Synapsis of loxP Sites by Cre Recombinase*

Cre recombinase catalyzes site-specific recombination between 34-bp loxP sites in a variety of topological and cellular contexts. An obligatory step in the recombination reaction is the association, or synapsis, of Cre-bound loxP sites to form a tetrameric protein assembly that is competent for strand exchange. Using analytical ultracentrifugation and electrophoresis approaches, we have studied the energetics of Cre-mediated synapsis of loxP sites. We found that synapsis occurs with a high affinity (Kd = 10 nm) and is pH-dependent but does not require divalent cations. Surprisingly, the catalytically inactive Cre K201A mutant is fully competent for synapsis of loxP sites, yet the inactive Y324F and R173K mutants are defective for synapsis. These findings have allowed us to determine the first crystal structures of a pre-cleavage Cre-loxP synaptic complex in a configuration representing the starting point in the recombination pathway. When combined with a quantitative analysis of synapsis using loxP mutants, the structures explain how the central 8 bp of the loxP site are able to dictate the order of strand exchange in the Cre system.

Site-specific recombinases from the tyrosine recombinase family are used by bacteria and yeast to mediate the integration, excision, resolution, and inversion of DNA segments to carry out a wide variety of biological functions (1,2). In the simplest systems, typified by the bacteriophage P1 Cre recombinase (3) and the Saccharomyces cerevisiae Flp recombinase (4), only the recombinase enzymes and DNA substrates containing 34-bp recombination sequences are required for efficient recombination. In others, such as the bacteriophage -integrase (-int) 3 (5) and the Escherichia coli XerC and XerD recombinases (6), efficient recombination requires more complex recombination sites and the activities of auxiliary proteins that tightly regulate the forward and reverse reactions. A remarkable property of both the simple and complex systems is their ability to efficiently synapse (associate) recombining sites that can be located far from one another on the same chromosome or, for bacteriophage integration, are in separate DNA molecules.
Synapsis of recombining sites has been visualized in tyrosine recombinase systems using a variety of complementary experimental approaches. For example, synaptic -int att and Cre-loxP complexes have been observed by electron microscopy (7,8) and as slower-migrating bands on non-denaturing polyacrylamide gels (9,10); atomic force microscopy has been used to probe synaptic complexes in the Int, Flp, and Cre systems (11,12); and single molecule light microscopy has recently been used to observe formation of synaptic complexes during -int recombination (13). In addition, crystal structures have been reported for synaptic Cre-DNA complexes (14,15) and for a truncated -int-DNA complex (16). Despite the fundamental importance of this initial step in the recombination reaction pathway, however, there have been no quantitative studies of the energetics of synapsis in these systems.
Cre recombinase has proven to be a useful model system for biochemical and structural investigations (17). During synapsis, Cre sharply bends the loxP sites to form an assembly that is stabilized by protein-protein interactions between Cre subunits bound to the associating sites (14,15). Once this complex is formed, the recombinase subunits are committed to cleavage and exchange of one of the two pairs of DNA strands. As shown in Fig. 1, however, this initial step in the recombination pathway is actually quite complex. There are four distinct synaptic complexes that can form between two Cre-bound loxP sites if the sites are arranged in an approximately co-planar configuration; two of the complexes are anti-parallel (I and II in Fig. 1c) and two are parallel (III and IV in Fig. 1c).
Structural, biochemical, and topological data strongly support a model in which only an anti-parallel alignment of sites will lead to efficient strand exchange (2,(17)(18)(19). One might predict that Cre-loxP site-specific recombination should be able to initiate recombination equally well from either of the anti-parallel synaptic configurations. However, several independent experimental approaches have demonstrated that Cre preferentially exchanges the loxP bottom strands (the black strands in Fig. 1) first to generate the Holliday junction (HJ) intermediate, and then exchanges the top (red) strands during HJ resolution to form recombinant products (19 -23). Indeed, recent data further indicate that preferential formation of com-plex II in Fig. 1c is responsible for this bias in the order of strand exchanges (23).
The structural and biochemical bases for this asymmetry in the Cre recombination pathway have only been partially addressed, leaving a number of unanswered questions. For example, why should one of the anti-parallel synapses be preferred over the other? What is the structure of this synaptic complex? What is the binding affinity associated with Cre-loxP synapsis? Is synapsis best viewed as an interaction between the protein surfaces from two rigid Cre-bound loxP sites, or do the dynamics of loxP bending also play a role? One of the primary gaps in our mechanistic understanding of the Cre-loxP recombination pathway has been the lack of an experimental system to quantitatively study the requirements for efficient synapsis. A second issue has been the lack of a structural model that represents the actual starting point in the recombination pathway. Crystal structures of synaptic complexes containing Cre bound to modified loxP sites (14,15) have provided a great deal of general insight but do not provide satisfactory explanations to these questions. Indeed, one of the reported Cre-DNA synaptic complex structures (15) argues against the currently accepted model for initiation of recombination.
Here, we report advances in understanding both the biochemistry and structural biology of synapsis in the Cre-loxP system. Using analytical ultracentrifugation and electrophoresis techniques, we have developed assays to measure equilibrium binding constants for Cre-mediated synapsis of loxP sites. Using these methods, we have examined the effects of pH, divalent ions, Cre mutants, and loxP modifications on synapsis. By using a cleavage-deficient Cre mutant that is able to synapse loxP sites as well as wild-type Cre, we have also determined two independent crystal structures of pre-cleavage Cre-loxP synaptic complexes that reveal the actual starting point for the recombination pathway. Together, the biochemical and structural data explain why Cre preferentially forms a bottom-strand cleavage synaptic complex to initiate recombination and provide new insights into this important first step in the recombination pathway.

EXPERIMENTAL PROCEDURES
Protein and DNA Purification-Wild-type Cre and Cre mutants were overexpressed and purified as described (24). The extinction coefficient for Cre at 280 nm was determined to be 49.1 mM Ϫ1 cm Ϫ1 by comparing protein absorbance before and after denaturation in 3 M guanidine hydrochloride (25). Oligonucleotides for ultracentrifugation experiments (loxPcent in supplemental Fig. S1) were synthesized by the Keck Facility at Yale University with the 5Ј-dimethoxytrityl group attached and purified by reversed-phase high-performance liquid chromatography as described (24). Following detritylation and concentration, oligonucleotides were annealed in 10 mM Tris-HCl, pH 8, 0.5 mM EDTA, 100 mM NaCl (annealing buffer) and further purified on a 5-ml type I hydroxyapatite column (Bio-Rad) with sodium/potassium phosphate elution to remove small amounts of excess single strand DNA. The extinction coefficient at 260 nm was determined experimentally as 661 mM Ϫ1 cm Ϫ1 for duplex 44-mer loxP sites by comparing the absorbance before and after total DNase digestion, and using literature values for the extinction coefficients of individual nucleotides. For crystallization, oligonucleotides were purified by reverse-phase high-performance liquid chromatography as described above, but the hydroxyapatite column step was omitted. For gel-based synapsis assays, oligonucleotides were gel-purified.
Analytical Ultracentrifugation-Sedimentation equilibrium ultracentrifugation experiments were performed at 20°C using a Beckman Optima XL-A analytical ultracentrifuge fitted with an AN-60 Ti rotor and with six-sector centerpieces. To determine synapsis affinities, Cre-loxP complexes were prepared at 0.6, 0.3, and 0.15 M 44-mer loxP site DNA and 1.8, 0.9, and 0.45 M Cre, respectively, in standard recombination buffer (20 mM sodium HEPES, pH 7.5, 150 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol; measured density of 1.004 g/cm 3 ). These ratios provide an excess of 0.5 Cre subunits per loxP half-site, which we found to be important in ensuring saturation of loxP sites throughout the experiment. Reference chambers were filled with either buffer alone or buffer containing 0.6, 0.3, and 0.15 M Cre, respectively. Similar binding constants resulted from both approaches, but the fits were slightly better when the concentration of excess Cre was matched in the reference chambers due to the small absorbance contribution of unbound Cre that affected the absorbance baselines. The data fit in Fig. 2 and supplemental Fig. S2 were measured using the latter approach. Samples were centrifuged at 6,000, 9,000, 12,000, and 15,000 rpm, and radial absorbance scans were measured in step mode at 260 nm after 12 and 14 h, at which time equilibrium had been achieved.
Radial absorbance data were fit to a monomer-dimer equilibrium model using the program SEDPHAT (26), with a fixed "monomer" molecular mass of 104,000 Da, based on two 38.5-kDa Cre subunits and one 27-kDa loxP-containing 44-mer duplex. The partial specific volume (v ) for Cre (0.73 cm 3 /g) was determined from the protein amino acid composition. The v value for loxP DNA (0.59 cm 3 /g) was determined from sedimentation equilibrium analysis of the unbound DNA duplex, which behaved as an ideal monomer. The v value of the Cre 2 loxP monomer (0.69 cm 3 /g) was estimated as the massweighted average of the partial specific volumes for Cre and the loxP site. An estimated error for the equilibrium constant was determined from a 1,000-iteration Monte Carlo simulation, as implemented in SEDPHAT.
To examine the effects of buffer composition on synapsis, Cre-loxP complexes were prepared in the appropriate buffers at 400 nM loxP and 1.2 M Cre. Centrifugation and absorbance scans were performed as described above, and relative synapsis association constants were estimated from global fits of the radial scans. These affinity constants are less well determined than the global fits involving three different Cre-loxP concentrations, but this approach allowed the simultaneous comparison of a large number of buffer conditions in the multichamber centerpieces.
Sedimentation equilibrium analysis of Cre alone was performed using 6 M Cre in 20 mM sodium Hepes, pH 7.5, 275 mM NaCl, 1 mM dithiothreitol, with or without 5 mM MgCl 2 . Samples were centrifuged at 12,000, 15,000, 18,000, and 21,000 rpm, and radial absorbance scans were measured at 280 nm after equilibrium had been reached at 14 h. Absorbance data fit well to a single species model corresponding to a Cre monomer.
Synapsis Gel Electrophoresis Assay-Ten picomoles of 54-mer top strand was 5Ј-end labeled with 32 P and annealed to the complementary bottom strand by slow cooling (1°C/min) in a thermal cycler. Synapsis reactions were carried out in 1ϫ NCB buffer (20 mM sodium Hepes buffer, pH 7.5, 5 mM MgCl 2 , 150 mM NaCl, 2 mM dithiothreitol, and 50 g/ml sheared salmon sperm DNA). Each reaction in a given titration contained 200 pM radiolabeled 54-mer loxP DNA, 1 M Cre or Cre mutant, and varying concentrations of unlabeled 54-mer DNA. Reactions were incubated for 20 min at 20°C, mixed with loading dye (0.001% bromphenol blue, 0.001% xylene cyanol, and 3% Ficoll) and loaded onto a 6% non-denaturing polyacrylamide gel (29:1 acrylamide:bisacrylamide) that had been pre-run at 10 V/cm for 45 min at 20°C, with constant temperature maintained by a circulating water bath. The gel polymerization and running buffer was 50 mM Hepes acid, 25 mM Tris base, at pH 7.0. Similar results were obtained when 1 mM EDTA was present in the buffers. After electrophoresis at 13 V/cm for 2 h at 20°C, gels were dried and quantitated by PhosphorImager analysis (Amersham Biosciences).
Standard electrophoretic mobility shift assays with wild-type Cre and 54-mer loxP DNA performed under identical conditions were used to verify the assignments of free DNA, DNA bound by one or two Cre subunits, and synaptic complex (as annotated in Fig. 3). The fraction synapsed (f) was calculated as the counts in the synaptic complex band divided by the sum of the synaptic complex counts plus the unsynapsed Cre 2 loxP counts in the corresponding lane and then plotted against the total DNA concentration. A monomer-dimer equilibrium model was utilized to fit the data using Equation 1, where c is the total concentration of 54-mer loxP site, and K is the monomer-dimer dissociation constant.  Fig. S1) in 20 mM sodium acetate, pH 5, 30% 2-methyl-2,4-pentanediol, and 20 mM CaCl 2 . Both orthorhombic and trigonal crystal forms grew from the same conditions. Diffraction data were measured at the Advanced Light Source beamline 8.2.1 using a Quantum4 charge-coupled device detector at 100 K and processed using the HKL suite (27). The trigonal crystal form was sensitive to radiation damage, resulting in a lower completeness and higher overall R sym value compared with the orthorhombic form (Table 3).
Both synaptic complex crystal forms were initially phased using molecular replacement with AMORE (28) and the recombinase-DNA half-sites from the covalent intermediate structure 1CRX (29) as a search model. The resulting starting models were then optimized using rigid-body refinement of the individual protein domains and the 13-bp inverted repeat DNA arms at 3-Å resolution, resulting in correlation coefficients of 0.88/0.81 and R-factors of 0.35/0.38 for the orthorhombic and trigonal crystal forms, respectively. The 8-bp spacer region encompassing positions 4Ј through 4 was omitted from the model until these DNA residues could be unambiguously fit into unbiased electron density maps. Iterative cycles of positional refinement with REFMAC (30) and model building in O (31) were performed as the resolution was gradually increased to the limits of the individual data sets.
The A -weighted 2F o Ϫ F c and F o Ϫ F c electron density maps showed clear density for the bases and for the backbone of the omitted central 8 bp of the loxP site. This region was initially modeled as (dAdU) 4 until the final stages of refinement to avoid bias in the choice of sequence directionality as described before (32). The sequence assignment was then made based on the presence or absence of difference map densities for Gua-N2 and Thy-C5, which uniquely identifies the direction of the pseudo-palindromic spacer. Electron density maps following refinement of the (dAdU) 4 model were consistent with only one direction for the loxP spacer, corresponding to a bottom strand cleavage configuration. Refinement results are summarized in Table 3, and coordinates for the orthorhombic and trigonal forms have been deposited in the Protein Data Bank, with accession codes 2HOF and 2HOI, respectively.

Cre-loxP Synapsis Is pH-dependent but Does Not Require
Divalent Cations-Our initial attempts to monitor Cre-loxP synapsis as a function of loxP concentration were based on native PAGE analysis. Slower migrating bands representing higher order complexes have been reported in the Cre system (9,33), and these bands were shown to contain primarily synaptic complexes formed by association of two Cre-bound loxP sites (9). When we attempted to extract binding constants using this approach, the resulting affinities were weaker than anticipated and the results were difficult to reproduce in separate experiments. After considering a number of plausible explanations (e.g. buffer requirements and stability of the synaptic complex during electrophoresis), we decided to first establish the basic requirements and affinities involved in Cre-loxP synapsis using sedimentation equilibrium analytical ultracentrifugation (hereafter referred to as AU).
The primary advantage of AU is the ability to monitor complex formation under equilibrium conditions using well defined experimental parameters, such as buffer components, pH, and temperature (34,35). The equilibria involved in synapsis of loxP sites are complex if cooperative binding of Cre subunits to the loxP site is also considered. To simplify the analysis, we performed AU experiments with 44-bp loxP sites and saturating concentrations of Cre, taking advantage of observations that Cre binds to the loxP site with a subnanomolar K d (36). Under these conditions, the loxP sites are fully occupied with Cre subunits, and synapsis can be modeled as a simple monomer-dimer equilibrium. In this model, two associating 104-kDa Cre 2 loxP complexes (monomers) associate to form a 208-kDa synaptic complex (dimer) (Fig. 1b). The 44-bp loxP site used (supplemental Fig. S1) dominates absorbance at 260 nm, with Cre contributing only a few percent of the signal at this wavelength. The AU experiment therefore monitors the oligomeric state of the loxP site, to a good approximation.
The results of an AU experiment for wild-type Cre-loxP synapsis are summarized in Fig. 2 and in supplemental Fig. S2. Three different loxP concentrations (each with saturating amounts of Cre) were monitored at four different rotor speeds, and the equilibrium binding constant was determined by a global fit to the resulting twelve radial distributions. The excellent fits and randomly distributed residuals indicate that the monomer-dimer model is a good description of loxP site association under these conditions. The experiment described in Fig. 2 was performed using our standard recombination assay buffer, which is at near neutral pH and contains both monovalent (sodium) and divalent (magnesium) cations. To test the effects of different buffer conditions on the ability of Cre to synapse loxP sites, we performed AU experiments using the buffer conditions shown in Table 1. We were particularly interested to learn if divalent ions are required for efficient synapsis. It was originally reported that divalent ions and small polyamines increase the efficiency of the overall Cre recombination reaction (3), and a role for divalent ions in stabilizing the synaptic complex has been suggested (14). We were therefore surprised to find that divalent ions have only a small effect on synapsis of loxP sites and that a physiological concentration of monovalent cations appears to be sufficient for this stage of the recombination pathway. A role for divalent ions or polyamines in facilitating cooperative Cre-loxP binding has also been suggested (37), which would not be manifested in these experiments, because we are forcing full occupancy of the loxP sites by providing Cre in excess.
We also wanted to determine if Cre-loxP synapsis is pH-dependent. As shown in Table 1, synapsis is efficient from pH 5.5 to 7.5, which is consistent with our previous observations that the in vitro Cre recombination reaction is robust over a wide pH range, including that of most restriction enzyme buffers, which are typically buffered at pH 7.5-8.0. At pH 8.5 (Table 1) and above (data not shown), Cre-loxP synaptic complexes become much less stable. We do not yet understand the cause of the breakdown in synapsis that occurs at pH 8.5, but ionization of  two Cys residues located in the helical core-binding domains of Cre could alter the ability of these domains to interact in the synaptic complex.
The negative effect of alkaline pH on synapsis provides an explanation for the difficulties we encountered in performing synapsis assays using native PAGE, because these types of gels are commonly run in buffers with pH 8 -9. Indeed, we had been using Tris borate-EDTA buffer systems at pH 8.5-9. As expected, AU synapsis experiments performed in 0.5ϫ or 1ϫ Tris borate-EDTA result in only small amounts of synaptic complex formation even at high loxP concentration (data not shown).
In addition to studying synapsis of loxP sites under a variety of conditions, we also used AU to determine the oligomeric state of Cre in solution. An early report based on gel filtration and glycerol gradient ultracentrifugation indicated that Cre was monomeric but that magnesium ions cause the sedimentation coefficient to increase (3). We conducted sedimentation equilibrium experiments for 6 M wild-type Cre using buffers composed of 20 mM sodium Hepes, pH 7.5, 275 mM NaCl, 1 mM dithiothreitol, both with and without 5 mM MgCl 2 . In both cases, we found that Cre is monomeric, with no evidence of dimer or higher order multimer formation (supplemental Fig.  S3). Similar results were obtained with buffers containing 125 mM NaCl. The increase in sedimentation coefficient previously observed in the presence of magnesium ions is therefore most likely due to a change in shape of the free protein in solution.
Based on the findings from the AU experiments described above, we modified the gel-based synapsis assay by using Tris-Hepes or Tris-Hepes-EDTA at pH 7-7.5 as a buffer system. Although a number of neutral pH buffers have been described for native PAGE (38), we favored a system where two buffering components bracket the desired pH. An example of this improved assay for wild-type Cre-loxP synapsis is shown in Fig.  3. The effective K d of 14 nM obtained from a fit of the binding data agrees reasonably well with the value of 10 nM obtained using AU. This indicates that the synaptic complexes formed are sufficiently stable under electrophoresis conditions to be able to extract useful quantitative binding data.
Cleavage Is Not Required for Efficient Synapsis-One possible concern in interpreting the synapsis experiments shown in Figs. 2 and 3 for wild-type Cre is that cleavage and strand exchange should be occurring at some rate during the course of the experiment. In particular, some fraction of the "dimer" complex representing synapsed loxP sites may also include a contribution from HJ intermediates that have formed between sites undergoing recombination. To address this issue, we measured the ability of cleavage-deficient active site mutants to synapse loxP sites using the gel-based assay. As shown in Table 2 and supplemental Fig. S4, the Cre K201A mutant is able to synapse FIGURE 3. Electrophoretic mobility shift assay to monitor wild type Cre-loxP synapsis. a, example of a synapsis titration with increasing concentration of unlabeled loxP site. Cre is present at saturating concentrations to ensure full occupancy of loxP sites at all loxP concentrations. b, fit of a monomer-dimer equilibrium model based on quantitation of the bands shown in a. The fraction synapsed and estimated errors were calculated from at least three independent titrations for wild-type Cre-loxP synapsis and for the results summarized in Table 2 and Fig. 7.

TABLE 1 Effects of pH and divalent ions on Cre-loxP synapsis
K d values were determined by sedimentation equilibrium ultracentrifugation of 400 nM Cre 2 loxP at four rotor speeds as described under "Experimental Procedures." K d values are relative to the standard buffer condition, which was also used for the experiment in Fig. 2. All buffers also contained 1 mM dithiothreitol. Sodium was used as counterion for Hepes and acetate buffers and chloride was used for Tris. loxP sites as well as wild-type Cre, despite being catalytically inactive for cleavage or recombination (23). An interesting difference observed in the synapsis of loxP sites by wild-type Cre versus the K201A mutant is a lower binding plateau (f max ϳ 0.75) at high concentrations of loxP site for the mutant. To determine if this is due to the inability of Cre K201A to cleave the loxP site, as opposed to an inherent property of this particular mutant, we tested the ability of this protein to synapse loxP sites containing a 5Ј-bridging phosphorothioate at the bottom strand scissile phosphate. Cre K201A can cleave this modified loxP site, due to the lowered pK a of the 5Ј-thiol leaving group versus the normal 5Ј-hydroxyl group (23). We found that Cre K201A synapsis of the phosphorothioate-containing site is nearly identical to that observed for wildtype Cre (supplemental Fig. S4 and Fig. 3). Thus, although cleavage is not a prerequisite for efficient synapsis of loxP sites, there is a subtle difference observed when a cleavage-competent recombinase enzyme is used.

Buffer
Although this result indicates that cleavage of loxP sites is not required for synapsis, we were still interested to know if HJ intermediates were present at significant steady-state levels under our experimental conditions in the case of wild-type Cre. To test this, synapsis binding assays were rapidly quenched with SDS/proteinase K at various time points, and the extracted DNA was analyzed by native PAGE. Using our standard recombination buffer conditions (Table 1), we were not able to detect HJ intermediates Ͼ1% of the total loxP DNA, whereas accumulation of HJ intermediates by the HJ-generating Cre A312T mutant (20) was clear (data not shown). This result is consistent with early reports that HJ intermediates are present at very low steady-state levels in wild-type Cre-loxP recombination, under standard buffer conditions (20).
In contrast to the K201A mutant, two other inactive Cre mutants, Y324F and R173K, are both severely defective in synapsis (Table 2). This was not anticipated in our earlier work, where we used these mutants bound to symmetric loxP variants to determine crystal structures of synaptic complexes (14). Presumably, the high concentrations used for crystallization were able to overcome the decreased affinity between mutant Cre-bound sites. Neither Tyr-324 or Arg-173 is directly involved in the proteinprotein interface formed between two associated loxP sites, therefore they must each affect synapsis indirectly. Arg-173 forms an intimate, double hydrogen bonding interaction with the scissile phosphate, which is thought to stabilize the transition states of the catalytic phosphoryl transfer steps during recombination (29). Mutation of the corresponding residue in Flp recombinase (Arg-191) to lysine causes altered DNA bending of the frt site (39). Given that the loxP sites are sharply bent in synaptic Cre-loxP crystal structures, a deficiency in DNA bending could be partly responsible for the synapsis defect of the R173K mutant.
The observation that Cre Y324F is defective in synapsis is more difficult to rationalize based on current   (14). One possible explanation for the synapsis defect is that the Tyr-324-phosphate interaction may be important in stabilizing the helix L-M-N region of the recombinase catalytic domain (29). This region plays key roles in both synapsis and catalysis, because the C-terminal helix-N forms domain-swapping interactions across the synaptic interface, and helices L and M contribute the conserved active site residues Trp-315 and Tyr-324, respectively. Indeed, the Cre ⌬C331 mutant (deletion from Glu-331 to the C terminus) is unable to synapse loxP sites even at relatively high loxP concentrations (Table 2). We also tested the Cre A36V substitution, which was the first synapsis-deficient Cre mutant reported (9). Ala-36 forms part of the protein-protein interface formed between core-binding domains in the synaptic complex and substitution of valine would be expected to disrupt this region of the structure, based on existing models. This mutant has wild-type loxP-binding activity (40) and is able to cleave suicide substrates (23). As expected, we found that Cre A36V is defective in synapsis, with K d Ͼ 500 nM.
Structural Basis for Selective DNA Bending in the Cre-loxP Synaptic Complex-To determine the structure of a Cre-loxP synaptic complex representing the start of the recombination pathway, we made use of our observations that Cre K201A does not cleave loxP, but forms synaptic complexes as efficiently as the wild-type protein. Using this Cre mutant and unmodified loxP site DNA, we were able to grow diffraction quality crystals of the Cre K201A-loxP complex in orthorhombic and trigonal crystal forms, with resolution limits of 2.4 and 2.6 Å, respectively ( Table 3). The orthorhombic form is isomorphous with the structure of the covalent Cre-DNA intermediate (29) and contains one Cre 2 -loxP complex in the crystallographic asymmetric unit. In this case, a crystallographic 2-fold symmetry axis relates the two Cre-bound loxP sites in an anti-parallel aligned synaptic complex. The trigonal crystal form contains the entire [Cre 2 -loxP] 2 synaptic complex in the asymmetric unit. As expected, this complex contains two anti-parallel loxP sites related by local (non-crystallographic) 2-fold symmetry.
Both structures were determined by straightforward molecular replacement and rigid body refinement of the protein domains and the 13-bp inverted repeats from the Cre/loxA structure (PDB code 1CRX (29)) where the central 8 bp of the loxP sites were removed to eliminate model bias in this region. The central loxP sequence was fit into electron density as dA-dU base pairs, which allowed for refinement of the structure without committing to a particular direction for the loxP site (32). This simple approach is possible because the central 8 bp of loxP are pseudo-palindromic with respect to purine and pyrimidine nucleotides. The final assignment of sequence was then made from inspection of electron density maps late in the refinement process, where we observed a unique pattern of Thy-C5 and Gua-N2 densities that allowed unambiguous identification of the loxP direction. Unbiased electron densities used in the sequence assignment are shown in supplemental Fig. S5. In both structures, the central region of loxP is very well ordered, with strong electron density for the sugar-phosphate backbone and for the bases.
The synaptic structures in the two crystal forms both reveal a bottom strand cleavage configuration, where the loxP site is sharply bent in the left half-site (defined in Fig. 1a) and the recombinase subunit poised to catalyze strand exchange is bound to the right half-site ( Fig. 4 and complex II in Fig. 1c). The identification of which strand is committed for cleavage and exchange in the synaptic complex is based primarily on geometric and stereochemical arguments (14). In the structures described here, the loxP bottom strand trajectory is ideally suited for exchange between the two halves of the synaptic complex to form the HJ intermediate, with no protein-DNA contacts that might inhibit this process. In contrast, the top strand is intimately engaged in a network of interactions with the recombinase subunit and would be unable to migrate to the opposing loxP site even if the corresponding scissile phosphate were cleaved in the synaptic complex. As described for the earlier structures, the "active" and "inactive" active sites are quite similar, with no large differences involving coordination of the scissile phosphate. Positioning of the Tyr-324 nucleophile does differ between the two active sites, a likely consequence of allosteric regulation of catalysis via the domain-swapped C-terminal helices (41).
The structures in the two crystal forms are nearly identical, with a total of three independent views of the synapsed loxP site (one from the orthorhombic form and two from the trigonal form). When the central 8 bp of the loxP site are compared, the all-atom r.m.s.d. between independent sites in the trigonal form is 0.4 Å and the r.m.s.d. for comparison of each of these with the orthorhombic form is 0.8 Å. Similarly, comparison of the three independent active and the three independent inactive recombinase subunits shows r.m.s.d. values of 0.4 -0.6 Å for well ordered C ␣ atoms in residues 40 -320. Most aspects of the recombinase subunits, the active sites, and the Cre-DNA interface are similar to that already described for previous Cre-DNA crystal structures. Here, we limit our discussion to the central region of the loxP site, which is most relevant to understanding why the bottom strand synaptic complex is favored to initiate recombination.
As shown in Fig. 4b, the loxP sites are bent sharply one base pair step away from the top strand scissile phosphate. There is a large negative roll associated with the bend, which is a conse- quence of the major groove opening toward the synaptic interface (Fig. 4a). There is a corresponding compression of the minor groove, which is centered very close to the T3Ј and T4Ј backbones (Fig. 4c). This bend occurs in the left half-site, which is proximal to the inactive recombinase subunit. The right halfsite adopts a relatively undistorted B-DNA conformation, which includes the scissile bottom-strand phosphate that would normally be cleaved to initiate recombination in this step of the reaction pathway. The geometry of the DNA bends observed in both the orthorhombic and trigonal crystal forms described here are similar to those previously observed in the structure of Cre bound to a symmetric lox site (14) but involve entirely different DNA sequences (see "Discussion").
The location and stereochemistry of the sharp loxP bend provide a compelling explanation for the observed preferential formation of a bottom-strand cleavage synapse. Compression of the minor groove would be expected to be more readily tolerated in the A/T-rich left half-site of loxP, relative to the right half-site. This argument is particularly striking when we consider that the center of the minor groove compression is very close to the 4Ј-scissile base (Fig. 4c), which implies that the minor groove functional groups of the 4 versus 4Ј bases (and to some extent the 3 versus 3Ј bases) should largely dictate whether Cre bends the left versus right half-site of loxP. Because right versus left half-site bending implies a difference in bend direction (i.e. complex I versus complex II in Fig. 1c), the resulting synaptic complexes will be poised for top strand or bottom strand cleavage, respectively. Previous observations that addition of a 2-amino group to Ade-4Ј to create a 2,6-diaminopurine-containing loxP eliminates the bottom strand cleavage preference (23) and that swapping of the 4Ј-scissile bases in the lox4 substrate also reverses the preference (21) strongly support this simple model.
Asymmetric Cre-loxP Contacts at the Scissile Bases-Only two recombinase residues interact directly with bases in the asymmetric central region of loxP (Fig. 5). Lys-201 normally interacts with the N3 group of the scissile 4 or 4Ј-purine base via the minor groove and activates the scissile O5Ј leaving group during catalysis (23). This residue has been mutated to alanine in the structures described here. The second interacting residue is Lys-86, which directly contacts the scissile bases via the major groove and is well defined in both structures. In the right half-site of loxP, Lys-86 hydrogen bonds directly to O6 of Gua4, whereas in the left half-site, Lys-86 hydrogen bonds directly to N7 of Ade4Ј. This asymmetric interaction at the ends of the 8-bp central region of loxP is similar to what was observed in the structure of an inactive Cre mutant bound to a loxP-derived HJ substrate (32). In that case, the resolution was higher (2 Å), and an additional water-mediated interaction was observed between the side chain of Lys-86 and the exocyclic N6-amino group of Ade4. We also observe a network of water-mediated interactions involving Lys-86 in the somewhat lower resolution structures described here, but the details of these networks differ slightly in the three independent representations available from the two crystal forms. The well defined set of direct interactions between Lys-86 and the scissile bases in the current structures support the sequence assignment in the central region, because the direct hydrogen bond observed to O6 of Gua4 (ϳ3 Å between Lys-N⑀ donor and Gua-O6 acceptor atoms) would not make sense if that base were adenine.
Despite the distinct hydrogen bonding interactions observed between Lys-86 and the scissile bases in the 4 and 4Ј positions (Fig. 5), this residue does not appear to play an important role in preferential formation of a synaptic complex poised for bottom-strand cleavage (21,23). This result can be rationalized in terms of the synaptic complex structures described here. The two alternative anti-parallel synapses differ by the bend directions of the sites and the DNA sequences distorted to achieve the bends (Fig. 1). However, the alternative complexes (I versus II in Fig. 1c) would be expected to have the same set of asymmetric interactions involving Lys-86. There is no obvious structural reason why the specific interaction of Lys-86 with Ade4Ј would somehow facilitate bending in the left loxP halfsite, or that the unique interaction with Gua4 would somehow disfavor bending in the right half-site.
In principle, Lys-201 could contribute to selecting the observed synaptic configuration (i.e. complex II rather than complex I in Fig. 1c). This residue interacts directly with the 4Ј-purine base in the minor groove where cleavage occurs but is excluded from making a similar interaction in the opposite halfsite of the synaptic complex, covalent intermediate, and HJ intermediate structures (Fig. 5 (17)). Since Lys-201 is substituted by alanine in the structures described here, we cannot establish the nature of its interactions with the loxP half-sites. It is clear, however, that Lys-201 is not required for preferential formation of this synaptic complex. The Cre K201A mutant displays the same bottom-strand cleavage preference with 5Ј-bridging phosphorothioate substrates and fluorescence resonance energy transfer experiments demonstrating a synaptic preference in solution made use of the K201A mutant to prevent cleavage of loxP sites (23).  (17) and are expected to be present in the wild-type Cre-loxP synaptic complex that initiates recombination.
Insights from Synapsis of Symmetric loxP Variants-A complication in interpreting an equilibrium binding constant for synapsis of loxP sites in terms of the specific structural model shown in Fig. 4 is illustrated in Fig. 1. In principle, four distinct synaptic complexes can form between two loxP sites. Two have a relative anti-parallel orientation (I and II in Fig. 1c) and differ by both the bend direction of the sites and by which strand is committed for cleavage and exchange in the synaptic complex. The other two are parallel complexes (III and IV in Fig. 1c), containing loxP sites with opposite bend directions. Although the bottom strand cleavage complex II has been shown to be favored over the top-strand complex I in solution (23), there is currently no experimental evidence to indicate that parallel complexes do not readily form in solution as well. Indeed, parallel synaptic complexes for both the Cre and Flp systems have been observed by electron microscopy (7,42). Thus, the synapsis affinity constant measured in an experiment with loxP sites represents a macroscopic K d , which includes contributions from individual equilibria involving the species shown in Fig. 1c.
To further simplify the interpretation of synapsis binding data and to test mechanistic hypotheses inferred from the Cre-loxP crystal structures, we compared the ability of Cre, Cre K201A, and Cre K86A to synapse the loxP site and the symmetric loxSL and loxSR sites (Fig. 6). The symmetric sites have two useful properties. First, they can form only one type of synaptic complex, because parallel and anti-parallel alignments are all the same. This allows for more direct interpretation of the interaction energies in terms of the DNA sequences involved. Second, because the DNA distortions involved in the synaptic loxP bend are localized to one of the two half-sites (as opposed to involving both half-sites), the two halves of the loxP spacer region can be separated and assayed independently by using symmetric sites. The results of these experiments are summarized in Fig. 7a.
Relative to loxP, synapsis of loxSR sites is less efficient by a factor of ϳ10 -15 for wild-type Cre, Cre K201A, and Cre K86A. This is the trend that one would expect, given the structural model for DNA bending shown in Fig. 4. The G/C-rich halfsites of loxSR are less able to accommodate the DNA bend observed in the A/T-rich left half site, because the 2-amino groups of Gua3 and Gua4 would be inserted into the tightly Core sequences of the modified sites are given in Fig. 6. Dissociation constants were measured using the EMSA method as shown in Fig. 3. compressed minor groove at the bend center. The previously reported crystal structures of the Cre R173K/loxSR and Cre Y324F/loxSR complexes indicate that this is exactly what happens when a synaptic complex is formed between loxSR sites. Presumably, there are no stereochemical alternatives for loxSR bending that would allow synapse formation with a lower energy.
In contrast to the results with loxSR, the synapsis of loxSL sites by Cre, Cre K201A, and Cre K86A is more efficient than synapsis of loxP by a factor of 3-4. In this case, both halves of the loxSL spacer region contain the sequence that readily accommodates a DNA bend in the synaptic complex. Because there are four identical ways of forming the synaptic complex from loxSL sites, the macroscopic K d value measured for loxSL synapsis should be roughly one-fourth the value of the microscopic K d for synapsis of loxP sites to form anti-parallel complex II in Fig. 1. This estimate is based on the assumption that the right half-site sequence of loxP (which is not significantly distorted from B-form DNA when the left half-site is bent) does not contribute substantially to the energetics of synapsis in complex II. If wild-type Cre-loxP synapsis is dominated by formation of complex II, then one might also expect to see a 4-fold difference in macroscopic K d values between loxP and loxSL, based on a simple statistical factor. The experimental ratios K d -loxP/K d -loxSL for the Cre K201A and Cre K86A mutants are both 3.7, suggesting that this is a reasonable model. The same ratio for wild-type Cre is 2.9.
A comparison of loxSL versus loxSR synapsis provides an estimate of the energy difference between bending the alternative loxP half-sites. Using the Cre K86A mutant synapsis data, the ratio of K d values is 41.8, corresponding to a free energy difference of ϳ2.2 kcal/mol at 25°C for bending at the A/T-rich left versus the G/C-rich right half-site of loxP. Because the Cre K86A mutant cannot make potentially discriminating interactions with the scissile bases in loxSL and loxSR, the assumption that the difference in synapsis affinity is primarily derived from DNA bending energetics seems reasonable. The calculated free energy differences for wild-type Cre and Cre K201A are 2.2 and 2.1 kcal/mol, respectively, indicating that Lys-86 interactions with the scissile bases do not have a large affect. Although this analysis is admittedly a simplification, it does show that a quantitative understanding of individual steps in the Cre-loxP sitespecific recombination can yield useful insights into fundamental mechanistic issues.
Synapsis of loxP Mutants-The loxSL and loxSR sites simplify the interpretation of synapsis affinities, but each of these sites involves three base pair changes relative to the loxP sequence (Fig. 6). To probe the synapsis of loxP sites with smaller changes, we tested the ability of Cre, Cre K201A, and Cre K86A to synapse a series of loxP sites modified at the scissile bases (Fig. 7b). The loxP-DAP site, for example, differs from loxP by only a single functional group. In this site, Ade-4Ј is replaced by 2,6-diaminopurine (DAP), and the complementary thymidine residue is left unchanged. The DAP/T base pair is therefore identical to an A/T base pair, except for the presence of an adenine 2-amino group and an additional DAP/T hydrogen bond (Fig. 5). If facile minor groove compression in the left loxP half-site is responsible for selecting which synaptic config-uration is favored, then the loxP-DAP site should de-stabilize complex II (Fig. 1c) due to the insertion of the exocyclic N2amino group into the narrowed minor groove. Indeed, loxP-DAP synapsis is less efficient than loxP for all three Cre variants (Fig. 7b). Interestingly, the decrease in synapsis affinity is most extreme for the Cre K86A mutant, despite the identical major groove composition in loxP and loxP-DAP.
We made a similar substitution in loxP to give the loxP-GG site, where the A/T base pair at the 4Ј position is replaced by G/C. Like loxP-DAP, the loxGG site would insert a 2-amino group into the minor groove near the bend center of the left half-site. However, loxGG also presents a different functional group for interaction with Lys-86 in the major groove. We observed a similar decrease in affinity for wild-type Cre and Cre K201A synapsis of loxP-GG (relative to the loxP site) as was observed for loxP-DAP. However, the decrease in synapsis affinity for Cre K86A-loxGG is only about a factor of three relative to loxP, rather than the factor of eight observed for the loxP-DAP site.
Both the loxP-DAP and loxGG sites were predicted to destabilize complex II by perturbing the left half-site sequence that most readily accommodates the bend shown in Fig. 4. We also tested loxP variants that would be predicted to improve the ability of the right half-site to accommodate the bend, leading to a more highly represented top-strand cleavage synaptic complex (i.e. I in Fig. 1). The loxAA site contains an A/T base pair in the 4-position of loxP, so that both the 4-and 4Ј-scissile bases are now adenine (Fig. 6). The lox4 site is a loxP variant where the bases in the 4 and 4Ј-position have been swapped (43). In both cases, wild-type Cre synapses these sites more efficiently than loxP, by about a factor of 2. Surprisingly, however, the Cre K201A mutant synapsis affinity is slightly decreased, and the Cre K86A mutant synapsis affinity is decreased by a factor of 2-3 for these modified sites. For both loxAA and lox4, we predicted that complexes I, III, and IV (Fig. 1c) would be more likely to contribute to synapsis observed in solution. Indeed, the lox4 site shows a complete reversal of preferential strand exchange order, cleaving and exchanging the top strands first during recombination (21). This suggests that synaptic species I is preferred over species II for lox4, although this has not yet been demonstrated experimentally.
The modest decrease in synapsis affinity for the Cre K86A mutant with loxAA and lox4 sites is difficult to explain based on existing structures of the Cre-loxP synaptic complex. The Lys-86 side chain interacts differently with scissile-A versus scissile-G, but it does not appear that the nature of these interactions differs depending on which base is present in the bent versus unbent half-site. However, it is important to note that a change in measured synapsis affinity could be due to either an altered stability of the synaptic complex, or to a change in stability of the unsynapsed sites, or both. It may be unreasonable to assume that changes in the loxP site will only affect the synaptic complex and not the unsynapsed sites.
Unfortunately, there is currently no structural information available for the unsynapsed Cre 2 loxP site, or for any corresponding full site among the tyrosine recombinases. However, there is evidence that the synapsed and unsynapsed Cre 2 loxP complexes must differ in structure. For example, DNA bending analysis of unsynapsed loxP sites indicates that the bend location and magnitude is different than that observed in structures of the synaptic complex (44). It has also been shown that the top strands are preferentially cleaved in isolated loxP sites that have not synapsed, an opposite bias to that observed upon synapsis at the start of the recombination pathway (23). These results imply that the loxP site is bent differently and may even be bent in the opposite direction prior to association with a partner site.

DISCUSSION
The biochemical studies described here were initially motivated by the need to have a more quantitative understanding of the energetics of loxP site synapsis. Although higher order complexes representing synapsed Cre-loxP assemblies have been visualized in gel-based assays (9,33), an assay capable of providing binding affinities has not been described. We found analytical ultracentrifugation to be a powerful tool for establishing the basic properties of synapsis in the Cre system, which we were then able to use to develop and validate a more practical gel-based approach. Although widely used to study protein-DNA interactions, the electrophoretic mobility shift assay is not routinely used to study interactions dominated by proteinprotein contacts, and it was essential to establish whether the gel-based assay accurately reports the equilibrium distributions of species in solution.
The 10 nM K d that we measured for synapsis of loxP sites explains a number of observations, including those from applications of Cre in DNA manipulations of transgenic systems. For example, Cre will excise genetic loci flanked by loxP sites when the sites are separated by megabases of sequence on a eukaryotic chromosome or on different chromosomes in site-directed translocation experiments from a previous study (45). The dissociation constant for wild-type Cre also agrees remarkably well with a previously estimated product dissociation K d based on kinetic analysis of an intramolecular excision reaction (36).
Our finding that divalent ions are not required for efficient Cre-loxP synapsis was somewhat surprising. Although there is a small, positive effect on synapsis for Mg 2ϩ , Ca 2ϩ , and Mn 2ϩ , synapsis occurs efficiently in buffers containing 150 mM NaCl without divalent ions. Initial crystal structures of Cre-loxP synaptic complexes revealed that the phosphate backbones of the synapsed loxP sites approach each other quite closely, with a minimum distance between non-bridging phosphate oxygen atoms of ϳ4 Å (14). Indeed, the new structures reported here confirm this close contact. Because divalent ions and polyamines were known to stimulate Cre-loxP recombination (3), it seemed reasonable to assume that a cationic species might bridge the close contact between phosphate backbones, thereby explaining this stimulatory effect. Interestingly, replacement of sodium by potassium as the monovalent cation also resulted in modest enhancement of synapsis that was similar in magnitude to the effects of divalent cations.
Our second goal in this work was to establish a structural model for the synaptic complex that forms at the start of Cre-loxP recombination. Because our initial structural models of this reaction intermediate made use of mutants that we later learned were defective in synapsis, we questioned whether these results were biased in some way. The initial studies also made use of a symmetric loxP variant that contained two identical G/C-rich right half-sites (i.e. loxSR), which left open the possibility that the synaptic complex with wild-type loxP could be different. Indeed, a third synaptic complex structure was subsequently described that contained a loxP sequence with non-bridging phosphorothioates at the scissile phosphates (15). This structure revealed a loxP bend that is quite different than that originally observed in the symmetric loxSR-containing complexes. However, the anti-parallel alignment of sites observed in this unique synaptic complex structure corresponds to a top strand cleavage configuration, which existing biochemical data strongly argue is not the preferred starting configuration for Cre-loxP recombination. Thus, despite the reports of three Cre-DNA crystal structures with uncleaved, synapsed sites, a structural model that could clearly explain the biochemical properties of this system was still lacking.
The new Cre-loxP synaptic complex crystal structures described here were formed with the cleavage-deficient, but synapsis-competent Cre K201A mutant. Indeed, the analysis of synapsis efficiencies of Cre mutants was a crucial step in being able to crystallize this intermediate. Three independent representations of the bent loxP site are virtually identical in these structures, revealing a bottom-strand cleavage synaptic complex that explains the existing biochemical data regarding reaction directionality in the Cre system. In particular, the tightly compressed minor groove adjacent to the inactive scissile base explains how the identity of the bases in the 4 and 4Ј positions can have such a dramatic influence on the directionality of the recombination pathway (21,23). Interestingly, our earlier attempts to determine the structure of a synaptic Cre-loxP complex included experiments using the inactive R173K mutant with wild-type loxP sites. However, this combination of Cre mutant and DNA site reproducibly resulted in crystallization of a Cre-Holliday junction complex, where junction DNA formed spontaneously from melting, and annealing of loxP ends was trapped by the recombinase (32). HJ formation from loxSR sites was less efficient with the Cre R173K mutant, leading instead to formation of a synaptic complex despite the unfavorable combination of a synapse-defective mutant and a poorly synapsing lox site (14).
It is interesting to note that each of the five synaptic Cre-DNA complex crystal structures (three previously described plus two described in this work) reveals a nearly co-planar arrangement of loxP sites. The four duplex segments of the complex that will ultimately become the "arms" of the HJ intermediate are arranged such that their ends trace out a parallelogram. A topological and mathematical analysis of site-specific recombination on circular DNA substrates has indicated that -int, Cre, and Flp share an intrinsic chirality of recombination (46). This chirality was interpreted as a right-handed crossing of sites in the synaptic complex that is maintained through the reaction to provide recombinant products with the same handedness. More recently, a topological analysis based on atomic force microscopy images following in vitro recombination by the Cre and Flp recombinases has supported the chiral recombination mechanism (11). In the latter work, a model was proposed in which a chiral HJ intermediate is responsible for the biased topological outcomes of recombination. In both cases, formation of a right-handed crossing of loxP sites would require that the loxP arms adopt a non-coplanar arrangement. This could be envisioned as a complex where the arms pass through the vertices of a distorted tetrahedron, which need only differ slightly from the planar arrangement observed in crystal structures.
The structural basis for the chirality of Cre-loxP recombination is not apparent from existing structural models of synaptic complexes and HJ intermediates, all of which show a planar arrangement of DNA arms (17). In principle, the lack of structural insights into this phenomenon could be due to crystal lattice effects that force the complexes into adopting misleading quaternary structures. Because crystals of protein-DNA complexes are often formed from repeating units of continuous DNA duplexes (47), this could be a general source of concern. However, the Cre-loxP synaptic complex crystal structures described here, as well as the Cre-HJ intermediate crystal structures that have been described (32,48,49), do not contain continuous DNA duplexes. In fact, there are no DNA-DNA contacts at all in either the orthorhombic or trigonal forms shown in Table 3. It therefore seems unlikely that a chiral crossing of loxP sites in the Cre-loxP synaptic complex has been masked by crystal packing effects in several independent experiments. The source of the observed chirality in the integrase family recombination is likely to lie in an intermediate stage of the pathway for which we do not yet have an adequate structural or biochemical understanding.
The work described here advances our understanding of the initial stages of Cre-loxP recombination, but these insights also impact related areas of investigation. One such area is the ongoing effort of a number of laboratories to modify Cre and/or the loxP site for more optimal use in genetic engineering and DNA manipulation applications. For example, because the initial report that mutant loxP sequences exist that will recombine with one another, but not with wild-type loxP (50), applications have been designed in a number of recombinase systems that make use of mutually exclusive pairs of sites. The recombinasemediated cassette exchange reaction is perhaps the most widely used example (51,52). In this context, it is interesting to note that altered loxP sites generally recombine less efficiently than wild-type sites (53). Of the loxP variants described here, the loxGG and loxAA sites have been shown to be extremely poor at recombination in vitro, and the lox4 site has been shown to be moderately defective relative to loxP (23).
A simplistic view of the recombination process might be that synapsis efficiency should be a primary limitation of the ability of Cre to recombine loxP sites with altered spacer sequences. If synapsis can be achieved, then one might argue that spacer sequences with similar G/C-content should then be able to undergo strand exchange to generate recombinant sites with efficiencies similar to loxP. This is clearly not the case. The loxAA and lox4 sites are synapsed more efficiently than loxP, but both are poor substrates for recombination. The formation of unproductive dead-end complexes (such as parallel synapses) and the inability to efficiently pass through formation and resolution of the HJ intermediate (19) are likely to be the primary reasons for the poor recombination behavior of loxAA and lox4, although this remains to be established. To under-stand and anticipate these effects in the design of alternative recombination sequences, it will be important to be able to dissect and understand in energetic terms the individual biochemical steps of the recombination pathway.