Control of ϕC31 integrase-mediated site-specific recombination by protein trans-splicing

Abstract Serine integrases are emerging as core tools in synthetic biology and have applications in biotechnology and genome engineering. We have designed a split-intein serine integrase-based system with potential for regulation of site-specific recombination events at the protein level in vivo. The ϕC31 integrase was split into two extein domains, and intein sequences (Npu DnaEN and Ssp DnaEC) were attached to the two termini to be fused. Expression of these two components followed by post-translational protein trans-splicing in Escherichia coli generated a fully functional ϕC31 integrase. We showed that protein splicing is necessary for recombination activity; deletion of intein domains or mutation of key intein residues inactivated recombination. We used an invertible promoter reporter system to demonstrate a potential application of the split intein-regulated site-specific recombination system in building reversible genetic switches. We used the same split inteins to control the reconstitution of a split Integrase-Recombination Directionality Factor fusion (Integrase-RDF) that efficiently catalysed the reverse attR x attL recombination. This demonstrates the potential for split-intein regulation of the forward and reverse reactions using the integrase and the integrase-RDF fusion, respectively. The split-intein integrase is a potentially versatile, regulatable component for building synthetic genetic circuits and devices.


INTRODUCTION
It has recently become possible to create computational and memory systems in cells (1)(2)(3) allowing us to foresee many new ways to enhance the applications of living organisms (4,5). Engineered cells could act as powerful biosensors with applications in health, environmental and industrial processes. As well as sensing components, it is necessary to process the received information and express it as outputs in the form of specific biological responses (3,6). However, the genetic switches and logic gates that have been constructed to date are based on a limited repertoire of biological component types (7), and there is a need for new systems that can be used to implement more elaborate and robust devices.
DNA site-specific recombination has been much exploited for rapid DNA assembly, and to build genetic switches and memory devices (4,(8)(9)(10). In a typical module, two recombination sites flank a promoter sequence. Expression of the recombinase promotes inversion of the orientation of the promoter sequence, thus switching between expression of two genes which are divergently transcribed from the module ( Figure 1A). One group of site-specific recombinases known as the serine integrases is especially suited for the construction of switching devices, particularly because these enzymes promote very efficient and highly directional recombination (11). For such modules to be useful, fully integrated components of the cell, activity of the recombinase must be tightly regulated, so that switching occurs only when other cellular conditions are fulfilled. Here we demonstrate a powerful new approach to regulation of serine integrase activity, in which the enzyme itself is assembled by intein-mediated fusion of two precursor components.
Inteins are naturally occurring autocatalytic systems that catalyse protein splicing reactions to generate active proteins from precursor polypeptides (12). Synthetic 'split in- Non-covalent association of the N-terminal intein (I N ) and the C-terminal intein (I C ) is followed by a trans-splicing reaction that covalently joins the N-terminal extein (E N ) and C-terminal extein (E C ). (C) Intein-mediated protein trans-splicing involves specific chemical steps requiring a catalytic nucleophile. In this example, cysteine residues are required to mediate the reactions.
teins' have been developed to carry out protein splicing in trans, covalently joining two proteins together for a wide range of biotechnological applications (13,14) ( Figure 1B). For example, Schaerli et al. (15) split T7 polymerase into two parts and fused each part to split-intein sequences so that the two parts were covalently joined together by transsplicing. Expression of each of the two individual parts was placed under the control of an inducible promoter to allow conditional expression of polymerase activity.
A key requirement for the construction of split-intein systems is the need to introduce an intein nucleophilic residue (typically Cys, Ser or Thr; Figure 1C) by mutation of the sequence of the target protein adjacent to the junction between the two 'exteins'. Such changes can potentially lead to reduction or loss of enzyme activity. However, inteins have been engineered that tolerate variations of these flanking residues, thereby minimizing the number of changes that need to be made in the target protein (16,17).
Conditional site-specific recombinase activation by assembly of split protein fragments has been achieved; for example, for the popularly used tyrosine recombinase Cre (18)(19)(20)(21)(22). Wang et al. (22) used a Cre split-intein system to reconstitute functional recombinase in transgenic mice. However, there are no reports to date of analogous systems using split serine integrases. The transposase TnpX (distantly related to the serine integrases) was split into two parts and some DNA-binding activity was reconstituted when both parts were present, but no recombination (transposition) activity was observed (23).
With the increasing importance of serine integrases as tools in synthetic biology, methods for signal-induced posttranslational regulation of integrase activity are becoming very desirable. Here, we report post-translational activation of site-specific recombination by reconstitution of functional C31 integrase using split intein-catalysed reactions.

Plasmids and DNA
The codon-optimized C31 integrase sequence was derived from pFM141 (24). Codon-optimized sequences of NpuN, a 102-amino acid residue split intein from Nostoc punctiforme DnaE and SspC, a 36-amino acid residue split intein from Synechocystis sp. DnaE, were from GeneArt (Invitrogen). Plasmids for constitutive expression of the Nterminal extein-intein fusion protein (Integrase-Ext N -I N ) in Escherichia coli ( Figure 2) were made by inserting proteincoding DNA sequences between NdeI and Acc65I sites in pMS140, a low-level expression vector with a pMB1 origin of replication (24). Plasmids for constitutive expression of the C-terminal intein-extein fusion protein (I C -Integrase-E C ) were made in a similar way by inserting the coding sequence between NdeI and Acc65I sites in pEK76, which has a p15a (pACYC184) origin (24,25). Detailed properties of the vectors used here for expression of the recombinases have been described in detail elsewhere (9,(24)(25)(26)(27). Plasmids for tet-inducible expression of the split integrase components were made by cloning the protein-coding DNA sequences between SacI and SalI sites in pTet (28), whilst those for ara-inducible expression were made by cloning the protein-coding sequences between SacI and Sal sites in pBAD (29).
The plasmid substrate for assessing recombination (deletion) by native C31 integrase or trans-spliced C31 integrase (pC31-delPB) was described in Olorunniji et al. (24). The sequences of C31 att sites are shown in Supplementary Table S1. The test substrate contains a galK gene flanked by attP and attB sites arranged in direct repeat (head-to-tail orientation) resulting in deletion of the galK gene upon integrase-catalysed recombination. The plasmid substrate for assessing recombination (inversion) activity (pC31-invPB) was made by cloning the invertible promoter device shown in Figure 3A in a pSC101 origin plasmid with kanamycin resistance selection.
The DNA sequences of all protein constructs and plasmid substrates used in this study are shown in Supplementary Table S2.

Recombination analysis
Assays for integrase-mediated recombination were carried out in E. coli strain DS941 (24). DS941 was first transformed with the test substrate plasmid (pC31-delPB or pC31-invPB). The substrate plasmid-containing strains were then either transformed with an integrase-expressing plasmid (pFM141), or co-transformed with two plasmids, Integrase-E N (residues 1-307) fused to the 102-amino acid residue Npu DnaE N (I N ) using the two-residue EY and the N-terminal Cys residue from Npu DnaE N . (iii) Integrase-E N -I N *; C31 integrase (1-307) fused to Npu DnaE N as described in (i) but with the EY residue at the splice site changed to GA to inactivate splicing activity. (iv) Integrase-EC; C31 integrase Cextein (residues 314-605). (v) I C -Integrase-E C ; C31 integrase C-extein (residues 314-605) fused directly to the Ssp DnaE C (I C ). Ssp DnaE C nucleophilic cysteine residue and the flanking residues required for splicing are highlighted (CFN). (vi) I C* -Integrase-E C ; C31 integrase C-extein fused to Ssp DnaE C as described in (i) but with the CFN residue at the splice site changed to ASA to inactivate splicing activity. each expressing part of the split integrase. The transformants were grown on selective MacConkey-galactose indicator plates (MacConkey agar base (Difco) supplemented with 1% galactose); kanamycin (50 g/ml) was included The open reading frame of Integrase-E N -I N was expressed from a pACYC184-based vector, and I C -Integrase-E C was expressed from a pBR322-based vector (24). Expression of the two polypeptides followed by intein-catalysed protein trans-splicing reconstitutes C31.Int ␣ . (C) Assay of recombination activities (galK colour assay) of integrase constructs on attP × attB deletion substrates (pC31-delPB). In these assays, cells containing the substrate plasmid (pC31-delPB) were transformed with the expression vectors indicated and grown for 20 h in selection media. For analysis of in vivo recombination products, plasmid DNA was recovered from cells (24) and separated by means of 1.2% agarose gel electrophoresis. (1) Substrate only (blank control); (2) FEM141: C31 Integrase; (3) FEM136: Integrase-E N -I N ; (4) FEM137: to select for the substrate plasmid and the recombination product, whilst ampicillin (50 g/ml) and/or chloramphenicol (25 g/ml) were included to select for the integrase expression plasmids. Integrase-mediated deletion of the galK gene results in pale-coloured (galK − ) colonies on the indicator plates, whereas red (galK + ) colonies suggest lack of recombination. In some cases, colonies in which lowlevel recombination has occurred maintain a red colour.
To determine the extent of recombination more accurately, plasmid DNA recovered from the cells was examined. Lbroth (1 ml) was added to the surface of each plate and the cells were suspended using a plate spreader. An aliquot of this suspension was used to inoculate L-broth (1 in 1000 Nucleic Acids Research, 2019, Vol. 47, No. 21 11455 dilution), and the culture was incubated overnight at 37 • C with kanamycin selection for the substrate and recombinant plasmids. Plasmid DNA was prepared using a Qiagen miniprep kit, and analysed by 1.2% agarose gel electrophoresis.

Inducible expression of integrase-intein fusions for transsplicing
In vivo expression of the integrase-intein fusion proteins under the control of arabinose and tet inducible promoters was carried out in strain DS941/Z1 (30), which consitutively expresses TetR, required for regulation of the pTet promoter. The DS941/Z1 strain was made competent by a standard calcium chloride method (31). The cells were transformed with the plasmid substrate pC31-invPB, then cultured for 90 min and selected on L-agar plates containing kanamycin (50 g/ml). A single colony was picked and grown in kanamycin-containing L-broth (5 ml) to make a stationary phase overnight culture. An aliquot of the stationary phase culture was then diluted into L-broth containing kanamycin and grown to mid-log phase. These cells were made 'chemically competent' (as above) and transformed with the two vectors containing the coding sequences of the intein-integrase fragments. The plasmid vector pFEM148 (ampicillin selection) expresses Integrase-E N -I N under the control of the pTet promoter. The second plasmid pFEM149 (chloramphenicol selection) expresses I C -Integrase-E C under the control of the pBAD promoter.
The transformant cells were cultured for 90 min, and selected on L-agar plates containing kanamycin (50 g/ml), ampicillin (100 g/ml) and chloramphenicol (25 g/ml). To carry out the recombination assays, a culture from a single colony was grown overnight in L-broth in the presence of the three antibiotics, to stationary phase. The culture was diluted further (1:100) and grown to mid-log phase (about 90 min), after which expression of the split-intein-integrase fragments was induced by the addition of anhydrotetracycline, aTc (0.1 g/ml), arabinose, Ara (0.2% w/v) or both. Glucose was added to 0.4% concentration to the cultures where it is required to turn off expression of the arabinose promoter. The induced cultures were grown for 24 h at 37 • C, after which the cultures were left at room temperature for 8 h. Next, aliquots of each culture (50 l) were diluted with 950 l phage buffer (10 mM Tris, pH 7.5, 10 mM MgCl 2 , 68 mM NaCl), and GFP expression was measured by means of fluorescence.

Fluorescence measurements
Fluorescence measurements were carried out on a Typhoon FLA 9500 fluorimager (GE Healthcare). Aliquots of the diluted cultures (200 l) were added to a 96-well plate, and the fluorescence of the expressed proteins was measured (GFP: excitation, 485 nm; emission, 520 nm and RFP: excitation, 532 nm; emission, 575 nm). To determine cell density for each sample, 50 l aliquots were diluted to 1000 l and the spectrophotometric absorbance was read at 600 nm. The GFP-independent background signals of the cells were determined by measuring the fluorescence of DS941/Z1 strain (containing the test substrate but without the split-inteinintegrase expression vectors). The background fluorescence was subtracted from the values measured for samples in the different treatment groups, after normalization using the cell density measurements.

Flow cytometry
Flow cytometry was used to measure single-cell fluorescence on a BD Accuri C6 instrument. Cells were washed in phosphate buffered saline and diluted to ∼10 6 cells/ml. For each sample, GFP ( Ex 488 nm; Em 533/30 nm) and RFP ( Ex 488 nm; Em 585/40 nm) fluorescence was recorded. The forward scatter threshold was lowered to 10 000 to ensure acquisition of bacteria, and gating was performed to tightly select the dense population of bacteria depicted on a log scale plot of forward versus side scatter. Samples were acquired at a slow flow rate of 14 l/min. Data were analysed using FlowJo™ software (Version 10.6.1).

Design of split-intein serine integrase
We designed our split-intein system for conditional expression of C31 integrase, a prototype serine integrase, based on previously reported systems (15). Since the natural DnaE split inteins used in this work require an invariant active site cysteine to remain in the spliced product protein (32)(33)(34), their use for split integrase reconstitution depends upon the identification of a short region of the protein sequence where insertion of a cysteine does not disrupt activity. In addition, for trans-splicing to be essential, the two extein components of the spliced integrase must not associate noncovalently to reconstitute a functional enzyme. We started by analysing the domain structure of serine integrases to determine where to split the protein, since a functional split intein could require introduction of mutations that would remain in the spliced protein product, if suitably placed natural Cys, Ser or Thr residues were unavailable. Based on sequence alignments of serine integrases ( Figure 2A) and published crystal structures (35), we split C31 integrase at the non-conserved region of the recombinase domain between the ␤9 and ␣I domains; this region includes a 10-12 residue loop which is absent in some related serine integrases (36) (Figure 2A). We therefore predicted that introduction of the required intein nucleophilic residue and any flanking residues would not have a deleterious effect on integrase activity. Furthermore, Lucet et al. (23) found that when the related large serine recombinase TnpX (from Clostridium perfringens) was split between the ␤9 and ␣I domains, the two fragments complemented each other to restore DNA binding (but not recombination) activity.
We then attached two well-characterised split-intein components to the C31 integrase sequences; Npu DnaE N , I N (102 amino acids) from Nostoc punctiforme DnaE, and Ssp DnaE C , I C (36 amino acids) from Synechocystis sp. DnaE. The Npu DnaE N variant contains a L22S change and the Ssp DnaE C variant has a P21R mutation (32). This pair was chosen based on previous reports that the mutations confer high trans-splicing activity at 37 • C in E. coli and tolerance to changes in amino acid sequence at the splicing junctions (15,32,37). Npu DnaE N (I N ) was fused to the C-terminus of the N-terminal moiety of C31 Integrase (Integrase-E N ), and Ssp DnaE C (I C ) was fused to the N-terminus of the Cterminal moiety of C31 Integrase (Integrase-E C ) ( Figure  2B). Also, residues 308-310 of Integrase-E N were changed from EGY to EY, since these sequences exist naturally at the extein-intein junction of Npu DnaE N , and Integrase-E N residues 311-313 were changed from RIQ to CFN, sequences found at the intein-extein junction of Ssp DnaE C (32,37). It is known that these flanking residues are involved in enhancing the splicing efficiencies of this pair of split inteins (33,37). In the reconstituted trans-spliced integrase (C31.Int ␣ ), the natural 6-residue sequence at positions 308 to 313, EGYRIQ, is replaced with the 5-residue sequence EYCFN. We made these changes to maximize the efficiency of the splicing reaction in order to optimize activity in E. coli. Hence the reconstituted trans-spliced integrase, C31.Int ␣ is one amino acid residue shorter than the wild-type C31 integrase. Since these changes are in a nonconserved region (see above), we predicted that they would not substantially affect recombination activity.

Intein-mediated reconstitution of functional C31 integrase
To assay in vivo recombination activity of our split-intein C31 integrase, we used a well-characterized colour-based galk assay ( Figure 3A; 26,27,38). The attP and attB recombination sites (Supplementary Table S1) on the substrate plasmid pC31-delPB are in a direct repeat orientation, so that recombination between them causes deletion of the galK gene. Pale-coloured (galK−) colonies on the indicator plates indicate recombination proficiency, whereas red (galK+) colonies indicate incomplete or zero recombination. The coding sequences for Integrase-E N -I N and I C -Integrase-E C (and also appropriate control proteins; see Figure 2C) were cloned into separate low-level expression vectors (24), as illustrated in Figure 3B. An E. coli strain containing the test substrate was transformed with these plasmids, and recombination activity was assessed by colony colour and by gel electrophoresis analysis of plasmid DNA recovered from the cells ( Figure 3C).
Neither Integrase-E N -I N ( Figure 3C, lane 3) nor I C -Integrase-E C ( Figure 3C, lane 4) on their own were able to catalyse attP x attB recombination. Furthermore, coexpression of the N-extein (Integrase-E N ) and C-extein (Integrase-E C ) components of the integrase (lane 5) gave no recombination. This important control shows that the intein-less precursor proteins Integrase-E N and Integrase-E C do not complement each other by non-covalent association to give recombination activity. This contrasts with reported split tyrosine recombinases, the components of which associate to reconstitute recombination activity without protein splicing (18)(19)(20). When the intein-tagged integrase exteins Integrase-E N -I N and I C -Integrase-E C were coexpressed, reconstitution of integrase recombination activity was observed (lane 6). The DNA analysis shown in the lower panel of Figure 3c (lanes 1-6) suggests that recombination by the trans-spliced integrase (C31.Int ␣ ) has proceeded to over 80%.
Our results show that the changes from the wild-type integrase sequence that had to be introduced at the splice site are compatible with recombination activity. As many other serine integrases have similarly non-conserved, vari-able lengths of amino acid sequence in the region of the protein between ␤9 and ␣I (Figure 2A), we predict that these enzymes could also be engineered to create active splitintein systems.

Trans-splicing is required for integrase recombination activity
The results shown in Figure 3C show that the Npu DnaE and Ssp DnaE intein moieties are required for reconstitution of recombination activity, but these experiments do not unambiguously establish the requirement for transsplicing. Some split inteins are known to associate tightly via non-covalent interactions (39), and this can lead to reconstitution of the split protein activity, without splicing (40). To test whether this applies in our split integrase system, we mutated the nucleophilic cysteine residue and rateenhancing flanking residues at the active sites of the two intein moieties (29) to render them catalytically inactive. We changed the junction residues 'EY' in the Integrase-E N moiety to 'GA' and the Cys residue in the Npu DnaE N moiety of Integrase-E N -I N to 'A' to give the mutated version Integrase-E N -I N * (Figure 2c). Similarly, the residues 'CFN' in the Ssp DnaE C moiety of I C -Integrase-E C were changed to 'ASA', to derive the mutated version I C* -Integrase-E C ( Figure 2C). No recombination activity was observed when active Integrase-E N -I N was co-expressed with inactive I C* -Integrase-E C ( Figure 3C, lane 7), nor when active I C -Integrase-E C was co-expressed with inactive Integrase-E N -I N * ( Figure 3C, lane 8), showing that reconstitution of recombination activity requires the catalytic activities of both split intein fragments. In this aspect, serine integrases differ from integrases of the tyrosine recombinase family where non-covalent association is sufficient to allow reconstitution of activity from split protein fragments (18)(19)(20).
The attP x attB in vivo deletion reactions were slower in cultures where recombination required post-translational trans-splicing to reconstitute functional integrase, when compared to recombination by native integrase (see Supplementary Figure S1). It is therefore likely that the rate of trans-splicing between Npu DnaE N and Ssp DnaE C in our cell cultures is significantly slower than the fast splicing reaction rate observed in vitro for these intein pairs (37). The reduced rate might be due to limiting amounts of the two precursor polypeptides expressed in the cells. It is possible that this delay is caused by the different rates of expression of the integrase fragments from the two different expression vectors used in this study, rather than the actual trans-splicing reaction in vivo.

Split intein-regulated integrase-catalysed inversion system
To demonstrate the potential application of our split intein C31 integrase in conditional expression of recombination activity, we designed an invertible genetic system based on an inversion substrate plasmid, pC31-invPB. Recombination between attP and attB sites in pC31-invPB inverts the orientation of a constitutive promoter sequence (Biobrick J23104), thereby switching expression from RFP to GFP ( Figure 4A). The E. coli strain DS941/Z1/pC31-invPB was co-transformed with two vectors, each expressing one of the two split-intein integrase fragments. Expression of Integrase-E N -I N was placed under the control of . Controlling the function of an invertible promoter system using split intein-regulated C31 integrase activity. (A) Design of a recombinaseoperated switch using an invertible promoter reporter system. The constitutive promoter sequence (cyan rectangle; Biobrick J23104) is flanked by C31 integrase attP and attB sites (grey arrows) that are arranged in inverted orientation. In its default state, the promoter constitutively drives the expression of a red fluorescent protein (rfp) gene (pink arrow). The T1 terminator sequence (red squares) located immediately upstream of the promoter prevents transcriptional read-through to the green fluorescent protein (gfp) gene (green arrow). Upon integrase-catalysed site-specific recombination, the orientation of the promoter is reversed to allow the expression of GFP and block RFP production. A Biobrick ribosomal binding site, RBS, (RBS B0034, blue rectangles), is positioned 5 of the rfp and gfp genes to drive optimal translation of the synthesized mRNAs. (B) Conditional expression of C31 integrase fragments and split intein-mediated reconstitution of activity. The two split-intein integrase fragments are expressed from two inducible promoter vectors. Arabinose (ara) induces expression of Integrase-E N -I N under the control of the pBAD promoter (pFEM149), whilst anhydrotetracycline (aTc) induces expression of I C -Integrase-E C under the control of the pTet promoter (pFEM148). Posttranslational trans-splicing of the split-intein integrase fragments generates the functional reconstituted integrase (C31.Int ␣ ). Catalysis of attP x attB the P BAD promoter, and expression of I C -Integrase-E C was placed under the control of the Ptet promoter ( Figure 4B). Co-expression of the split-intein fragments (and thus recombination) is dependent on the presence of both of the inducers arabinose and anhydrotetracycline (aTc) ( Figure  4C). No recombination (GFP expression) was observed unless both split-intein components were expressed (when aTc and arabinose were added to the growth medium; column IV. The split-intein regulated serine integrase can be deployed as an effective tool for transient inversion of genetic regulatory modules (Figure 4). The ability to reconstitute integrase activity from inactive protein fragments using split inteins could allow faster control of site-specific recombination by specific activation of intein splicing. This could be achieved by means of light-sensitive protein domains that regulate intein splicing (39,40) or small molecule ligands that act directly on the split inteins (41) to control the protein-protein association step that precedes protein transsplicing. Fast, accurate switching on of enzyme activity in vivo could be achieved if methods for efficient conditional activation of protein trans-splicing were available (42). An engineered photo-activatable gp41-1 split intein system was recently shown to work in bacterial cells (43), demonstrating the feasibility of this approach. Others have demonstrated the use of pH changes (44) and temperature (45) as tools for regulating intein functions in vitro. In our system, conditional splicing would enable fast activation of recombination by reconstituting integrase from the precursors which are already present. Further development of these technologies can enhance capacity for building orthogonal logic gate components for conditional gene expression regulation and genome engineering applications, and add to the existing tools for programmable cellular functions (46,47).

Intein-mediated assembly of integrase-RDF fusion recombinase and catalysis of attR x attL recombination
We recently showed that recombination between attR x attL sites can be catalysed efficiently by artificial proteins in which the recombination directionality factor (RDF) is fused to the integrase using a short peptide linker. The fusion recombinase C31.integrase-gp3 catalysed attR x attL recombination efficiently (24) ( Figure 5A). This system allows more predictable regulation of integrase-catalysed conversion of attP/attB to attR/attL and vice versa, with potential applications in building genetic switches and recombinase-based counters.
To demonstrate the use of our split-intein regulated system in attR x attL recombination, we split C31.integrase-  Non-covalent association of Integrase-E N -I N and I C -Integrase-E C -gp3 via the intein domains (I N and I C ) is followed by a trans-splicing reaction that gives the fusion recombinase C31.Int ␣ -gp3. (C) Assay of recombination activities (galK colour assay) of integrase constructs on attR × attL deletion substrate (pC31-delRL). The test substrate is identical to pC31-delPB ( Figure 3A), except that the attP and attB sites are replaced by attR and attL respectively. (D) The in vivo recombination assay and analysis are as described in Figure 4C. gp3 between the ␤9 and ␣I domains of the integrase (Figure 5B), at the same position as described above for C31 integrase itself. To test for recombination activity, we used a test substrate (pC31-delRL) similar to the attP x attB substrate (pC31-delPB) but with the attP and attB sites replaced by attR and attL, respectively (Supplementary Table  S1). The sites are arranged in the head-to-tail orientation such that deletion of the galK gene leads to a reaction product that is smaller in size than the starting substrate ( Figure  5C). As expected, neither integrase ( Figure 5D, lane 2) nor co-expression of the Integrase-E N -I N and I C -Integrase-E C (lane 4) were able to catalyse attR x attL recombination. In contrast, co-expression of the intein-tagged integrase N-extein (Integrase-E N -I N ) and C-extein-gp3 (I C -Integrase-E C -gp3) results in efficient attR x attL recombination (lane 5). The extent of recombination by the trans-spliced fusion recombinase (C31.Int ␣ -gp3) was lower than that catalysed by native C31.integrase-gp3 (lane 3), in line with our observations for wild-type integrase and its trans-spliced analogue (see above).

Time courses of recombination by wild-type and trans-spliced integrase
To characterize further the effects of the EYCFN mutations introduced into the reconstituted integrase (C31.Int ␣ ) after trans-splicing, we carried out a time course comparison of recombination catalysed by reconstituted and wild-type integrase. To do this, we made a plasmid expressing the expected product of trans-splicing, with the EYRIQ sequence in wild-type C31 integrase replaced with EYCFN. We refer to this construct as C31.Int ␤ . Recombination activity was measured using the invertible promoter reporter system (see Figure 4). The extent of recombination was monitored by measuring the progress of switching from RFP to GFP expression, using single-cell flow cytometry (Figure 6A). Escherichia coli DS941 cells containing the pC31-invPB substrate were transformed with appropriate vector plasmids expressing wild-type C31 integrase, C31.Int ( Figure 6B); full length C31 integrase with the EYCFN mutations, C31.Int ␤ ( Figure 6C); and reconstituted transspliced C31 integrase, C31.Int ␣ ( Figure 6D). Cells were cultured with appropriate antibiotic selection and sampled at 8, 9, 10, 10.5, 11, 11.5, 12 and 24 h. The expression of GFP following recombination by wild-type integrase was observable early and peaked at about 10.5 h; a slight further increase in GFP expression was observed at 24 h (Figure 6B). The progress of the reaction by the C31 integrase with the EYCFN mutations, C31.Int ␤ , was slightly slower, showing a steady increase in GFP expression from 9 to 11.5 h and an even higher level of expression after 24 h. The mutations thus caused a slight decrease in integrase activity ( Figure 6C). In contrast, increase in GFP expression was noticeably slower when protein trans-splicing (C31.Int ␤ ) was required to reconstitute the integrase and activate recombination activity ( Figure 6D). This is consistent with our interpretation of the lower activity of trans-spliced integrase in the experiment shown in Figure 3.

CONCLUSION
Serine integrases promote efficient directional DNA sitespecific recombination, and thus have major potential applications in genome engineering and metabolic pathway engineering (9,11,(48)(49)(50)(51). They have also been incorporated into the design of cellular state machines and biocomputing devices (4,8,11). Here we have demonstrated a novel method for regulation of serine integrase activity by making activity dependent on post-translational trans-splicing of two integrase extein components. Our experiments demonstrate that the split-intein regulated recombination system can potentially be used to toggle between two DNA states in which the forward reaction is catalysed by a serine integrase, and the reverse reaction by an integrase-RDF fusion. In a practical application, reconstitution of the integrase and the integrase-RDF fusion could be mediated by two pairs of orthogonal split inteins. Several types of split inteins have been described in the literature (52,53), and further characterization of their properties would make them available as orthogonal functional parts. Recently, mutations of the Npu DnaE split intein have been identified that affect the specificity, reaction rates, and tolerance to changes in the flanking residues (16,17). Use of these intein variants that tolerate a wider range of amino acid residues around the splice site would require fewer changes to the extein sequence of the target integrase, thereby reducing the potential risk of introducing deleterious mutations and making it easier to find suitable split positions. The assays were carried out using the invertible promoter system described in Figure 4, in which integrase-catalysed inversion switches expression from RFP to GFP. (B) Wild-type C31 integrase. (C) Full length C31 integrase with 'EYCFN' mutations (C31.Int ␤ ). (D) Trans-spliced C31 integrase formed in situ (C31.Int ␣ ). Integrases were expressed from appropriate plasmid vectors as follows: (B) C31 Integrase, pFEM141; (C) EYCFN C31 Integrase, pFEM204; (D) Splicing precursor proteins Integrase-E N -I N , pFEM136 and I C -Integrase-E C , pFEM137. The extent of recombination at the indicated time points was monitored by measuring the amount of GFP formed.