Article No. jmbi.1999.3150 available online at http://www.idealibrary.

com on

J. Mol. Biol. (1999) 294, 373387

Structure-specific Binding of the Two Tandem HMG Boxes of HMG1 to Four-way Junction DNA is Mediated by the A Domain
Michelle Webb and Jean O. Thomas*
Cambridge Centre for Molecular Recognition and Department of Biochemistry University of Cambridge, 80 Tennis Court Road, Cambridge CB2 1GA, UK We have investigated the nature of the ``structure-specic'' binding of the tandem A and B HMG boxes of high mobility group protein 1 (HMG1) to four-way junction DNA. AB didomain binding favours the open, planar form of the junction, as shown by reaction with potassium permanganate. Site-directed cleavage of the DNA by a 1,10-phenanthroline-copper moiety attached to unique natural or engineered cysteine residues in the A or B domain shows that the two linked HMG boxes are not functionally equivalent in four-way junction binding. The A domain of the didomain binds to the centre of the junction, mediating structurespecic binding; the concave surface of the domain interacts with the widened minor groove at the centre, contacting one of the four strands of the junction, and the short arm comprising helices I and II and the connecting loop protrudes into the central hole. The B domain makes contacts along one of the arms, presumably stabilising the binding of the didomain through additional non-sequence-specic interactions. The isolated B domain can, however, bind to the centre of the junction. The preferential binding of the A domain of the AB didomain to the centre correlates with our previous nding of a higher preference of the isolated A domain than of the B domain for this structurally distinct DNA ligand. It is probably at least partly due to the higher positive surface potential in the DNA-binding region of the A domain (in particular to an array of positively charged side-chains suitably positioned to interact with the negatively charged phosphates surrounding the central hole of the junction) and partly to differences in residues corresponding to those that intercalate between bases in other HMG box/DNA complexes.
# 1999 Academic Press

*Corresponding author

Keywords: HMG box; N-(1,10-phenanthroline-5-yl)iodoacetamide; potassium permanganate; site-specic DNA cleavage

Introduction
The relatively abundant high mobility group proteins HMG1 and 2 (Johns, 1982) in vertebrate cell nuclei bind to DNA with structure-specicity, essentially irrespective of sequence, as rst reported for binding to four-way DNA junctions (Bianchi et al., 1989). The proteins contain two homologous HMG-box domains (A and B) of about 80 amino acid residues whose structures have been determined (Weir et al., 1993; Read et al.,
Abbreviations used: HMG1, high mobility group protein 1; MALDI, matrix-assisted laser desorption ionisation. E-mail address of the corresponding author: jot1@bioc.cam.ac.uk
0022-2836/99/47037315 $30.00/0

1993; Hardman et al., 1995) and an acidic C-terminal tail. The individual HMG-box domains, like HMG1 and 2, bind to four-way junctions (Bianchi et al., 1992; Teo et al., 1995), bend and distort linear DNA in vitro (Paull et al., 1993), bind preferentially to distorted DNA such as cisplatin adducts (Chow et al., 1995) and cause DNA supercoiling and tros & stabilization of DNA loops (Teo et al., 1995; S Reich, 1998). HMG1 and 2 have been implicated in a number of cellular transactions involving DNA (reviewed by Bustin & Reeves, 1996) and appear to act as architectural factors, stimulating the assembly of nucleoprotein complexes (Grosschedl et al., 1994; Grosschedl, 1995). They may facilitate the binding of certain sequence-specic proteins to their target DNA by both protein-protein interactions and DNA bending (see, for example,
# 1999 Academic Press

374 Boonyaratanakornkit et al., 1998; Hiom & Gellert, 1998). Single HMG boxes occur in several relatively abundant proteins that bind to DNA without sequence-specicity as well as in a number of transcription factors that bend their DNA targets. NMR structures have been determined for the HMG boxes of the non-sequence-specic proteins HMG-D from Drosophila (Jones et al., 1994) and NHP6A from yeast (Allain et al., 1999), and for the sequence-specic HMG boxes of lymphoid enhancer factor, LEF-1, and the male sex determining factor, SRY, complexed to short oligonucleotides (Love et al., 1995; Werner et al., 1995). The latter showed that DNA bending occurs through binding of the concave surface of the L-shaped HMG box to the minor groove. Partial intercalation of a hydrophobic residue in helix I of the HMG box (Met in LEF-1, Ile in SRY) between two bases causes opening of the groove and permits formation of multiple contacts with the N-terminal extended strand and helices I and II, with kinking of the DNA toward the major groove. In the recently determined X-ray crystal structure of the non-sequence-specic A domain of HMG1 bound to a short cisplatinated oligonucleotide (Ohndorf et al., 1999), which shows a generally similar mode of binding but with the protein bound asymmetrically with respect to the bend, the intercalating residue is Phe37 in the loop between helices I and II (immediately adjacent to the N terminus of helix II). The corresponding residues in HMG-D (Val32; Balaeff et al., 1998) and yeast NHP6A (Phe48; Allain et al., 1999) were also proposed to intercalate in modelled structures of the respective protein/ DNA complexes, together with methionine residues (Met13 and Met29, respectively) corresponding to the intercalating Met in LEF-1, and to introduce kinks into the DNA. Interestingly, NHP6A binds in reverse orientation to that found in the SRY HMG box/DNA complex and is shifted by one base-pair. HMG boxes occur in two or more copies in some proteins other than HMG1 and 2, for example, four to six copies (depending on species) in the RNA polymerase I transcription factor UBF (Jantzen et al., 1990; Hisatake et al., 1991; Bachvarov & Moss, 1991) and two copies in the mitochondrial transcription factors mtTF1 (Parisi & Clayton, 1991) and ABF2 (Difey & Stillman, 1992), all of which bind to DNA with low sequence specicity, and the structure-specic recognition protein SSRP1 (Bruhn et al., 1992). NMR studies of the free AB didomain of HMG1 showed that the two HMG-box domains are essentially independent (Grasser et al., 1998); an AB didomain/linear oligonucleotide complex was not sufciently well dened for structure determination (unpublished results). We report here a study of the binding of the tandem HMG boxes of HMG1 (the AB didomain) to four-way junction DNA, in which the lack of sequence-specicity of the protein is not an obstacle to formation of a high-afnity complex.

Binding of HMG1 to Four-way Junction DNA

The conformation of the four-way junction is largely dependent on metal ions, which favour the ``stacked X'' conformation over the open, planar conformation in which the arms are maximally separated (Lilley & Clegg, 1993). The AB didomain, like the isolated A domain (Pohler et al., 1998), favours the open form. We have mapped the interaction of the AB didomain with the junction using a site-directed cleavage approach in which the protein is converted, at selected positions, into a ``chemical nuclease'' by a combination of protein engineering and chemical derivatization; this approach has been useful in the study of the interaction of several DNA-binding proteins with linear DNA (Ebright et al., 1992; Pendergrast et al., 1992; Sutton et al., 1993; Dumoulin et al., 1996). The results show clearly that one of the two HMG boxes of the didomain, the A domain, mediates structure-specic binding to the four-way junction.

Results
The AB didomain (residues 1-164) of HMG1 contains both of the HMG-box domains, linked by a short basic region (Grasser et al., 1998). We have used two approaches to investigate the interaction of the AB didomain with the four-way junction. First, we have probed with potassium permanganate, which reacts predominantly with pyrimidine bases in single-stranded DNA (thymine being considerably more reactive than cytosine). This approach has been used to probe for conformational changes in DNA upon protein binding (SasseDwight & Gralla, 1989; White & Lilley, 1997), and recently to show that the isolated A domain of HMG1 binds preferentially to the open conformation of the four-way junction and converts the stacked X to the open form (Pohler et al., 1998). Second, we have carried out site-directed chemical mapping of the DNA in the AB/junction complex. In this approach (Sigman, 1990; Ebright et al., 1990) a single redox-activated 1,10-phenanthroline-copper DNA cleavage group is incorporated at dened positions in the protein, by treatment with N-(1,10phenanthroline-5-yl)iodoacetamide to derivatise the thiol groups of unique cysteine residues, either natural or engineered into the protein. The reactivity of a four-way junction to potassium permanganate is increased upon binding of the AB didomain Potassium permanganate was used to probe the reactivity of thymine bases in a four-way junction in the presence and absence of the AB didomain, and in the presence and absence of Mg2. Addition of the didomain to the junction (Figure 1(a)) in either the open (Mg2) or stacked X conformation (Mg2) increases the reactivity of a single thymine near the centre of the junction in both strand 1 and strand 2 (Figure 1(b); compare lanes 3 and 2, and lanes 5 and 4), showing (i) that binding of the didomain, like the isolated A domain (Pohler et al.,

Binding of HMG1 to Four-way Junction DNA

375

Figure 1. Potassium permanganate probing of a four-way DNA junction complexed with the AB didomain of HMG1. (a) The four-way junction showing the position of the reactive thymine bases. (b) A complex of AB and fourway junction DNA labelled on either strand 1 or strand 2 was subjected to permanganate treatment, in the presence (lanes 3) or absence (lanes 5) of MgCl2, and the DNA was cleaved and analysed in a 20 % polyacrylamide/8 M urea gel; autoradiograms are shown. Lanes 2 and 4 show the corresponding results in the absence of protein. Lanes 1 (marker), strand 1 or strand 2 subjected to a Maxam-Gilbert (A G) sequencing reaction. Arrows indicate bands generated by cleavage at the reactive thymine bases.

1998), favours the open conformation of the junction; and (ii) that protein binding to the open form causes some degree of further opening (compare lanes 5 and 4). As expected, the reactivity of free DNA is greater in the absence of Mg2 (compare lanes 4 and 2). Conversion of the didomain into a ``chemical nuclease'' and mapping of its interaction with four-way junction DNA A chemical-cleavage reagent (copper/phenanthroline) was attached to the thiol groups of unique cysteine residues in the A or B domain of the AB didomain (Grasser et al., 1998), as well as in the isolated A (amino acid residues 1-83) and B (88-164), and larger B0 (84-184), domains (Teo et al., 1995), and cleavages in the four-way junction DNA due to protein binding were mapped. The cysteine residues were naturally occurring unique cysteine residues (in the case of Cys105 in B and B0 ), or unique cysteine residues (Cys22 and Cys105 in AB; Cys44 in the A domain) left after mutagenesis to convert other cysteine residues to alanine (or, in one case, serine), or cysteine residues introduced at selected positions after deleting all the natural cysteine residues. Figure 2(a) shows the positions of the mutations in the didomain sequence. The structure of LEF-1 bound to its cognate recognition sequence (Love et al., 1995) and the NMR structures of the A and B domains (Weir et al., 1993; Hardman et al., 1995) were used as a basis for selection of residues for conversion to cysteine. Residues were chosen (Figure 2(b)) that are predicted to be close to the DNA but not directly

located at the protein-DNA interface; the locations of the resulting cysteine residues in the structures of the A and B domains are shown in Figure 2(c). Four single-cysteine mutants of AB were generated (Figure 2(a) and (b)): ABmut1, a product of the initial mutagenesis experiments to remove two of the three native cysteine residues, leaving a single Cys in helix I of the A domain; ABmut2 (with a single Cys residue engineered into helix III of the A domain); ABmut3 (with a single natural Cys residue in helix I of the B domain); and ABmut4 (with a single Cys residue engineered into helix III of the B domain). ABmut2 and ABmut4 have cysteine residues in corresponding positions in the A and B domain, respectively (Figure 2(c)). CD spectroscopy suggested no signicant structural changes in any of the mutants, and all bound to four-way junction DNA with essentially the same afnity as wild-type AB as judged by gel-retardation assays (data not shown). Other singlecysteine proteins studied (Figure 2(b)) were wildtype B and B0 , and the A domain mutant (Amut; Cys22 ! Ser) used previously for NMR studies (Hardman et al., 1995), which contains a natural cysteine residue (Cys44). Derivatisation with N-(1,10-phenanthroline-5yl)iodoacetamide was greater than 95 % for all the proteins containing single cysteine residues, except B0 (70 %), as judged by a combination of matrixassisted laser desorption ionisation (MALDI) mass spectrometry, UV spectroscopy and thiol titration with 5,50 -dithio-bis(2-nitrobenzoic acid). Derivatised wild-type A and AB which contain, respectively, two and three cysteine residues were heterogeneous due to partial intramolecular oxi-

376

Binding of HMG1 to Four-way Junction DNA

Figure 2. Single cysteine mutants in the HMG boxes. (a) The positions of mutations in the AB didomain. Ala ! Cys and Cys ! Ala mutations are indicated by vertical lines below the main sequence. Native cysteine residues that were not removed to construct unique cysteine AB mutants are indicated by *. Cys22 (underlined) was mutated to an alanine in the didomain and a serine in the isolated A domain (Hardman et al., 1995). The positions of the a-helices in the NMR structures of the isolated A and B domains (Hardman et al., 1995; Weir et al., 1993, respectively) are shown above the sequence. (b) Summary of the various mutants constructed to produce proteins containing single cysteine residues. (c) Structures of the A and B domains of rat HMG1 (Hardman et al., 1995; Weir et al., 1995) showing the positions of the unique cysteine residues (lled circles). The residues are numbered according to their positions in the didomain sequence. The drawing was generated using MOLSCRIPT (Kraulis, 1991).

dation of the two cysteine residues (Cys22 and Cys44) in the A domain and were not investigated further. All the derivatised proteins were able to bind to the four-way junction as judged by gelretardation assays (Figure 3). Modied B0 , ABmut1 and ABmut3 bound with lower afnity than the unmodied proteins (Figure 3(a), (c) and (f), respectively), presumably due to steric hindrance by the phenanthroline moiety, but the conclusions from the cleavages with these proteins were entirely consistent with those for derivatised proteins that bound with high afnity (see below). To map site-directed cleavages, complexes of derivatised proteins were formed with four-way junctions that were uniquely 32P-end-labelled on

each of the four strands in turn. The complexes and free DNA were separated in polyacrylamide gels, as for the gel retardation assays (but omitting EDTA), and the cleavage reaction (see Materials and Methods) was carried out in the gel. Both the complexed DNA and the free DNA were eluted and analysed (Figure 4(a)-(d)). Control experiments (data not shown) showed that cleavage was absolutely dependent upon the modication of the protein. To accommodate differences in loading (necessitating different exposure times for some autoradiograms; see the legend to Figure 4) the radioactivity in the gels was quantied using a PhosphorImager and the data normalised so that all the cleavage data could be directly compared.

Binding of HMG1 to Four-way Junction DNA

377

Figure 3. Proteins derivatised with N-(1,10-phenanthroline-5-yl) iodoacetamide still bind to four-way junction DNA. Gel retardation assays contained 500 nM 32P-labelled synthetic four-way junction and 50, 100, 150, 300 and 500 nM underivatised (lanes 2-6) or derivatised (lanes 8-12) proteins. (a) ABmut1, (b) ABmut2, (c) ABmut3, (d) ABmut 4, (e) Amut, (f) B0 . The samples were analysed in 6.5 % polyacrylamide gels; autoradiograms are shown. Lanes 1 and 7 in all panels contain free DNA.

Figure 5 shows the gel scans for ABmut2 and B0 , and the positions of the cleavages on the junction for these and all the other single-cysteine proteins. The primary sites of cleavage for all the derivatised proteins are clustered around the centre of the junction; cleavage generally occurs at more than one nucleotide position, possibly due to exibility in the linker between the cysteine residue and the phenanthroline moiety, as well as normal thermal motion. ABmut1 and B0 both contain unique native cysteine residues in helix I (Cys22 of the A domain and Cys105 of the B domain, respectively) and both cleave strand 1 (Figure 4(a)) and strand 2 (Figure 4(b)) of the four-way junction. The results for the minimal B domain are similar to those for B0 (data not shown). The efciency of cleavage of strand 1 is essentially the same

for ABmut1 and B0 . However cleavage of strand 2 is greater for B0 than for ABmut1, showing that Cys105 of B0 is closer to strand 2 than Cys22 of AB, with a similar proximity of both cysteines to strand 1. In contrast, no cleavage of strand 1 (Figure 4(a)) was observed for didomain mutants containing a modied cysteine residue in helix III of either the A or B domain (ABmut2 containing Cys63 and ABmut4 containing Cys147) or for Amut with a cysteine residue (Cys44) in helix II of the A domain. Efcient cleavage occurred at strand 2 (Figure 4(b)) for all the derivatised proteins, with the exception of ABmut3, where no cleavage was observed, and ABmut4, where the cleavage was much reduced. Strand 3 (Figure 4(c)) was cleaved by ABmut2, ABmut3, ABmut4 and Amut, although ABmut4 gave less efcient cleavage than the other

378

Binding of HMG1 to Four-way Junction DNA

Figure 4 (legend opposite)

proteins, strand 2 Cys63 of closer to

particularly ABmut2. Cleavage of both and strand 3 by ABmut2 shows that AB is close to both these strands, but strand 3 as shown by more extensive

cleavage of this strand (Figures 4(c) and 5(a)). The converse is true for Amut, where strand 2 is more highly cleaved. None of the derivatised proteins cleaved strand 4 (Figure 4(d)).

Binding of HMG1 to Four-way Junction DNA

379

Figure 4. The 1,10-phenanthroline/copper-mediated site-directed cleavage of four-way junction DNA by the AB didomain and single domain derivatives. Four-way junction DNA with a unique 32P label on (a) strand 1, (b) strand 2, (c) strand 3 or (d) strand 4 was incubated with each of the derivatised proteins. Reactions were analysed in 20 % polyacrylamide/8 M urea gels; autoradiograms are shown. In each panel: rst lane, MaxamGilbert (A G) sequencing reaction for the strand; second lane, free DNA; third lane, the product of cleavage by a derivatised protein as indicated. Exposure times are roughly the same in all cases except for the gels for strand 3 cleaved by ABmut1, ABmut2, ABmut4 and B0 , where roughly three-fold longer exposure was necessary because the marker lanes were weak. (The bands just below the DNA in both the free and bound lanes (particularly for strands 1 and 3 in panels a, c), are present in the free oligonucleotides immediately after labelling and are probably due to a small proportion of incomplete oligonucleotides produced during synthesis (or, less likely, to degradation).)

Based on the cleavage data, we deduce that the A domain binds to the widened minor groove at the centre of the junction, mediating structurespecic binding and, perhaps surprisingly, contacts one of the four apparently structurally equivalent arms selectively (Figure 6) (see Discussion). The position and orientation shown is the only one

compatible with the cleavage data, avoiding steric between the protein backbone and clashes at 42 A DNA. The C-terminal end of helix I, the N-terminal end of helix II and the connecting loop protrude into the central cavity, with the extended N-terminal strand of the domain and the C-terminal helix (helix III) pointing away from the centre, running

380

Binding of HMG1 to Four-way Junction DNA

Figure 5. Summary of sitedirected cleavage data. (a) and (b) Complete data for ABmut2 (Cys in the A domain) and B0 , respectively. Right, PhosphorImager scans of the gels for the reactions of ABmut2 and B0 with each strand, adjusted for different sample loadings. In each case the upper scan is of the Maxam-Gilbert (A G) sequencing reaction; middle scan, cleavage reaction on free DNA; lower scan, cleavage reaction containing the derivatised proteins. Left, diagrammatic representation of the positions of cleavage in the four-way junction (the lled blocks are the positions of highest cleavage for each strand; the thin lines represent positions that are cleaved less extensively). (c) Summary of the positions of the cleavages due to Amut and ABmut1 (Cys in the A domain); ABmut3 and ABmut4 (Cys in the B domain).

almost parallel with the junction arm formed by strands 2 and 3 (Figure 6(a) and (b)). The cleavage results for the single domains B0 (Cys105 in helix I) and Amut (Cys44 in helix II) are consistent with binding of both of the isolated HMG boxes to the centre of the junction (Figure 6(c) and (d)) in the same way as the A domain in the AB didomain.

In contrast, the B domain of the AB didomain is unable to bind in the same way as the A domain, based on two pieces of evidence. First, since the cysteine residues in helix III of the A domain (ABmut2; Cys63) and the B domain (ABmut4; Cys147) of AB are in structurally equivalent positions (Figure 2(c)), the cleavage results for these two proteins would be expected to be identical if

Binding of HMG1 to Four-way Junction DNA

381

Figure 6. A model for the location of the A domain of the AB didomain of HMG1 on the open form of four-way junction DNA, based on the 1,10-phenanthroline/copper-mediated cleavage data shown in Figures 4 and 5. (a) The orientation of the A domain of AB on the four-way junction. Strands 1, 2, 3 and 4 are shown in blue, green, red and yellow, respectively. The B domain makes further contacts along the arm of the junction formed by strands 2 and 3, as shown schematically in Figure 7. (b) The position of the phenanthroline moieties on two unique cysteine residues (shown here together in a single structure for convenience) in the A domain of AB. (c) The position of derivatised Cys44 in the isolated A domain. (d) The isolated B domain (Cys105) bound as for the A domain, as indicated by the cleavage results. (If the B domain of AB were bound in the same way, Cys105 and Cys147 in mutants ABmut3 and ABmut4, respectively (shown in a single structure for convenience) would be placed as shown, but this is incompatible with the cleavage results for these mutant didomains (see the text).) The Brookhaven accession codes for the coordinates of the A and B domains are 1HMF and 1AAB, respectively; the co-ordinates for the phenanthroline moiety were obtained from phenanthroline-modied murine adipocyte lipid-binding protein (Brookhaven accession code 1A18). The unpublished co-ordinates of the four-way junction in the open form (Hargreaves et al., 1998) were provided by Dr J. A. Rafferty. Docking of the proteins on to the four-way junction and modelling of the phenanthroline moieties on to the proteins were done manually using QUANTA. Non-bonded contacts between the backbone of the and above. The drawing was generated using MOLSCRIPT (Kraulis, 1991) and Raster protein and the DNA were 2 A 3D (Merrit & Bacon, 1997).

the AB didomain could bind to the four-way junction with either the A or the B domain interacting at the centre. However, relatively little cleavage of strand 2 or strand 3 is observed for ABmut4 compared to ABmut2 (Figure 4(b) and (c)), showing that the B domain of AB does not bind in the same way as the A domain. (Figure 6(d) shows where Cys147 of ABmut4 would lie if the B domain of AB bound in the same way as the isolated B domain.) Second, since B0 and ABmut3 have a cysteine residue (Cys105) in the same position within the B domain, ABmut3 would be expected to cleave strands 1 and 2 if the B domain of AB bound to the centre of the junction as deduced above for the A domain of AB (and the isolated B

domain). However, ABmut3 cleaves only strand 3. It is therefore likely that the B domain of AB binds in the minor groove of the junction arm formed by strands 2 and 3, with helices I and II oriented toward the centre of the junction, accounting for the strong cleavage of strand 3 and lack of cleavage of strands 1 and 2 by ABmut3. The position of the two domains is shown schematically in Figure 7. The representation of the B domain with the junction arm does not show any DNA bending that might occur. The A and B domains have strikingly different surface potentials in the DNA-interacting regions (Figure 8) which might be a major factor in the different preferences of the two domains of the

382

Binding of HMG1 to Four-way Junction DNA

Discussion
HMG1 and 2 bind to DNA with little or no sequence specicity but show appreciable afnity for distorted structures such as four-way junction DNA (Bianchi et al., 1989), which appears to be a universal ligand for HMG boxes and indeed a number of other ``architectural'' proteins (Hill et al., 1999; Zlatanova & van Holde, 1998). There is at present no structural information available on how tandem HMG boxes bind to any DNA. Our aims were to determine whether one or other (or both) of the two HMG-box domains is responsible for structure-specic binding to four-way junction DNA and to dene the location of the AB didomain of HMG1 on a junction. The rst consideration in interpretation of the site-directed cleavage results was the conformation of the junction in the protein-DNA complex; in the absence of protein this exists in a Mg2-dependent equilibrium between open and stacked X forms (Lilley & Clegg, 1993). Permanganate cleavage (Figure 1) showed that binding of the AB didomain to the four-way junction DNA under conditions that stabilise either the stacked X or the open conformation of the junction results in an increase in the reactivity of thymine bases at the centre of the junction, showing that the didomain promotes the open form, like the isolated A domain (Pohler et al., 1998). Enhanced reactivity even in the absence of magnesium, conditions that already favour junction opening, suggests that the protein results in further displacement of the equilibrium position toward the open conformation; alternatively, protein binding promotes a conformation distinct from both the stacked X and open structures, in which the bases at the centre of the junction are unstacked or possibly even unpaired, as suggested for the yeast CCE1 resolving enzyme (White & Lilley, 1997). Unpairing is evident in the recently determined structures of the Cre recombinase/ junction complex (Gopaul et al., 1998) and the RuvA/DNA complex with two RuvA tetramers bound to a junction (Roe et al., 1998), although not in a RuvA/DNA complex containing one tetramer (Hargreaves et al., 1998).

Figure 7. Schematic representation of the AB didomain of HMG1 bound to four-way junction DNA, based on the 1,10-phenanthroline/copper-mediated cleavage data shown in Figures 4 and 5. The A domain (A) binds selectively to the junction point, mediating structurespecic binding and the B domain (B), shown schematically positioned near the minor groove, binds to the junction arm formed by strands 2 and 3. The linker region between the two domains is represented schematically by the broken line. The drawing was generated using MOLSCRIPT (Kraulis, 1991) and Raster 3D (Merrit & Bacon, 1997).

didomain for a four-way junction. The A domain is generally more positively charged (shown in blue in Figure 8) in the helix I/helix II regions than the B domain and in the model based on the cleavage data (Figure 6(a) and (b)) is positioned on the junction such that a ``ring'' of several basic side-chains is located close enough to the DNA at the centre of the junction to make electrostatic contacts with the surrounding phosphate backbones. Other differences between the A and B domains might also contribute to their different preferences for fourway junctions (see Discussion).

Figure 8. Surface potential of the A and B domains of rat HMG1, displayed using the program GRASP (Nicholls et al., 1991). Blue, positive potential; red, negative potential.

Binding of HMG1 to Four-way Junction DNA

383 HMG1 and Phe in the B domain. In the complex of the A domain with (pre-bent) cisplatin-modied DNA the intercalating residue is Phe37, immediately adjacent to the N-terminal end of helix II (Ohndorf et al., 1999). The importance of this residue in binding to the four-way junction (another pre-distorted substrate) is not known but it may be signicant that, in the model shown in Figure 6(a), Phe37 packs against the base in strand 2 of the arm formed by strands 2 and 3, at the centre of the junction (not shown); in the B domain the corresponding residue is Ile. That the didomain binds specically to one arm (strands 2 and 3) at the centre of the junction even though, in the open conformation of the junction, there are four potentially equivalent binding sites on the four arms, may reect the local conformational exibility of the DNA. Strand 2 contains the conformationally exible dinucleotide step 50 TG-30 (El Hassan & Calladine, 1996) close to the centre of the junction and may offer the greatest possibility for distortion near the centre on protein binding (consistent with the permanganate cleavage results; Figure 1). Moreover, distortion of the DNA at, and abutting, the centre of the junction caused by binding of the A domain could conceivably provide a preferential binding site for the B domain of the didomain on the same arm. Other possible causes of the preferential selection of one of the four arms would include further conformational changes in the junction upon protein binding or possibly some degree of sequencespecic recognition of one of the four arms, for example a contribution of the stacking of Phe37 in the A domain against the base (thymine) at the junction end of the arm. Recent hydroxyl radical footprinting of HMG1 on a different four-way junction is consistent with the cleavage results shown here in that binding occurs around the junction point, but in that case all four arms were reported to show (unequal) protection (data shown for one strand of one arm; Hill et al., 1999). The nature of the binding of the A domain deduced here (Figure 6(a)) has much in common with the binding of other HMG boxes to DNA in structurally dened complexes (Love et al., 1995; Werner et al., 1995; Ohndorf et al., 1999) in that the concave face makes extensive contacts with a widened minor groove. However, the cleavage data indicate that the HMG box is positioned on the four-way junction in such a way that the relative orientation of the concave surface and the minor groove is slightly different from that in complexes with linear DNA, as a consequence of binding of the end of the short arm in the central cavity, which is stabilised by a constellation of positive charges in the arm. It is perhaps not surprising that there should be some differences between protein binding to the four-way junction in which minor groove widening is inherent in the structure where the four strands meet at the centre (von Kitzing et al., 1990) and binding to a widened minor groove at a kink in linear DNA.

The site-directed cleavage results show that when the AB didomain of HMG1 binds to the four-way junction the mode of binding of the two boxes is different, as shown schematically in Figure 7 (the interdomain linker, which is not represented in the NMR structures of the A and B domains (Hardman et al., 1995; Weir et al., 1993), is shown as a broken line). The A domain binds at the centre of the junction with the short arm of the HMG box protruding into the hole in the centre, and mediates the structure-specic binding of the didomain (Figure 6(a) and (b)). The B domain is excluded from the centre and binds quite differently, making contacts with the minor groove of the junction arm formed by strands 2 and 3 (Figure 7), possibly bending it (not shown), although the data do not bear on this. Binding of the second domain contributes to the overall binding energy, consistent with the higher afnity of the didomain than of the individual HMG-box domains for the four-way junction, as well as other DNA ligands (Grasser et al., 1998). DNA contacts with the B domain are evidently important, since derivatisation of Cys105 in helix I of the B domain in ABmut3 reduced its afnity for the four-way junction (Figure 3(d)). There is nothing intrinsically different about the B domain, however; in the absence of the A domain the isolated B domain is able to bind to the centre of the junction, as shown by the cleavage data. Several factors might contribute to the different preferences of the A and B domains of the didomain for a four-way junction and the preferential binding of the A domain to the centre. First, differences in the number, nature and distribution of charged surface residues (Figure 8), particularly in the short arm of the HMG box, might be a major factor. Such differences might be reected in the higher afnity of the isolated A domain than of the B domain for four-way junction DNA, and in the marked preference of the A domain, even compared with the B0 domain (which binds with a higher afnity than A), for four-way junctions in competition assays with supercoiled and linear plasmid DNA (Teo et al., 1995). Basic residues in helices I and II are also important in the assembly of DNA ``cross-overs'' consisting of two linear duplexes (Hu et al., 1998). Second, a slight difference in the orientation of helices I and II (Hardman et al., 1995) in the A and B domains might also be relevant. Mutation of Pro31 in the loop between the helices to Ala, which might alter the orientation of the helices, reduced the afnity of the isolated A domain for four-way junction DNA (Falciola et al., 1994). A further contribution to the differences between the A and B domains might come from the nature of the amino acid in the position corresponding to the intercalating Ile in helix I of the HMG box of SRY (which is responsible for DNA bending), replacement of which by Ala abolished binding to linear DNA but enhanced the afnity for four-way junction DNA (Weiss et al., 1997). The corresponding residue is Ala in the A domain of

384 A model for the interaction of the A domain with four-way junction DNA has been proposed based on the structure of the HMG box of LEF-1 bound to a 15 bp oligonucleotide containing its binding site, and bending it by intercalation of a methionine side-chain (Pohler et al., 1998). The model has broad features in common with the mode of interaction proposed here based on the chemical cleavage data but there are also some signicant differences. In both cases the A domain binds primarily to the widened minor groove at the centre of the junction. In the model described by Pohler et al. (1998) the short arm of the Lshaped structure, comprising helices I and II, again projects into the open centre of the junction, but much less deeply than in the model shown in Figure 6 in which electrostatic contacts of basic residues in helices I and II with phosphates around the centre of the junction play a much greater role in positioning of the protein, and constraints imposed by intercalation of the residue corresponding to the Met in LEF-1 (Ala in the A domain of HMG1) do not apply. The long arm (the extended N terminus and helix III) again points away from the centre in the model of Pohler et al., as in the model shown in Figure 6, but is positioned out of the plane of the arms of the junction, rather than in the plane as suggested by the experimental cleavage data and shown in Figure 6, and would be incompatible with the observed cleavage of strand 3 by ABmut2 (Figures 4 and 5). In summary, this study has dened the way in which the two tandem HMG boxes of HMG1 bind to four-way junction DNA, which is in its open form. Only one box of the linked pair, the A domain, can bind to the structure-specic feature(s) at the centre of the junction: a widened minor groove and a central cavity. We see no evidence for alternative binding of the B domain of the AB didomain to the centre, although in principle this should be possible because the two domains are joined by a exible linker; NMR spectroscopy (Grasser et al., 1998) and differential scanning calorimetry (Ramstein et al., 1999; unpublished results) show that they do not interact in the free protein. That structure-specic binding to the junction is through the A domain rather than the B domain is probably due at least partly to the greater number of basic residues in A suitably positioned to form electrostatic interactions with DNA, although other factors (such as the packing of an aromatic side-chain against a base at the centre of the junction in the A domain) may also operate. In intact HMG1, reduction of the afnity of the two HMG boxes for four-way junctions (as well as for linear DNA) in the presence of the acidic tail is presumably due either to general electrostatic repulsion or to direct interaction of the tail with one (or both) of the HMG boxes (Ramstein et al., 1999). Although four-way junction DNA is almost certainly not the natural ligand for HMG1 and 2, these experimental observations on tandem HMG boxes bound to this distorted DNA are likely to be

Binding of HMG1 to Four-way Junction DNA

relevant to the architectural role of HMG1 and 2 in the stabilization of nucleoprotein complexes in vivo. They also show clearly that the two HMG-box domains in the didomain show distinctly different behaviour (as do the isolated HMG-box domains in earlier studies; Teo et al., 1995), which is likely to be relevant. One possibility is that if one HMG box is principally concerned with structure-specic DNA binding, as in the case of the four-way junction, the second box (in addition to stabilizing the complex through non-sequence-specic interactions) might be a target for protein-protein interactions, e.g. with sequence-specic proteins involved in transcription (e.g. Zwilling et al., 1995; Zappavigna et al., 1996; Boonyaratanakornkit et al., 1998) or recombination (e.g. Hiom & Gellert, 1998). This might be the means by which non-sequencespecic proteins such as HMG1 can be recruited to promote the assembly and/or stabilization of nucleoprotein complexes at specic DNA sequences.

Materials and Methods

Protein expression and purification Wild-type and mutant (see below) protein domains were expressed in Escherichia coli BL21(DE3) cells and puried to homogeneity from crude cell extracts by ionexchange and hydrophobic chromatography as described (Weir et al., 1993; Grasser et al., 1998). Protein concentrations were determined by automated amino acid analysis.

Preparation of labelled four-way junction DNA The following 30 bp oligonucleotides were annealed as described (Weir et al., 1993) to form the four-way junction: strand 1, 50 -GAATTCAGCACGAGTCCTAACGCCAGATCT-30 ; strand 2, 50 -AGATCTGGCGTTAGGTGA TACCGATGCATC-30 ; strand 3, 50 -GATGCATCGGTATCAGGCTTACGACTAGTG-30 ; strand 4, 50 -CACTAGTC GTAAGCCACTCGTGCTGAATTC-30 . The corrected extinction coefcients at 260 nm of the oligonucleotides were determined using the observed hypochromicity following digestion with (Sigma) snake venom phosphodiesterase (Kallansrud & Ward, 1996) and were used to determine the concentrations of oligonucleotide solutions. One nmol of each of the four oligonucleotides was radiolabelled using phage T4 polynucleotide kinase (New England Biolabs) and [g-32P]ATP (Amersham), and unincorporated ATP was removed using a Sephadex G-25 spin column. The labelled oligonucleotides were ethanol precipitated and resuspended in 400 ml of the same unlabelled oligonucleotide at a known concentration, generally 5-10 mM. Junctions uniquely labelled on one strand were formed by incubating equimolar amounts of the four strands, only one of which was labelled, in 10 mM Tris-HCl (pH 7.8), 50 mM NaCl at 95  C for ve minutes and cooling slowly to 4  C. The concentration of the labelled four-way junction was determined from the absorbance at 260 nm using a corrected extinction coefcient, determined as above.

Binding of HMG1 to Four-way Junction DNA Potassium permanganate cleavage An equimolar mixture of protein and four-way junction DNA uniquely labelled on one of the four strands was incubated in 20 mM Tris-HCl (pH 8), 50 mM NaCl, 2.5 mg calf liver tRNA (Boehringer Mannheim; included to aid the subsequent precipitation of small DNA fragments) and where stated, 100 mM MgCl2, in a total volume of 25 ml. The reaction was initiated by the addition of 5 ml of 10 mM KMnO4 and quenched after one minute with 2 ml of 2-mercaptoethanol. The DNA was ethanol precipitated, resuspended in 100 ml of 1 M piperidine and heated to 95  C for 30 minutes to induce cleavage at the sites of oxidation. Piperidine was removed by precipitation of the DNA rst with 1 ml of butan-1-ol (AR grade) and then with 1 ml of butan-1-ol containing 0.1 % (w/v) SDS. The dried samples were dissolved in 10 ml of formamide and the DNA was analysed by electrophoresis in a 20 % polyacrylamide/8 M urea gel containing 1 TBE (89 mM Tris, 89 mM boric acid, 2 mM EDTA), alongside markers of individual oligonucleotides subjected to an (A G) chemical sequencing reaction (Maxam & Gilbert, 1980). Gels were vacuumdried and autoradiographed at 70  C using pre-ashed Fuji RX lm. Site-directed mutagenesis The gene for the wild-type rHMG1 didomain (AB) in pT7-7 (Grasser et al., 1998) was mutated using the QuikChange mutagenesis kit (Stratagene) as recommended by the supplier. After each round of mutagenesis, to replace a cysteine residue with alanine or vice versa, the DNA sequence of the new mutant was checked. Four single-cysteine mutants of AB were generated: ABmut1 and ABmut3, products of the initial mutagenesis experiments to remove some of the native cysteine residues; and ABmut2 and ABmut4 in which unique cysteine residues were selectively introduced into a cysteine-free construct in which all the original cysteine residues had been replaced by alanine. ABmut1 and ABmut3 contain a single cysteine residue at position 22 of the A domain and position 105 of the B domain, respectively; ABmut2 and ABmut4 contain a single cysteine at position 63 of the A domain and position 147 of the A and B domain, respectively. N-(1,10-phenanthroline-5-yl)iodoacetamide derivatisation and DNA cleavage Proteins (5-20 mM) were dialysed extensively into 10 mM sodium phosphate (pH 7), 50 mM NaCl, to remove dithiothreitol (DTT) which is present in the standard storage buffer (Weir et al., 1993; Hardman et al., 1995; Grasser et al., 1998) and treated with a 30-fold molar excess of N-(1,10-phenanthroline-5-yl)iodoacetamide (Molecular Probes) in 3 % (v/v) dimethylformamide (DMF), 10 mM sodium phosphate (pH 7), 50 mM NaCl. (N-(1,10-phenanthroline-5-yl)iodoacetamide was dissolved in DMF to a concentration that would result in a nal concentration of 3 % DMF in the reaction mixture.) The samples were incubated at 23  C for three hours and then dialysed into 10 mM Tris-HCl (pH 7.8), 50 mM NaCl. To check the derivatised protein for DNA binding in a gel-retardation assay, 32P-end-labelled four-way junction (500 nM) was incubated with 50, 100, 150, 300 and 500 nM protein in 10 mM Tris-HCl (pH 7.8), 50 mM NaCl, 1 mM DTT, 5 % (v/v) glycerol, incubated at 4  C

385
for 15 minutes, and analysed by electrophoresis in a 6.5 % polyacrylamide gel in 1 TBE at 150 V at 4  C for ca. two hours. Gels were then vacuum dried and autoradiographed overnight at 70  C. To ensure that the DNA cleavage reactions on the derivatised proteins were being carried out on an electrophoretically homogeneous species, reactions were performed within the gel matrix. In a typical experiment an equimolar mixture of protein and labelled four-way junction DNA was incubated for 15 minutes at 23  C in 10 mM Tris-HCl (pH 7.8), 50 mM NaCl, 5 % glycerol, 1 mM DTT, and then loaded on to a 6.5 % polyacrylamide gel containing 89 mM Tris, 89 mM boric acid. After electrophoresis as above, gels were incubated in 20 mM Tris-HCl (pH 7.8) for ve minutes at 23  C. To initiate nucleolytic activity, copper sulphate was added to a nal concentration of 40 mM followed by 3-mercaptopropionic acid (Sigma) to a nal concentration of 0.5 mM. After 30 minutes at 23  C the reaction was quenched with 7 mM neocuproine hydrate (Sigma). Regions of the gel containing free DNA and the proteinDNA complex, located by exposure of the gel to X-ray lm for two hours at 4  C, were excised and the gel slices were incubated overnight at 42  C in 400 ml of elution buffer (0.5 M ammonium acetate, 1 mM EDTA, 0.1 % SDS, 10 mM MgCl2). The eluted products were ethanol precipitated, dissolved in 95 % (v/v) formamide and analysed by denaturing gel electrophoresis in a 20 % polyacrylamide/8 M urea gel containing 1 TBE, alongside markers of individual oligonucleotides subjected to an (A G) chemical sequencing reaction (Maxam & Gilbert, 1980). The gels were dried under vacuum without xing for one hour and autoradiographed at 70  C using pre-ashed Fuji RX lm. They were also scanned on a PhosphorImager (Molecular Dynamics) and the data analysed using Image Quant software, with normalisation to accommodate variation in radioactivity loaded in different samples.

Acknowledgements
We thank Dr John Rafferty (Krebs Institute, University of Shefeld) for supplying the unpublished co-ordinates of the open form of the four-way junction, and Drs Helen Mott and Kathryn Phillips for help with the molecular graphics programs used to generate Figures 6 and 7. Amino acid analysis, amino acid sequencing, oligonucleotide synthesis, automated DNA sequencing and mass spectrometry were carried out in the Cambridge Centre for Molecular Recognition (CCMR), which is supported by the Biotechnology and Biological Sciences Research Council of the UK (BBSRC) and the Wellcome Trust. This work was supported by the BBSRC and by the Ernest Oppenheimer Fund of the University of Cambridge.

References
Allain, F. H.-T., Yen, Y.-M., Masse, J. E., Schultze, P., Dieckmann, T., Johnson, R. C. & Feigon, J. (1999). Solution structure of the HMG protein NHP6A and its interaction with DNA reveals the structural determinants for non-sequence-specic binding. EMBO J. 18, 2563-2579. Bachvarov, D. & Moss, T. (1991). The RNA polymerase 1 transcription factor xUBF contains 5 tandemly

386
repeated HMG homology boxes. Nucl. Acids Res. 19, 2331-2335. Balaeff, A., Churchill, M. E. A. & Schulten, K. (1998). Structure prediction of a complex between the chromosomal protein HMG-D and DNA. Proteins: Struct. Funct. Genet. 30, 113-135. Bianchi, M. E., Beltrame, M. & Paonessa, G. (1989). Specic recognition of cruciform DNA by nuclear protein HMG1. Science, 243, 1056-1059. Bianchi, M. E., Falciola, L., Ferrari, S. & Lilley, D. M. J. (1992). The DNA binding site of HMG1 protein is composed of two similar segments (HMG boxes), both of which have counterparts in other eukaryotic regulatory proteins. EMBO J. 11, 1055-1063. Boonyaratanakornkit, V., Melvin, V., Prendergast, P., Altmann, M., Ronfani, L., Bianchi, M. E., Taraseviciene, L., Nordeen, S. K., Allegretto, E. A. & Edwards, D. P. (1998). High-mobility group chromatin proteins 1 and 2 functionally interact with steroid hormone receptors to enhance their DNA binding in vitro and transcriptional activity in mammalian cells. Mol. Cell. Biol. 18, 4471-4487. Bruhn, S. L., Pil, P. M., Essigman, J. M., Housman, D. E. & Lippard, S. J. (1992). Isolation and characterization of human cDNA clones encoding a high mobility group protein that recognises structural distortions to DNA caused by binding of the anticancer agent cisplatin. Proc. Natl Acad. Sci USA, 89, 2307-2311. Bustin, M. & Reeves, R. (1996). High mobility-group chromosomal proteins: architectural components that facilitate chromatin function. Progr. Nucl. Acid Res. Mol. Biol. 54, 35-100. Chow, S. C., Barnes, C. M. & Lippard, S. J. (1995). A single HMG domain in high-mobility-group 1 protein binds to DNAs as small as 20 base-pairs containing the major cisplatin adduct. Biochemistry, 34, 2956-2964. Difey, J. F. X. & Stillman, B. (1992). DNA binding properties of an HMG1-related protein from yeast mitochondria. J Biol. Chem. 267, 3368-3374. Dumoulin, P., Ebright, R. H., Knegtel, R., Kaptein, R., Granger-Schnarr, M. & Schnarr, M. (1996). Structure of the Lex A repressor-DNA complex probed by afnity cleavage and photo-cross-linking. Biochemistry, 35, 4279-4286. Ebright, R. H., Ebright, Y. W., Pendergrast, P. S. & Gunsekara, A. (1990). Conversion of a helix-turnhelix motif sequence-specic DNA binding protein into a site-specic cleavage reagent. Proc. Natl Acad. Sci. USA, 87, 2882-2886. Ebright, Y. W., Pendergrast, P. S. & Ebright, R. H. (1992). Incorporation of an EDTA-metal complex at a rationally selected site within a protein: application to EDTA-iron DNA afnity cleaving with catabolite gene activator protein (CAP) and Cro. Biochemistry, 31, 10664-10670. El Hassan, M. A. & Calladine., R. C. (1996). Propellertwisting of base-pairs and the conformational mobility of dinucleotide steps in DNA. J. Mol. Biol. 259, 95-103. Falciola, L., Murchie, A. I. H., Lilley, D. M. J. & Bianchi, M. E. (1994). Mutational analysis of the DNA binding domain A of chromosomal protein HMG1. Nucl. Acids Res. 22, 285-292. Gopaul, D. N., Guo, F. & Van Duyne, G. D. (1998). Structure of the Holliday junction intermediate in Cre-loxP site-specic recombination. EMBO J. 14, 4175-4187.

Binding of HMG1 to Four-way Junction DNA Grasser, K. D., Teo, S.-H., Lee, K.-B., Broadhurst, R. W., Rees, C., Hardman, C. H. & Thomas, J. O. (1998). DNA-binding properties of the tandem HMG boxes of high-mobility-group protein 1 (HMG1). Eur. J. Biochem. 253, 787-795. Grosschedl, R. (1995). Higher-order nucleoprotein complexes in transcription: analogies with site-specic recombination. Curr. Opin. Cell Biol. 7, 362-370. Grosschedl, R., Giese, K. & Pagel, J. (1994). HMG domain proteins architectural elements in the assembly of nucleoprotein structure. Trends Genet. 10, 94-100. Hardman, C. H., Broadhurst, R. W., Raine, A. R. C., Grasser, K. D., Thomas, J. O. & Laue, E. D. (1995). Structure of the A domain of HMG1 and its interaction with DNA as studied by heteronuclear threeand four-dimensional NMR spectroscopy. Biochemistry, 34, 16596-16607. Hargreaves, D., Rice, D. W., Sedelnikova, S. E., Artymiuk, P. J., Lloyd, R. G. & Rafferty, J. B. (1998). Crystal structure of the E. coli RuvA with bound resolution. Nature DNA Holliday junction at 6 A Struct. Biol. 5, 441-446. Hill, D. A., Pedulla, M. A. & Reeves, R. (1999). Directional binding of HMG-I(Y) on four-way junction DNA and the molecular basis for competitive binding with HMG-1 and histone H1. Nucl. Acids Res. 27, 2135-2144. Hiom, K. & Gellert, M. (1998). Assembly of a 12/23 paired signal complex: a central control point in V(D)J recombination. Mol. Cell, 7, 1011-1019. Hisatake, K., Nishimura, T., Maeda, Y., Hanada, K., Song, C.-Z. & Muramatsu, M. (1991). Cloning and structural analysis of cDNA and the gene for mouse transcription factor UBF. Nucl. Acids Res. 19, 4631-4637. Hu, C.-H., Wang, J.-M. & Tseng, H.-B. (1998). The rst high-mobility-group box of the upstream binding factor assembles a cross-over DNA junction by basic residues. Biochem. J. 333, 51-58. Jantzen, H.-M., Admon, A., Bell, S. P. & Tjian, R. (1990). Nucleolar transcription factor hUBF contains a DNA-binding motif with homology to HMG proteins. Nature, 344, 830-836. Johns, E. W. (1985). The HMG Chromosomal Proteins, Academic Press Inc., London. Jones, D. N. M., Searles, M. A., Shaw, G. L., Churchill, M. E. A., Ner, S. S., Keeler, J., Travers, A. A. & Neuhaus, D. (1994). The solution structure and dynamics of the HMG-box motif of HMG-D from Drosophila melanogaster. Structure, 2, 609-627. Kallansrud, G. & Ward, B. (1996). A comparison of measured and calculated single and double stranded oligodeoxynucleotide extinction coefcients. Anal. Biochem. 236, 134-138. Kraulis, P. J. (1991). MOLSCRIPT - a program to produce both detailed and schematic plots of protein structures. J. Appl. Crystallog. 24, 946-950. Lilley, D. M. J. & Clegg, R. M. (1993). The structure of the four-way junction in DNA. Annu. Rev. Biophys. Biomol. Struct. 22, 299-328. Love, J. J., Xiang, L., Case, D. A., Giese, K., Grosschedl, R. & Wright, P. E. (1995). Structural basis for DNA bending by the architectural transcription factor LEF-1. Nature, 376, 791-795. Maxam, A. M. & Gilbert, W. (1980). Sequencing endlabeled DNA with base-specic chemical cleavages. Methods Enzymol. 65, 499-559.

Binding of HMG1 to Four-way Junction DNA Merritt, E. A. & Bacon, D. J. (1997). Raster3D: photorealistic molecular graphics. Methods Enzymol. 227, 505-524. Nicholls, A., Sharp, K. & Honig, B. (1991). Protein folding and association: insights from the interfacial and thermodynamic properties of hydrocarbons. Proteins: Struct. Funct. Genet. 11, 281-296. Ohndorf, U-M., Rould, M. A., He, Q., Pabo, C. O. & Lippard, S. J. (1999). Basis for recognition of cis platin-modied DNA by high-mobility group proteins. Nature, 399, 708-712. Parisi, M. A. & Clayton, D. A. (1991). Similarity of human mitochondrial transcription factor 1 to high mobility group proteins. Science, 252, 965-968. Paull, T. T., Haykinson, M. J. & Johnson, R. C. (1993). The non-specic DNA-binding and -bending proteins HMG1 and HMG2 promote the assembly of complex nucleoprotein structures. Genes Dev. 7, 1521-1534. Pendergrast, P. S., Chen, Y., Ebright, Y. W. & Ebright, R. H. (1992). Determination of the orientation of a DNA-binding motif in a protein DNA complex by photo-cross-linking. Proc. Natl Acad. Sci. USA, 89, 10287-10291. Pohler, J. R. G., Norman, D. G., Bramham, J., Bianchi, M. E. & Lilley, D. M. J. (1998). HMG box proteins bind to four-way DNA junctions in their open conformation. EMBO J. 17, 817-826. Ramstein, J., Locker, D., Bianchi, M. E. & Leng, M. (1999). Domain-domain interactions in high mobility group protein 1 (HMG1). Eur. J. Biochem. 260, 692-700. Read, C. M., Cary, P. D., Crane-Robinson, C., Driscoll, P. C. & Norman, D. G. (1993). Solution structure of a DNA-binding domain from HMG1. Nucl. Acids Res. 21, 3427-3436. Roe, M. S., Barlow, T., Brown, T., Oram, M., Keeley, A., Tsaneva, I. R. & Pearl, L. H. (1998). Crystal structure of an octameric RuvA-Holliday junction complex. Mol. Cell, 2, 361-372. Sasse-Dwight, S. A. & Gralla, J. D. (1989). KMnO4 as a probe for lac promoter DNA melting and mechanism in vivo. J. Biol. Chem. 264, 8074-8081. Sigman, D. S. (1990). Chemical nucleases. Biochemistry, 29, 9097-9105.

387
Sutton, C. L., Mazumder, A., Chen, C. B. & Sigman, D. S. (1993). Transforming the Escherichia coli Trp repressor into a site-specic nuclease. Biochemistry, 32, 4225-4230. tros, M. & Reich, J. (1998). Formation of large nucleoS protein complexes upon binding of the high-mobility-group (HMG) box B-domain of HMG1 protein to supercoiled DNA. Eur. J. Biochem. 251, 427-434. Teo, S.-H., Grasser, K. D. & Thomas, J. O. (1995). Differences in the DNA-binding properties of the HMGbox domains of HMG1 and the sex-determining factor SRY. Eur. J. Biochem. 230, 943-950. von Kitzing, E., Lilley, D. M. J. & Dieckmann, S. (1990). The stereochemistry of a four-way DNA junction: a theoretical study. Nucl. Acids Res. 18, 2671-2683. Weir, H. M., Kraulis, P. J., Hill, C. S., Raine, A. R. C., Laue, E. D. & Thomas, J. O. (1993). Structure of the HMG box motif in the B-domain of HMG1. EMBO J. 12, 1311-1319. Weiss, M. A., Ukiyama, E. & King, C. Y. (1997). The SRY cantilver motif discriminates between sequence- and structure-specic DNA recognition: alanine mutagenesis of an HMG box. J. Biomol. Struct. Dynam. 15, 177-184. Werner, M. H., Ruth, J. R., Gronenborn, A. M. & Clore, G. M. (1995). Molecular-basis of human 46X,Y sex reversal revealed from the 3-dimensional solution structure of the human Sry-DNA complex. Cell, 81, 705-714. White, M. F. & Lilley, D. M. J. (1997). The resolving enzyme CCE1 of yeast opens the structure of the four-way junction DNA. J. Mol. Biol. 266, 122-134. Zappavigna, V., Falciola, L., Citterich, M. H., Mavilio, F. & Bianchi, M. E. (1996). HMG1 interacts with Hox proteins and enhances their DNA-binding and transcriptional activation. EMBO J. 15, 4981-4991. Zlatanova, J. & van Holde, K. (1998). Binding to fourway junction DNA: a common property of architectural proteins? FASEB J. 12, 421-431. nig, H. & Wirth, T. (1995). High mobility Zwilling, S., Ko group protein 2 functionally interacts with the POU domains of octamer transcription factors. EMBO J. 14, 1198-1208.

Edited by K. Nagai (Received 19 July 1999; received in revised form 25 August 1999; accepted 26 August 1999)