Sin recombinase from Staphylococcus aureus: synaptic complex architecture and transposon targeting

Authors


Summary

The Sin recombinase from Staphylococcus aureus builds a distinctive DNA-protein synaptic complex to regulate strand exchange. Sin binds at two sites within an 86 basepair (bp) recombination site, resH. We propose that inverted motifs at the crossover site, and tandem motifs at the regulatory site, are recognized by structurally disparate Sin dimers. An essential architectural protein, Hbsu, binds at a discrete central site in resH. Positions of Hbsu-induced DNA deformation coincide with natural targets for Tn552 integration. Remarkably, Sin has the same topological selectivity as Tn3 and γδ resolvases. Our model for the recombination synapse has at its core an assembly of four Sin dimers; Hbsu plays an architectural role that is taken by two resolvase dimers in models of the Tn3/γδ synapse.

Introduction

Serine recombinases can rearrange DNA sequences with exquisite topological selectivity. The ability to recognize the alignment and connectivity of the two crossover sites is thought to be encoded in the unique protein-DNA architecture of the active synapse. The synapse structures incorporate the two crossover sites, each bound by a recombinase dimer, and additional regulatory elements. In the Gin and Hin inversion systems, a separate en-hancer element binds the architectural protein FIS. In the Tn3 and γδ resolution systems, binding sites for two additional resolvase dimers are located adjacent to each crossover site (see Fig. 7); a total of six resolvase dimers are thought to assemble an interwound synapse, trapping three supercoils between the crossover sites (reviewed in Stark and Boocock, 1995; Grindley, 2002). The structures of these synaptic complexes are unknown. Current models differ in more or less fundamental respects (Rice and Steitz, 1994; Yang and Steitz, 1995; Merickel et al., 1998; Sarkis et al., 2001) and decisive structural data are not available. We chose to study Sin, a previously uncharacterized serine recombinase from Staphylococcus aureus, because it appeared to have an unusual recombination site; we reasoned that this system would provide new insights into synapse architecture and function.

Figure 7.

Crossover and regulatory binding sites in the recombination sites of selected resolvases and Hin invertase. Boxes represent imperfect copies of 12-bp sequence motifs that correspond to known or presumptive recognition sites for resolvases (R) and Hin invertase (10-bp motifs); white boxes represent poorer matches to the consensus. Alternative configurations of recognition motifs were previously proposed for Tn552 (Rowland and Dyke, 1989) and ISXc5 (Liu et al., 1998); our assignments are based on comparisons of many related sites. Hatched boxes represent the start codons of recombinase genes (thick lines). The first basepair of each repeat motif is numbered, beginning at the left end of each recombination site.

Serine recombinases have major roles in horizontal gene transfer in Gram-positive pathogens such as S. aureus; they mobilize the Mec element of methicillin-resistant strains (MRSA), and provide resolvases for type II transposons such as Tn552 (β-lactam resistance) and Tn1546 (vancomycin resistance) (Rowland and Dyke, 1990; Katayama et al., 2000; Grindley, 2002). Sin was first identified as a potential recombinase, of unknown function, encoded by the multiresistance plasmid pI9789 (Fig. 1A; Rowland and Dyke, 1989; Paulsen et al., 1994). Many other large staphylococcal plasmids, including pSK1, encode close relatives of Sin (Fig. 1A; Paulsen et al., 1994). Sin is not associated with a transposon or an invertible DNA segment.

Figure 1.

Figure 1.

res sites and transposon targets.

A. The sin regions of pI9789 and pSK1, and natural integration sites of Tn552 and Tn4002 described previously (Rowland and Dyke, 1989; Paulsen et al.,1994) (not to scale). Tn552 and Tn4002 are very close relatives. The region of similarity between pI9789 and pSK1 is hatched; the encoded Sin proteins are 92% identical. Recombination sites for the recombinases Sin (res, resΔ) and BinR (resR) are represented by triangles. Flanking genes encode resistance to heavy metals (cad, ars), trimethoprim (dfrA, carried by Tn4003), and antiseptics and disinfectants (ORF188; qac). pI9789 sin was formerly designated bin3.

B. res sites, and predicted res sites, for selected Gram-positive recombinases. resH is the synthetic site used in our in vitro studies with pI9789 Sin. Other sites are upstream of genes for: pI9789 Sin, pSK1 Sin, four Sin-like proteins, two β recombinases and resolvases of pLH1 and Tn1546. The 12-bp sequence motifs in sites I and II of resH (sites I-L, I-R, IIa and IIb), and repeat motifs in the other sites, are boxed. Sequence identity between pI9789 and pSK1 is indicated by vertical lines. Site I-L of resH is from pSK1, and the rest is identical to pI9789, except for a single nucleotide change creating the EcoRI site (wild-type sequence is in upper case). Target sites for Tn552-family transposons are highlighted (yellow; lower case); the site II-proximal target sites in pI9789 resΔ and pSK1 res correspond to the transposon insertions shown in Fig. 1A. The 67-bp imperfect palindrome in pI9789, pSK1 and Enterococcus faecalis CH116 is indicated by purple and red arrows above the resH sequence. Notes to sequences: (1) the site I-proximal Tn552‘target site’ is believed to be the end-point of an adjacent deletion (Rowland and Dyke, 1989); we assume a 6-bp target region; (2) insertions into the site I- and site II-proximal target sites account for the structures of pIP1663 and pIP1085 respectively (Derbise et al., 1995); (3) the deduced sequence before insertion of Tn552. Sources of sequences: Staphylococcus aureus pI9789 (X16298); pSK1 (L23109); pIP1066 (Derbise et al., 1995); E. faecalis CH116 (U39859; Rice et al., 1996b; L.B. Rice and S.H. Marshall, personal communication); S. aureus‘21 kb plasmid’ (gi 9621478); E. faecalis pAM373 (AE002565); Streptococcus pyogenes pSM19035 (X64695); E. faecalis pAMβ1 (AF007787); Lactococcus helveticus pLH1 (AJ222725); E. faecium Tn1546 (M97297).

Figure 1.

Figure 1.

res sites and transposon targets.

A. The sin regions of pI9789 and pSK1, and natural integration sites of Tn552 and Tn4002 described previously (Rowland and Dyke, 1989; Paulsen et al.,1994) (not to scale). Tn552 and Tn4002 are very close relatives. The region of similarity between pI9789 and pSK1 is hatched; the encoded Sin proteins are 92% identical. Recombination sites for the recombinases Sin (res, resΔ) and BinR (resR) are represented by triangles. Flanking genes encode resistance to heavy metals (cad, ars), trimethoprim (dfrA, carried by Tn4003), and antiseptics and disinfectants (ORF188; qac). pI9789 sin was formerly designated bin3.

B. res sites, and predicted res sites, for selected Gram-positive recombinases. resH is the synthetic site used in our in vitro studies with pI9789 Sin. Other sites are upstream of genes for: pI9789 Sin, pSK1 Sin, four Sin-like proteins, two β recombinases and resolvases of pLH1 and Tn1546. The 12-bp sequence motifs in sites I and II of resH (sites I-L, I-R, IIa and IIb), and repeat motifs in the other sites, are boxed. Sequence identity between pI9789 and pSK1 is indicated by vertical lines. Site I-L of resH is from pSK1, and the rest is identical to pI9789, except for a single nucleotide change creating the EcoRI site (wild-type sequence is in upper case). Target sites for Tn552-family transposons are highlighted (yellow; lower case); the site II-proximal target sites in pI9789 resΔ and pSK1 res correspond to the transposon insertions shown in Fig. 1A. The 67-bp imperfect palindrome in pI9789, pSK1 and Enterococcus faecalis CH116 is indicated by purple and red arrows above the resH sequence. Notes to sequences: (1) the site I-proximal Tn552‘target site’ is believed to be the end-point of an adjacent deletion (Rowland and Dyke, 1989); we assume a 6-bp target region; (2) insertions into the site I- and site II-proximal target sites account for the structures of pIP1663 and pIP1085 respectively (Derbise et al., 1995); (3) the deduced sequence before insertion of Tn552. Sources of sequences: Staphylococcus aureus pI9789 (X16298); pSK1 (L23109); pIP1066 (Derbise et al., 1995); E. faecalis CH116 (U39859; Rice et al., 1996b; L.B. Rice and S.H. Marshall, personal communication); S. aureus‘21 kb plasmid’ (gi 9621478); E. faecalis pAM373 (AE002565); Streptococcus pyogenes pSM19035 (X64695); E. faecalis pAMβ1 (AF007787); Lactococcus helveticus pLH1 (AJ222725); E. faecium Tn1546 (M97297).

Sin is only distantly related to known resolvases and DNA invertases (e.g. Hin, Gin), although sequence alignment implies that it has a structural fold essentially the same as that of γδ resolvase (31% identical to pI9789 Sin) (Yang and Steitz, 1995). Its closest, yet still distant, homologues (39% identity) are the β recombinases from Gram-positive plasmids pSM19035 and pAMβ1, which are unusual in that they function both as resolvases and invertases, bind to only two sites within res, and require an architectural protein for activity (e.g. Bacillus subtilis Hbsu or Escherichia coli HU) (Alonso et al., 1995a; Petit et al., 1995; Rojo and Alonso, 1995; Janniere et al., 1996).

DNA sequences immediately upstream of sin appear to be favoured integration sites for Tn552-family transposons (Fig. 1A; Gillespie et al., 1988; Rowland and Dyke, 1989; Paulsen et al., 1994). We suspected that this region contains the recombination site for Sin, but its sequence did not resemble previously characterized sites. Recently, it was shown that certain transposons are ‘res site hunters’ (Minakhina et al., 1999); they integrate at or near diverse res (resolution) sites if the cognate resolvase is present. These reports challenge the assumption that resolvases and transposases function independently, and raise the question of how res sites might be designated as transposition targets.

Here we show that Sin from pI9789 is a functional resolvase which depends on an HU-like protein (e.g. Hbsu) for activity. We map the phosphodiester bonds that are cut by Sin, characterize binding sites for Sin and Hbsu within a minimal res site and show that Tn552 targets specific points within res. We demonstrate the topological selectivity of recombination and present a simple model of the synapse. Our model suggests that Hbsu takes the place of two of the resolvase dimers needed in the Tn3/γδ system, and leads us to propose a common structural basis for the regulation of strand exchange by resolvases.

Results

Recombination by Sin in vitro and in vivo

We suspected that a recombination site for Sin lies upstream of sin in pI9789, in a region containing a 67 basepair (bp) palindrome and a transposition target site for Tn552 (Fig. 1). However, this region was found not to function as a recombination site for Sin in our assays (see below). We therefore compared it with the corresponding region of pSK1, which encodes a Sin protein 92% identical to that of pI9789 (Fig. 1A). The two DNA sequences diverge close to the centre of a potential crossover site for Sin (site I) (Fig. 1B). We hypothesized that pSK1 contains a complete recombination site (res), whereas pI9789 contains a truncated site (resΔ) lacking the left arm of site I (site I-L). A new res site for pI9789 Sin, resH, was synthesized (Fig. 1B). Like the natural res site of pSK1, resH (86 bp) contains four imperfect copies of a 12-bp motif: two in inverted repeat (sites I-L and I-R) and two in direct repeat (sites IIa and IIb). Site I-L of resH comes from pSK1; the remainder is from pI9789. The region of pI9789 and pSK1 that is targeted by Tn552-family transposons is between sites I and II (Fig. 1B).

Sin from pI9789 was overexpressed in Escherichia coli and purified (see Experimental procedures). In vitro, Sin resolves a supercoiled plasmid substrate containing two directly repeated resH sites (pSB312; Fig. 2A). A substrate containing pI9789 resΔ sites was not detectably recombined (data not shown). Resolution requires stoichiometric amounts of Sin, and is absolutely dependent on the presence of a chromatin-associated protein, such as B. subtilis Hbsu (Fig. 2A). Hbsu could be substituted by a related E. coli protein, HU or IHF, although recom-bination was less efficient with IHF (≈ 5% resolved after 2 h at 1 μg ml–1 of IHF). FIS, an unrelated DNA-bending protein (Kostrewa et al., 1991), did not stimulate Sin-mediated recombination (data not shown). The importance of site II for resolution was shown by deleting it from one or both of the resH sites in pSB312: a resH× site I substrate was recombined very inefficiently (<5% resolved after 20 h), and a site I × site I substrate was inactive (data not shown). Sin was active as a recombinase in E. coli DS941 (HU+ IHF+recF). A substrate with directly repeated resH sites was resolved in the presence of a Sin-expressing plasmid, but resolution was not detectable with a resΔ substrate (data not shown).

Figure 2.

In vitro recombination by Sin.

A. Resolution of pSB312 (3915 bp). Products are shown digested with XhoI. Reactions (2 h) contained Sin (≈ 3 μg ml–1) and Hbsu (0, 0.1, 0.2, 0.4, 1 and 2 μg ml–1); 1.2% agarose gel. nr, non-recombinant (3633, 282 bp) restriction fragments; r, recombinant (2541, 1374 bp) restriction fragments.

B. Inversion of pSB313 (3915 bp). Products are shown digested with XhoI. Reactions (20 h) contained Sin (≈ 12 μg ml–1) and Hbsu (20 μg ml–1); 0.7% agarose gel. nr and r, non-recombinant (2705, 1210 bp) and recombinant (3633 bp) restriction fragments; the small recombinant fragment (282 bp) is not shown.

C. Resolution of pSB312. Products are shown uncut, and nicked with DNase I. Reactions (2 h) contained Sin (≈ 6 μg ml–1) and Hbsu (2 μg ml–1); 0.7% agarose gel. sc, supercoiled; oc, open circular; 2, 2-noded catenane; scd, supercoiled dimer.

D. Recombination of pSB328 (resH-AT ×resH-AT; 3915 bp) and pSB329(resH×resH-AT; 3915 bp). Products are shown nicked with DNase I. Reactions as in (C). 2, 4, 8, 12: topological node numbers of products, assigned using catenane and knot markers made with Tn3 resolvase.

In C and D, additional minor species (not annotated) are half-nicked forms of the catenane, linear forms released by overdigestion with DNase I (pSB312, pSB328), and products of multiple rounds of strand exchange (pSB312, pSB328, pSB329) (see Stark et al., 1991). In the diagrams of the plasmid substrates, resH and resH-AT (see text) are represented by arrowheads.

E. Combinations of resH and resH-AT sites in the plasmid substrates.

F. Proposed topology of pSB329 recombination. resH and resH-AT are represented by boxes; the central 2 bp of each site I, and the Sin-mediated cleavages (Fig. 3B), are shown. In resH-AT, the altered basepair is in bold type. Synapsis of resH and resH-AT traps three interdomainal supercoils. A double round of strand exchange (360° right-handed rotation) generates a non-recombinant, 4-noded knot. A 180° round of strand exchange would give a recombinant 2-noded catenane with two mismatched basepairs (not shown).

Strand cleavage by Sin

To analyse the topology of Sin recombination (as shown in Fig. 2C–F), we first needed to determine the positions of strand cleavage, and hence the overlap region for strand exchange (Fig. 3). For various serine recombinases, DNA double-strand cleavage intermediates can be trapped under altered reaction conditions. We investigated the strand cleavage activities of Sin using a buffer containing 40% ethylene glycol. Under these conditions, Sin alone made double-strand breaks in supercoiled plasmids containing one or two copies of resH or resH site I (Fig. 3A). With two-site substrates, the main product at low concentrations of Sin was linearized plasmid (data not shown), implying that cleavage at the two sites is not concerted. At high concentrations of Sin, the linear DNA fragments were re-ligated, making recombinant concatemers (Fig. 3A). Site I substrates were cut and re-ligated as efficiently as resH substrates (Fig. 3A, and data not shown). The strand cleavage/re-ligation activity therefore resides in the Sin protein bound at site I; Hbsu and site II of resH, which are essential for the resolution reaction, are not required.

Figure 3.

Strand cleavage by Sin.

A. Reaction in ‘cleavage buffer’ of a plasmid substrate containing a single copy of resH site I. Linear cleavage product (2532 bp; 1) is ligated by Sin to form linear concatemers (2, 3, 4, 5, etc.). Reactions (24 h) contained Sin (≈ 24 μg ml–1); 1.2% agarose gel. M, 1 kbp ladder (BRL); sc, supercoiled; oc, open circular.

B. Mapping the cleavage sites in resH. The sources of the labelled DNA strands seen in the gel are explained in the diagram, which shows the 65-bp site I-containing BamHI–EcoRI fragment from pSB312, and the sizes of the 5′-end-labelled strands (*) resulting from cleavage by Sin (open triangles) and RsaI (filled triangles). Cleavage by Sin gave fragments of 38 nt (top strand) and 33 nt (bottom strand) (lane 3). RsaI gives an exact marker for bottom strand cleavage by Sin (lanes 4 and 5). The reference ladder (R) was a 5′-end-labelled oligonucleotide of different sequence (38 nt, and shorter synthesis products). See Experimental procedures.

The Sin-mediated strand cleavages were mapped by analysing the products on denaturing gels (Fig. 3B). In common with γδ resolvase, Gin and Hin (Reed and Grindley, 1981; Klippel et al., 1988; Johnson and Bruist, 1989), Sin cuts the DNA at the centre of site I with a 2-bp stagger, leaving overhanging 3′-OH ends. We infer that Sin is covalently linked to the 5′-ends (via the conserved serine-9 residue), as the cleavage products had reduced mobility on SDS-agarose gels before treatment with proteinase K (data not shown).

Topological selectivity of Sin recombination

Sin-mediated recombination is dependent on (i) the relative orientation of the resH sites and (ii) the substrate topology. Whereas supercoiled pSB312 (resH sites in direct repeat) was resolved efficiently (Fig. 2A), recom-bination of supercoiled pSB313 (resH sites in inverted repeat) was very inefficient, and required a high concentration of Hbsu (10–20 μg ml–1; Fig. 2B). Recombination of linear forms of pSB312 and pSB313 was also detected at high concentrations of Hbsu; for these linear substrates, the extent of recombination (≈10% in 24 h; data not shown) did not depend on the relative orientation of the resH sites. Thus, Sin preferentially recombines directly repeated resH sites in a supercoiled substrate.

The topologies of recombination products can be analysed by nicking with DNase I, and provide information about the wrapping of the recombination sites in the synapse (see Discussion). In the presence of Sin and Hbsu, the major resolution product from pSB312 was a supercoiled 2-noded catenane (Fig. 2C), as seen in the Tn3/γδ resolution system. There was no evidence of the complex product topologies typical of ‘random collision’ synapsis. A substrate identical to pSB312 except for a mutation in the central dinucleotide of each resH site I (AC to AT; ‘resH-AT’) also gave a 2-noded catenane (pSB328; Fig. 2D); the mutation in the central dinucleotide (Fig. 3B) did not impair recombination.

When we tested a substrate containing one resH site and one resH-AT site (pSB329), the products were all non-recombinant knots, the simplest being the 4-noded knot (Fig. 2D). The 4-noded knot is the product expected if Sin uses the same synapse and strand exchange topology as Tn3/γδ resolvase, and if the non-complementarity in the 2 bp overlap regions of site I induces a double round of strand exchange (Fig. 2F). A synapse with three interdomainal supercoils is in fact the only topology that could give a 2-noded catenane in the first round and a 4-noded knot in a processive second round of strand exchange. The more complex knots (Fig. 2D) could result from multiple double rounds of strand exchange (see Stark et al., 1991; Stark and Boocock, 1994). In the Tn3/γδ system, the product topologies reflect the interwinding of res sites II and III (Wasserman et al., 1985; Stark et al., 1989; Kilbride et al., 1999). Given that resH is 28 bp shorter than Tn3/γδres, contains fewer recombinase binding sites and interacts with both Sin and Hbsu (see below), it is striking that the reactions show the same topological selectivity.

Binding sites for Sin and Hbsu

Sin binds specifically to resH in a gel assay (Fig. 4). At limiting Sin concentrations, a resH DNA fragment gave four retarded complexes. A fragment containing site I of resH gave two retarded complexes, as did a fragment containing site II; Sin showed a similar affinity for sites I and II. The data are consistent with binding of a monomer of Sin to each of the 12-bp motifs (Fig. 1B; see Blake et al., 1995), and with the co-operative binding of monomers to form dimers at site I and at site II. The fast-migrating minor complexes seen only at saturating Sin concentrations (‘x’ in Fig. 4) have not been further characterized. Experiments with a pI9789 fragment containing the entire sequence between binR and sin (Fig. 1A) did not reveal any further binding sites (data not shown). Hbsu bound non-specifically to resH in the gel assay; stable co-complexes with Sin were not detected (data not shown).

Figure 4.

Binding of Sin to resH, site I and site II. For each DNA fragment (118, 109 and 91 bp respectively), the estimated concentrations of Sin in the reactions were 0, 0.7, 1.5, 3, 6 and 12 μg ml–1. U, unbound DNA; 1–4, resH-Sin complexes; x, additional resH-Sin complexes reproducibly observed at higher concentrations of Sin.

Sin protects two regions of resH from DNase I, corresponding to sites I and II (Fig. 5A and C). Both sites were partially protected at limiting Sin concentrations, consistent with Sin having a similar affinity for each site, or with cooperative binding. At site I, the DNase I footprint of Sin includes short regions of accessibility near the outer edge of each binding motif, offset by several basepairs between the top and bottom strands (Fig. 5C). Similar regions of accessibility are seen in the footprints of other resolvases (e.g. Grindley et al., 1982). In the crystal structure of γδ resolvase bound at res site I, these are regions at which the DNA binding domains interact with basepairs in the major groove of the DNA, and the minor groove is slightly widened (Yang and Steitz, 1995). We suggest that a dimer of Sin binds in a similar way to resH site I. This is supported by dimethyl sulphate (DMS) methylation protection data: a conserved G base at position 2 in each half-site was strongly protected, indicating major groove contacts, and a flanking A base showed enhanced methylation, possibly indicating distortion of the minor groove (Fig. 5B and C).

Figure 5.

Figure 5.

Footprinting of resH.

A. DNase I footprinting of resH. Motifs I-L, I-R, IIa and IIb are indicated. Bonds in the spacer that are hypersensitive in the presence of Sin and Hbsu are arrowed. Estimated protein concentrations: bottom strand: 0, 1.5, 3 and 6 μg ml–1 of Sin, 10 μg ml–1 of Hbsu; top strand: 6 μg ml–1 of Sin, 10 μg ml–1 of Hbsu.

B. DMS methylation protection footprinting of resH, showing cleavage at methylated G bases (top strand; bottom strand data not shown). The G bases at position 2 of motifs I-L, IIa and IIb that are strongly protected by Sin are arrowed. Estimated protein concentrations: 6 μg ml–1 of Sin; 10 μg ml–1 of Hbsu.

C. Summaries of DNase I and DMS methylation protection footprinting data. The 12-bp sequence motifs in sites I and II (boxed; blue arrows), the inverted repeat in the spacer between sites I and II (red arrows) and the centre of symmetry in the spacer (oval) are shown. Superimposed on resH are the inferred 6-bp target regions of Tn552 in pI9789 (yellow highlight; see Fig. 1B). DNase I protection: bonds are shown as: strongly protected from DNase I (▪), weakly protected (half square), hypersensitive (arrows), unassigned owing to weak cutting by DNase I (○), or unprotected (no symbol). Hbsu alone slightly protected the entire fragment; only the enhancements are shown. DMS methylation protection: data for G bases (squares) and A bases (circles) are shown only where an effect was detected. Bases are shown as: unprotected (unfilled symbols), strongly or weakly protected from methylation (filled and half-filled symbols respectively), or enhanced for methylation (arrows). To the right of site II, the dotted line indicates the point at which the sequences of the fragments used to obtain top and bottom strand data diverged.

Figure 5.

Figure 5.

Footprinting of resH.

A. DNase I footprinting of resH. Motifs I-L, I-R, IIa and IIb are indicated. Bonds in the spacer that are hypersensitive in the presence of Sin and Hbsu are arrowed. Estimated protein concentrations: bottom strand: 0, 1.5, 3 and 6 μg ml–1 of Sin, 10 μg ml–1 of Hbsu; top strand: 6 μg ml–1 of Sin, 10 μg ml–1 of Hbsu.

B. DMS methylation protection footprinting of resH, showing cleavage at methylated G bases (top strand; bottom strand data not shown). The G bases at position 2 of motifs I-L, IIa and IIb that are strongly protected by Sin are arrowed. Estimated protein concentrations: 6 μg ml–1 of Sin; 10 μg ml–1 of Hbsu.

C. Summaries of DNase I and DMS methylation protection footprinting data. The 12-bp sequence motifs in sites I and II (boxed; blue arrows), the inverted repeat in the spacer between sites I and II (red arrows) and the centre of symmetry in the spacer (oval) are shown. Superimposed on resH are the inferred 6-bp target regions of Tn552 in pI9789 (yellow highlight; see Fig. 1B). DNase I protection: bonds are shown as: strongly protected from DNase I (▪), weakly protected (half square), hypersensitive (arrows), unassigned owing to weak cutting by DNase I (○), or unprotected (no symbol). Hbsu alone slightly protected the entire fragment; only the enhancements are shown. DMS methylation protection: data for G bases (squares) and A bases (circles) are shown only where an effect was detected. Bases are shown as: unprotected (unfilled symbols), strongly or weakly protected from methylation (filled and half-filled symbols respectively), or enhanced for methylation (arrows). To the right of site II, the dotted line indicates the point at which the sequences of the fragments used to obtain top and bottom strand data diverged.

At site II, our data are consistent with a different mode of binding, with Sin recognizing directly repeated motifs (IIa and IIb; Figs 1B and 5C), not inverted motifs as at site I: (i) the dyad symmetry evident in the DNase I footprint at site I is not seen at site II: the footprint at site IIa resembles those at sites I-L and I-R, but at site IIb the pattern of accessibility differs and less flanking DNA is protected (Fig. 5A and C); (ii) on the top strand, two bonds at equivalent positions in motifs IIa and IIb are relatively accessible to DNase I (* in Fig. 5C); (iii) G base methylation is markedly enhanced near the centre of site II, but not site I (Fig. 5B and C); and (iv) the G bases at position 2 of motifs IIa and IIb are both strongly protected from methylation by DMS, as in motifs I-L and I-R (Fig. 5B and C).

Binding of Sin alone did not alter the DNase I accessibility of the spacer between sites I and II, suggesting that Sin did not constrain the spacer in a stable DNA loop. However, when Hbsu was added, the spacer was protected on both strands, and showed strongly enhanced cleavage by DNase I at two bonds, 10 bp apart, in equivalent positions in the 7-bp inverted motifs (red arrows in Fig. 5C). These hypersensitive sites, which may indicate DNA distortion, correlate with transposition target sites in pI9789 resΔ and pSK1 res (Figs 1B and 5C; see Discussion). The same two bonds, and several others in resH, showed slightly enhanced cleavage by DNase I in the presence of Hbsu alone, but specific protection was not detectable; likewise, no effects of Hbsu alone on DNA methylation by DMS were detected. In the presence of Sin and Hbsu, two pairs of adenines opposite the positions of enhanced DNase I cleavage were partially protected from methylation, consistent with Hbsu contacting, or distorting, the minor groove (Fig. 5C). These data suggest that Hbsu binds strongly at a specific location between sites I and II in the presence of Sin, but weakly and less specifically in the absence of Sin. Hbsu had very little effect on DNase I cleavage in the regions footprinted by Sin, suggesting that it does not significantly perturb Sin binding at sites I and II.

Discussion

Sin functions as a resolvase

In vitro, Sin selectively resolves supercoiled substrates containing directly repeated resH sites (Fig. 2A and B). Thus, its likely function in vivo is to stabilize S. aureus plasmids by resolving multimers, as previously suggested (Paulsen et al., 1994). Although Sin from pI9789 is active, the associated resΔ site is functionally defective in vitro and in E. coli. The truncation of resΔ at the centre of the crossover site (likewise the presumptive res site upstream of entinv; Fig. 1B) suggests that Sin also mediates DNA rearrangements using cryptic res sites. The DNA inversion reaction catalysed by Sin is extremely in-efficient in vitro (Fig. 2B); there is no evidence that inversion is biologically relevant. DNA inversion in vitro has been reported for the β recombinase, and is believed to have a role in plasmid maintenance (Alonso et al., 1995a). At present, the topology of the Sin-mediated inversion reaction, and hence the protein-DNA structure(s) that might be involved, are not known.

Sin differs from characterized transposon-encoded resolvases in its requirement for an accessory DNA bending protein (Fig. 2A). In this respect, it resembles the β recombinases (Alonso et al., 1995a; Petit et al., 1995), and the DNA invertase Hin, which requires HU when the recombination enhancer is in its natural loca-tion (Haykinson and Johnson, 1993). Staphylococcus aureus encodes a protein 75% identical to the B. subtilis Hbsu protein used in our in vitro system (Kuroda et al., 2001).

Binding of Sin at resH

Sin appears to recognize inverted binding motifs at site I and directly repeated motifs at site II (Fig. 5C). At site I, the data are consistent with binding of a Sin dimer in a conformation similar to γδ resolvase (Yang and Steitz, 1995; Fig. 6A. At site II, two Sin monomers could conceivably bind in tandem to the directly repeated motifs IIa and IIb, creating a novel dimer interface. However, similar patterns of Sin-DNA complexes were seen with sites I and II (Fig. 4); we therefore suggest that the monomers at site II form a dimer resembling that at site I, except that the orientation of the DNA-binding domain at site IIb is reversed (Fig. 6A). This would require substantial conformational flexibility in the region linking the catalytic and DNA-binding domains. Crystal structures and NMR spectroscopy of γδ resolvase show that the linker region is flexible (Liu et al., 1993; Yang and Steitz, 1995; and references therein); Sin contains 11 additional residues in this region (Leschziner et al., 1995). A flexible linker between the dimerization and DNA-binding domains is thought to enable binding to inverted or directly repeated half-sites by other homodimeric proteins, including TRF1 (Bianchi et al., 1999) and the arabinose repressor (Carra and Schleif, 1993).

Figure 6.

Models for binding and synapsis.

A. Proposed mode of Sin binding to inverted and directly repeated motifs at resH sites I and II respectively. The orientations of the N-terminal catalytic domain and C-terminal helix–turn–helix domain are shown. At the dimer interface is the long E-helix. The complex at site I is a cartoon of the γδ resolvase co-crystal structure (Yang and Steitz, 1995).

B. A speculative model for the Sin synapse; the proteins and DNA are shown approximately to scale. For clarity, only one resH site is shown as double helical DNA. The two Sin + Hbsu-dependent DNase I hypersensitive bonds seen when footprinting resH are indicated (T, top strand; B, bottom strand; Fig. 5C). The regions targeted by Tn552-family transposons in resH-like sites (Fig. 1B) are shown (Tn).

C and D. The model for the Sin synapse compared with a model for the Tn3/γδ resolvase synapse (Sarkis et al., 2001). Rectangles with arrowheads represent the binding motifs in the DNA. Dimers of recombinase catalytic domains are represented by dominoes; the DNA binding domains are omitted. The contact between the long edges of two dominoes represents a hypothetical synaptic interface between dimers. The contacts between corners of adjacent dominoes (e.g. light and dark dominoes at sites I and II respectively, in the Sin model) correspond to the crystallographic 2,3′ interface between γδ resolvase dimers.

Recognition of either direct or inverted repeats by a flexible dimer may be important for many serine recombinases. The β recombinases, for example, bind at two sites within res; the structure of site II was enigmatic (Petit et al., 1995; Rojo and Alonso, 1995). We note that footprinting and mutagenesis data for β recombinase (Rojo and Alonso, 1995; Canosa et al., 1997) are consistent with binding of a dimer to potential direct repeat motifs at site II (Fig. 1B). Thus, the Sin and β recombinases probably bind to a similar, distinctive array of direct and inverted motifs within res, although the DNA sequences are dissimilar. We have identified the same pattern of repeat motifs in the presumed res sites of some other Gram-positive resolvases (e.g. Tn1546; Fig. 1B). We also note that there are directly repeated motifs at site II of res in some ‘conventional’ resolution systems, which do not require an HU-like protein as a cofactor (e.g. Tn552 and ISXc5; Fig. 7).

Binding of Hbsu

Our footprinting data provide the first direct evidence for site-specific binding of an HU-like protein near the crossover site of a serine recombinase. HU-like proteins act as architectural factors in diverse higher-order nucleoprotein complexes (reviewed in Nash, 1996); specialized methods are often needed to map their binding sites (Lavoie et al., 1996; Aki and Adhya, 1997). There is evidence that Hbsu binds with high affinity to the β recombinase res site (Canosa et al., 1996 and references therein), although binding of Hbsu or HU to res was not detectable by footprinting (Rojo and Alonso, 1995; Petit et al., 1995).

Our data indicate that when Sin occupies sites I and II of resH, Hbsu binds at a specific location in the intersite spacer (Fig. 5). The size of the DNase I footprint in the spacer is compatible with binding of an IHF-like protein. Hbsu is 87% identical to the homodimeric Bacillus stearothermophilus HU protein, which has the same structural fold as IHF (White et al., 1999; and references therein). Hbsu is thus predicted to bind and bend DNA in a similar way to IHF. IHF induces a U-turn in DNA, most of the bending being at two sharp kinks with widened minor grooves approximately one helical turn apart (Rice et al., 1996a). A similar distortion of the spacer DNA in resH by Hbsu would account for the two DNase I-hypersensitive bonds 10 bp apart (Fig. 5C).

High affinity binding of Hbsu occurs only when Sin is bound at resH. This suggests that Hbsu is recruited by direct interactions with Sin, or through recognition of Sin-mediated local distortion or looping of the spacer DNA. Sequence-dependent flexibility of the spacer DNA, which contains inverted 7-bp motifs, could influence the exact position of Hbsu binding.

Topological selectivity

Our topological analysis of resolution by Sin (Fig. 2C and D) implies that strand exchange takes place in a synaptic complex topologically equivalent to that assembled by Tn3/γδ resolvase, with three interwound negative supercoils trapped between the crossover sites. This synapse topology seems to be characteristic of enzymes that function as resolvases, including Tn3, Tn21 and γδ resolvases (reviewed in Stark and Boocock, 1995), Tn552 resolvase (S.-J. Rowland, unpublished), β recombinase (J.C. Alonso, personal communication), ISXc5 resolvase (Liu et al., 1998), and the unrelated tyrosine recombinase XerC/D (Colloms et al., 1997). The assembly of an interwound synapse may be a topological device for selecting appropriate pairs of res sites in cis within supercoiled DNA, as we have discussed for Tn3 resolvase (Stark and Boocock, 1995). Consistent with this idea, when Sin acts on linear DNA substrates, it recombines resH sites in direct or inverted repeat with a similar, low efficiency.

A model for the synapse

We deduced that the topology of the DNA in the Sin and Tn3/γδ resolvase synapses is the same. However, the protein-DNA architectures are necessarily different: resH contains two, not three, recombinase binding sites and also binds Hbsu (Fig. 7). We now propose a simple model of the synapse that satisfies the severe constraints imposed by the compact organization of resH (Fig. 6B). We assume that Sin and Hbsu interact with DNA in similar ways to γδ resolvase and IHF respectively (Yang and Steitz, 1995; Rice et al., 1996a). The essential features of the model are: (i) the two Sin dimers at site II synapse; (ii) Hbsu bends the spacer DNA in an IHF-like manner; and (iii) the two Sin dimers at site I synapse, trapping three highly condensed supercoils in the synaptic com-plex. Molecular modelling (not shown) confirms that a loop of the 31-bp spacer between sites I and II could accommodate a DNA duplex and Hbsu, as required. The positions of Sin + Hbsu-dependent DNase I hypersensitivity in resH (Fig. 5C), which we interpret as DNA kinks, are shown in Fig. 6B. Although we have no evidence that the resH fragments were synapsed in our footprinting assays, the model predicts that the conformation and nuclease sensitivity of the spacer would be little altered by synapsis.

We propose that the Sin dimers at sites I, and at sites II, synapse ‘back-to-back’, with the DNA on the out-side. This hypothetical dimer–dimer interface is structurally plausible (P. Rice, personal communication; Rice and Steitz, 1994), and includes a cluster of residues that are altered in hyperactive Tn3/γδ resolvase mutants (Arnold et al., 1999; J. He and W. Marshall Stark, in preparation); Sin contains eight additional residues in this region. The same interface features in a ‘domain-swapping’ model for strand exchange (M.R. Boocock, X. Zhu and N.D.F. Grindley, in preparation), and in a new model for the Tn3/γδ resolvase synapse (Sarkis et al., 2001), but not in earlier structure-based synapse models (Rice and Steitz, 1994; Murley and Grindley, 1998).

In our model (Fig. 6B and C), the Hbsu-induced U-turn in the spacer brings the Sin dimers at sites I and II of partner resH sites into contact. We propose that this contact is equivalent to the crystallographic 2–3′ interface between γδ resolvase dimers (Rice and Steitz, 1994) and, therefore, predict that mutations in this region of Sin will impair synapsis, as shown for γδ resolvase (Murley and Grindley, 1998).

In our model (Fig. 6B), Hbsu and Sin are not placed in close proximity, suggesting that direct contacts between these proteins may not be needed. This is consistent with our observation that distantly related architectural proteins, IHF and HU, can substitute for Hbsu. Recombination was less efficient with IHF, presumably because the spacer in resH does not resemble an IHF binding site. The β recombinase is stimulated by HMG 1, which is unrelated to Hbsu and binds on the outside of a DNA bend, suggesting that, in this system also, the DNA bending protein does not need to contact the recombinase (Alonso et al., 1995b).

Regulation of recombination

Characterized resolution systems differ greatly in the number and arrangement of regulatory binding sites that are assembled into a functional synapse (Fig. 7). Within these synapses, is there a common architectural unit that regulates strand exchange?

Previously, we suggested that six resolvase dimers assemble the Tn3/γδ synapse, forming tetramers that pair site II with site III in trans, and site I with site I (Stark et al., 1989). In an alternative proposal, based on experiments with hybrid res sites, the two dimers at site III are synapsed, as are the dimers at site I, whereas the dimers at site II are primarily involved in DNA bending (Soultanas and Halford, 1995). This model is similar to our proposal for Sin: two pairs of dimers are synapsed, and Hbsu bends the spacer DNA (Fig. 6B and C). In a current structure-based model of the Tn3/γδ synapse (Sarkis et al., 2001; Fig. 6D), three pairs of dimers, at site I, at site III, and at site II, are synapsed and interact with each other in a filament.

In the Sin synapse model (Fig. 6B and C), the DNA is wrapped around a structural core of four Sin dimers. We suggest that these four dimers specify the topology of the synapse, trap the three interwound supercoils between the crossover sites, and are required to activate strand exchange. A similar unit, comprising the four resolvase dimers at sites I and III, is present in the Tn3/γδ resolvase synapse model (Fig. 6D). In this system, strand exchange is thought to require contacts between dimers at sites I and III (Murley and Grindley, 1998). Comparison of the models suggests that the resolvase dimers at site II of Tn3/γδres have a role analogous to that proposed for Hbsu: they comprise an architectural unit that is peripheral to the structural core of the synapse, and may induce substantial DNA bends that guide the crossover sites into position (Fig. 6C and D). We speculate that other resolvases of the serine recombinase family (see Fig. 7) are also regulated by elaboration of the minimal core architecture used in the Sin system.

Transposon target selection

The region upstream of sin is a target for Tn552-family transposons; out of 10 natural isolates in the current sequence database, seven are inserted into resH-like sites (Fig. 1A; Gillespie et al., 1988; Rowland and Dyke, 1989; Paulsen et al., 1994; S.-J. Rowland et al., our unpublished observations). We can now propose a specific structural basis for this targeting. Insertions map to one of two locations between sites I and II (Fig. 1B). In each case, the deduced transposase cleavage site that is closest to the centre of the spacer is opposite one of the two bonds in resH that are hypersensitive to DNase I in the presence of Sin and Hbsu (Fig. 5C); we interpreted these hypersensitive bonds as positions where Hbsu induces sharp bends or kinks in the DNA (Fig. 6B). We suggest that in vivo, the transpososome recognizes a specific distortion in the DNA that is stabilized by the recombinase-dependent binding of an HU-like protein. Protein-induced DNA distortions are thought to be targeted by other transposable elements, including retroviruses (Muller and Varmus, 1994) and Tn7 (Kuduvalli et al., 2001).

Targeting of transposition to res sites is potentially a significant factor in transposon and plasmid evolution, and has been seen with Tn552-family elements and Tn5053/5090-family elements (carrying integron gene cassettes), which have related transposases. In in vivo experiments, Tn5053 and Tn5090 targeted res sites similar to those of Tn21 and ISXc5, but not Tn3/γδ (Fig. 7); targeting required the presence of the cognate resolvase (Minakhina et al., 1999; Kamali-Moghaddam and Sündstrom, 2000). Tn552-family elements have also targeted res sites that do not resemble resH, including resR in pI9789 (Fig. 1A), which is similar to the res site of Tn552 itself (Fig. 7). The essential structural features of the targets remain to be determined; indeed there may be complex structural and evolutionary relationships between the transposon ‘hunters’ and the res site ‘prey’ (Minakhina et al., 1999).

Tn552 transposase (TnpA) has been purified and characterized in vitro (Rowland et al., 1995; Leschziner et al., 1998). In the presence of TnpA, an excised Tn552 inserts essentially randomly into DNA sequences (Griffin et al., 1999). We are currently attempting to reconstitute the targeting of Sin res sites by Tn552 in vitro.

Experimental procedures

Plasmids

Substrates were constructed by cloning synthetic resH sequences (MluI–BglII fragment; Fig. 1B) into a standard cloning vector, using methods described by Kilbride and colleagues (Kilbride et al., 1999). Site I-containing plasmids contain the MluI–EcoRI segment of resH, and the site II sequence used in binding experiments is the EcoRI–BglII segment of resH. Plasmid sequences are available on request from S.-J. Rowland.

Plasmids were transformed into Escherichia coli DS941 (recF, lacZΔM15, lacIq, StrR) or DH5α (BRL); pSB313 was maintained in E. coli SURE (Stratagene). Supercoiled plasmid DNA was purified using an alkaline lysis method followed by caesium chloride/ethidium bromide centrifugation. DNA concentrations were estimated spectrophotometrically.

Purification of Sin

pI9789 sin (with an Nde I site at the start codon) was subcloned in the expression vector pKET-3a (Rowland et al., 1995). An exponential culture of E. coli BL21(DE3)[pLysS] containing this plasmid was induced by adding IPTG (0.4 mM, for 3 h). Cells were broken by freeze-thaw cycles in buffer A. The insoluble fraction was washed with (i) buffer A and (ii) buffer B. Sin was dissolved in buffer C, renatured by dialysis against buffer B and then precipitated by dialysis against buffer D. Sin was redissolved in buffer E, bound to SP Sepharose (Pharmacia) and eluted in buffer F. Sin was renatured by dialysis against buffer G, precipitated by dialysis against buffer H and redissolved in buffer G. The final fraction (>95% pure, ≈ 0.5 mg ml–1; as estimated by SDS–PAGE) was diluted with an equal volume of 100% glycerol (v/v) and stored at –20°C. The buffers used were: (A) 100 mM NaCl; 20 mM Tris-HCl (pH 7.5), 10 mM MgCl2, 1 mM DTT, 0.5 mM spermine, 0.1 mM EDTA, 0.2 mg ml–1 of PMSF; (B) buffer A, but 2 M NaCl; (C) buffer A, but 6 M urea, and no NaCl; (D) buffer A, but 40 mM NaCl; (E) 6 M urea, 25 mM NaCl, 20 mM Tris-HCl (pH 7.5), 1 mM DTT, 0.1 mM EDTA; (F) buffer E, but 200 mM NaCl; (G) 2 M NaCl; 20 mM Tris-HCl (pH 7.5), 1 mM DTT, 0.1 mM EDTA; (H) buffer G, but 20 mM NaCl.

Enzyme reactions

Purified Sin (≈0.25 mg ml–1) was diluted with 10 mM Tris-HCl (pH 8.2), 0.5 mM EDTA, 0.5 mM DTT, 1 M NaCl, 50% (v/v) glycerol. Purified Hbsu (200 or 250 μg ml–1) was diluted with 25 mM Tris-HCl (pH 7.5), 250 mM NaCl, 0.1 mM EDTA, 50% (v/v) glycerol. Typically, 1 μl of diluted Sin was added to 20 μl of buffer containing 12–20 μg ml–1 of plasmid substrate and 1–20 μg ml–1 of Hbsu; the final reaction buffer contained 60–70 mM NaCl (from the added protein fractions). Reactions were incubated at 37°C for the time indicated. Recombination reactions were stopped by rapid heating to 70°C for 5 min; cleavage reactions were stopped by adding glycerol loading buffer (50% glycerol, 10 mM Tris-HCl (pH 8.0), 200 μg ml–1 of proteinase K, 0.1% SDS; 0.01% bromophenol blue, 0.01% xylene cyanol).

The buffers used were: (1) for recombination reactions: 50 mM Tris-HCl (pH 8.2), 5 mM MgCl2, 5 mM spermidine HCl, 0.1 mM EDTA, 10% (v/v) glycerol; (2) for cleavage reactions: 50 mM Tris-HCl (pH 8.2), 1 mM EDTA, 40% (v/v) ethylene glycol.

A typical DNase I nicking reaction was initiated by adding, to a 10 μl sample, 1.1 μl of a solution containing 0.5–2 μg ml–1 of DNase I, 3 mg ml–1 of ethidium bromide, 50% (v/v) glycerol, 150 mM NaCl and 10 mM Tris-HCl (pH 7.5). After incubating for 30 min at 20°C, 2.5 μl of glycerol loading buffer (as above) was added, and the reaction was extracted with 30–50 μl of 1:1 phenol:chloroform, and loaded onto a 0.7% agarose gel.

All samples were incubated with glycerol loading buffer (as above) before electrophoresis on agarose gels (essentially as described by Stark et al., 1991).

Gel binding assays

Sin (0.5 μl) was added to 10 μl of buffer [50 mM Tris-HCl (pH 8.2), 5 mM MgCl2, 0.1 mM EDTA, 0.1 mM DTT, 10% glycerol, 10 μg ml–1 of supercoiled carrier DNA (pTZ19R; Pharmacia)] containing end-labelled DNA fragment (≈ 50 cps); the added Sin contributed ≈ 50 mM NaCl. When Hbsu (1 μl) was included in the assay mixture, only 0.4 μl of Sin was added. Reactions were incubated for 10 min at 20°C, and then at 0°C before loading onto 6% polyacrylamide gels (Rowland et al., 1995).

DNase I footprinting

Sin (0.2 μl) was added to 4 μl of buffer (as for gel binding, but with 0.01% Triton X-100) containing end-labelled DNA fragment (≈ 500 cps) and Hbsu (0.2 μl). Reactions were incubated for 5 min at 20°C, before adding 1 μl of DNase I (0.5 μg ml–1 in 10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 50% glycerol) and incubating for a further 10 min at 20°C. Stop solution was added (3.5 μl of 94% formamide, 25 mM EDTA, 0.05% bromophenol blue, 0.05% xylene cyanol FF), and a 3 μl aliquot was mixed with 1 μl of a mixture containing 86% formamide, 10 mM EDTA and 440 μg ml–1 of yeast tRNA. Samples were heated to 85°C before loading onto 6% sequencing gels (run using ‘glycerol-tolerant’ buffer: 89 mM Tris, 29 mM taurine, 0.54 mM EDTA; USB Sequenase handbook).

Dimethyl sulphate (DMS) methylation protection footprinting

Sin (1 μl) and Hbsu (1 μl) were added to 18 μl of buffer (final composition as for DNase I footprinting, but without DTT) containing labelled DNA fragment (≈ 1000 cps). The mixture was incubated for 5 min at 20°C, and DMS was added (2 μl of 10% aqueous). After 5 min at 20°C, the reaction was quenched with β-mercaptoethanol (1 μl), and 50 μl of TE buffer (pH 8.0) was added. The DNA was extracted with 1:1 phenol:chloroform and chloroform (70 μl each), and precipitated with ethanol and carrier tRNA. For cleavage at methylated G bases, the DNA was resuspended in 5 μl of double-distilled (dd)H2O, and 50 μl of 1 M piperidine was added. The mixture was heated (at 90°C, for 30 min), and the DNA was lyophilized, and ‘washed’ three times by lyophilization from 10 μl of ddH2O. For cleavage at methylated A and G bases, the DNA was resuspended in 20 μl of ddH2O, and 20 μl of 0.2 M HCl was added before incubating at 0°C for 2 h. After adding 11.7 μl of 1 M NaOH, the mixture was heated (at 90°C, for 30 min). MOPS was added (11.7 μl of 2 M), and the DNA was precipitated with ethanol and carrier tRNA. Redissolved samples were mixed with formamide gel loading buffer and loaded onto sequencing gels (as above).

Mapping the cleavage sites in resH

Sin (5 μl of 0.25 mg ml–1) was added to pSB312 (12 μg ml–1) in 100 μl of cleavage buffer. The reaction was incubated at 37°C for 20 h, and was stopped by adding proteinase K and SDS (500 μg ml–1; 0.1%). The DNA was purified (Wizard resin; Promega) and digested with BamHI + EcoRI; the DNA fragments were labelled using T4 polynucleotide kinase before running on a 10% sequencing gel. The identities of the labelled cleavage products were established by cutting with RsaI (Fig. 3B) and HincII (between BamHI and RsaI in resH; data not shown).

Acknowledgements

We are grateful to J. Alonso, M. Watson, P. Rice and R. Kahmann for gifts of Hbsu, HU, IHF and FIS respectively. We also thank L. Rice and S. Marshall for communicating unpublished sequence data, and K. Dyke for critical comments on the manuscript. This work was funded by the Wellcome Trust.

Ancillary

Advertisement