 2c), we found that the extended bRNAs were compatible with both excision and insertion (Extended Data Fig. 6b). When both ends were extended to mimic read-through from the upstream transcription start site into the recombinase-coding sequence, the excision and insertion efficiencies were reduced approximately two-fold relative to the minimal 177-nt bRNA (Extended Data Fig. 6b). Together, these results indicate that read-through transcription from flanking genes can produce functional bRNA transcripts that support the excision of IS621 elements from the E. coli genome. Overall, these biochemical data show that the IS621–bRNA complex can mediate excision, albeit with substantially lower efficiency than insertion.

Because the handshake guide (HSG) dinucleotides in the bRNA (positions 81–82 and 166–167) can base-pair with DNA substrates either before or after top-strand exchange² (Extended Data Fig. 4), we reasoned that this pairing might modulate the reaction directionality. We investigated how the HSG–DNA base pairing affects excision and insertion, using wild-type and engineered bRNA scaffolds with four different HSGs. As reported previously², insertion was facilitated by HSG–DNA base pairing after top-strand exchange, as in the natural reaction (Fig. 1h, i and Extended Data Fig. 7a). Notably, we observed an inverse trend in the relationship between the HSG sequences and the excision efficiencies (Fig. 1h,i and Extended Data Fig. 7a), suggesting that the top strands of the RD-containing LH and the RT-containing RH are recognized by the DBL RDG and TBL RTG, respectively, before top-strand exchange (Fig. 1g). Moreover, these findings indicate that in both insertion and excision, the optimal HSG configuration promotes a transition from fewer to more base pairs in the complex after top-strand exchange. We refer to the HSG configurations that maximize RNA–DNA base pairing before top-strand exchange as pre-TSE and those that maximize pairing after exchange as post-TSE. Together, these findings indicate that (1) the IS621–bRNA complex mediates both excision and insertion, (2) the HSG configuration modulates the favourability of top-strand exchange, thus determining the overall efficiency and directionality of recombination, and (3) insertion is generally more efficient than excision (for example, the efficiencies of insertion and excision were about 70% and 1%, respectively, with their post-TSE configurations) (Fig. 1h).

To determine the relative contributions of bRNA expression and sequence to the excision efficiency, we constructed plasmids encoding the IS621 element with a T7 promoter upstream of the LE in both wild-type bRNA (pre-TSE) and engineered bRNA (post-TSE) contexts. The engineered bRNA carries the post-TSE HSG configuration, RTG/RDG extensions and an LTG G53A substitution (to enable canonical A–T base pairing), which collectively promote the excision reaction (Extended Data Fig. 7b). We transformed E. coli BL21(DE3) with these plasmids and assessed the excision efficiency by qPCR, quantifying the post-excision junctions relative to the total plasmid. With the wild-type bRNA, the T7 promoter drove an approximately 100-fold increase in excision efficiency relative to the construct lacking a T7 promoter (Extended Data Fig. 7c), consistent with bRNA expression contributing to excision frequency. Without a T7 promoter, the engineered bRNA increased excision approximately 13,000-fold over the wild-type bRNA under the same conditions (Extended Data Fig. 7c). Adding the T7 promoter to the engineered bRNA did not further increase the excision efficiency (Extended Data Fig. 7c). Consistent with these qPCR results, both the circular intermediate and the post-excision junction were detected by PCR (Extended Data Fig. 7d,e). Together, these results indicate that the bRNA sequence, rather than its expression level, is the primary determinant of the excision efficiency, and that this barrier can be overcome by engineering the bRNA to favour strand exchange, with no additional benefit from increased transcription.

Although the HSG configurations modulate the relative efficiencies of both reactions, insertion is generally more efficient than excision. To explain why the IS621–bRNA complex mediates these reactions with different efficiencies, we determined the cryo-EM structure of the IS621 recombinase bound to the bRNA and its DNA substrates (referred to as the excision complex). Similar to the insertion reaction², excision was enhanced by the engineered bRNA scaffold with the post-TSE HSG and DNA substrates carrying four-nucleotide mismatches at positions 2–5 in the top strand, designed to locally destabilize the DNA duplex and thereby promote base pairing with the bRNA guides (Extended Data Fig. 8a,b). Furthermore, both insertion and excision were inhibited by the bRNA with the pre-TSE HSG (Extended Data Fig. 8b). Previously, we used bRNAs containing the pre- and post-TSE HSGs to determine the structures of the insertion complexes in the pre- and post-strand exchange states, respectively². Therefore, we initially reconstituted the excision complex in the pre-strand exchange state by mixing the IS621 recombinase, the engineered bRNA with the pre-TSE HSGs (A81/U82/G166/U167C), and the mismatch-containing LH and RH DNA substrates, and then determined its cryo-EM structure at 2.5-Å resolution (Fig. 2a–c, Extended Data Figs. 4 and 8a, Supplementary Fig. 1a–e and Supplementary Video 1).

The excision complex consists of four IS621 recombinase molecules (IS621.1–IS621.4), the TBL and DBL modules of the bRNA, and two DNA molecules (LH and RH) (Fig. 2b,c). As in the insertion complex², the recombinase dimers (IS621.1/2 and IS621.3/4) bind the TBL and DBL, respectively, which in turn recognize the two DNA substrates through RNA–DNA base pairing (Fig. 2a–c and Supplementary Fig. 2a,b). However, the DNA substrates adopt strikingly different conformations within the insertion and excision complexes. During insertion, the tDNA and dDNA are bent into U-shaped conformations within each dimer² (Extended Data Fig. 9a). During excision, the LH and RH instead adopt extended, linear conformations that cross between both dimers, producing an X-shaped arrangement (Fig. 2c and Extended Data Fig. 9b). Furthermore, the DNA substrates are recognized by the TBL and DBL differently in these two complexes. In the insertion complex, the tDNA and dDNA are recognized by the TBL and DBL within each dimer, respectively² (Extended Data Fig. 9a). By contrast, in the excision complex, the LH and RH are recognized by both the TBL and DBL, resulting in each dimer interacting with both substrates simultaneously (Fig. 2c and Extended Data Fig. 9b). As in the insertion complex², the catalytic S241 loops in IS621.2 and IS621.4 are ordered and interact with the D102 loops in IS621.3 and IS621.1, respectively, creating composite active sites to cleave the top strands of both DNA substrates and form 5′-phosphoserine intermediates (Supplementary Fig. 2a,b). The corresponding S241 loops in IS621.3 and IS621.1, which would mediate bottom-strand cleavage, are disordered in this structure (Supplementary Fig. 2a,b). Thus, this structure represents the top-strand cleaved, pre-strand exchange state during the excision reaction.

The cryo-EM structure of the excision complex explains how the IS621–bRNA complex recognizes the excision DNA substrates (LH and RH). As in the insertion complex², the TBL and DBL adopt similar pseudo-symmetric architectures and are recognized by the protein (Fig. 2d–f and Supplementary Fig. 3a). The TBL LTG and the DBL LDG base pair with the LH LT and the RH LD, respectively, whereas the TBL RTG and the DBL RDG base pair with the RH core–RT and the LH core–RD, respectively (Fig. 2d–f and Supplementary Fig. 3a). As expected, the HSG regions in the TBL (positions 81–82) and DBL (positions 166–167) base pair with the RH (A6–T7) and the LH (G6–C7), respectively, thereby preventing top-strand exchange. Unlike in the insertion complex², nucleotides U49, U73, G122 and U156, which flank the four guide segments, form stacking interactions with the DNA nucleobases at positions 3 and 3*, helping the DNA strands pass across the TBL- and DBL-bound dimers (Supplementary Fig. 3a,b). Moreover, in the excision complex, U160 in the RDG adopts a different conformation compared with that in the insertion complex and stacks with A6* in the RH, contributing to DNA binding (Supplementary Fig. 3a,b).

During the insertion process, the TBL- and DBL-bound recombinase dimers separately bind only the tDNA and dDNA, respectively, and then come together to form the tetrameric complex² (Extended Data Fig. 10a). By contrast, the excision complex structure suggests that the TBL- and DBL-bound recombinase dimers can bind both the LH and RH substrates (Extended Data Fig. 10b). Microscale thermophoresis measurements showed that the TBL- and DBL-bound dimers each bind both LH and RH with comparable affinity, whereas they preferentially bind the tDNA and dDNA, respectively (Extended Data Fig. 10c), indicating the different assembly processes between the two complexes. Together, these data show how the IS621–bRNA complex recognizes the LH and RH sequences during the excision process.

As in the insertion complex², the top strands of the RH and LH substrates are cleaved adjacent to their CT core motifs (C8–T9) within the composite active sites (RuvC.1–Tnp.4 and RuvC.3–Tnp.2), generating free 3′-OH groups on the T9 nucleotides of both substrates and covalent 5′-phosphoserine intermediates with S241 of Tnp.4 and Tnp.2, respectively (Fig. 3a–c and Supplementary Fig. 2b). Consistently, our in vitro recombination experiments with fluorescently labelled DNA substrates showed the presence of the top-strand-cleaved intermediates during both the excision and insertion reactions (Extended Data Fig. 8b). The top-strand-cleaved intermediates accumulated to similar levels in both reactions, yet the final recombination product was markedly less abundant during excision (Extended Data Fig. 8b). This suggests that strand exchange, rather than initial cleavage, is the rate-limiting step for excision and accounts for much of the efficiency difference between the two reactions. Together, these data indicate that the top strands of the RH and LH substrates are cleaved adjacent to the CT core, forming 5′-phosphoserine intermediates and 3′-OH groups, which enable subsequent strand exchange (Fig. 3d).

To further investigate the excision mechanism, we determined the 2.6-Å resolution cryo-EM structure of the excision complex in the post-strand exchange state, using the engineered bRNA with the post-TSE HSGs (A81G/U82C/G166A/U167), which facilitate HSG–DNA base pairing after top-strand exchange (Fig. 4a–c, Supplementary Fig. 4a–e and Supplementary Video 2). As expected, the top strands of the LH and RH are cleaved adjacent to the CT core and exchanged, forming new TBL–LH and DBL–RH base pairs (Fig. 4a–d). After exchange, the RDG retains its base pairing with positions 10–14 of the LH top strand, while forming new base pairs with the incoming core (C8–T9) of the RH top strand (Fig. 4a,c). Immediately adjacent to this core interaction, the HSG region (A166–U167) of the DBL base pairs with T7–A6 of the incoming RH strand, stabilizing the post-strand exchange state. Similarly, the RTG remains bound to positions 10–14 of the RH top strand, while forming new base pairs with the incoming core (C8–T9) of the LH top strand, with the HSG region of the TBL (G81–C82) base pairing with C7–G6 of the LH (Fig. 4a,c). These structural observations indicate that, similar to the insertion reaction², the top strands of the DNA substrates are exchanged after strand cleavage during the excision reaction.

In this study, we showed that bRNA is expressed at low levels from linear IS621 elements in E. coli cells and that the IS621 recombinase–bRNA complex mediates the recombination between the two junction sequences flanking the element, the LH (LT–RD) and RH (LD–RT), enabling the excision of the IS621 element from the genome (Fig. 5a). Our biochemical data showed that the IS621–bRNA complex catalyses excision with much lower efficiency than insertion. Furthermore, the cryo-EM structures provided insights into how the IS621–bRNA complex recognizes the excision DNA substrates to mediate their recombination. These findings suggest that excision is normally suppressed in E. coli, but once a rare excision event occurs, the resulting circular intermediate reconstitutes the RE–LE junction promoter, boosting recombinase and bRNA expression. As the natural system is inherently configured to favour insertion, this would preferentially drive transposition into a new target site.

A structural comparison between the excision complex and the insertion complex² showed mechanistic similarities between the IS621-mediated excision and insertion processes (Fig. 5b, Supplementary Fig. 5 and Supplementary Video 3). Initially, the IS621–bRNA complex recognizes DNA substrates for both excision (LH and RH) and insertion (dDNA and tDNA) primarily through identical RNA–DNA base pairing at four guide regions. In both processes, the top strands of the DNA substrates are cleaved at the CT core within the RuvC–Tnp active sites, forming 5′-phosphoserine intermediates and 3′-OH groups on the top strands and allowing top-strand exchange. It is likely that the bottom strands are subsequently cleaved at the other RuvC–Tnp active sites during excision, as observed during insertion². These structural findings highlight the ability of the IS621 bridge system to mediate programmable recombination by similar but distinct stepwise catalytic mechanisms: excision (that is, recombination within a single double-stranded DNA (dsDNA) molecule with LT–RD and LD–RT) and insertion (that is, recombination between two dsDNA molecules with LT–RT and LD–RD).

Notable mechanistic differences exist between the excision and insertion reactions. First, the DNA substrates adopt different conformations in the two complexes. The insertion complex contains bent, U-shaped tDNA and dDNA substrates, whereas the excision complex binds linear LH and RH substrates, which form an X-shaped structure (Fig. 5a,b and Supplementary Video 3). As the transition from bent to linear conformations in the cleaved DNA top strands is likely energetically favourable, top-strand exchange would occur more efficiently during insertion than during excision, consistent with our biochemical data (Extended Data Fig. 8b). These conformational differences probably contribute to the different efficiencies of the two processes. Moreover, our structural and biochemical data indicate the distinct assembly mechanisms of the IS621–bRNA–DNA tetrameric complex in excision and insertion. During insertion, IS621 recombinase dimers bound to TBL/tDNA and DBL/dDNA assemble into a functional tetrameric complex². By contrast, during excision, tetramers can form through multiple pathways, because TBL- and DBL-bound recombinase dimers can each bind both LH and RH substrates (Extended Data Fig. 10c). Productive excision complexes form when a TBL-bound recombinase dimer engaged with the LH and RH assembles with a DBL-bound recombinase dimer engaged with the RH and LH, respectively (Fig. 5a). However, alternative, non-productive combinations may occur and could reduce the overall excision efficiency. Although two TBL-bound recombinase dimers could simultaneously bind to the LH and RH sequences in the natural IS621 locus, they cannot form a productive complex, preventing recombination and thereby reducing excision in E. coli cells (Fig. 5a). Our biochemical data showed that the DBL alone can support excision (Extended Data Fig. 6a), suggesting that recombinase dimers bound to DBL–LH and DBL–RH can form a productive complex, in which the DNA substrates probably adopt U-shaped rather than X-shaped conformations because of a steric clash between the linear LH and RH substrates. Furthermore, in the natural IS621 system, the LH (LT–RD) and RH (LD–RT) sequences are located at both ends of the elements on the same DNA molecule (genomic DNA), whereas the target (LT–RT) and donor (LD–RD) sequences reside on separate DNA molecules (genomic DNA and circular intermediates) (Fig. 5a). These positional differences in the recognition sequences are also likely to affect the efficiencies of the excision and insertion processes.

Typically, RNA-guided enzymes such as CRISPR–Cas9¹⁶ recognize nucleic-acid targets by base pairing with an RNA guide and cleave them within the guide–target duplex. By contrast, the IS621 bridge recombinase recognizes two DNA substrates by base pairing with the four RNA guide segments in the bRNA, and then a portion of the substrates (positions 6–9 in the top strands) must dissociate from the guides after top-strand cleavage for the subsequent strand exchange (Fig. 5b and Supplementary Fig. 5). As the CT core dinucleotides (positions 8–9 in the top strands) are invariant and always base pair with the guide segments, the extent of the base pairing between the top strand of the DNA substrates at positions 6–7 and the HSG dinucleotides of the bRNA (TBL positions 81–82 and DBL positions 166–167) serves as the primary determinant of the recombination efficiency (Fig. 5b and Supplementary Fig. 5). Our previous study showed that HSG–DNA base pairing before and after strand exchange inhibits and facilitates the insertion reaction, respectively². In this study, we found that HSG–DNA base pairing also affects the efficiency of the excision reaction, highlighting a common mechanistic principle between excision and insertion. The wild-type bRNA contains AU (positions 81–82 in TBL) and GU (positions 166–167 in DBL) as the HSG sequences, whereas the wild-type tDNA and dDNA have GC and AT at positions 6 and 7, respectively (Fig. 5b and Supplementary Fig. 5). The wild-type HSG sequences form only one Watson–Crick base pair with the DNA before strand exchange, but three after exchange. This asymmetry drives the insertion reaction forward while simultaneously disfavouring excision, consistent with our biochemical data. Together, these observations suggest that, apart from the low bRNA expression and the limited excision activity of the IS621 complex, the native HSG–DNA base-pairing pattern substantially contributes to suppressing excision in E. coli cells. This base-pairing pattern is highly conserved among other IS110 elements², supporting the role of HSGs as key regulators of the IS110 life cycle.

This study clarifies the programmable recombination mechanism of IS110 bridge systems and shows that the optimal bRNA designs for insertion are inherently unfavourable for excision—a principle essential for maximizing the persistence of genome modifications. The engineering principles presented here offer a practical guide for optimizing bRNA scaffolds to favour either insertion or excision. Several modifications, including the split bRNA system and RTG/RDG extensions, have already improved recombination efficiency in human cells^12,13, underscoring the translational relevance of our findings.

RNA sequencing and analysis

E. coli BL21(DE3) cells (NEB: C2527H), which lack endogenous copies of IS621 elements, were transformed with plasmids encoding either the RE–LE junction or the LE–recombinase–RE region of the IS621 element. The cells were plated on LB (lysogeny broth) agar with kanamycin and grown overnight at 37 °C. Colonies were scraped from the plates, and RNA was extracted using a Direct-zol RNA miniprep kit (Zymo Research). To assess the expression of the IS621 elements encoded in the E. coli genome, RNA was extracted using a Direct-zol RNA miniprep kit from One Shot E. coli Mach1 T1^R cells (F– φ80lacZΔM15 ΔlacX74 hsdR(r_K–, m_K+) ΔrecA1398 endA1 tonA) (Thermo Fisher Scientific, C862003) harvested from bacterial plates 24 h after plating at 37 °C.

RNA-seq was carried out on a NovaSeq X Plus sequencer in collaboration with Novogene Research Services (Novogene). Paired-end RNA-seq reads from each sample were aligned to the respective plasmid and genomic reference sequences using Burrows–Wheeler Aligner-MEM¹⁷. Aligned reads were filtered using SAMTools¹⁸ for reads that originated from the strand that encodes the bRNAs. For each locus of interest on the reference sequences, we used the Pysam Python module to count the coverage depth at the single-base level from these filtered reads, with a quality threshold of 0. As the E. coli Mach1 genome has three separate IS621 loci that are highly homologous with one another, many of the RNA-seq reads that mapped to these regions were expected to lack unique mapping to one of the three loci. Thus, among the reads that were mapped to any of the IS621 loci, pairs of reads in which either pair had a mapping quality of zero (meaning the read can map equally well to other parts of the reference sequence) were classified as ‘nonspecific’ reads, and the coverage from these reads was counted separately from the reads that mapped specifically to one of the three IS621 loci.

Detection of plasmid-based IS621 excision by PCR

Plasmids containing elements with the wild-type bRNA (pre-TSE for excision) or the engineered bRNA (post-TSE for excision) were transformed into E. coli BL21(DE3), which lacks endogenous copies of IS621 elements, ensuring that all detected excision events originate from the plasmid-borne construct. For constructs lacking a T7 promoter upstream of the bRNA, bacteria were plated on LB agar with kanamycin and grown overnight at 37 °C. For constructs with a T7 promoter, plates were supplemented with 0.07 mM IPTG. Colonies were scraped from plates, and plasmid DNA was extracted using a QIAprep Spin Miniprep Plus Kit (Qiagen). Primers were designed to span the post-excision (LT–RT) junction site left after element excision from the plasmid and yield a 110-bp product. The same primers used to detect circularized IS621 intermediates from the Mach1 genome were also used to confirm the production of circular intermediates from plasmid-borne constructs, yielding a 725-bp product. All PCRs were performed with the same protocol described below for the detection of native IS621 excision and were fractionated on 2% agarose gels and visualized with SYBR Gold (Thermo Fisher Scientific). All PCR products were sequence-verified by Sanger sequencing.

Detection of circular intermediates and post-excision sites

Genomic DNA was isolated from freshly purchased aliquots of E. coli strains in liquid culture using Zymo Quick DNA Miniprep Plus Kits (Zymo), according to the instructions of the manufacturer. Whole-genome sequencing was performed at 100× coverage using the Plasmidsaurus whole-genome nanopore sequencing service, which confirmed that the One Shot Mach1 E. coli T1^R strain harbours three copies of IS621, whereas the E. coli BL21(DE3) strain contains zero copies. Bacteria were plated on LB agar and grown overnight at 37 °C. Colonies were then scraped from the plates, and total DNA was extracted using a Zymo Quick DNA Miniprep Plus Kit. Primers were designed to match all three IS621 recombinase-coding sequences in the Mach1 genome, such that PCR amplification would occur only on excised and circularized IS621 sequences from any locus. Moreover, primers were designed to base pair with the junction remaining post-excision at all three loci encoding IS621, such that amplification with a downstream primer would yield a specific product. All PCRs were performed using Platinum SuperFi 2× MasterMix (Thermo Fisher Scientific) with the following protocol: 98 °C for 120 s; 25 cycles of 98 °C for 10 s, 65 °C for 15 s and 72 °C for 15 s; and 72 °C for 2 min. PCR products were subsequently fractionated on 2% agarose gels and visualized with SYBR Gold.

In vitro recombination measurement by qPCR

DNA substrates (51-bp LH, 121-bp RH, 86-bp tDNA and 86-bp dDNA) were purchased from IDT as ssDNA and annealed by heating to 95 °C followed by slow cooling to 4 °C over 1 h in a thermocycler. DNA substrates (0.25–0.5 µM) were mixed with the IS621–bRNA complex (10 µM) in 20 µl buffer containing 20 mM Tris-HCl, pH 7.5, 300 mM NaCl, 5 mM MgCl₂ and 1 mM dithiothreitol (DTT), and then the reactions were incubated at 37 °C for 2 h. The reactions were quenched by the addition of 50 mM EDTA, treated with 8 µg RNase A (NEB) at 50 °C for 1 h, and then treated with 3 units of Proteinase K at 37 °C for 1 h. After RNA and protein digestion, the DNA was purified using DNA Clean & Concentrator-5 (Zymo Research) and eluted with 75 °C nuclease-free water. For qPCR analysis, reaction dilutions were prepared, and qPCR was performed on a LightCycler 480 II instrument (Roche) using primers and PrimeTime qPCR probes (final concentration 0.5 µM) purchased from IDT, along with TaqMan Fast Advanced qPCR mix. In vitro recombination products are expected to have an unligated bottom strand, with the two nucleotides adjacent (3′) to the core mismatched with the top strand, after the bottom strand exchange step of the reaction mechanism. Therefore, to accurately quantify the reaction efficiency, considering the polymerase processivity through lesions and the primer binding near nicks, qPCR signals for different reactions were compared with standard curves generated using known quantities of 121-bp RH (LD–RT, insertion product) DNA and 86-bp target (LT–RT, excision product) DNA containing an unligated bottom strand with the expected mismatches adjacent to the core. These standards were prepared by annealing the top strand of RH/target (121/86 nt) with two oligos (56- and 65-nt RH and 56- and 30-nt target) constituting the bottom strand.

In vivo excision efficiency measurement by qPCR

To determine the excision efficiency in E. coli using the wild-type (pre-TSE for excision) and the engineered bRNA (post-TSE for excision) encoded within the LE of the IS621 element, 100 ng of sequence-verified plasmid, confirmed by nanopore sequencing to lack pre-excised molecules in the input population, was transformed into chemically competent BL21(DE3) cells, which were recovered in SOC (super optimal broth) medium for 1 h at 37 °C, plated onto kanamycin-selective agar and grown overnight at 37 °C. All colonies from a single transformation plate were bulk-harvested by scraping into liquid medium, providing a population-level average of excision efficiency across thousands of independent clonal lineages and minimizing bias from stochastic colony-to-colony variation. Total plasmid DNA was extracted from each pooled sample using a QIAprep Spin Miniprep Plus Kit (Qiagen), and qPCR was carried out on 400 ng of the extracted plasmid. Primers and probes were designed to amplify the KanR resistance marker (HEX) as a measure of total plasmid, as well as the post-excision junction (LT–RT, FAM) on plasmids that had undergone excision of the IS621 element. Multiplexed TaqMan qPCR was performed on a control plasmid containing both the post-excision junction and the KanR marker to determine the relative amplification efficiencies of the two probes. This inter-channel correction factor was applied when calculating the excision efficiency from experimental samples. At least three independent transformations were performed to confirm the reproducibility of the population-level measurement.

Protein and RNA preparation

The IS621 recombinase protein was prepared as described previously². Briefly, the IS621 gene was cloned into a modified pFastBac1 vector encoding an N-terminal His₆–Twin-Strep tag and an HRV3C cleavage site. Sf9 cells were infected with the baculovirus and cultured at 27 °C for 48 h. The cells were harvested and lysed, and the soluble fraction was purified by Strep-Tactin affinity chromatography followed by size-exclusion chromatography. The purified protein was concentrated and stored at −80 °C, in buffer containing 20 mM HEPES-NaOH, pH 7.5, 500 mM NaCl, 2 mM MgCl₂, 1 mM DTT and 10% glycerol, until use. The bRNAs were transcribed in vitro with T7 RNA polymerase and purified by 10% denaturing (7 M urea) polyacrylamide gel electrophoresis (Supplementary Table 1).

In vitro recombination assays with fluorescently labelled DNA substrates

DNA substrates (38-bp LH, 44-bp RH, 38-bp tDNA and 44-bp dDNA) were purchased from Eurofins Genomics. The LH and tDNA were labelled with Cy5 at the 5′ ends of the top strand (Supplementary Table 1). The DNA substrates (0.4 µM; LH and RH for excision or tDNA and dDNA for insertion) were mixed with the pre-incubated IS621–bRNA complex (2 μM) in 10 µl buffer containing 20 mM Tris-HCl, pH 7.5, 300 mM NaCl, 5 mM MgCl₂ and 1 mM DTT, and then the reactions were incubated at 37 °C for 1 h. The reaction mixture was mixed with Proteinase K (Nacalai Tesque) and incubated at 95 °C for 2 min in denaturing buffer (7 M urea). The samples were analysed on an 18% TBE–urea denaturing gel, and fluorescent signals were imaged using FUSION Solo S (Vilber Bio Imaging).

Excision complex preparation

The excision complex was reconstituted by mixing the purified IS621 recombinase, a 177-nt bRNA (177 nucleotides plus a 5′ GGG leader for in vitro transcription) and DNA substrates (LH and RH) containing 4-nt mismatches on the top strands. The bRNAs with pre-TSE and post-TSE were used for reconstitution of the excision complex in the pre- and post-strand exchange states, respectively. The reconstituted complex was purified by size-exclusion chromatography on a Superose 6 Increase 10/300 column (Cytiva), equilibrated with buffer (20 mM Tris-HCl, pH 7.5, 300 mM NaCl, 5 mM MgCl₂ and 1 mM DTT). The purified excision complex was concentrated to 0.5–1 mg ml⁻¹ using an Amicon Ultra-4 centrifugal filter unit (MWCO 50 kDa; Millipore). Protein concentrations were determined using the Pierce 660 nm Protein Assay Reagent.

Cryo-EM analysis

For cryo-EM data collection, Quantifoil Holey Carbon Grids (R1.2/1.3, Au, 300 mesh) (SPT Labtech) were glow-discharged in low-pressure air at a 10 mA current in a PIB-10 (Vacuum Device). The excision complex solution was applied to freshly glow-discharged grids using a Vitrobot Mark IV system (Thermo Fisher Scientific) at 4 °C, with a waiting time of 10 s and a blotting time of 6 s under 100% humidity conditions. The grids were plunge-frozen in liquid ethane cooled by liquid nitrogen.

The grids were transferred to a Titan Krios G3i TEM (Thermo Fisher Scientific), running at 300 kV and equipped with a Gatan Quantum-LS Energy Filter (GIF) and a Gatan K3 Summit direct electron detector. Imaging was performed at a nominal magnification of 105,000×, corresponding to a calibrated pixel size of 0.83 Å per pixel (px). Each movie was dose-fractionated to 50 frames at a dose rate of 7.8 e⁻ px⁻¹ s⁻¹ at the detector in the correlated double sampling mode, resulting in a total accumulated exposure of 50 e⁻ Å⁻² of the specimen. The data were automatically acquired using the image-shift method in the EPU software (Thermo Fisher Scientific), with a defocus range of −0.8 μm to −2.0 μm.

The data were processed using the cryoSPARC v.4.4.0 software package¹⁹. The dose-fractionated movies were aligned using Patch Motion Correction, and the contrast transfer function (CTF) parameters were estimated using patch-based CTF estimation. For the excision complex with the pre-TSE bRNA, particles were automatically picked using Blob Picker and Template Picker, followed by reference-free two-dimensional classification to curate particle sets. The particles were further curated by several cycles of Heterogeneous Refinement. The best class particle set was refined using Homogeneous Refinement, yielding a map at 2.71-Å resolution. Reference-Based Motion Correction followed by Non-Uniform Refinement with optimization of the CTF value yielded a map at 2.44 Å resolution (according to the Fourier shell correlation (FSC) = 0.143 criterion)²⁰ using a tight mask; however, the final reported resolution was determined to be 2.53 Å based on the FSC calculated using the softer refinement mask. For the excision complex with the post-TSE bRNA, particles were automatically picked using Template Picker followed by reference-free two-dimensional classification. The particles were further curated by several cycles of Heterogeneous Refinement. The best class particle set was refined using Non-Uniform Refinement, yielding a map at 2.62 Å resolution. Reference-Based Motion Correction followed by Non-Uniform Refinement with optimization of the CTF value yielded a map at 2.41 Å resolution (according to the FSC = 0.143 criterion) using a tight mask; however, the final reported resolution was determined to be 2.55 Å based on the FSC calculated using the softer refinement mask. The local resolution was estimated by BlocRes in cryoSPARC.

Model building and validation

The models of the excision complexes were manually built with COOT²¹, using the IS621 insertion complex model (PDB ID: 8WT6) as the initial model, and then refined using Servalcat²² against unsharpened half-maps. The models were validated using MolProbity²³. The statistics of the three-dimensional reconstruction and model refinement are summarized in Extended Data Table 1. The cryo-EM density maps were calculated with UCSF ChimeraX²⁴, and molecular graphics figures were prepared with CueMol (http://www.cuemol.org).

Microscale thermophoresis

Microscale thermophoresis (MST) was performed using a Monolith NT.115 Pico Series instrument (NanoTemper Technologies) with premium capillaries. The IS621 recombinase was labelled using a RED-MALEIMIDE 2nd Generation cysteine-reactive kit (NanoTemper Technologies), according to the instructions of the manufacturer. The labelled protein was diluted in buffer containing 20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 5 mM MgCl₂, 1 mM DTT and 0.01% Tween 20, and ligands were prepared by dilution in the same buffer. For ligand preparation, DNA was purchased from IDT and annealed in buffer containing 10 mM Tris, pH 8.0, 5 mM MgCl₂ and 5 mM KCl. To determine the affinities of the IS621–RNA complex for the DNA substrates (tDNA, dDNA, LH and RH), the IS621–RNA complex (20 nM) was incubated with serial dilutions (0.3 nM to 10 µM) of each relevant DNA ligand. MST measurements were performed at 37 °C with 8% LED excitation and medium MST power in the Pico-RED excitation mode. Data were analysed using the NanoTemper MO.affinity analysis software package, and raw data were plotted on Prism for visualization.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

1. Durrant, M. G. et al. Bridge RNAs direct programmable recombination of target and donor DNA. Nature 630, 984–993 (2024).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

2. Hiraizumi, M. et al. Structural mechanism of bridge RNA-guided recombination. Nature 630, 994–1002 (2024).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

3. Siguier, P., Gourbeyre, E., Varani, A., Ton-Hoang, B. & Chandler, M. Everyman’s guide to bacterial insertion sequences. Microbiol. Spectr. https://doi.org/10.1128/microbiolspec.mdna3-0030-2014 (2015).

4. Perkins-Balding, D., Duval-Valentin, G. & Glasgow, A. C. Excision of IS492 requires flanking target sequences and results in circle formation in Pseudoalteromonas atlantica. J. Bacteriol. 181, 4937–4948 (1999).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

5. Müller, C., Lauf, U. & Hermann, H. The inverted repeats of IS1384, a newly described insertion sequence from Pseudomonas putida strain H, represent the specific target for integration of IS1383. Mol. Genet. Genomics 265, 1004–1010 (2001).

   Article  PubMed  Google Scholar 

6. Choi, S., Ohta, S. & Ohtsubo, E. A novel IS element, IS621, of the IS110/IS492 family transposes to a specific site in repetitive extragenic palindromic sequences in Escherichia coli. J. Bacteriol. 185, 4891–4900 (2003).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

7. Partridge, S. R. & Hall, R. M. The IS1111 family members IS4321 and IS5075 have subterminal inverted repeats and target the terminal inverted repeats of Tn21 family transposons. J. Bacteriol. 185, 6371–6384 (2003).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

8. Prosseda, G. et al. Plasticity of the P_junc promoter of ISEc11, a new insertion sequence of the IS1111 family. J. Bacteriol. 188, 4681–4689 (2006).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

9. Higgins, B. P., Carpenter, C. D. & Karls, A. C. Chromosomal context directs high-frequency precise excision of IS492 in Pseudoalteromonas atlantica. Proc. Natl Acad. Sci. USA 104, 1901–1906 (2007).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

10. Higgins, B. P., Popkowski, A. C., Caruana, P. R. & Karls, A. C. Site-specific insertion of IS492 in Pseudoalteromonas atlantica. J. Bacteriol. 191, 6408–6414 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

11. Curcio, M. J. & Derbyshire, K. M. The outs and ins of transposition: from Mu to Kangaroo. Nat. Rev. Mol. Cell Biol. 4, 865–877 (2003).

    Article  CAS  PubMed  Google Scholar 

12. Perry, N. T. et al. Megabase-scale human genome rearrangement with programmable bridge recombinases. Science 391, eadz0276 (2026).

    Article  CAS  PubMed  Google Scholar 

13. Pelea, O. et al. Programmable genome editing in human cells using RNA-guided bridge recombinases. Science 391, eadz1884 (2026).

    Article  CAS  PubMed  Google Scholar 

14. Siddiquee, R., Pong, C. H., Hall, R. M. & Ataide, S. F. A programmable seekRNA guides target selection by IS1111 and IS110 type insertion sequences. Nat. Commun. 15, 5235 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

15. Parés-Guillén, E., Yuste, L., Rojo, F. & Moreno, R. The ISPpu9 insertion sequence of Pseudomonas putida KT2440 generates various circular intermediates enabling modular transposition. Nucleic Acids Res. 53, gkaf668 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

16. Jinek, M. et al. A programmable dual-RNA-guided DNA endonuclease in adaptive bacterial immunity. Science 337, 816–821 (2012).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

17. Li, H. et al. The sequence alignment/map format and SAMtools. Bioinformatics 25, 2078–2079 (2009).

    Article  PubMed  PubMed Central  Google Scholar 

18. Danecek, P. et al. Twelve years of SAMtools and BCFtools. GigaScience 10, giab008 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

19. Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14, 290–296 (2017).

    Article  CAS  PubMed  Google Scholar 

20. Rosenthal, P. B. & Henderson, R. Optimal determination of particle orientation, absolute hand, and contrast loss in single-particle electron cryomicroscopy. J. Mol. Biol. 333, 721–745 (2003).

    Article  CAS  PubMed  Google Scholar 

21. Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. D Biol. Crystallogr. 66, 486–501 (2010).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

22. Yamashita, K., Palmer, C. M., Burnley, T. & Murshudov, G. N. Cryo-EM single-particle structure refinement and map calculation using Servalcat. Acta Crystallogr. D Struct. Biol. 77, 1282–1291 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

23. Williams, C. J. et al. MolProbity: more and better reference data for improved all-atom structure validation. Protein Sci. 27, 293–315 (2018).

    Article  CAS  PubMed  Google Scholar 

24. Pettersen, E. F. et al. UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci. 30, 70–82 (2021).

    Article  CAS  PubMed  Google Scholar 

25. Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

Download references

Author notes

1. These authors contributed equally: Masahiro Hiraizumi, Januka S. Athukoralage, Nicholas T. Perry, Eisuke Tsujimoto, Nami Shiojiri

Authors and Affiliations

1. Department of Chemistry and Biotechnology, Graduate School of Engineering, The University of Tokyo, Tokyo, Japan

   Masahiro Hiraizumi, Eisuke Tsujimoto, Nami Shiojiri, Naoto Nagahata & Hiroshi Nishimasu

2. Arc Institute, Palo Alto, CA, USA

   Januka S. Athukoralage, Nicholas T. Perry, Lauren Lee, Gwanggyu Sun, Matthew G. Durrant, Sita S. Chandrasekaran, Silvana Konermann & Patrick D. Hsu

3. Department of Bioengineering, University of California, Berkeley, Berkeley, CA, USA

   Nicholas T. Perry, Matthew G. Durrant & Patrick D. Hsu

4. San Francisco Graduate Program in Bioengineering, University of California, Berkeley, Berkeley, CA, USA

   Nicholas T. Perry

5. Department of Biochemistry, School of Medicine, Stanford University, Stanford, CA, USA

   Silvana Konermann

6. Structural Biology Division, Research Center for Advanced Science and Technology, The University of Tokyo, Tokyo, Japan

   Keitaro Yamashita & Hiroshi Nishimasu

7. Innovative Genomics Institute, University of California, Berkeley, Berkeley, CA, USA

   Patrick D. Hsu

8. Center for Computational Biology, University of California, Berkeley, Berkeley, CA, USA

   Patrick D. Hsu

9. Inamori Research Institute for Science, Kyoto, Japan

   Hiroshi Nishimasu

Contributions

M.H., E.T. and N.S. performed biochemical and structural analyses with assistance from N.N., K.Y. and H.N.; J.S.A. performed in vitro recombination and binding experiments with assistance from L.L. and S.S.C.; N.T.P. performed E. coli recombination and RNA-seq experiments and M.G.D. and G.S. performed computational analysis; H.N. wrote the manuscript with assistance from M.H., J.S.A., N.T.P., M.G.D., E.T., K.Y., S.K. and P.D.H.; S.K., P.D.H. and H.N. supervised the research.

Corresponding authors

Correspondence to Patrick D. Hsu or Hiroshi Nishimasu.

Competing interests

P.D.H. acknowledges outside interest in Stylus Medicine, Terrain Biosciences and Monet AI as a co-founder, serves on the board of directors at Stylus Medicine and the scientific advisory board at Amgen, and is a venture partner at Thrive Capital. M.G.D. acknowledges outside interest in Stylus Medicine. N.T.P., M.G.D. and P.D.H. are inventors on patent application no. PCT/US23/82192 submitted by the Arc Institute that covers bridge recombinase in human cells. M.H., J.S.A., N.T.P., M.G.D., P.D.H. and H.N. are inventors on patent application no. PCT/US25/12632 submitted by the Arc Institute that covers bRNA engineering. All other authors declare no competing interests.

Peer review information

Nature thanks Chase Beisel, Chunyi Hu, Anuja Kibe and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Reporting Summary (download PDF )

Peer Review File (download PDF )

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions