Mechanism of Tn3 Resolvase Recombination in Viuo*

To determine the physiologically important features of site-specific recombination by Escherichia coli Tn3 resolvase we analyzed the salient properties of the reaction in vivo. A two-plasmid system in which one plasmid served as substrate while the other encoded both resolvase and a thermolabile repressor of resolvase transcription provided controlled, synchronous recombination. Recombination proceeded rapidly and was promoted by (-) DNA supercoiling. The structures of the in vivo recombination products were predomi-nantly the same as those previously identified in vitro. By examination of the products of successive rounds of recombination of a four-site substrate, we ruled out a tracking mechanism for site alignment. Inversion and plasmid fusion occurred in vivo at a much lower rate than resolution but ultimately reached a higher extent than found in vitro. We propose that inversion and fusion exploit topologically interlinked substrates that occur at low levels in vivo. This proposal is supported by the unexpected specificity of fusion. Our data imply that supercoiled DNA, the resolvase synaptic complex, and the mechanism of strand exchange are fundamen- tally similar in vivo and in vitro, but that the repertoire of resolvase substrates and products is expanded in vivo by the action of other enzymes that alter DNA topology.

This site specificity of resolvase may be explained by topological effects. Prior to DNA exchange, the two res sites come together in an elaborate nucleoprotein structure called the synaptic complex, which contains three tightly interwound (-) supercoils (7,8). The synaptic complex contains a particular interwound geometry of the res sites which is favored by negative supercoiling of directly repeated sites, knotting of inverted sites, and multiple catenation of sites on separate DNA rings.
It is important to establish the properties of resolvase action in uiuo, as i n uitro studies show that the requirement for negative supercoiling and the unidirectionality of the reaction may be defeated somewhat by changes in reaction conditions (6,(8)(9)(10). As well, inside of cells the ionic composition differs from in vitro conditions and the supercoiled structure of the substrate DNA is altered by the binding of a number of proteins (11,12). In turn, the comparison of the properties of resolvase i n vivo and in uitro can provide information about the cellular milieu.
Toward these ends, we developed a system that provides controlled, efficient resolvase expression and recombination in Escherichia coli cells. From a comparison of the products formed in vitro and in uiuo, we conclude that resolvase recombination in the cell displays three of the topological properties previously determined by rigorous in vitro biochemical analyses: 1) The highly preferred substrate has directly repeated sites contained on a single molecule; 2) the product of the reaction is almost exclusively a singly linked catenane; and 3) recombination is promoted by negative supercoiling. From these observations and from the structure of products from multiple rounds of recombination, we conclude that the mechanism of site synapsis and strand exchange are the same in viuo as established in uitro. However, the intracellular environment did influence aspects of the reaction. The product catenanes were instantly converted to free circles by the action of DNA gyrase present in the cell. Inversion and intermolecular reactions did occur i n vivo, albeit slowly, and we propose that these reactions result from the transient formation of knotted and catenated substrates, respectively, in the cell.

1 2042
Mechanism of Tn3 Resolvase Recombination in Vivo gene (Cam') was isolated. The overhangs were repaired with DNA polymerase, and synthetic BamHI linkers were ligated to the ends. The resultant fragment was cyclized with DNA ligase, transformed into MG1655.10, purified, and reopened at the BamHI site by cleavage with XhoII. The tnpR gene under control of the pL promoter was isolated from pMK17 (3) as follows. The unique BamHI site in tnpR was protected from XhoII cleavage by methylation with BamHI methylase. The plasmid was then cleaved with XhoII and the largest fragment (1570 bp) was purified, ligated into the XhoII-cut pACYC184 derivative, and transformed into W3101X. The plasmid containing the tnpR gene in the same transcriptional orientation as the Cam' gene was then isolated. This plasmid, designated pJBRES, was linearized with XhoII after methylation protection of the BamHI sites. A 2400-bp EglII fragment containing the clts gene was isolated from pRK248clts (16), inserted into the linear pJBRES, and transformed into MG1655.10. The plasmid containing clts in the opposite transcriptional orientation with respect to tnpR was isolated and designated pJBREScI (Fig. 1A). The plasmid pA2ARI (Fig. 1B) contains two directly repeated res sites and is a head-to-tail dimer of plasmid pA (4), except that one of the two EcoRI sites was destroyed by partial restriction, filling in of the overhangs, and recyclization. Plasmid pA4AR12 is a naturally occurring head-to-tail dimer of pA"AR1 formed in E. coli. To simplify nomenclature for the four-site substrate and its complex products, pA'AR12 will be referred to as pA'. Plasmid pRR55 (Fig. IC) contains two res sites in inverted orientation (2).
Enzyme Reactions"T4 DNA ligase, DNA polymerase, BamHI methylase, and restriction enzymes were purchased from New England Biolabs. DNA was labeled with [a-"PJdCTP using an Amersham Corp. multiprime labeling kit according to the manufacturer's instructions. The preparation of resolvase has been descrihed elsewhere (3). Standard resolvase reactions were performed a t 37 "C for 1 h in 20-p1 mixtures containing 20 mM Tris-HC1, pH 7.5, 150 mM NaCI, 10 mM MgCls, 0.2 pg of DNA, and 0.1 pg of resolvase. Reactions with T 4 topoisomerase 11, obtained from Bruce Alberts, University of California, San Francisco, were carried out as described (7,17). DNA was nicked with DNase I to remove supercoils before analysis of knot and catenane structure by high resolution agarose gel electrophoresis (12).
Assays of Recombination in Viuo-Cells containingpJBRESc1 (Fig.  1A) and a resolvase substrate plasmid were grown at 28 "C in 25 ml of L-broth supplemented with 30 pg/ml chloramphenicol to maintain selection for pJBREScI. As well, either 50 pg/ml ampicillin or 15 pg/ ml tetracycline was added to maintain the pA2ARI and PA' plasmids or plasmid pRR55, respectively. At a density of 70 Klett units, the cells were shifted to 42 "C for the indicated times to induce resolvase and then returned to 28 "C. T o inhibit DNA gyrase, 10 pg/ml norfloxacin (obtained from Merck, Sharp and Dohme Laboratories) was added. 1.5-ml samples were removed at various times and processed as described (12).
Gel Electrophoresis-Published procedures were used for standard and high resolution one-dimensional gel electrophoresis (4, 12), twodimensional gel electrophoresis (3), and filter hybridizations (12). Resolvase substrate and product DNA were selectively visualized by probing filters with an [a-"2P]dCTP-labeled 768-bp XhoII fragment from pBR322 (coordinates 3225-3993). Resolvase expression was analyzed by electrophoresis of an extract of cellular proteins through a n SDS-containing 12.5% polyacrylamide gel (18) followed by staining with Coomassie Blue. Autoradiograms and stained protein gels were traced on a Hoefer (model GS300) scanning densitometer to quantify recombination and protein expression, respectively.

RESULTS
Resolvase-mediated Resolution i n Vivo-To control resolvase expression i n uiuo, we constructed the plasmid pJBREScI ( Fig. 1A) containing both the resolvase gene, tnpR, expressed from the XpLpromoter, and clts, which encodes a thermolabile repressor of pL. A recA-E. coli strain devoid of the Tn3 and y b transposons (MG1655.10) was transformed sequentially with pJBREScI and pA'ARI ( Fig. l B ) , a substrate plasmid for resolvase. Resolvase expression was repressed during growth of the cells a t 28 "C. Shifting the culture to 42 "C inactivated the repressor and rapidly led to high levels of resolvase synthesis. Returning the induced cells to 28 "C repressed further tnpR transcription (19) while allowing re- combination to continue. The use of short induction times of usually just a few minutes' duration synchronized recombination. Recombination was assayed by cutting at the single EcoRI site in pA2ARI, which converts the substrate into full length linear DNA and the resolved products into half-length linear and circular molecules, and separating the DNA species by gel electrophoresis (Fig. 2 A ) . Negligible recombination was detected prior to induction, demonstrating the effectiveness of repression of transcription from pL by the clts product at 28 "C. However, after only a 2.5-min induction a t 42 "C, 35% of the substrate was recombined. This corresponds to about four plasmids recombined per min/cell, assuming 30 copies of the plasmid substrate/cell. If, after the 2.5-min induction, the cultures were returned to 28 "C for 17.5 min, 93% of the substrate was resolved.
The extent of catenation of the i n uiuo recombination products was determined by nicking the samples with DNase I to remove supercoiling, and then visualizing these relaxed forms by high resolution gel electrophoresis (Fig. 2B). We detected only a trace of catenated product, in contrast to the extensive formation of singly linked catenanes in vitro (compare lanes 8 and 9). Since DNA gyrase efficiently unlinks multiply interlocked catenanes in viuo ( E ) , we inhibited gyrase activity by addition of the drug norfloxacin (20)  end of the induction period. After a 2.5-min induction and 17.5 min of additional incubation in the presence of the inhibitor, we recovered a substantial amount of singly linked catenane (lane 5). We believe that most if not all of the unlinked product circles also detected arose from decatenation of the recombinants formed before gyrase was fully inhibited, because 35% of the substrate was recombined at the time of norfloxacin addition (lane 2). Accordingly, addition of norfloxacin just 30 s earlier (at the end of a 2.0-min induction) increased the ratio of catenanes to free circles (lane 4 ) . A 1.5min induction period did not permit adequate recombination (lune 3), presumably because it did not allow sufficient time for expression of resolvase. We determined the pattern of minor products generated in uiuo by overexposing the filter blot. We now detected products (lune 4 ' ) above the contaminant level of knots in the substrate (lune 1 ') which comigrated with the characteristic ladder of products of processive recombination (5) detected in in uitro reactions (lune 9). Because the structure of the products of recombination establishes uniquely both the mechanism of DNA exchange and the degree of supercoiling in the synaptic complex (21), we conclude that these are the same in uiuo and in uitro.
Norfloxacin was required to block decatenation by DNA gyrase so that the extent of linkage of the products of recombination could be established. However, the inhibitor also reduced the extent of recombination. For example, addition of norfloxacin at 2.5 min reduced recombination by 15% as measured at the end of the 20-min assay (Fig. 2 A , Fig. 2B,  lanes 5 and 8). The reduction in recombination was even greater (24%) when norfloxacin was added immediately after a 2.0-min induction (Fig. 2B, lunes 4 and 7). Because gyrase inhibition occurred after the induction was terminated by downshift of the growth temperature to 28 'C, the diminution of recombination was not due to reduced transcription of tnpR. In addition, norfloxacin had no direct effect on resolvase recombination in uitro (data not shown). Instead, we attribute the drop in recombination to the relaxation of the plasmid substrate by topoisomerase I which results after inhibition of gyrase supercoiling activity (12) and the requirement for negative supercoiling for resolvase function (4).
res Site Selection in Vivo-A plasmid such as pA4 with four directly repeated res sites can undergo three successive rounds of recombination. As shown in Fig. 3, there are two possible products in each round. If recombination occurs at adjacent sites in the first round, the product is a singly linked catenane, designated pA3. A, in which a ring with three res sites is linked to a ring with a single site. Exchange via opposite sites would instead yield pA2.A2, a singly linked catenane with two sites in each ring. The possible second round products consist of a ring with two res sites linked to two rings each with a single site (p+.A2. A) and a ring with a single res site linked to another such ring and a ring with 2 sites (PA. A -A2). The two possible products of the third and final round, pisoA and pnA, are, respectively, branched and straight chain catenanes of four rings, each of which contains a single res site.
The distribution of the products formed in each round of recombination provides key information about the mechanism of res site selection. For example, tracking models in which resolvase bound to one res site slides along the DNA until a second site is captured were originally considered to explain the direct site specificity of resolvase (2,4). With a 4site substrate a tracking mechanism would generate only the products on the left-hand side of Fig. 3 (pA3. A, pA. A'. A, and pisoA). The discovery of substantial amounts of the other products in vitro in the second and third rounds provided the first evidence against tracking models (4). For reasons which are still not understood, resolvase demonstrated a marked preference for adjacent sites in the first round, such that the product pA3 -A was 20 times more abundant than pA2. A2 (4).
We constructed a derivative of pA4 that lacked two of the four EcoRI sites to simplify product analysis. The resolvase recombination products generated in uitro from this substrate were nicked by DNase I and displayed by high resolution gel electrophoresis. The result is the characteristic irregularly spaced ladder of bands (Fig. 4, lane 1 ) whose topological structure (indicated at the left of the figure) was previously determined by a combination of high resolution gel electro-

Mechanism of Tn3 Resolvase Recombination in Vivo
FIG. 3. Recombination products formed from a substrate containing four res sites. The resolvase substrate pA4, which contains four res sites (bold arrows; not to scale) in direct orientation and two EcoRI sites ( R I ) , is shown. For simplicity, the positions of the EcoRI sites in the products are omitted. Thin arrows denote possible pathways for formation of the products. Superscripts indicate the number of pA copies in a substrate or catenane; a center dot indicates catenation, pn a straight chain catenane, and piso a branched chain catenane. A tracking mechanism predicts that only the pathway on the left would be used. phoresis and electron microscopy (4). As before, there is a marked preference for adjacent site recombination in the first round. Subsequently, sites are selected nearly randomly.
We then performed the analogous experiment in vivo to study the physiological mechanism of site selection. Cells harboring pA4 and pJBREScI were induced for 2 min by a shift to 42 "C after which the cells were returned to 28 "C for 28 min and the DNA analyzed. Addition of norfloxacin at 1.5 min proved optimal for blocking decatenation while still allowing multiple rounds of recombination. Under these conditions, the products of the second and third rounds (Fig. 4,  lune 2 ) had the same electrophoretic mobilities as the products formed in uitro (Fig. 4, lane 1 ).
The possible first round products, pA2. A2 and pA3. A, were not resolved. To try to determine their abundance and to confirm the structure of the subsequent products we used twodimensional gel electrophoresis. A gel lane containing a ladder of nicked pA4 recombination products duplicate to that shown in Fig. 4, lane 2, was excised and the DNA was digested in situ with EcoRI. Electrophoresis in the second dimension yielded a characteristic pattern of spots consistent with the previous band assignments (Fig. 4). From the position of the two possible first round products (pA3 .A and pA2.A2), the linear lA3 (9600)) ZA2 (6400)) and 1A (3200) forms were generated as well as nicked nA (3200) species from monomer rings lacking a restriction site. From the second round products (PA. A2. A and pA . A. A2), the 1A2, IA, and nA forms were produced. The third round products (pisoA and pnA) resulted in only the 1A and the nA forms, as expected.
We estimated the proportion of pA3.A and pA2.A2 by analysis of the two-dimensional pattern as explained in the Resolvase recombination in MG1655.10 cells harboring pJBREScI and pA4 was induced as described for Fig. 2. After 1.5 min, 10 pg/ml norfloxacin was added to portions of the culture and the induction period was continued for an additional 0.5 min. The cells were then returned to 28 "C for 28 min. Plasmid DNA was isolated, nicked with DNase I, and displayed by high resolution gel electrophoresis (lane 2). A duplicate lane was excised from the gel, the DNA therein was digested in situ with EcoRI; the products were subjected to electrophoresis in a second dimension (left-to-right). Filter hybridization and autoradiography were used to visualize the products displayed in both dimensions. Lane 1 contains a reference ladder of PA' recombination products generated in vitro. On the left is indicated the positions of the pA4 recombination products schematized in Fig.

At the top is shown the position of the linear (1) and nicked (n)
EcoRI products. The minor spots adjacent to the lA3 band are unidentified, but may be highly relaxed topoisomers of pA4, which are refractory to DNase I nicking.

TABLE I
Distribution of pA4 recombination products The frequency of resolvase products is given for the three successive rounds of recombination of pA4 for the experiment shown in Fig. 4. Also shown are the product distributions expected if the res sites synapse randomly in each round without regard to distance between sites and site order. The ratio of pA3.A to pA2.A2 generated in the first round of recombination was estimated from the results of the two-dimensional gel. The first dimension resolved pA3.A only partially from pA2.A2, but after restriction in situ with EcoRI the products in the second dimension help distinguish the two species. The lA2 restriction product consists of two overlapping spots, one derived from pA3.A, and the other derived from the faster moving pA2.A2 (4). We calculated the amount of pA2.A2 by subtracting the contribution of pA3.A to the total amount of lA2. The ratio of IA to nA provides a control for the accuracy of the method, and this ratio was within 15% of the expected 2:l value. In addition, the in vitro first round product distributions calculated by this method agree to within 1% of the data previously determined by gel electrophoresis and electron microscopy (4).  Table I. The distributions of the second and third round products were determined by densitometric scans of one-dimensional gels. These results are summarized in Table  I. The primary conclusion is that the distribution of the products of recombination in the second and third rounds is essentially the same in vivo and in vitro. This shows that tracking by resolvase does not occur in vivo. The products of the first round of recombination in vivo are also consistent with this conclusion. These products appear to show little or no bias toward adjacent site selection, but the incomplete resolution of the two possible products, even after two-dimensional gel electrophoresis, makes the precise quantification uncertain.

Resolvase-mediated Inversion and Intermolecular Recombination in
Vivo-With a standard supercoiled substrate, purified resolvase acts efficiently only on res sites oriented as direct repeats contained within the same DNA molecule (2,3). It has been reported that high resolvase expression in vivo leads to recombination between inverted res sites (inversion), but that recombination between sites on different molecules (intermolecular recombination or fusion) remains rare under these conditions (1,2). In order to assess inversion and fusion under our in vivo conditions, the inverted-site substrate pRR55 (Fig. 1C) was used in conjunction with the pJBREScI expression plasmid.
Inversion was measured by cleavage of the DNA with EcoRI, followed by gel electrophoresis and filter hybridization. As evidenced by the appearance of the 3200-bp product fragment, inversion of pRR55 was clearly detectable only after 30 min of induction of resolvase (Fig. 5A, lane 2). Thereafter, we observed a slow but linear increase in recombination (lanes 3   and 4 ) , at the rate of 3-5% of the substrate recombined per h. We can make only a rough comparison of this rate to the rate of resolution, because after the short induction times used in the resolution experiments, inversion was undetectable. After a 2.5-min induction, 35% of pA2ARI was resolved ( Fig. 2A). We estimate by linear extrapolation that 0.2% of pRR55 was inverted after the same induction period. Therefore, in vivo, resolution was about 175 times faster than inversion. Inversion, like resolution, was inhibited by the addition of norfloxacin (data not shown); all the experiments shown here were done in the absence of the inhibitor.
To assay for intermolecular recombination, a portion of each DNA sample analyzed for inversion in Fig. 5A was displayed by gel electrophoresis without restriction with EcoRI. The amount of the fusion product, dimeric pRR55, increased from 1 to 20% by 30 min, and over the next 90 min the amount remained constant (Fig. 5B, lanes 1-4). Because the dimer can be resolved by recombination back to monomer pRR55, we suggest that a steady-state of dimer formation and breakdown was achieved. Because of differences in the sequences flanking the two res sites in pRR55, two distinct intermolecular products can be formed (Fig. 6). In pathway a, recombination between sites in different positions in pRR55 generates a nonsymmetrical dimer. In pathways b and c, a perfect head-to-tail dimer is formed by recombination between res sites in equivalent contexts. The proportion of the two alternatives can be determined by cleavage at the single BamHI site in pRR55. Restriction of the symmetric dimer with BamHI releases two fragments the size of pRR55, whereas cleavage of the asymmetric dimer produces two fragments, one larger and one smaller than pRR55 (Fig. 6). After cleavage with BamHI, nearly all products (>99%) were pRR55 sized fragments (Fig. 5C). Remarkably, all the dimer detected in vivo was formed via equivalent-site exchange.  (Fig. IC) was the substrate, and resolvase expression was induced as described for Fig. 2, except that the culture was held a t 42 "C. At the times indicated, plasmid DNA was isolated, processed as described, and displayed by gel electrophoresis and filter hybridization. A , inversion was assayed by cleaving the DNA with EcoRI. On the right is indicated the positions in the gel of the large substrate restriction fragment (4980 bp) and the large product fragment (3200 bp) which hybridize to the DNA probe used. The correct induction time in lane 1 is 0 min. B, detection of intermolecular dimer formation. Portions of the same DNA samples assayed in A were subjected to gel electrophoresis without EcoRI restriction. On the right is indicated the position of substrate monomer (pRR55) and product dimer (pRR55'). C, distinction of symmetric and asymmetric dimeric product by digestion with BarnHI. On the right is indicated the position of the symmetric dimer product fragment (ZRR55) and the asymmetric product fragments (6950 and 4950). Nicked pRR55 (nRR55) resulted from uncompleted digestion of pRR55 by BamHI. The percentage of recombination shown in A and B was determined from densitometric scans of the autoradiograms. nd, not detected.
To measure the amount of resolvase that was produced during this experiment, the cellular proteins of the cultures were displayed by SDS-polyacrylamide gel electrophoresis (Fig. 7). By 30 min of induction (lane 2), resolvase was easily detected; by 120 min (lane 4 ) , it constituted 4% of the total cellular protein. This corresponds to a resolvase to substrate ratio of about 60001, assuming that the weight of protein in a single E. coli cell is 1.56 x g (22) and that the copy number of the substrate is 30/cell.

DISCUSSION
We have undertaken a topological analysis of the resolvase reaction in vivo to determine its physiologically important properties. Topological properties are ideal for the study of recombination in vivo, as they are unchanged during the isolation and processing of the DNA. We devised a system to provide controlled, synchronous resolvase recombination in  Moreover, because the activity of resolvase is highly sensitive to the structure of the substrate DNA, it appears that DNA supercoils in vivo and in vitro are functionally equivalent, despite the large differences in ligands.
The models for resolvase site specificity that best fit the in vitro data focus on the interwound (plectonemic) shape of supercoiled DNA (6,7,9,23 1,2,28). The limitations of both approaches were that long time periods, even days, were needed for the assays and that the exact levels of resolvase and the immediate products of the reaction were unknown. Using our controlled recombination system, we were able to measure in this report the amount of resolution, inversion, and fusion using a variety of substrates, as well as to determine the structure of the immediate resolution products.
I n uiuo resolution is rapid and efficient. The rate of about four plasmids/min/cell resolved at 28 "C is 4 times greater than the rate we measured previously for Int recombination i n uiuo (12). The products of resolvase action, singly linked catenanes and smaller amounts of more complex forms resulting from processive recombination, are the same in uiuo and in vitro to the limits of resolution of gel electrophoresis. This implies that both in vivo and in vitro, the synaptic complex has three (-) plectonemic supercoils and that the strand exchange mechanism introduces one positive supercoil and converts two substrate supercoils into a catenane interlink; any deviation from these specifications would lead to different products (21).
If, in uiuo, the DNA linking deficit were expressed entirely in the form of solenoidal supercoils, the exclusion of mulitply linked catenanes as products of recombination could be a trivial result of DNA structure, rather than the outcome of a controlled site-alignment mechanism. However, our previous experiments with XInt recombination in uiuo demonstrated that about 40% of the linking deficit in plasmid DNA is in the form of plectonemic supercoils, and that complex catenanes with variable numbers of links were generated by Int (12). Thus, in uiuo, resolvase relies on this fraction of plectonemic supercoils to supply specificity in the reaction but does not allow free supercoils to be converted to catenane interlinks.
Under standard conditions in vitro, the resolvase reaction is proportional to the superhelical density of the substrate (4). However, under modified reaction conditions, linear or relaxed substrates will recombine, albeit with loss of certain aspects of site specificity (6,9,10). I n uiuo, inhibition of DNA gyrase by norfloxacin reduced the extent of resolvase recombination. The earlier the inhibitor was added, the greater was the diminution of recombination. This reduction of the amount of recombination coincides with the decline in effective DNA supercoiling associated with inhibition of gyrase supercoiling activity by norfloxacin. The majority of the substrate was recombined in the first 5 min (Fig. 2), and about one third of plasmid supercoils were relaxed within 5 min of norfloxacin addition (12). We conclude that, in uiuo, plecto-Mechanism of Tn3 Resolvase Recombination in Vivo 2047 nemic (-) supercoiling promotes recombination by resolvase. We measured the distribution of products formed i n uiuo resulting from the three rounds of recombination of a foursite substrate (Table I). The products formed in the second and third rounds of recombination are inconsistent with a tracking mechanism. In addition, the first round products do not show as high a preference for reaction of adjacent sites as that observed in uitro (Ref. 4 and Fig. 4), but the extent of the bias is not clear. In addition, adjacent site preference i n uitro is largely eliminated by use of low resolvase concentrations.' This finding, taken with the difference between the results in uiuo and in uitro, implies that adjacent site preference is not a fundamental aspect of resolvase action. Reactions between both inverted and intermolecular res sites were detected in uiuo. These results are most easily explained by consideration of the following two observations. First, with catenated (7) and knotted (6) substrates, the constraints against intermolecular recombination and inversion, respectively, are bypassed i n uitro. The right-handed intertwinings of knots or catenanes provide a local configuration equivalent to that of (-) plectonemic supercoils. Second, both knots and catenanes occur naturally i n uiuo.
Knots can be both formed and untied by recombination and by topoisomerases (29) thus maintaining a low, strainspecific, steady-state level. A few percent of our resolvase substrate occurs in knotted form (Fig. 2B, lane 1') and this is approximately the level of inversion measured i n uitro and early in the i n uiuo assay. The total extent of inversion i n uiuo should slowly increase with time as new knots are formed and recombined knots are disentangled, and slow continuous inversion rate i n uiuo is exactly what we find. High continuous tnpR expression was required, perhaps to ensure saturating levels of resolvase. The hypothesis that knotting is required for inversion i n vivo may be tested by measuring resolvasemediated inversion in strains that have unusually high levels of knots, resulting from topoisomerase mutations (30-32) or from simultaneous Int-mediated inversion.
We propose that the intermolecular resolvase reaction is also a result of altered topology in the substrate. Multiply linked catenanes are normal intermediates in DNA replication and are generated by incomplete segregation of daughter strands at the termination of the process (reviewed in Ref.

) .
Because replication catenanes have right-handed intertwines, res sites on the separate rings can adopt the correct geometry for synaptic complex formation (7). Moreover, since the rate of DNA replication is constant under our assay conditions, the catenated replication intermediates that we postulate are the substrates for recombination, would be generated at a constant rate. The chief advantage of this model is that it explains the exclusive formation of symmetrical head-to-tail dimers by resolvase. As shown in Fig. 6C, the rings of the replication catenane are oriented in the same direction and thus the res sites in equivalent positions can synapse in the correct parallel orientation without distortion of the DNA. For the nonequivalent res sites to synapse properly, one must be looped back and this should be energetically unfavorable.
We conclude that resolvase can be "tricked into performing the forbidden reactions of inversion and fusion by interlinking of the DNA substrate. The rate of these ancillary reactions, however, remains small compared to resolution. Thus, to preserve its biological specificity, resolvase relies heavily on topoisomerases to minimize the amount of the intracellular DNA present in a tangled form.
Our system provides a method for the controlled intracellular generation of singly linked catenanes via resolvase action. Because topologically linked DNA molecules are unavoidable intermediates of DNA replication and recombination, decatenation is an essential function in both prokaryotic and eukaryotic organisms (reviewed in Ref. 5 ) . Our data show that DNA gyrase i n vivo efficiently decatenates both multiply (12) and singly linked catenanes (Fig. 2B). The latter event allows the products of resolvase to replicate semi-conservatively and thereby to complete the process of transposition.