The EcoRV modification methylase causes considerable bending of DNA upon binding to its recognition sequence GATATC.

The EcoRV methyltransferase modifies DNA by the introduction of a methyl group at the 6-NH position of the first deoxyadenosine in GATATC sequences. The enzyme forms a stable and specific complex with GATATC sequences in the presence of a nonreactive analogue, such as sinefungin, of its natural cofactor S-adenosyl-L-methionine. Using circular permutation band mobility shift analysis (in which the distance between the GATATC sequence and the end of the DNA is varied) of protein-DNA-cofactor complexes we have shown the methylase induces a bend of just over 60° in the bound DNA. This was confirmed by phasing analysis, in which the spacing between the GATATC site and a poly(dA) tract is varied through a helical turn, which showed that the orientation of the induced curve was toward the major groove. There was no significant difference in the bend angle measured using unmethylated GATATC sequences and hemimethylated sequences which contain GATATC in one strand only. These are the natural substrates for the enzyme. The EcoRV endonuclease, a very well characterized protein, served as a positive control. DNA bending by this protein has been previously determined both by crystallographic and solution methods. The two proteins bend DNA toward the major groove but the bend angle produced by the methylase, slightly greater than 60°, is a little larger than that observed with the endonuclease, which is approximately 44°.

The EcoRV methyltransferase modifies DNA by the introduction of a methyl group at the 6-NH 2 position of the first deoxyadenosine in GATATC sequences. The enzyme forms a stable and specific complex with GATATC sequences in the presence of a nonreactive analogue, such as sinefungin, of its natural cofactor S-adenosyl-L-methionine. Using circular permutation band mobility shift analysis (in which the distance between the GATATC sequence and the end of the DNA is varied) of protein-DNA-cofactor complexes we have shown the methylase induces a bend of just over 60°in the bound DNA. This was confirmed by phasing analysis, in which the spacing between the GATATC site and a poly(dA) tract is varied through a helical turn, which showed that the orientation of the induced curve was toward the major groove. There was no significant difference in the bend angle measured using unmethylated GATATC sequences and hemimethylated sequences which contain G 6-Me ATATC in one strand only. These are the natural substrates for the enzyme. The EcoRV endonuclease, a very well characterized protein, served as a positive control. DNA bending by this protein has been previously determined both by crystallographic and solution methods. The two proteins bend DNA toward the major groove but the bend angle produced by the methylase, slightly greater than 60°, is a little larger than that observed with the endonuclease, which is approximately 44°.
Bending of DNA is fairly common and can be due to: 1) sequence-specific intrinsic curvature Hagerman, 1990Hagerman, , 1992Trivinov, 1991), for example with poly(dA)⅐poly(dT), which induces curvature toward the minor groove; 2) the binding of small ligands (Feuerstein et al., 1990); and 3) the binding of proteins. One of the earliest, structurally characterized, complexes in which proteins induce bending of DNA was the nucleosome (Richmond et al., 1984). However, DNA curvature does not only occur in these macromolecular structures. A large number of proteins involved in important cellular processes such as replication, recombination, or transcription are known to produce bending of the double helix (Travers, 1990). An excellent example is the prokaryotic catabolite gene activator protein (CAP protein), which produces a 90°b end, as measured both in crystals (Schultz et al., 1991) and solution (Wu and Crothers, 1984;Kim et al., 1989;Zinkel and Crothers 1990). The importance of the bend produced by this protein has been demonstrated by its replacement with intrinsically bent DNA; this is enough to activate transcription in vitro (Bracco et al., 1989). Inducing the bending of DNA in order to activate transcription has been reported for other transcriptional proteins (Perez-Martin and Espinosa, 1991), and this seems to be a very common feature for this class of protein. The bend angles obtained using solution methods range from 30°, for Cro protein binding to O R 1 operator, to 112°f or the GalR protein binding to O I operator (Kim et al., 1989).
The interactions of type II restriction enzymes with their target sites can also result in DNA bending. Crystal structures of the EcoRI (Kim et al., 1990;Rosenberg, 1991) and EcoRV (Winkler et al., 1993;Kostrewa and Winkler, 1995) endonucleases with their cognates sequences (GAATTC and GATATC, respectively) show that the bound DNA is distorted and both enzymes induce considerable bends. In both cases studies in solution have confirmed the bending of the DNA observed in the crystalline state (Thompson and Landy, 1988;Douc-Rasy et al., 1989;Stöver et al., 1993;Vipond and Halford, 1995). Bend angles of 50°and 44°have been reported for RI and RV, respectively. However, the crystal structure of the cognate DNA(CAGCTG)-PvuII complex  shows no significant DNA curvature, even though this nuclease shares structural similarities with EcoRV. Other data, obtained using only solution methods, showed that RsrI bent DNA in a very similar manner to EcoRI (Aiken et al., 1991). These two enzymes are perfect isoschizomers. SmaI and XmaI (imperfect isoschizomers that recognize the sequence CCCGGG) induce bending of 32°and 40°, respectively. Interestingly the DNA is bent in opposite directions, with SmaI toward the major groove and with XmaI in the direction of the minor groove (Withers and Dunbar, 1993).
There is much less structural data available for type II DNA methyltransferases. Two deoxycytidine methylases, HhaI (Cheng et al., 1993;Klimasauskas et al., 1994) and HaeIII (Reinisch et al., 1995), have been crystallized as complexes with DNA. Both structures show that the target deoxycytidine base is completely flipped out of the helix, but there is little or no bending to the nucleic acid. One deoxyadenosine methylase, TaqI (Labahn et al., 1994), has also had its structure determined albeit in the absence of an oligonucleotide. We are unaware of experiments describing solution studies in which any DNA curvature, induced by complexation with a DNA methyltransferase, has been measured. In this study we investigate DNA bending induced by the EcoRV methyltransferase. This is a type II deoxyadenosine methylase which adds CH 3 groups to the first deoxyadenosine in GATATC sequences (Nwosu et al., 1988). We have recently shown that this methylase binds specifically to GATATC sequences with the following order of affinity; hemimethylated DNA Ͼ unmethylated DNA Ͼ dimethylated DNA. Sinefungin, an analogue of the cofactor Sadenosyl-L-methionine (AdoMet) 1 increased binding by a factor of 20, and ternary enzyme/cofactor/substrate (un-or hemimethylated DNA) migrated slightly more rapidly than the equivalent product complexes containing dimethylated DNA. The enzyme copurified with traces of AdoMet and so was capable of methylating small quantities of DNA without the addition of this cofactor . Several of the methylase-DNA contacts were determined using ethylation and methylation interference analysis . The formation of a specific protein-DNA complex means that solution methods, such as the circular permutation band mobility shift assay (Wu and Crothers, 1984;Liu-Johnson et al., 1986;Kim et al., 1989;Crothers et al., 1991;Zwieb and Adhya, 1994) and helical phasing analysis (Zinkel and Crothers, 1987;Crothers et al., 1991;Gartenberg et al., 1990;Withers and Dunbar, 1993), can be used to see if the methylase bends DNA. The two approaches have been used with both unmethylated and hemimethylated GATATC sites to reveal details of DNA bending by the EcoRV methylase and to elucidate the means by which this protein recognizes its target sequence.

EXPERIMENTAL PROCEDURES
Materials-The purification of the EcoRV methyltransferase has been described previously (Nwosu et al., 1988;. The gel filtration column used in the second step was a Superdex G-75 10/30 (Pharmacia, St. Albans, UK) instead of Protein-Pak 300W (Waters-Millipore). The EcoRV endonuclease, purified as previously described (Luke et al., 1987;Newman et al., 1990), was a generous gift from Sarah A. Cullinane (University of Newcastle upon Tyne, UK). The AdoMet analogue, N-methyl-AdoMet, was kindly provided by Professor Michael Blackburn (University of Sheffield, UK). Restriction enzymes were purchased from NBL (Northumbria Biological Laboratories, Cramlington, UK) and Taq DNA polymerase was from Boehringer Mannheim (Lewes, UK). Oligonucleotides primers for PCR amplification were synthesized on an Applied Biosystems 381A DNA synthesizer using the phosphoramidite method (Newman et al., 1990;Connolly 1991Connolly , 1992Waters and Connolly, 1994) with reagents from either Cruachem (Glasgow, Scotland) or Applied Biosystems (Warrington, UK). The phosphoramidite of 6-methyldeoxyadenosine was purchased from Pharmacia.
Construction of DNA Fragments for Bending Analysis-A set of six different 141-bp DNA fragments, with the EcoRV site "permutated" (i.e. placed at different distances from the end of the fragment) were constructed by PCR amplification of a pBend3 derivative (Zwieb and Adhya, 1994) which had an additional insert of 20 bp between the XbaI and SalI sites (Vipond and Halford, 1995). This was kindly provided by I. B. Vipond and S. E. Halford (University of Bristol, UK). The primers used in this amplification were: 5Ј-GGATCCCCCGGGCTGCAGG-3Ј and 5Ј-AGGTCGACGGTATCGATAA-3Ј. pBend3 has two, tandemly repeated, copies of a polylinker sequence (that includes a GATATC EcoRV site) on either side of the unique XbaI and SalI sites. Therefore the subsequent digestion of the PCR product with either XhoI, NheI, EcoRI/ SalI, XbaI/HindIII, KpnI, or SmaI generate the six 141-bp fragments with the EcoRV site permutated along the DNA (see Fig. 1).
A second set of four different 150-bp DNA fragments, containing a permutated EcoRV site, were generated by the PCR amplification of pAD10; a plasmid that contains a chemically synthesized DNase I gene which has a unique EcoRV site within its sequence (Worrall and Connolly, 1990;Doherty et al., 1993). In the first instance four pairs of primers, 15 bp long, were used to generate the four 150-bp fragments. The positions at which the primers bound to the template, and hence the products produced, are shown in Fig. 2. The second four pairs of primers were used to give products containing a hemimethylated, permutated, EcoRV site (i.e. G 6-Me ATATC and GATATC on the two strands). One of each of the pairs of the primers was of varying length and contained a G 6-Me ATATC sequence. The second was 15 bp long and was selected to maintain the product length at 150 bp (see Fig. 2).
For the phasing analysis, five different M13 derivatives which contain poly(dA) tracts (Withers and Dunbar, 1993) were used. These have the poly(dA) tracts phased with a polylinker and were a generous gift of J. C. Dunbar (Wayne State University, Detroit, MI). Unfortunately the polylinker does not contain an EcoRV site. Therefore a partially complementary pair of oligonucleotides (AATTCCGATGATATCC and GGCTACTATAGGGGCC, both written in the 5Ј 3 3Ј direction), containing an EcoRV site and EcoRI/XmaI cohesive ends, were cloned between the EcoRI and XmaI sites in the M13 derivatives, as illustrated in Fig. 3. The plasmids were transfected into Escherichia coli TG1competent cells, and the resulting reisolated plasmids had their sequences confirmed by automated DNA sequencing. These plasmids were then used as templates for PCR reactions with a common forward primer (CACTTTATGCTTCCGGCT) and five different reverse primers (phasing oligonucleotide 1, AGTCACGACGTTGTAAAAA; 2, TCAC-GACGTTGTAAAACG; 3, CGACGTTGTAAAACGACG; 4, ACGTTGTA-AAACGACGGC; 5, GTTGTAAAACGACGGCCA). This produces five fragments of the same length (206 bp) with a GATATC EcoRV site phased with the poly(dA) tract (see Fig. 3).
Polymerase Chain Reaction-PCR was carried out using a Techne Thermal Cycler from Techne (Cambridge) Ltd., Cambridge, UK. Each reaction contained Ϸ5 nM plasmid DNA and Ϸ0.35 M concentrations of the primers in a 100-l volume of 10 mM Tris-HCl, pH 8.3, 15 mM MgCl 2 , 50 mM KCl, 10 g of gelatin and 60 M of each dTTP, dCTP, dGTP, and ␣-[ 32 P]dATP (3000 Ci/mmol). The amount of Taq DNA polymerase added was 2.5 units (as defined by the supplier, Boehringer) and the DNA was subjected to 1 cycle of 94°C, 3 min; 55°C, 2 min; 72°C, 1 min; and 30 cycles of 94°C, 1 min; 55°C, 3 min; and 72°C, 2 min. The amplified fragments were separated in a 10% nondenaturing polyacrylamide gel and visualized by autoradiography. The required band was excised and eluted overnight at 37°C in Tris-EDTA buffer with continuous shaking and the DNA fragments separated from buffers and gel contaminants by ethanol precipitation (Sambrook et al., 1989).
Band Mobility Shift Experiments-The interaction of the EcoRV methylase with DNA was carried out in volumes of 20 l containing: 50 mM Hepes-NaOH, pH 7.0, 1 mM EDTA, 100 mM NaCl, 5 mM dithiothreitol, and 1 g of acetylated bovine serum albumin. 1.5 mM levels of either the AdoMet analogue N-methyl-AdoMet or 0.2 mM sinefungin, Ϸ100 pM amounts of each amplified DNA, and 100 nM concentrations of the methylase were used . The binding of EcoRV endonuclease was carried out in 20-l volumes containing: 50 mM Tris-HCl, pH 7.0, 100 mM NaCl, 10 mM ␤-mercaptoethanol, 1 g of acetylated bovine serum albumin, 0.1 mM EDTA, and 5 mM CaCl 2 (Vipond and Halford, 1995). The concentration of each amplified DNA was ϳ25 pM and 10 nM levels of the enzyme was used. The free DNA and complexes were separated in either 7 or 8% of a nondenaturing polyacrylamide gel and detected by autoradiography as has been previously reported . For the experiments with the endonuclease 5 mM Ca 2ϩ was included in the gels, and the EDTA was omitted (Vipond and Halford, 1995).
Determination of Bend Angles-The angle of curvature induced in the DNA by protein binding was calculated using the empirical equation of Thompson and Landy (1988); M/E ϭ cos(␣/2) (where M and E correspond to the electrophoretic mobility for the DNA-protein complexes when the binding site is at the middle and at the end of the DNA fragment, respectively).

Design of DNA Fragments for Bending Analysis-
We have used a number of oligonucleotides containing a single GATATC site to study DNA bending by the EcoRV methylase. The first set, illustrated in Fig. 1, are derived from the bend plasmid series (Zwieb and Adhya, 1994) and have been extensively used, for example with the EcoRV endonuclease (Vipond and Halford, 1995). The "permutation" of the EcoRV site along the DNA fragment was achieved by the PCR amplification of the plasmids followed by the digestion of the amplified fragments with the restriction enzymes shown in Fig. 1. The six 141-bp oligonucleotides produced have a single EcoRV site at a position that varies relative to the center and the ends of the molecule. The only disadvantage with these oligonucleotides is that the disposition of the restriction sites does not allow a dead-center location of the GATATC sequence. In addition the oligonucleotides, shown in Fig. 2, have been prepared and used. These were synthesized by the PCR amplification of a fragment that contains a unique GATATC sequence. Using the short forward and reverse pairs of primers 1-4 gives the "normal," i.e. unmethylated, product fragments 1-4 shown in this figure. Product 1 contains a centrally located EcoRV sequence and with products 2-4 this site moves toward the end of the oligonucleotide. More importantly this system can be used to prepare hemimethylated oligonucleotides containing a single 6-methyldeoxyadenosine residue at the G 6-Me ATATC site. For this it is necessary to use the longer forward primers 1-4, containing a G 6-Me ATATC sequence, with the appropriate reverse primers as shown in Fig. 2. The EcoRV methylase adds CH 3 groups to the first deoxyadenosine in GATATC sequences (Nwosu et al., 1988) and it is thought that hemimethylated DNA, that transiently arises following replication, is the true substrate for the enzyme. Finally we have made use of a set of five oligonucleotides in which the GATATC site is phased, through a helical turn, with a poly(dA) tract (Fig. 3). These sequences were used by Withers and Dunbar (1993) to study DNA bending with the XmaI and SmaI endonucleases. An EcoRV site (absent in the original derivatives) was introduced by inserting an appropriate complementary pair of oligonucleotides between the EcoRI and XmaI sites (Fig. 3). The plasmids produced were used as PCR substrates with a common forward primer and different reverse primers, for each of the derivatives, selected so as to compensate for the slight differences in length in the EcoRV/poly(dA) tract sequences. This results in the five products of identical length (206 bp) shown in Fig. 3, which have an EcoRV site and a poly(dA) tract separated through one turn of the helix. The center of the EcoRV/poly(dA) run occurs about 45% from one end of the fragment and 55% from the other, i.e. approximately central.
DNA Bending by the EcoRV Methylase-As mentioned in the Introduction, the EcoRV methylase copurifies with small amounts of AdoMet that can add CH 3 groups to GATATC sequences in the preincubation prior to band mobility shift analysis. This can be prevented under certain conditions, mainly a combination of the inclusion of sinefungin, a very strong competitive inhibitor of AdoMet, and the use of small quantities of the methylase . However, in this study we have selected cofactor and methylase levels that allow a degree of methylation to take place to give a mixture of un-, hemi-, and dimethylated oligonucleotides. This allows the bending of both the ternary complexes containing un/hemimethylated and dimethylated DNA to be simultaneously assessed. The phasing experiments (using the oligonucleotides shown in Fig. 3) have been carried out in the presence of 0.2 mM sinefungin, an amount sufficient to ensure the 20-fold increase in the binding of GATATC-containing sequences. However, sinefungin has recently become unavailable commercially. Therefore, the bending experiments (using the oligonucleotides shown in Figs. 1 and 2) have been performed with 1.5 mM levels of the analogue N-methyl-AdoMet. N-Methyl-AdoMet is a poorer analogue than sinefungin, 2 but the higher levels used ensured the 20-fold increase in affinity of cognate oligonucleotides.
Initially we evaluated any DNA bending produced by the binding of the methylase to GATATC sequences using the circular permutation band mobility shift assay (Wu and Crothers, 1984;Liu-Johnson et al., 1986;Kim et al., 1989;Crothers et al., 1991). The six fragments shown in Fig. 1 were bound to the methylase, in the presence of N-methyl-AdoMet, and the complexes formed subjected to nondenaturing gel electrophoresis. The results obtained are illustrated in Fig. 4A. This figure clearly shows that fragments containing a centrally located GATATC site give complexes with lower electrophoretic mobility than do those with the EcoRV site at the end of the DNA. Furthermore, there is a good correlation between the degree of retardation and the nearness of the GATATC site to the center of the fragment as shown in Fig. 4B, a graphical representation of the gel data. Close inspection of Fig. 4A shows that the retarded band is actually composed of two just-resolved species (this is most apparent in lane 5). However, the separation of the two bands is much less obvious here than in our earlier publication , possibly because of the much longer oligonucleotides used in this study. The two bands arise because of the presence of AdoMet that copurifies with the methylase and adds methyl groups to the first deoxyadenosine in the GATATC sequence to generate hemi-and dimethylated species. For Fig. 4B the mobility has been calculated using the leading edges of all the bands. We take this to represent the faster bands, which, as described earlier  represent un-and hemimethylated oligonucleotides, i.e. substrates for the methylase. However, very similar results were seen if the trailing edges were used for mobility determination. This is a measure of the slower bands, which represent dimethylated oligonucleotide products. The mobility of a DNA fragment through a gel is lowered as the distance between its two ends decreases and this most commonly arises because of DNA bending. Furthermore the more centrally located the bend the smaller the end to end distance and, in consequence, the greater the retardation on gels (Wu and Crothers, 1984;Hagerman, 1990;Crothers et al., 1991). The results shown in Fig. 4, A and B, are therefore consistent with the methylase bending of the DNA.
It is possible to calculate the degree of bending using the empirical equation described by Thompson and Landy (1988), in which the mobility of fragments containing the bend at the center and at the end on the molecules are compared. The fragments we have used do not contain an exactly centered GATATC site but nevertheless comparing the two most central sites (Fig. 1, fragments 3 and 4) with the two most peripheral (fragments 1 and 6) gave a bend angle of 62 Ϯ 3°(number of determinations ϭ 4). This angle was produced regardless of which combination of the two central and peripheral sites were used to evaluate M and E. An angle of 62°was found for both the faster, substrate, complexes and the slower, product, com-FIG. 1. Oligonucleotide fragments containing a permutated EcoRV site. These were produced by PCR amplification and subsequent restriction endonuclease digestion (enzymes used shown) of a pBEND3 derivative. The position of the EcoRV site is shown, relative to the ends of the fragments. These fragments were used to study DNA bending by the EcoRV methylase and gave the results shown in Fig. 4. plexes. This demonstrates that the differences in mobility between enzyme/un-and hemimethylated DNA complexes and those that comprise enzyme/dimethylated DNA cannot be due to a change in DNA bending.
As a positive control we have used the binding of the very well characterized EcoRV endonuclease to the DNA fragments, in the presence of Ca 2ϩ . The results are also shown in Fig. 4, A and B, and quantitative analysis, exactly as above, gave a bending angle of 47 Ϯ 3°(number of determinations ϭ 4). The slower migration of the endonuclease-DNA complexes compared to those of the methylase-DNA are consistent with the different molecular weights of the two proteins: endonuclease 58 kDa (comprised of a dimer of subunit molecular weight 29 kDa), methylase a monomer of 32 kDa. Using the same oligo-nucleotides, Vipond and Halford (1995) reported a bend angle of 53°Ϯ 4°and with a different set of oligonucleotides, Stöver et al. (1993) found an angle of 44 Ϯ 4°. A figure of 55°has been seen by crystallography (Winkler et al., 1993;Kostrewa and Winkler, 1995). The similarity of the results we obtain with the endonuclease, to those seen by several investigators, validates the data and conclusions drawn with the relatively uncharacterized methylase.
We have confirmed that the EcoRV methylase bends DNA by using a second set of oligonucleotides, shown in Fig. 2. Once again in this set of experiments N-methyl-AdoMet was used as cofactor. One of the four 150-bp oligonucleotides (fragment 1) contains an exactly center GATATC sequence, and in the others this is moved toward the end of the DNA. Using G 6-Me -

FIG. 2. The production of 150-bp fragments containing either an un-or a hemimethylated
EcoRV site at various distances along the oligonucleotide. These were prepared by PCR amplification of part of a DNase I gene (shown in bold type) containing a single GATATC site (highlighted). Pairing of the short forward (3) primers (containing only normal bases) 1, 2, 3, and 4 with the reverse (4) primers 1, 2, 3 and 4, respectively, gave the fragments 1, 2, 3, and 4 which contain an unmethylated EcoRV site at the locations shown. Pairing of the longer forward (3) primers (containing a single 6-methyldeoxyadenosine, shown by an asterisk (*)) 1, 2, 3, and 4 with the reverse (4) primers 1, 2, 3, and 4 gave the fragments 1, 2, 3, and 4, containing a hemimethylated EcoRV site at the positions illustrated. All primers are shown in italic type. The results found with these oligonucleotides are shown in Fig. 5.   FIG. 3. The DNA fragments used for phasing analysis. These were prepared by inserting an EcoRV site between the EcoRI and XmaI sites of the five phasing plasmids described by Withers and Dunbar (1993). This gives the five plasmids shown, each of which contains a GATATC site at a defined distance from a poly(dA) tract. These distances, measured from the center of the EcoRV site (between the thymidine and deoxyadenosine bases) to the middle deoxyadenosine of the poly(dA) tract are given. The five plasmids were then amplified by PCR to give the 206-bp fragments illustrated which were used to confirm DNA bending by the EcoRV methylase and gave the results shown in Fig. 6. ATATC containing PCR primers an analogous set, containing a hemimethylated EcoRV site, was also prepared. The results found with these oligonucleotides are shown in Fig. 5. With these fragments the resolution of enzyme/substrate (un-and hemimethylated DNA) and enzyme/product (dimethylated DNA) was poor, and the leading edges have been used to evaluate fragment mobility. By analogy with the above experiment we assume that the bands seen in Fig. 5, using the unmethylated oligonucleotides, are actually a mixture of un-, hemi-, and dimethylated species. Once again we observe that the more central the location of the EcoRV site, and hence any bend induced on methylase binding, the more retarded the complex runs on gels (Fig. 5, A and B). Quantitation using the Thompson-Landy equation and fragments 1 (center) and 4 (end) gives a bend angle of 60 Ϯ 3°(four determinations) in excellent agreement with the values produced above. When samples of these unmethylated fragments were reisolated, following a normal pre-gel shift incubation with the methylase, they could be gel-shifted by the endonuclease (not shown). This confirms that at least some of the DNA remains fully unmethylated, because hemimethylated species are not gel-shifted by the endonuclease (see later). The EcoRV endonuclease was again used as a positive control, and the data for the unmethylated oligonucleotides shown in Fig. 5, A and B, give a bend angle of 46 Ϯ 3°. Almost identical results were seen when gel shifts were carried out with the methylase and hemimethylated oligonucleotides (Fig. 5, A and B), and the bend angle found was 61 Ϯ 3°. In this instance the material in the shifted band will consist of hemi-and dimethylated DNA, and unmethylated material cannot be present. These experiments confirm the bending angle determined above but also show that methylase/un-and hemimethylated DNA complexes have the nucleic acid bent to the same degree. Thus all three possible methylase-nucleic acid complexes (unmethylated, hemimethylated, and dimethylated) bend DNA to the same extent, just over 60°. It is interesting that the hemimethylated oligonucleo-tides shown in Fig. 2 were not gel-shifted by the endonuclease (not shown). Under the conditions used, i.e. in the presence of Ca 2ϩ , an analogue of the normal cofactor Mg 2ϩ , the endonuclease binds specifically to GATATC sequences and rejects all others (Vipond and Halford, 1995). The function of methylation is to protect the host DNA from endonuclease action, and the lack of binding of hemimethylated (and presumably dimethylated) DNA, in the presence of cations, presumably facilitates this.
Confirmation of DNA Bending and Determination of Its Direction by Phasing Analysis-In the above discussion we have assumed that the retardation of centrally placed GATATC sequences on binding to the methylase is due to DNA bending. However, it is well known that DNA conformational effects other than bending can give rise to the gel mobilities of the sort observed above (Zinkel and Crothers, 1987;Gartenberg et al., 1990;Crothers et al., 1991). In order to discriminate between DNA bending and alternative conformational features and to determine the bend direction we have used phasing analysis. For this purpose, the DNA fragments shown in Fig. 3 were used. These contain a GATATC site phased (i.e. separated over a helical turn) with an intrinsically curved poly(dA) tract. The results seen with these fragments and both the methylase and endonuclease (which once again acted as a positive control) are shown in Fig. 6. In these experiments sinefungin, rather than N-methyl-AdoMet was used as cofactor. From the gels (Fig. 6A) and their associated graphical representations (Fig. 6B) it can be seen that, for both enzymes, the mobility of the DNA-enzyme complex is dependent on the spacing between the

FIG. 4. Gel retardation of DNA fragments containing a permutated GATATC site by the EcoRV methylase and endonuclease.
A, autoradiographs of the gels obtained. The lanes 1-6 correspond exactly to the fragments 1-6 shown in Fig. 1. B, plots of relative mobility (migration of protein-DNA complex/migration of free DNA) versus distance of EcoRV site from the 5Ј-end of the DNA fragment.

FIG. 5. Gel retardation of DNA fragments containing a GATATC site at various positions along a 150-bp oligonucleotide by the EcoRV methylase and endonuclease.
A, autoradiographs of the gels obtained. The lanes 1-4 correspond to the fragments 1-4 shown in Fig. 2. The experiments were performed with the methylase and both un-and hemimethylated (indicated by an asterisk (*)) fragments. With the endonuclease only the unmethylated fragment was used. The hemimethylated one did not form a complex, and hence did not show a gel shift, with this protein. B, plots of the relative mobility (migration of protein-DNA fragment/migration of free DNA) versus distance of the EcoRV site from the 5Ј-end of the fragment. With the methylase, the triangles refer to hemimethylated DNA and the circles to unmethylated. GATATC sequence and the poly(dA) tract. This behavior is consistent with the results seen in the preceding experiments being due to protein-induced DNA bending. Examination of Fig. 6 shows that, for the methylase and the endonuclease, the maximum electrophoretic mobility, corresponding to the least bending, occurs in track 3, i.e. when the GATATC sequence is separated by 32.5 bp from the center of the poly(dA) tract. B-DNA has 10.5 bp per turn of the helix and sequences are "in phase," i.e. on the same side of the helix, when separated by an exact multiple of this number, which in this experiment will occur at the 31.5-bp separation. Poly(dA) tracts bend DNA toward the minor groove, and thus when the protein-induced DNA bends are very nearly in phase with this intrinsically curved sequence, the overall bending is reduced. This means that the two proteins must bend the DNA in the opposite sense to deoxyadenosine tracts, that is, in the direction of the major groove. Conversely the lowest mobility, and hence most bending, takes place in tracks 1 and 5, that is, at spacings of 27.5 and 36.5 bp (the GATATC and poly(dA) will be exactly out of phase at 26.25 and 36.75 bp). Here the two bends are on opposite sides of the helix and, although directed toward the major and minor grooves, have the same orientation and so increase overall curvature. Although the mobility differences are small, and maximal and minimal mobility occur at nearly, rather than exactly, in and out of phase, the conclusion that the EcoRV endonuclease bends DNA toward the major groove is fully supported by x-ray structural data (Winkler et al., 1993;Kostrewa and Winkler, 1995). Once again this increases the reliability that can be placed on the methylase experiments.

DISCUSSION
When the EcoRV methylase binds to its GATATC target site, the DNA is bent toward the major groove by just over 60°. The degree of bending is identical, irrespective of whether the GATATC site is un-, hemi-, or dimethylated. The hemimethylated species is the true physiological substrate for the enzyme although the protein is also capable of de novo CH 3 group addition to unmethylated sequences. Dimethylated oligonucleotides represent enzyme-product complexes. The different gel mobility of the enzyme-substrate and enzyme-product complexes, observed here and previously , are most probably due to a protein conformational change, given that the bending of the DNA does not alter. The parallel, positive control experiments, carried out with EcoRV endonuclease, gave a bend angle of about 47°, in excellent agreement with earlier crystallographic (Winkler et al., 1993;Kostrewa and Winkler, 1995) and solution studies (Stöver et al., 1993;Vipond and Halford, 1995). It is also apparent that the two proteins, which recognize the same GATATC sequence, both bend DNA toward the major groove, although the methylase induces more curvature than does the endonuclease.
It has been extensively reported that the binding to DNA of a great number of proteins introduces curvature into the double helix. This is often observed with repressor/activator proteins that must bind specifically to particular DNA sequences (Wu and Crothers, 1984;Kim et al., 1989;Schultz et al., 1991;Aggarwal et al., 1988;Curran 1991a, 1991b) or enzymes, especially restriction endonucleases, that must catalyze a chemical transformation with high fidelity at their target sites (Thompsom and Landy, 1988;Douc-Rasy et al., 1899;Rosenberg, 1991;Aiken et al., 1991;Stöver et al., 1993;Winkler et al., 1993;Vipond and Halford, 1995;Kostrewa and Winkler, 1995;Withers and Dunbar, 1993). In all of these cases the bending of the DNA serves to increase the specificity of the proteins and to aid in the selection of target sequences and in the discrimination against noncognate ones. Proteins achieve selectivity for particular DNA sequences by a variety of mechanisms that include: 1) direct readout, where proteins recognize the bases (Seeman et al., 1976) and 2) indirect readout, where a particular, nontypical, DNA structure is recognized (Otwinowski et al., 1988;Matthews, 1988;Brennan and Matthews, 1989). It is known that many DNA sequences have intrinsic curvature or other distortions, and it is thought that this can used in the recognition process (Heitman, 1992). Thus it has been suggested that the GAATTC EcoRI recognition sequence is intrinsically distorted and that this is enhanced on binding to the EcoRI restriction endonuclease (Thomas et al., 1989). A variant on this theme is the concept of a DNA sequence that is easy to bend or distort. Here a bendable cognate sequence can easily adapt to a complementary protein binding site, whereas stiff noncognate sequences cannot. This seems to occur with DNase I, where rigid poly(dA)⅐poly(dT) tracts are poorly cut, whereas more flexible sequences are able to adopt the bound, bent, conformation and so are better substrates (Drew and Travers, 1984;Suck et al., 1988;Weston et al., 1992). It should be recognized that bending of DNA also serves to correctly line up the protein and nucleic acid partners and so enhance protein-DNA interaction. Thus direct and indirect readout are intimately linked. Formation of interactions between the bases and the protein may produce DNA curvature, but equally the ability to adopt a bent conformation may enhance favorable contacts between the two macromolecules. Finally, although DNA bending is common with proteins that act on specific DNA sequences, it is not universal. Thus certain repressors, e.g. the repressor (Jordan and Pabo, 1988), the 434 Cro protein (Wolberger et al., 1988) and the PvuII restric- FIG. 6. Gel retardation of DNA fragments containing a GATATC site at varying separations from a poly(dA) tract. A, autoradiographs of the gels obtained. The lanes 1-5 correspond to the fragments 1-5 shown in Fig. 3. Two bands are once again visible for the methylase that correspond to un-and hemimethylated DNA (fast band) or dimethylated DNA (slow band). B, plots of the relative mobility (migration of protein-DNA complex/migration of free DNA versus spacing between the center of the GATATC site and the poly(dA) tract. tion endonuclease  have a bound DNA that is essentially straight. These proteins are able to achieve discriminations as good as those that utilize DNA bending.
It is likely that DNA methylases, which must add CH 3 groups accurately to specific sequences, would utilize the mechanisms outlined above to generate specificity. However, there is little structural data for these proteins. Two protein-DNA structures, for the type II DNA-deoxycytidine methylases HhaI (Cheng et al., 1993;Klimasauskas et al., 1994) and HaeIII (Reinisch et al., 1995), have been solved by crystallography. With HhaI the target deoxycytidine is flipped out of the double helix and into the active site but hardly any other distortion, and no bending, takes place. Base flipping is also seen with HaeIII, and in addition the other bases in the recognition site are also severely distorted. However, once again DNA bending is not observed. Both these methylases have similar recognition sites (HhaI, GCGC; HaeIII, GGCC), containing only (G/C) base pairs. A structure of the type II deoxyadenosine methylase TaqI (Labahn et al., 1994), has been solved but this is for the apoenzyme lacking bound DNA. The distance from the putative DNA binding site to the AdoMet cofactor suggested base flipping. However, in the absence of bound DNA it was not possible to draw any conclusions about DNA bending. Based on comparisons of several AdoMet-dependent enzymes, it has been proposed that all DNA methylases have a common, two domain, topology (Schluckebier et al., 1995). The AdoMet and catalytic functions are present in the N-terminal domain, which is extremely similar for the enzymes studied. A more variable C-terminal domain, the target recognition domain, is involved in making specific interactions with the cognate DNA sequence. Support for this proposal has come from the recent successful modelling of the structure of a type I methylase EcoKI (Dryden et al., 1995). Again flipping of the target deoxyadenosine base was required to bring it close to AdoMet. With so little data it is difficult to generalize about methylase-DNA structures. It does seem likely that all of the proteins will have the two-domain structure, with a common catalytic domain, and will flip out the target base, although this has still to be unequivocally demonstrated for deoxyadenosine methylases. However, within the framework of these common elements the rest of the DNA cognate sequence could be handled very differently. Thus some methylases, such as HhaI and HaeIII, may be able to achieve selectivity without inducing DNA curvature. Others, like EcoRV, build DNA bending into the recognition mechanism. In this regard most of the repressors mentioned above use a common DNA recognition element, the helix-turn-helix motif. Some, but not all of them, cause DNA bending. Support for the bendability of the GATATC, and related, sequences has come from the study of several protein-DNA structures. With the EcoRV restriction endonuclease the 55°b end, toward the major groove, is centered on the TpA step (Winkler et al., 1993;Kostrewa and Winkler, 1995), which has a sharp kink. It has long been recognized that the two base pairs in TpA steps, in particular, and YpR steps, in general, are poorly stacked (Klug et al., 1979;Calladine, 1982;Dickerson, 1983). This gives these steps high flexibility and deformability and, in particular, the ability to adopt high positive roles and become untwisted (Suzuki and Yagi, 1995). Deformations to protein-bound DNA structures are thus very often associated with TpA steps, e.g. Trp repressor (Otwinowski et al., 1988;Shakked et al., 1994), met repressor (Somers and Phillips, 1992), TATA-binding protein (Kim et al., 1993a(Kim et al., , 1993b and the ␥␦ resolvase (Yang and Steitz, 1995). The latter two proteins produce DNA bending very similar to the EcoRV endonuclease, despite the three proteins having absolutely no structural sim-ilarities. In all of these cases the key seems to be the instability of the TpA step, which provides a low energy barrier to any DNA structural perturbations that must take place on binding the protein. We would thus propose that the bending seen with the EcoRV methylase is centered on the TpA step of the GATATC sequence. A kink, at this step, directed toward the major groove, will cause this groove to become deeper and narrower and is also likely to cause unwinding of the helix. This is exactly what is seen with the EcoRV endonuclease. We have recently shown that the methylase makes many of its contacts to the GATATC bases via the major groove . The consequence of the motion we have described is actually to make the major groove determinants of the bases less accessible and with the endonuclease the middle TpA step is not contacted by the protein directly (Winkler et al., 1993;Kostrewa and Winkler, 1995). However, these bases can be efficiently recognized, indirectly, by facilitating the distortion and, also by the formation of new non-Watson-Crick hydrogen bonds or dipolar interactions between them (Winkler et al., 1993;Kim et al., 1993a). The problem of accessibility of the target deoxyadenosine base, which must be placed near the AdoMet, can also be solved by swinging it out of the helix, as for HhaI. This means that the cofactor does not need to find its way into the smaller major groove. Finally, we have also shown that important contacts are made between the methylase and the phosphates pNpNpGpA on each DNA strand. These lie to the 5Ј-ends of the symmetrical GATATC site . As pointed out for the TATA-binding protein (Kim et al., 1993a), the bending we describe tends to bring bordering phosphates from each strand closer to each other and provide strong polar anchoring interactions which will help to stabilize the complex.