The protein–protein interactions required for assembly of the Tn3 resolution synapse

The site‐specific recombinase Tn3 resolvase initiates DNA strand exchange when two res recombination sites and six resolvase dimers interact to form a synapse. The detailed architecture of this intricate recombination machine remains unclear. We have clarified which of the potential dimer–dimer interactions are required for synapsis and recombination, using a novel complementation strategy that exploits a previously uncharacterized resolvase from Bartonella bacilliformis (“Bart”). Tn3 and Bart resolvases recognize different DNA motifs, via diverged C‐terminal domains (CTDs). They also differ substantially at N‐terminal domain (NTD) surfaces involved in dimerization and synapse assembly. We designed NTD‐CTD hybrid proteins, and hybrid res sites containing both Tn3 and Bart dimer binding sites. Using these components in in vivo assays, we demonstrate that productive synapsis requires a specific “R” interface involving resolvase NTDs at all three dimer‐binding sites in res. Synapses containing mixtures of wild‐type Tn3 and Bart resolvase NTD dimers are recombination‐defective, but activity can be restored by replacing patches of Tn3 resolvase R interface residues with Bart residues, or vice versa. We conclude that the Tn3/Bart family synapse is assembled exclusively by R interactions between resolvase dimers, except for the one special dimer–dimer interaction required for catalysis.


| INTRODUC TI ON
The resolvases of the bacterial transposons Tn3 and γδ are closely related DNA site-specific recombinases and founder members of the serine recombinase family Stark, 2014).
They function to resolve cointegrate intermediates of replicative transposition (Grindley, 2002), and are characterized by stringent topological selectivity: recombination (resolution) is licensed only when two recombination sites (res) are oriented in direct repeat (head to tail) within a supercoiled DNA molecule. Each res has three binding sites for resolvase dimers ( Figure 1a); one of these (site I, the crossover site) is centred on the bonds that are broken and rejoined during recombination. In a productive reaction, the two res sites and six bound resolvase dimers assemble into an interwound synapse ( Figure 1b). Within this complex, the dimers at site I interact and undergo large conformational changes to form a tetramer, which catalyses DNA strand exchange by a rotational mechanism (Stark et al., 1989;Li et al., 2005;Rice, 2014). Four more resolvase dimers, bound at sites II and III (the accessory sites), do not participate in catalysis but have essential regulatory and architectural roles within the synapse. Resolvase can mediate synapsis of two sets of accessory sites when site I is deleted, whereas site I-bound wild-type (wt) resolvase dimers do not make a stable synapse, or promote recombination, in the absence of the accessory sites (Watson et al., 1996;Grindley et al., 2006). However, "activated" resolvase gain-of-function mutants have been characterized, which can synapse and recombine two copies of isolated site I (Arnold et al., 1999;Sarkis et al., 2001;Burke et al., 2004;Olorunniji et al., 2008;Rice, 2014).
Tn3/γδ resolvase (185/183 amino acid residues) has two domains ( Figure 2). The ~45-residue C-terminal domain (CTD) recognizes and binds to sequence motifs at the ends of each dimer-binding site in res. The ~140-residue N-terminal domain (NTD) contains the active site for DNA strand cleavage and rejoining; it also contains all the surfaces currently known to mediate subunit interactions (Grindley, 2002). A specific interaction between NTDs, seen in crystal structures of γδ resolvase, was shown to form both the dimer that binds to res and the solution dimer (Sanderson et al., 1990;Hughes et al., 1993;Rice and Steitz, 1994a;Yang and Steitz, 1995;Figure 2a).
Using activated resolvase variants (see above) several structures of the catalytic tetramer have been solved, without DNA or in complex with site I DNA that has been cleaved by the enzyme to create double-strand breaks (Nöllmann et al., 2004;Li et al., 2005;Kamtekar et al., 2006;Figure 2d). Formation of the tetramer involves substantial rearrangement of the dimer structure and new contacts between the two dimers.
Several other types of interactions between resolvase subunits were observed in the crystal structures (Sanderson et al., 1990), but only one of these has been shown to have biological relevance; here, we call it the R (for Regulatory) interface (previously referred to as the 2-3′ interaction/interface). The R interface is not required for binding of resolvase dimers to DNA, nor is it involved directly in catalysis at site I, but it is required for assembly of the synapse.
Mutations at any of four key R interface residues (R2, R32, K54, and E56; Figure 2a,c) can abolish recombination (Hughes et al., 1990), and there is evidence that activity of the catalytic tetramer at site I requires R interactions between the resolvase subunits bound at site I and subunits bound at the accessory sites (Grindley, 1993;Burke et al., 2004;Li et al., 2005). Previous analysis has defined relationships between the positions of resolvase subunits on the res DNA within the synapse and requirement for R proficiency (Murley and Grindley, 1998). A number of structural models of the synapse incorporating R interactions have been proposed (Rice and Steitz, 1994b;Murley and Grindley, 1998;Sarkis et al., 2001;Grindley, 2002;Rowland et al., 2002;Mouw et al., 2008;Rowland et al., 2009).
However, the involvement of additional protein-protein interfaces has not been ruled out.
Productive synapsis by Sin resolvase, a related serine recombinase (~30% identity; Rowland et al., 2002), has been shown to require a similar R interface, involving two residues (F52 and R54) that align with Tn3/γδ resolvase R interface residues K54 and E56 (Mouw et al., 2008;Rowland et al., 2009;Rice, 2014). The Sin and γδ resolvase R interfaces are structurally remarkably similar, despite the different amino acid residues involved. Sin-mediated synapsis of resH recombination sites was shown to require yet another specific interface, between the CTDs of Sin dimers. Sin mutagenesis and crystallographic data led to the first detailed structural model F I G U R E 1 Site-specific recombination by Tn3 resolvase. (a) The Tn3 res site (114 bp) contains three resolvase-binding sites (I, II, and III; the lengths of the sites and inter-site spacers are shown). Each binding site is recognized by a resolvase dimer (simplified cartoon on right). The NTDs, CTDs, and R interface regions are labeled. The staggered red line indicates the staggered break (with 2-nucleotide overhangs) when the DNA strands are cut during recombination. (b) Synapsis and recombination. Two res sites, directly repeated in a supercoiled plasmid, are brought together by interactions between res-bound resolvase dimers, forming a synapse. Three negative supercoil nodes are stabilized by interactions between res sites II and III (larger grey shape) and are trapped by interactions between the dimers bound at site I (smaller grey shape), forming a catalytic tetramer. This tetramer then catalyses recombination by breaking, exchanging and rejoining DNA strands. The DNA product is a 2-noded catenane of a complete recombination synapse, incorporating CTD-CTD interactions ( Figure 2b) between Sin dimers bound at the accessory site (site II) and R interactions between subunits at site I and site II (Mouw et al., 2008;Rowland et al., 2009). The Sin and Tn3/γδ catalytic tetramers are structurally similar (Keenholtz et al., 2011;Trejo et al., 2018), so synaptic site I -site I interactions were modeled using the γδ resolvase-site I synaptic structure 1ZR4. A speculative model of the Tn3/γδ synapse of two full res sites incorporating a similar CTD-CTD interface was suggested (Mouw et al., 2008), but there is no current experimental evidence for this type of CTD interaction in the Tn3/γδ system (Hughes et al., 1990;Grindley, 1993).
High-resolution structural analysis of the complete synapse for any regulated site-specific recombination system has not been achieved to date. Biochemical characterization of the network of protein-DNA and protein-protein interactions in such complexes is also very challenging. For the Tn3 resolution synapse, there are 12 identical resolvase subunits which form six dimers (three dimers bound to each res). To probe the resolvase architecture in the synapse, we need to be able to identify the dimers bound at two or more specific sites, to assess whether and how they interact. Our chosen strategy required two types of resolvase that form structurally similar dimers on their binding sites in res, but are otherwise "orthogonal"; that is, each resolvase dimer recognizes only its own binding site sequences in a two-res substrate plasmid, and can make interactions with dimers of its own type, but not with the other type.
By assaying for recombination proficiency, we would then be able to determine whether any particular arrangement of dimers can form the network of interactions needed to build a productive synapse.
We therefore sought a recombinase which is similar to Tn3 resolvase in its structure and synapse architecture, but which differs substantially in its DNA sequence recognition specificity and in surfaces implicated in synapse assembly. A previously uncharacterized resolvase from the sandfly-borne human pathogen Bartonella bacilliformis (GenBank ABM45303.1; hereinafter called Bart resolvase) was selected as being potentially suitable. Bart and Tn3 resolvases have only 49% of amino acid identity, but can be aligned with no gaps (Figure 3a) along with other related proteins that together form a "Tn3/γδ/Bart family" of serine resolvases ( Figure S2). The Bart and Tn3 res sites are similarly arranged, but the sequence motifs predicted to be recognized by the resolvase CTDs are very different (Figure 3b,c), suggesting that F I G U R E 2 Interfaces between resolvase subunits. (a) Resolvase dimer bound at site I of res (pdb 1GDT; Yang and Steitz, 1995). The NTDs of the two resolvase subunits are colored salmon and light pink; the NTD arm regions (119-140) are rendered as backbone. R interface residues are colored yellow, green, and blue (see part C). The CTDs are colored to distinguish the "hook" (gold; residues 141-145) from the helix-turn-helix (HTH) module (light orange; residues 146-183). CTD surface residues mutated to rule out a hypothetical CTD-CTD interface are highlighted in red (149,150,153,157,167; further details in Figure S1). The numbering (1-7) of the DNA base pairs bound by the light pink subunit corresponds to that shown in Figure 3c (see also Figure S1b). (b) Sin resolvase CTD-CTD synapsis interface at site II (pdb 2R0Q; Mouw et al., 2008). The CTD hook (gold; residues 147-151) is distinguished from the HTH module (light orange or green; residues 152-200), as in part A. The NTDs are not shown. (c) The R interface between two γδ resolvase dimers (salmon and light pink, light and dark olive; pdb 2RSL; Rice and Steitz, 1994a). The inset shows the main sidechains involved in the R interface. The canonical R interface tetrad is shown in yellow (R2, K54) and green (R32, E56). Ten surface residues (colored yellow and blue) were replaced to switch the interface specificity to mimic Bart resolvase (for details, see text and Figure S5). The three residues colored green (R32, E56, and D59) were not altered, as these are identical in Tn3/γδ and Bart resolvase. (d) γδ resolvase catalytic tetramer in a cleavage complex with two copies of site I (pdb 1ZR4: Li et al, 2005). Note that the R interface residues (colored as in part c) do not overlap other functional interfaces in the tetramer. Subunits are colored and rendered as in a and c (a) each CTD type (Tn3 or Bart) would specifically recognize its cognate motif. Furthermore, the Tn3 and Bart CTDs have very low amino acid sequence identity (22%), so it is likely that any synaptic CTD-CTD or NTD-CTD interactions would be disrupted by substitution with a CTD of the other type. The Bart and Tn3 resolvase proteins also differ significantly at the R interface. Two of the key Tn3 resolvase R interface residues (R2 and K54) are different in Bart resolvase (A2 and R54); one of these differences, when introduced into Tn3/γδ resolvase as an R2A mutation, is known to disrupt the interface and block recombination (Hughes et al., 1990;Wenwieser, 2001). We therefore hypothesized that Bart and Tn3 resolvases both make R interactions, but that their R interface surfaces would be incompatible. We further speculated that it might be possible to substitute the R interface residues of Tn3 resolvase with the equivalent Bart resolvase residues and vice versa, thereby giving one resolvase the R interaction specificity of the other.
Here, we present experiments that rule out any required proteinprotein interactions of the CTDs. We report the design and testing of a modular synthetic res site, which can be tailored to support either Tn3 or Bart resolvase-mediated recombination by changing the DNA sequence motifs recognized by the CTDs. Using hybrid res sites containing both Tn3 and Bart sequence motifs, and hybrid Tn3/Bart resolvases, we demonstrate a requirement for specific R interface interactions by resolvase dimers at each of the three res binding sites (I, II, and III), and conclude that these, together with the interaction at site I that forms the catalytic tetramer, are necessary and sufficient for productive synapse assembly. We anticipate that the technologies developed in this work will help us to map the full network of protein-protein interactions in the Tn3 resolution synapse.

| Synapsis-defective mutants of Tn3 resolvase
Several types of interfaces between resolvase subunits and dimers have been hypothesized in published models of the Tn3/γδ resolution synapse (see Introduction). In preliminary experiments to identify amino acid residues that might contribute to synaptic F I G U R E 3 Tn3 and Bart resolvases and res sites. (a) Amino acid sequence alignment of Tn3 and Bartonella (Bart) resolvases. The 24 C-terminal residues of the natural Bart resolvase (Genbank ABM45303.1) were deleted in the Bart resolvase used in all our experiments ( Figure S1). Secondary structure elements are indicated by narrow boxes and labels above the sequences, and the CTDs are in a wide box.
Mutations of Tn3 resolvase CTD residues that were made in order to test for CTD synapsis are indicated in green. The Tn3 resolvase R interface residues R2, R32, K54, and E56, and the catalytic serine S10, are denoted by bold numbers above the sequences. The T B and B T NTDs have altered residues at the ten positions indicated by blue or red letters, respectively (see Results section and Figure S5). (b) res sites. resT and resB are synthetic sites for Tn3 and Bart resolvases. They differ only in the six motifs (bold red or blue arrows) recognized by the helix-turn-helix (HTH) part of the CTD. Other sequences derive from WT Tn3 res (pale red), except the restriction sites used to construct the various resT/resB hybrids (boxes labeled E: EcoRI, and B: BglII) and two base pairs flanking site II in resB (pale blue). The full sequences of the sites are in Figure S3. To reduce off-target binding, resT and resB motifs were made fully uniform at positions 1-5, by substituting all the noncanonical bases (yellow bars) present in the WT sites. The DNA motifs contacted by the NTD arm and the CTD hook modules (small open and closed arrowheads, respectively) are similar in the Tn3 and Bart sites (see part C and Figure S3). (c) CTD target selectivity. The sequence logos compare base frequencies (linear scale) in the six half-sites of the synthetic and natural res sites (excluding positions 11-13 of site IIR and site IIIL; see Figure S3). Crosses indicate bases that distinguish the resT and resB motifs: the corresponding single base change from γδ to Bart at these positions reduces CTD affinity by ≥100-fold (positions 2, 4, 6) or 9-fold (position 5) (Rimphanitchayakit and Grindley, 1990). Filled squares indicate bases present in all motifs interfaces, we screened a library of Tn3 resolvase random mutants, using E. coli assays similar to those described previously for isolating synapsis mutants of Sin (Mouw et al., 2008;see Experimental Procedures). These assays identify mutants that are proficient in binding at site I, but defective in synapsis of sites II and III (and thus defective in recombination). In contrast to our analogous Sin screen, which identified synapsis-defective mutations at the CTD synaptic interface but not at the NTD R interface (see Introduction), all the effective Tn3 resolvase mutations mapped to the NTD--to the four canonical R (2-3′) residues R2, R32, K54, and E56, and also the non-R residue E128 (Supplementary Information SI 1). The synapsis defect of the E128K mutant might be because it fails to bind properly at res site III, as has been shown for the γδ resolvase E128K mutant (Hatfull et al., 1987) and which we have observed in binding assays in vivo (data not shown). In summary, these mutagenesis-screening results are consistent with an essential role for the R interface in assembling the Tn3 resolution synapse, and provided no evidence that CTD synaptic interactions are important. Further evidence against synaptic CTD interactions is presented below.

| A uniform res site design for Tn3 and Bart resolvases
The strategy outlined in the Introduction, to probe inter-subunit interactions in the synapse, requires a uniform res site that can be customized to bind Tn3 or Bart CTDs at specific positions simply by altering the CTD-binding DNA sequence motifs. The predicted natural Bart res site is very similar to that of Tn3 res in its organization of binding sites and "spacer" sequences ( Figure S3). We therefore designed two synthetic 114-bp res sites, resT, and resB, for the Tn3 and Bart resolvases, respectively, which differ only in the 7-bp sequences recognized by the helix-turn-helix (HTH) motifs of the CTDs, the other sequence elements all having Tn3 character ( Figure 3b). The design included sites for restriction endonucleases to facilitate the construction of hybrid res sites. The Bart and Tn3 CTD-binding motifs differ at four positions ( Figure 3c).
Previous analysis indicates that the Tn3 resolvase CTD should not recognize the Bart motifs (Rimphanitchayakit and Grindley, 1990), and we predicted that likewise, the Bart resolvase CTD would not recognize the Tn3 motifs.

| Activities of Tn3 and Bart resolvases on resT and resB substrates
The alignable parts of the Tn3 and Bart resolvase CTDs are only 22% identical, and the Bart CTD is 28 residues longer at its C-terminus ( Figure 3a). This C-terminal extension is not needed for recombination activity, and we truncated it by 24 amino acids, making the two CTDs similar in size. All of the experiments described below used this truncated Bart CTD.
Recombination activity was assayed in E. coli. The resolvase proteins were expressed in cells containing a two-res substrate plasmid with Tn3-and/or Bart-type binding sites ( Figure 4a). The two res sites flank a galK indicator gene that is deleted by resolvase-mediated recombination (resolution) (Burke et al., 2004;Rowland et al., 2009).
Efficient (>90%) resolution results in white colonies on MacConkey/ galactose agar indicator plates, whereas red colonies indicate lower efficiency (or no) resolution (Arnold et al., 1999;Burke et al., 2004;. and Bart resolvases cannot complement each other to resolve a resT × resB substrate (presumably because they cannot make required synaptic protein-protein interactions).

| NTD-CTD hybrid resolvases: CTD interactions are not required for synapsis
To dissect the role of the resolvase CTD in synapsis, we made two Tn3- coli. The test substrate plasmid contains two res sites (resT, resB, or modified versions as described in the text) in direct repeat, flanking a galK marker gene. In E. coli cells (strain DS941; galK − ) containing appropriate resolvase expression plasmids, recombination between the two res sites creates two product circles. The circle containing galK has no origin of replication and is lost during subsequent cell divisions. Colonies in which recombination (deletion) was efficient (approaching 100%) lack GalK and are thus uncolored ("white") on galactose-MacConkey agar indicator plates, whereas colonies with inefficient or zero recombination retain GalK activity and are red. Intermediate recombination efficiency gives pink colonies. Cells from a pool of >100 red or pink colonies were cultured further and plasmid DNA was recovered to measure % recombination, as described in Experimental Procedures. These % values therefore report on a later time point, when the substrate has been exposed to the resolvases in the cells for a further ~10 cell generations. The indicator plate pictures show assays 6, 10, and 7 from Figure 6a ("white," "pink," and "red" colonies, respectively). (b) Specificity of recombination of resT and resB substrates by Tn3, Bart, and hybrid resolvases. Tn3 and Bart resolvases efficiently resolve only their fully cognate substrates (lines 1 and 2). They do not complement each other to resolve a resT × resB substrate (line 3). Hybrid resolvases B-T and T-B efficiently resolve substrates recognized by their CTDs (lines 4 and 5). The resT × resB substrate is resolved efficiently when two resolvases with the same type of NTD, but different CTDs are expressed (lines 6-8).
Enzymes are named according to the identities of the NTD and CTD domains, in the order NTD-CTD. For example, B-T resolvase has a Bart NTD (B) attached to a Tn3 CTD (T). "+" in pale yellow boxes indicates pale-colored ("white") colonies in the in vivo assay, and thus, 100% or near-100% resolution of the test substrate; magenta boxes indicate red colonies, for which % recombination was estimated as described in the Experimental Procedures section (numerical values in the boxes; see Figure 4a) with its CTD). Nor did the two hybrid resolvases complement each other to recombine this substrate (Figure 4b To confirm these conclusions, we created two more hybrids, with the Tn3 resolvase NTD linked to the CTDs of two other well-characterized resolvases, Tn21 and Tn552. These CTDs interact weakly with Tn3 DNA sequence motifs, despite having low amino acid sequence similarity to Tn3 resolvase (Halford et al., 1985;Rowland and Dyke, 1989;Grindley, 2002;Figures S2 and S4). The hybrid resolvases (with Tn3 NTD and either Tn21 or Tn552 CTD) act like wt Tn3 resolvase in our assays; they recombine resT × resT but not resB × resB, and they complement T-B resolvase to recombine the resT × resB substrate ( Figure S4). Any essential Tn3 CTD-CTD interactions would have to be replaceable with Tn3-Bart, Tn552-Bart, or Tn21-Bart interactions in the above experiments, which seems very unlikely.
As a further test, we introduced multiple mutations on the solvent-accessible (non-DNA-binding) surface of the Tn3 resolvase CTD (Figures 2a and 3a). Pairs or triplets of amino acid residues were mutated to the negatively charged residue glutamate, which should disrupt any CTD-CTD interactions by charge repulsion as well as surface non-complementarity (Supplementary Information SI 2; Figure S1). However, these mutants were all indistinguishable from wt Tn3 resolvase in E. coli assays for recombination and for binding at the individual sites I, II, and III (data not shown).
We conclude that, unlike the Sin synapse (see Introduction), the Tn3/Bart synapse does not involve synaptic interactions between the CTDs.

| Resolvase dimers targeted to specific binding sites
To test whether resolvase dimers with Tn3 or Bart resolvase CTDs can be targeted to specific binding sites within res, we made hybrid res sites containing all six possible arrangements of Tn3 and Bart resolvase dimer-binding sites. These are denoted as btt, tbt, etc. The letters in each triplet indicate binding sites I, II, and III in that order; t represents a dimer-binding site with Tn3 motifs, and b represents a dimer-binding site with Bart motifs.
We then constructed four test substrate plasmids, which are similar to the resT × resT substrate except that one or both res sequences are hybrids with one b site instead of a t site (btt, tbt, ttb; Analogous results were obtained with substrates that are similar to resB × resB except that one or both res sequences are hybrids containing a t binding site (Figure 5b). However, some of these hybrid res substrates were fully or substantially resolved when just a single resolvase with Bart CTDs (B-B or T-B) was expressed (lines 1 and 2). This suggests that the Bart CTD can bind weakly to t sites, aided by the extensive cooperative NTD interactions within the synapse.
Such relaxed sequence specificity has been proposed to be a general feature of protein-DNA interactions in nucleoprotein complexes (Hall and Halford, 1993).
We also used these hybrid-res substrates to target catalytically defective resolvase mutants to specific dimer-binding sites. The active site serine (S10 in Tn3 resolvase) attacks the scissile DNA phosphodiester at the center of res binding site I during strand exchange; mutating it to alanine abolishes recombination activity while preserving DNA binding and subunit interactions (Olorunniji and Stark, 2009). Therefore, resolution should be inhibited if S10A NTDs are targeted to one or both copies of site I, but should remain efficient if S10A NTDs are targeted to sites II or III. This was the observed result with both Tn3 and Bart resolvase S10A mutants (Figure 5a, line 4; Figure 5b, line 4). These data also clearly indicate that when Tn3 CTDs are present, they out-compete, and thus exclude, Bart CTDs that may bind weakly at non-cognate t sites in the hybrid-res substrates (see above).
In summary, resolvase dimers can be targeted to specific binding sites in hybrid res substrates, according to the identity of their CTDs.
Resolution remains efficient when catalysis-defective S10A mutant NTDs are targeted to sites II and/or III, but is inhibited by S10A NTDs targeted to site I, confirming targeting specificity.

| All resolvase dimers in the synaptic complex must make R interactions
A key justification for our choice of Bart resolvase was the prediction that Bart resolvase NTDs would be unable to make effective R (2-3′) interactions with Tn3 resolvase NTDs (see Introduction). R interactions between catalytic and regulatory dimers (i.e., dimers bound at site I and the accessory sites II-III, respectively) are required for resolution by Tn3/γδ resolvase (Hughes et al., 1990;Grindley, 1993;Wenwieser, 2001;Burke et al., 2004). In our experiments, resolution efficiency was greatly reduced when the NTDs bound at site I and at the accessory sites were different (see Figure 6b, line 4), supporting the hypothesis that the Tn3 and Bart R surfaces are functionally incompatible.
To examine further the role of the R interface in synapse assembly, we made R interface mutants. The Tn3 resolvase R interface mutant R2A (Hughes et al., 1990;Wenwieser, 2001) has no resolution activity on the resT × resT substrate (see Figure 6a The res sites are each named by three lower-case letters representing the binding sites I, II, and III (t = Tn3 CTDbinding site, b = Bart CTD-binding site).
In part a, all or most of the binding sites are t-type; in part b most sites are b-type.
The results are depicted as described in the Figure 4 legend. Single-residue mutations are indicated by superscripts on the relevant NTD labels, and a black patch or black cross in the cartoons. For further details and interpretation, see main text strikingly comparable to the "reverse" experiment with mutant Tn3 resolvase NTDs described above.
In summary, our assays indicate that resolvase dimers at all three binding sites in res make R interactions, but the effect of R interface mutants at site II is less pronounced than at sites I and III. We also conclude that Bart and Tn3 resolvases may use similar R interfaces to build their respective synapses.

| Swapping the R interaction specificities of Tn3 and Bart resolvases
We predicted that, if only R interactions are involved in building the synapse, it might be possible to make the Tn3 NTD function like a Bart NTD, and vice versa, by changing only R interface residues.
Starting with no concrete information about the critical residues and interactions of the Bart R interface, we made systematic profile comparisons of Tn3-like and Bart-like resolvase sequences, relating these to the structure of the γδ resolvase R interface (Figure 2c). We then designed mutations of residues on a small surface patch of each protein, comprising the canonical R interface residues and other nearby residues predicted to be involved. For each protein, these comprise a total of 10 substitutions in parts of three surface loops near the R interface, sparing adjoining residues that pack in the core of the NTD (see Figure 2c, Figure  We also performed these experiments "in reverse"; that is, T B or B T NTDs were targeted to res binding sites II and III, and unmod- Some early models of the synapse (Krasnow et al., 1983;Boocock et al., 1986;Sarkis et al., 2001;Rowland et al., 2002) proposed that dimers bound at the res accessory sites (II and III) might mediate synaptic interactions via the "catalytic" dimer-dimer interface (as at site I) or another type of NTD-NTD interface. To test these proposals, we used the substrates tbb × ttt and btt × ttt (Figure 6c). Bart NTDs We conclude that, apart from the well-characterized catalytic tetramer-forming interactions at site I of res, the productive synapse is assembled using only R interactions between resolvase dimers.

| D ISCUSS I ON
We used Tn3 and Bart resolvases, and Tn3-Bart hybrids, to target distinct types of resolvase dimers to distinct binding sites in the resolvase recombination site res, allowing us to map protein-protein interactions within the synaptic recombination intermediate. Our results lead us to conclude that the architecture of the productive synapse is achieved by resolvase subunits making three types of essential protein-protein interactions.
(1) At each binding site of res, F I G U R E 6 Recombination by resolvases with R interface patch mutations. (a) Resolvases with R interface specificity-swap patch mutations, T B -T and B T -B, resolve resT and resB substrates, respectively, indicating restoration of functional R interfaces (lines, 3, 5, 8, and 10). In contrast, single-residue "swap" mutations at the R interface abolish recombination (lines 2 and 7). The four relevant "white colony" results from Figure 5 are shown here again for comparison (lines marked with *). (b) R interactions between dimers at the crossover sites and at the accessory sites of res. NTDs attached to a Bart CTD (p2, left-hand cartoon in each pair) are targeted to the crossover site (site I), whereas NTDs attached to a Tn3 CTD (p1, right-hand cartoon in each pair) are targeted to the accessory sites II and III, in a btt × btt substrate. The two-resolvase data are arranged into three sets of four (rows 1-4, 9-12, and 15-18, above the dashed grey line), to facilitate informative comparison of reactions; in each set, each resolvase pair differs from the one above or below by substitution of one NTD. The small boxes (ticked or crossed) indicate whether or not the R interfaces targeted to the crossover sites have the same "flavour" (i.e., Tn3 or Bart) as the R interfaces targeted to the accessory sites, and are therefore predicted to make a functional interaction. The "swapped specificity" T B and B T NTDs function preferentially together with Bart and Tn3 NTDs, respectively. One-resolvase controls are below the dashed grey line in the relevant column. See main text for further details and interpretation. The data are from single assays, all conducted in parallel under identical conditions (except controls 5-8 and 13). The error values represent the intrinsic RMS sampling error in determining % recombination by counting (see Experimental Procedures). Triplicate assays (three independent experiments) for two examples gave mean ± standard error as follows: 3.28 ± 0.09% (reaction 4) and 31.3 ± 2.3% (reaction 12). (c) No evidence for essential non-R synaptic interactions between dimers at sites II/III. T B resolvase complements B-B resolvase to restore resolution activity on a tbb × ttt substrate.
There is no rescue with a btt × ttt substrate. See text for details the CTDs of two resolvase subunits recognize and bind to DNA sequence motifs at the ends, and the NTDs interact to make a dimer ( Figure 2a). (2) The dimers bound at two interacting res sites make R interactions with each other (Figure 2c), thus, intertwining the two DNA sequences and holding them together.
(3) The two dimers bound at the two copies of site I interact and rearrange, forming a catalytic tetramer (Figure 2d).
The two key innovations that allowed us to reach these conclusions were our characterization of the Bart resolvase recombination system and our "patch mutagenesis" strategy to remodel the R interface. The Bart resolution system was chosen for this work because of its potential as an "orthogonal" yet structurally similar system to that of Tn3 (see Introduction). A crucial success in our experimental strategy was the achievement of almost complete DNA sequence specificity switching of Tn3 and Bart resolvases by swapping their CTDs. There had been previous attempts to alter the DNA sequence specificity of serine recombinases related to Tn3 resolvase, by mutations or substitution of the CTD with one from a different recombinase Avila et al., 1990;Grindley, 1993;Boocock et al., 1995;Murley and Grindley, 1998;Schneider et al., 2000), but none of these variant proteins had completely switched specificity from the original target DNA sequence to a new one. More complete specificity changes have been obtained by replacing the resolvase CTD with structurally unrelated zinc finger or TALE DNA-binding domains (Akopian et al., 2003;Gaj et al., 2013;Stark, 2014), but the substantially altered structures/ sizes of these hybrid proteins make them entirely unsuitable for mapping interactions within the natural Tn3 synapse.
Mutation of just one amino acid residue at the R interface of either Tn3 or Bart resolvase to the other's residue at that position (R2A and R54K respectively) was sufficient to render the resolvase inactive; but remarkably, for each resolvase, activity was restored by adding a further nine mutations, to create a Bart-like R surface on Tn3 resolvase, and a Tn3-like R surface on Bart resolvase. When the recombination synapse contained mixtures of Tn3 and Bart resolvase NTD dimers, these patches of mutations were sufficient to give one type of resolvase NTD the functionality of the other. Our results thus imply that the R interface is solely responsible for holding the two res sites together in the synapse, until formation of the catalytic resolvase tetramer at site I.
We note that our conversion of the Tn3 resolvase R interface to Bart specificity (T B ) was more effective than conversion of the Bart R interface to Tn3 specificity (B T ) (e.g. Figure 6a, compare lines 3 and 5; 8 and 10). This is unsurprising to us because the mutations of Tn3 resolvase were informed by high-resolution crystal structures (see Introduction), whereas we have no structural data for Bart resolvase.
The interface mutations were entirely based on a priori structural analysis and prediction; the effectiveness of the switching could undoubtedly be improved by iterative mutagenesis-screening methods.
Synapsis of the accessory sites (II and III; Figure 1a) can occur in the absence of site I and is thought to precede the site I-site I interaction (Boocock et al., 1986;Watson et al., 1996). This II-III synaptic complex has also been used in synthetic recombination systems, to impose topological selectivity on other recombinases (Kilbride et al., 1999(Kilbride et al., , 2006Grainge et al., 2000;Olorunniji et al., 2012).
Involvement of "catalytic-type" tetramers ( Figure 2d) in synapsis at sites II and III was previously ruled out by protein crosslinking data (Murley and Grindley, 1998;Wenwieser, 2001), and our new results imply that only R interactions between DNA-bound dimers are required.
Sin resolvase is distantly related to Tn3/γδ resolvase and has a very different organization of protein-binding sites in its res sequence (Rowland et al., 2002). Biochemical and crystallographic analysis of Sin-DNA complexes revealed a synaptic CTD-CTD interface at the res accessory sites (Figure 2b) and essential R-type contacts between subunits bound at the accessory sites and site I. It was also shown that the R interface plays no part in accessory site synapsis. These features were incorporated into a structural model of the complete synapse (Mouw et al., 2008;Rowland et al., 2009) Previous studies with γδ resolvase identified essential R interactions within the synapse, and models of the synapse architecture postulated R interactions between pairs of dimers bound at sites II and III in the same res site (Grindley, 1993;Murley and Grindley, 1998;Sarkis et al., 2001). Our results support crucial synaptic roles for the R interface, involving dimers bound at all three sites in res, and furthermore imply that the R interface is the only "new" interface made in the synapse by dimers bound at sites II and III. Thus, we deduce that the R interface makes the primary bridging contacts between the two res sites in the synapse.
The effects of R interaction deficiency at site II are generally less than at site I or site III, consistent with earlier proposals that the role of the resolvase dimers bound at site II might be primarily "architectural" (including bending the path of the DNA double helix; Blake et al., 1995;Soultanas et al., 1995), whereas the R interaction between sites I and III might be critical for events leading to catalysis of DNA strand exchange. The site I-site III R interaction might simply position the site I-bound resolvase dimers optimally for strand exchange, or it might have an additional allosteric role to stimulate catalytic activity (see for example Grindley et al., 2006).
Our results have significantly clarified the nature of the resolvase interactions in the recombination synaptic complex, but do not yet allow us to build a complete structural model that is fully supported by experimental evidence. Completion of this task will require an "interaction map" for each resolvase subunit within the synapse, and will therefore require resolvase subunits (rather than dimers) to be targeted to specific loci in substrate res sites. This is likely to be more challenging because of the highly cooperative protein-protein interactions in the synapse, but we are confident that the tools and methodologies presented here will allow us to achieve the necessary targeting specificity.

| Recombination substrates
Recombination sites resB and resT (synthesized by GeneArt) were cloned in the polylinker vector pMTL23 (Chambers et al., 1988), and hybrid sites were constructed by fragment swaps using the EcoRI and BglII restriction sites between sites I, II, and III (Figure 3b, Figure S3).
The in vivo resolution substrates are low copy number plasmids (pSC101 origin) similar in construction to pGal (res × res) (Burke et al., 2004; see also Prorocic, 2009 All these plasmids ("p2" in Figures) contain the following sequence upstream of the resolvase start codon: CTTTGTTATATACATATG.
The upstream sequences containing the ribosome binding site in the p1 and p2 plasmids were different so as to give compatible resolvase expression levels. Tn3 resolvase coding sequences are as described (Burke et al., 2004). The original Bart resolvase sequence (GenBank ABM45303.1) is from Bartonella bacilliformis KC583. The Bart resolvase coding sequences used here were synthesized (by GeneArt).
Both Tn3 and Bart resolvase coding sequences have restriction sites placed at regular intervals to allow for future fragment exchanges.
Random mutations in resolvase reading frames were introduced by PCR-based methods (Burke et al., 2004).

| In vivo recombination assays
The complementation assay has been described Mouw et al., 2008). In brief, chemically competent cells of the E. coli strain DS941 containing one or two resolvase expression plasmids (see above) were transformed with substrate plasmid DNA, and colonies were grown on selective MacConkey/galactose agar indicator plates (Arnold et al., 1999;Burke et al., 2004). Pale-colored ("white") colonies indicate near-complete resolution; that is, the galK gene has been deleted from all or most (typically >90%) copies of the multi-copy substrate plasmid (confirmed by gel electrophoresis and/ resolution means that we counted >1,000 red colonies and ≤1 white colonies; ">99.9%" resolution means >1,000 white colonies and zero red colonies. In Figure 6b, ± sampling errors (SE) are given, as: (1 − p)/n), where r and w are the counts of red and white colonies, respectively, p = w/(w + r), and n = w + r. Quantitative comparisons discussed in the text are from sets of assays done in parallel (using identical growth conditions, etc.), including all data in Figure 4b (except rows 3,7), 5a (except row 5), 6a, 6b (except rows 5-8,13), and 6c. The standard error in nine representative triplicate assays done as independent experiments was between 0.1% and 2.4% recombination (% recombination in the range 0.8%-71%).

| Design of R interface patch mutations
The aim was to modify a small surface patch of the Tn3 or Bart NTD, to switch the interface specificity ( Figure 2 and Figure S5). The two mutants described (T B and B T ) represent an initial test of a structure-informed "one-step" interface redesign; they are not products of stepwise optimization or selection. No structures are available for Bart resolvase; our redesign of the interfaces relied only on structures for the γδ resolvase interface (Figure 2c; Rice and Steitz, 1994a) and the very similar Tn3 resolvase interface (Montaño and Rice, unpublished). Bart resolvase differs from Tn3/γδ resolvase at two of the four canonical R interface residues: R2 is replaced by A2, and K54 by R54. R2 is at the heart of the Tn3 and γδ R interfaces: it contacts D59′ across the interface, and stacks between R2′ and R32 ( Figure 2c). Although we hypothesize that the role of R54 in the Bart R interface might be similar to that of R2 in the Tn3 resolvase interface (and analogous to the role of R54 in the Sin R interface), our designs assume only that the NTDs share similar core structures.
To evaluate candidate residues for mutation, we systematically inspected side-chain contacts in the γδ resolvase interface (Figure 2c), and compared profiles of residue frequencies in groups of resolvases similar to Bart (A2, R54) or to Tn3/γδ resolvase (R2, K54). These two groups seem to represent two primary "flavours" of R interface in the diverse Tn3/Bart resolvase family ( Figure S2).
The profile comparisons highlighted additional residues with a potential role in interface specificity ( Figure S5). A key difference between Bart and Tn3/γδ resolvases is in surface loops 2 and 4 (positions 29-32 and 55-58). In γδ resolvase, only R32 and E56′ make close contact across the interface, while residues 29-31 and 57′ make limited contacts and were not highlighted in loss-of-function screens (Figure 2c). Bart-group resolvases retain R32 and E56, but have long conserved side-chains at positions 29-30 and 57 ( Figure S5), suggesting that these residues might engage in the interface. At position 53, the Bart group typically have small residues (Ala or Gly), while in γδ resolvase, two bulky M53 sidechains pack on the interface (Figure 3c). A smaller sidechain might allow closer approach of the B-helices, shifting the entire interface. We replaced M53 with A53 (as in the Bart group profile), not G53 (from Bart resolvase), to minimize any risk of destabilizing the α-helix. Following similar logic, we replaced a total of ten surface residues, with zero net change in the predicted surface charge ( Figure S5). Changes were made only at residues directly involved in the γδ 2-3′ interface, residues supporting these (I3 and F4), or residues assigned as potential Bart interface contacts (29-31); the underlying foundation of NTD core residues is mostly conserved.
For the patch mutant Bart NTD (B T ), we made ten changes (all to Tn3 residues: Figure S5). Eight of these changes are the exact reverse of those made in the Tn3 NTD (B T ). One extra change was made (L52R, to preserve amphipathic character in helix B, breaking a long run of hydrophobic residues--LQLLLMK--and adding one unit of charge). At L3 no change was justified. Neither of these residues is likely to make any direct interface contacts (see Figure 2c).

| Bioinformatics
Sequence alignments and profile analyses made extensive use of Jalview (Waterhouse et al., 2009), and standard NCBI tools.