doi:10.1006/jmbi.2001.4799 available online at http://www.idealibrary.com on J. Mol. Biol. (2001) 310, 501±508

COMMUNICATION

DNA Binding and Cleavage Selectivity of theEscherichia coli DNA G:T-mismatch Endonuclease(vsr Protein)

Rebeca Gonzalez-Nicieza, David P. Turner and Bernard A. Connolly*

Department of Biochemistryand Genetics, The University ofNewcastle, Newcastle uponTyne NE2 4HH, UK

E-mail address of the correspondb.a.connolly@ncl.ac.uk

Abbreviations used: .

0022-2836/01/030501±8 $35.00/0

The Escherichia coli vsr endonuclease recognises T:G base-pair mismatchesin double-stranded DNA and initiates a repair pathway by hydrolysingthe phosphate group 50 to the incorrectly paired T. The gene encodingthe vsr endonuclease is next to the gene specifying the E. coli dcm DNA-methyltransferase; an enzyme that adds CH3 groups to the ®rst dCwithin its target sequence CC[A/T]GG, giving C5MeC[A/T]GG. Deamina-tion of the d5MeC results in CT[A/T]GG in which the ®rst T is mis-pairedwith dG and it is believed that the endonuclease preferentially recognisesT:G mismatches within the dcm recognition site. Here, the preference ofthe vsr endonuclease for bases surrounding the T:G mismatch has beenevaluated. Determination of speci®city constant (kst/KD; kst � rateconstant for single turnover, KD � equilibrium dissociation constant)con®rms vsr's preference for a T:G mismatch within a dcm sequence i.e.CT[A/T]GG (the underlined T being mis-paired with dG) is the best sub-strate. However, the enzyme is capable of binding and hydrolysingsequences that differ from the dcm target site by a single base-pair (dcmstar sites). Individual alteration of any of the four bases surrounding themismatched T gives a substrate, albeit with reduced binding af®nity andslowed turnover rates. The vsr endonuclease has a much lower selectivityfor the dcm sequence than type II restriction endonucleases have for theirtarget sites. The results are discussed in the light of the known crystalstructure of the vsr protein and its possible physiological role.

# 2001 Academic Press

Keywords: vsr endonuclease; DNA T:G mismatches; protein-DNArecognition; DNA sequence-speci®city

*Corresponding author

The mismatched base-pair T:G can arise in DNAfrom misincorporation during replication or, to agreater extent, from the spontaneous hydrolyticdeamination of 5-methyldeoxycytosine (d5MeC) inexisting base-pairs.1,2 In bacteria, d5MeC is intro-duced into DNA by the post-replicative action ofDNA-dC methyltransferases, which may functionas one component of a restriction-modi®cation sys-tem or be a protein of uncertain function such asthe Dcm methyltransferase from Escherichia coli. Tocounteract the mutagenic potential of T:G mis-matches, arising from deamination of d5MeC:Gbase-pairs, many bacteria possess a ``very shortpatch'' (VSP) mismatch repair pathway character-

ing author:

ised by the presence of a ``very short-patch repair''endonuclease, the product of the vsr gene.3 ± 7 Thevsr endonuclease cleaves the phosphodiester bond50 to the mismatched T.8,9 producing a nick con-taining 50 phosphate and 30 hydroxyl termini.Repair is completed by the 50-30 exonuclease andDNA polymerase activities of DNA polymerase Iand DNA ligase.7,8,10 The vsr genes are invariablyassociated with a gene coding for a DNA dC-meth-yltransferase; thus, in E. coli, the vsr and dcm genesoverlap.11 ± 13 The E. coli Dcm methyltransferaseadds CH3 groups to the second dC in the sequenceCC[A/T]GG producing C5MeC[A/T]GG.14 ± 16

Deamination at the methylated base producesCT[A/T]GG, giving a T:G mismatch within theDcm methytransferase recognition site. It isassumed that T:G mismatches within this sequencecontext are the preferred targets for the vsr endo-

# 2001 Academic Press

502 DNA Recognition by the E. coli vsr Protein

nuclease. In several other bacterial species, the vsrgene is near that of a modi®cationmethyltransferase.7,17 By analogy with the E. colivsr endonuclease, these enzymes are expected toact on T:G mismatches within the recognitionsequence of the methyltransferases with whichthey are associated.

Recently, our group18 used single turnoverkinetics and gel retardation analysis to determinethe speci®city constant for the vsr endonucleasewith a number of oligonucleotides. Only foursub-strates were found: T:G mismatch (hemi-methylated dcm sequence) >U:G mismatch (hemi-methylated dcm sequence) >T:G mismatch(unmethylated dcm sequence) > U:G mismatch(unmethylated dcm sequence). A hemi-methylateddcm sequence has the dC opposite the underlinedG (CT[A/T]GG) converted to d5MeC, an expectedconsequence of Dcm methylase activity. This studyrepresented one of the ®rst quantitative biochemi-cal characterisations of a vsr endonuclease anddemonstrates, as expected, that the enzyme showsthe highest preference for T:G mismatches within ahemi-methylated dcm context (expected to be thephysiological substrate). Intriguingly, bindingassays showed that less than 1 % of the puri®edvsr protein was capable of binding to its targetsequence. This phenomenon has yet to be fullyexplained but may re¯ect the need for the presenceof MutL (or other proteins) for ef®cient interactionof the vsr endonuclease with DNA. MutL has beenshown, in vitro, to stimulate binding of the vsrendonuclease to its target sequence19 and toincrease hydrolysis rates.20 Genetic studies haveindicated that an absence of MutL greatly reducesvsr-catalysed repair of G:T mismatches.5,21,22

Thus far there has been little quantitative investi-gation of the in¯uence that the sequence surround-ing the T:G mismatch has on E. coli vsrendonuclease activity. How selective is the nucle-ase for CT[A/T]GG sequences (mismatched Tunderlined) and what are the consequences ofaltering the remaining four bases, derived from thedcm recognition site? Biochemical studies with pur-i®ed protein showed that substitution of severalbases within the dcm recognition sequence gavereasonable substrates;8,23 the hierarchy of sequencepreference observed being in general agreementwith the results of in vivo experiments.3,4,24 Onebiochemical investigation23 used multiple substratekinetics to derive relative second-order rate con-stants (ki), allowing a comparison of cognate dcmsites with sequences altered at one or two pos-itions. However, it is dif®cult to partition selectiv-ity measured by multi-substrate kinetics intocontributions arising from binding and catalysis, asthe individual parameters giving rise to the overallki value (which is a sum of many rate constants)are not easy to deconvolute. Furthermore, thesestudies were carried out with vsr endonucleaselacking 12 or 14 amino acid residues from theamino terminus; a region later shown by X-raycrystallography to be important for interaction

with DNA.9,25 In order to better elucidate the targetsite selectivity of the vsr endonuclease, we havenow determined the speci®city constants for theinteraction of the vsr endonuclease with dcm starsequences. By analogy with type II restrictionendonucleases,26 star sequences are de®ned as hav-ing a single base-pair altered compared to thenatural target site. As a measure of speci®city con-stant we have used the parameter kst/KD, where kst

is the catalytic rate constant for a single turnoverreaction and KD is the equilibrium dissociationbinding constant. This is the most powerful metricfor studying the selectivity of proteins acting uponnucleic acids,27 ± 29 allowing discrimination to beproportioned between binding and catalytic steps,and giving rates unin¯uenced by the productrelease step. It is likely that product release is aslow step with the vsr endonuclease; enzyme-pro-duct complexes have been found to be relativelystable.18 Indeed, this complex may serve to recruitthe next enzyme in the repair pathway. Neverthe-less, slow product release means that steady-statemethods cannot be used to assess accurately thediscrimination that vsr shows between cognateand star substrates.

The oligonucleotides used in this study areshown in Table 1 and are based on a prior study ofthe vsr endonuclease from this group.18 Previously,we showed that the ®rst entry in table 1, an oligo-nucleotide containing a T:G mismatch in a hemi-methylated dcm context (X � d5MeC) constitutedthe best substrate for vsr. An oligonucleotide witha T:G mismatch in an unmethylated dcm sequence(X � dC, entry 2 in Table 1) was also a reasonablesubstrate. These two oligonucleotides constitutethe controls, i.e. cognate dcm sequences, for thepresent study. Our earlier publication showed thata T:G mismatch in a scrambled dcm sequence, i.e.one having all four bases ¯anking the mismatchedT changed, resulted in an oligonucleotide that wasneither bound nor hydrolysed by the vsr endonu-clease. Here, oligonucleotides that contain dcm starsequences have been evaluated for binding andcutting by the vsr endonuclease. The starsequences maintain the T:G mismatch, in either ahemi-methylated or unmethylated context, but theother base-pairs are altered individually, to givethe set of oligonucleotides shown in Table 1 (thealtered base-pairs are underlined).

Initially, we attempted to measure the binding ofthe star sequences to vsr by direct titration i.e. add-ing increasing amount of enzyme to the oligonu-cleotides and observing any formation of a protein-DNA complex by gel retardation analysis. Allbinding assays were carried out in the absence ofthe essential metal ion cofactor, Mg2�, and in thepresence of EDTA, conditions that prevent sub-strate hydrolysis.18 Previously, we were able toobtain KD values for good vsr substrates, such asentries 1 and 2 in Table 1, by direct titration.18

However, we were unable to observe satisfactorygel shifts with the star oligonucleotides. The twooligonucleotides in which the ®rst base-pair had

Table 1. Interaction of the vsr endonuclease with oligodeoxynucleotides containing a T:G mismatch within a dcm siteand star sequences that vary by one base-pair from the dcm site

Figure 1. Determination of KD values for the bindingof vsr to oligonucleotides containing a T:G mismatch ina star dcm sequence (i.e. a sequence differing by onebase-pair from the cognate dcm sequence, Table 1) usingcompetitive titration and gel shift analysis. Assays werecarried out in 20 ml of 25 mM Hepes-NaOH (pH 7.5),100 mM NaCl, 0.5 mM DTT, 5 mM CaCl2, 0.1 mg/mlbovine serum albumin. vsr endonuclease (1 mM deter-mined by UV absorbance; this equates to 9 nM activeprotein, assuming that only 0.9 % of the protein canbind DNA) was pre-incubated with 1 nM of a32P-labelled oligonucleotide containing a T:G mismatchwithin a cognate dcm sequence (entry1, Table 1) for 15minutes at 37 �C. Increasing amounts of unlabelled com-petitor oligonucleotide, containing a T:G mismatchwithin a star dcm sequence, were added and after afurther 15 minutes incubation the mixture was analysedby non-denaturing gel electrophoresis as described.18

The top panel shows an autoradiogram of the gel pro-duced by adding increasing amounts of the oligonucleo-tide containing the star sequence TTAGG/AGTCC (topstrand written in the 50-30 direction, bottom strand writ-ten in the 30-50 direction; sequence shown in dark bluein the lower panel). The lower panel shows an analysisof the data obtained for the star oligonucleotides Thecounts in the free and bound bands were obtained byphosphoro-imaging of gels and used to produce plots ofthe percentage 32P-labelled dcm-oligonucleotide boundagainst concentration of unlabelled star oligonucleotide.The plots shown were produced using Scientist (Version2, MicroMath Scienti®c Software) and assuming compe-tition between the labelled probe and star oligonucleo-tides, an active vsr concentration of 9 nM and a KD

describing the interaction of vsr with the dcm oligonu-cleotide of 0.74 nM.18 With CTAAG/GGTTC (entry 7,Table 1) poor ®ts (not shown) were produced due toweak binding and the KD could be obtained onlyapproximately. The star oligonucleotides are identi®edby colour coding using the nomenclature in Table 1 (i.e.only the ®ve bases derived from the dcm sequence areshown and both strands (top 50-30, bottom 30-50) aregiven; the base-pair changed from the cognate dcmsequence is underlined; 5 � d5MeC). These plots allowthe determination of KD describing the interaction of thestar dcm sequences with vsr, which are summarised inTable 1.

504 DNA Recognition by the E. coli vsr Protein

been changed (entries 3 and 4, Table 1), gave avery smeared gel shift from which it provedimpossible to extract KD values; all other oligonu-cleotides (entries 5-9, Table 1) gave no visibleshifted band. However, all the star substrates werehydrolysed by vsr (Figure 2, Table 1), implyingthat they must bind; presumably, the lack of a gelshift results from binding too weak to be observedby direct titration. Therefore, competition titration,in which a pre-formed complex of the vsr endonu-clease and a 32P-labelled oligonucleotide containinga T:G mismatch within a hemi-methylated dcm site(entry 1, Table 1) were challenged with increasingamounts of star oligonucleotides, was used. Asshown in Figure 1, all of the star oligonucleotideswere able to displace the 32P-labelled sequencefrom vsr, although high concentrations wererequired. Analysis of the displacement curves iscomplicated by our earlier observation, based onreverse titrations (where increasing amounts ofoligonucleotide are added to protein) and burstkinetic analysis, that only about 1 % of puri®ed vsris capable of binding to DNA.18 Low bindingcapacity may arise because of a requirement foradditional proteins, such as MutS and MutL, topromote ef®cient interaction of the vsr endonu-clease with DNA. Previously, we demonstratedthat the displacement of the 32P-labelled T:G mis-match probe with an identical sequence lacking theradiolabel (entry 1 in Table 1) could be ®tted onlyassuming, in agreement with reverse titrations andburst kinetics, that just 0.9 % of the vsr was activein DNA binding. Making this assumption, the KD

for a T:G mismatch in a hemi-methylated dcm site(entry 1, Table 1) was found to be 0.74 nM, usingboth direct and competition titrations.18 The con-ditions used here for competitive titrations withstar sequences (1 mM vsr, as measured by UVabsorbance spectroscopy; 1 nM 32P-labelled oligo-nucleotide) are very similar to those used pre-viously (1 mM vsr, as measured by UV absorbancespectroscopy; 2.5 nM 32P-labelled oligonucleotide).As before, data could not be ®tted using a vsr con-centration of 1 mM. However, taking the active vsrlevels to be 9 nM (0.9 %) and using a KD of0.74 nM for the radiolabelled oligonucleotide gavegood ®ts to a model describing a competitive dis-placement of the 32P-labelled oligonucleotide bythe star sequences (Figure 1). Exceptionally, entry 7(CTAAG) was found to interact very weakly withvsr and satisfactory ®ts could not be obtained. TheKD values found for the star sequences are sum-marised in Table 1; all bind to vsr more weaklythan the cognate sequence (taken as a T:G mis-match within a hemi-methylated dcm sequence,entry 1) with binding af®nities reduced to between0.5 and 9 %. Earlier, it was shown that a T:G mis-match within a hemi-methylated dcm sequencebound to the vsr threefold more tightly than themismatch in an unmethylated dcm sequence.18 Thistrend is maintained with the star sequences; inevery case two- to threefold better binding is seen

Figure 2. Hydrolysis of oligonu-cleotides containing a G:T mis-match within a star dcm sequence(Table 1) by the vsr endonucleaseunder single turnover conditions.Assays were carried out in 100 mlof 25 mM Hepes-NaOH (pH 7.5),100 mM NaCl, 0.5 mM DTT,10 mM MgCl2, 0.1 mg/ml bovineserum albumin. vsr endonuclease(2.5 mM determined by UV absor-bance; this equates to 22.5 nMactive protein, assuming that only0.9 % of the protein can bind DNA)was incubated with 1 nM of 32P-labelled oligonucleotide at 37 �C.Aliquots (5 ml) were withdrawn atvarious times, added to ``denatur-ing'' buffer the stop the reactionand analysed by denaturingPAGE.18 The top panel shows anautoradiogram of the gel produced

when the oligonucleotide containing the star sequence TTAGG/AGT5C (top strand written in the 50-30 direction, bot-tom strand written in the 30-50 direction; sequence shown in light blue in the graphs) was incubated with the vsrendonuclease for the times given. The graphs show an analysis of the data found with all the star sequences. Thecounts in the substrate and product bands were determined by phosphoro-imaging and the data produced ®tted to a®rst order rate equation using Gra®t (version 3.09a, Erithacus Software). The star oligonucleotides are identi®ed bycolour coding using the nomenclature in Table 1 (i.e. only the ®ve bases derived from the dcm sequence are shownand both strands (top 50-30, bottom 30-50) are given; the base-pair changed from the cognate dcm sequence is under-lined; 5 � d5MeC). The rate constants (kst) describing the single turnover reaction of each of the star sequences aresummarised in Table 1.

DNA Recognition by the E. coli vsr Protein 505

in hemi-methylated as compared to unmethylatedcontexts.

The single turnover rate constants for the vsrwith the star sequences have been evaluated. Ide-ally, such a measurement requires that the enzymeis in excess of the DNA (achieved here by using2.5 mM vsr as measured by UV absorbance, corre-sponding to 22.5 nM ``active'' vsr and 1 nM oligo-nucleotide) and that all the nucleic acid is bound tothe protein. The second condition can be achievedreasonably well for tight binding substrates e.g. forentries 1 and 2 in Table 1 (T:G mismatches inhemi-and unmethylated dcm contexts), the amountof oligonucleotide bound to vsr is 97 % and 90 %,respectively. Unfortunately, the weaker binding ofthe star sequences meant it was impossible toachieve complete DNA binding in these cases. Forthe most weakly bound substrates, e.g. entries 6and 9 in Table 1 (CTCGG and CTAGT), only about20 % of the DNA will be bound to the vsr.Although this ®gure could be increased by usingmore vsr, the 2.5 mM amounts utilised (giving avsr concentration competent for DNA binding of22.5 nM) approach solubility limits and the concen-tration of puri®ed protein stocks. Therefore manyof the kst values determined may contain a contri-bution arising from the second-order reactiondescribing the binding of the two macromolecules.Figure 2 shows a typical time-course for thehydrolysis of one of the star sequences, as deter-mined by denaturing gel electrophoresis followedby autoradiography. This Figure also gives data

analysis, in which the information obtained fromgel electrophoresis has been ®tted to a single expo-nential to obtain kst values for all the starsequences. The rate constants obtained are sum-marised in Table 1. All the star sequences areturned over more slowly than the oligonucleotidecontaining the cognate site, analogous to theirweaker binding. However, the reduction in kst

values is generally less severe than the decrease inthe corresponding KD. There is a correlationbetween the two parameters as, in general, themore poorly a star sequence is bound, the slower itis hydrolysed. All sequences, whether star or cog-nate, are cut more rapidly when present in a hemi-methylated as opposed to an un-methylated dcmsite, by factors ranging from two- to sixfold.

The speci®city constants (kst/KD) for the oligo-nucleotides are also given in Table 1, as are thepercentage values for the ratio (kst/KD)A/(kst/KD)C

(where A represents a star oligonucleotide and Cthe cognate dcm sequence). This ratio de®nes thediscrimination that the vsr endonuclease exhibitsfor two oligonucleotides; the lower the value, thebetter the enzyme is able to distinguish betweencognate and nearly cognate sites. As can be seenfrom the Table, the vsr endonuclease is able to tol-erate base-pair substitutions at every positionwithin the dcm sequence, and all of the star oligo-nucleotides were capable of acting as substrates.However, in every case, speci®city constant ratioswere reduced to values that varied between 0.01and 4 %, and both weaker binding and slower

Table 2. Comparison of the substrate properties of oli-gonucleotides containing cognate and star dcm sites asdetermined previously23 and here

Previously,23 the selectivity of the vsr endonuclease was stu-died using multi-substrate kinetics. The cognate oligonucleotidecontained a T:G mismatch within an unmethylated dcm site(entry 1); star sites contained a one base change (entries 4, 6and 9, altered bases underlined, the numbers correspond tothose in Table 1). Relative rate constants (ki) were determinedand the results found with certain oligonucleotides, correspond-ing to those we have also studied, are summarised in theTable (ki for the cognate site is set to 100 % and star sites ratesare determined relative to this value). To enable a comparisonwith previous work,23 the results we obtained with commonoligonucleotides (based on kst/KD and given in Table 1) havebeen re-formulated. The oligonucleotide containing a T:G mis-match within an unmethylated dcm site has been set as the100 % control (cf. Table 1, where a T:G mismatch in a hemi-methylated dcm site is the 100 % control) and the star oligonu-cleotides have been referenced to this value. Note, the sequencesurrounding the central dcm site differs between the two stu-dies; similarly, the 100 % reference values, assigned to cognatecontrols, are unlikely to be the same in both cases.

506 DNA Recognition by the E. coli vsr Protein

hydrolysis contributed to the reduction. To acertain extent, these results can be rationalised byreference to a vsr-oligonucleotide co-crystalstructure.25 Most of the bases within the dcm rec-ognition site make direct hydrogen bond contactsto the protein through the major groove. Inter-actions include K89 to the ®rst (C:G) base-pair,E116 to the third (A:T) base-pair, and R120 to thefourth (G:C or G:5MeC) base-pair. Furthermorethese bases and the ®fth (G:C) base-pair also inter-act with the protein via water-mediated contacts inthe major groove. Additionally, amino acid resi-dues at the N terminus of vsr make contact withthe bases through the minor groove. Alteration ofthe bases would perturb these interactions andlead to reductions in both KD and kst with a conse-quent drop in speci®city constant. It is instructiveto compare the selectivity of vsr with type IIrestriction endonucleases, as both classes of pro-teins cut the phosphodiester backbone of DNA inresponse to a de®ned sequence. Almost all restric-tion endonucleases show star activity i.e. cut atsequences one base-pair different from their targetsites.26 However, star activity is very low and often

needs sub-optimal conditions (e.g. Mn2� ratherthan Mg2�, organic co-solvents, altered pH) to bemeasured. There have been a few studies in whichstar and cognate sequences have been comparedunder identical conditions, as carried out here forvsr. The most comprehensive involved the EcoRIrestriction endonuclease, where each base-pair inthe GAATTC target site was altered system-atically.27 The speci®city constants, measured usingkst/KD, (the same parameter used here) werereduced to between 1.5 � 10ÿ3 and 2.9 � 10ÿ8 % ofthe wild-type values. Similar conclusions werereached using steady-state kinetics in conjunctionwith binding assays.30 A number of investigationswith the EcoRV restriction endonuclease have alsoshown drastic reductions in speci®city constantsfor star sequences; values of 1 � 10ÿ5 to 1 � 10ÿ6 %relative to wild-type being typical;31 ± 33 withBamHI ®gures of between 3 � 10ÿ5 and 3 � 10ÿ8 %are observed (L. Jen-Jacobson, University of Pitts-burgh, personal communication). Thus it appearsthat the vsr endonuclease is much more tolerant ofbase-pair substitutions than are restriction endonu-cleases. This may arise because much of the DNArecognition by vsr, particularly of the T:G mis-match, arises from the intercalation of three aro-matic amino acid residues into the DNA helix.25

Thus, vsr may be designed to recognise a mis-match principally by intercalation; the contactsbetween the protein and the bases ¯anking themismatch (outlined above) providing only a rela-tively limited, when compared to restriction endo-nucleases, degree of selectivity for the target site.

Previous studies of vsr speci®city, carried outin vitro, have shown that bases at positions 1, 3and 5 (CTAGG) within the dcm site can each besubstituted with any of the other three bases, togive reasonable substrates.23 This investigationused only unmethylated dcm sites (rather than thehemi-methylated and unmethylated site investi-gated here); nevertheless, certain oligonucleotidesare common between the two studies, allowing acomparison as shown in Table 2. While reasonableagreement for the drop in substrate properties isobtained in one case (entry 4, TTAGG), in twoothers (entries 6 and 9, CTCGG and CTAGTrespectively), our measurements indicate muchworse substrates (by a factor of about 100) than theresults reported previously.33 The discrepancy mayarise from measurement of different parameters;kst/KD in this study, relative second-order rate con-stants (ki) earlier. Although the meaning of kst/KD

is clear, ki is poorly de®ned. It may represent rela-tive Michaelis constants i.e. ratios of kcat/KM

(although this will depend on the exact conditionsused) and so include a contribution from productrelease steps. As product release steps are notmeasured using kst/KD, the two metrics do notnecessarily give the same results. The earlierstudy33 used a vsr containing a truncation of 12 or14 amino acid residues from the N terminus. Crys-tal structures have shown that these residues inter-act with the bases in the dcm site25 and so the full-

DNA Recognition by the E. coli vsr Protein 507

length protein, used here, may be expected to havea higher selectivity. Finally, no absolute (asopposed to ratios) values of kinetic constants arequoted,33 making assessment of the relative qualityof the vsr preparations dif®cult. The in¯uence thatbase-pair 4 (CTAGG) has on vsr activity has notbeen investigated previously using biochemicalapproaches, although in vivo work has suggestedthat it plays a key role.24,34 This G:C base-pair isclearly important; even the relatively minor switchbetween G:C and G:5MeC (the latter is expected tobe the natural substrate due to Dcm methyltrans-ferase activity) produces notable effects, with themethylated base-pair being preferred by some ten-fold (18, Table 1). Converting the base-pair at pos-ition 4 to A:T gave a very poor substrate (Table 1)in which binding, even using competitive titration,was hard to measure accurately. Although an accu-rate speci®city constant was not obtainable,changes at this location clearly gives a worse sub-strate than alterations at other sites. Neverthelessbase-pair 4, while highly signi®cant for target siterecognition by vsr, is not absolutely essential, asmodi®cation still yields a substrate, albeit a verypoor one (Table 1).

We show that individually changing any of thefour bases in the dcm site, that ¯ank the mis-matched T, gives oligonucleotides that remain sub-strates for the vsr endonuclease. Furthermore, theresults give the ®rst rigorous thermodynamicmeasurement of the degree to which vsr prefersthe CTAGG sequence and so allows a comparisonwith other DNA-recognising enzymes. Substitutionof the target site bases invariably results in bothweaker binding (KD increases) and slower hydroly-sis (kst decreases). However, the overall drop inspeci®city constants (kst/KD) are relatively modest,especially when compared with type II restrictionendonucleases, paradigms for sequence-speci®cendonucleases. Previous investigations bothin vitro8,23 and in vivo24,34 have shown a somewhatrelaxed selectivity for the enzyme. Why does vsrhave low speci®city? The physiological role of theenzyme is thought to be the repair of T:G mis-matches arising from deamination of 5MeC:G base-pairs, with d5MeC in DNA arising from the actionof the dcm methyltransferase. While this methylaseundoubtedly prefers CC(A/T)GG sequences (meth-ylating the underlined dC),14,16 its selectivity hasnot been probed comprehensively and it may becapable of acting on star sequences. There is in vivoevidence that the Dcm (or EcoRII or HpaII) methyl-transferases are capable of slow methylation atnon-cognate sequences; with EcoRII in vitro staractivity has been demonstrated.16 Thus, the Dcmmethyltransferase may generate 5MeC:G base-pairs(and following deamination T:G mismatches) indcm star sequences; the vsr endonuclease wouldrequire a matching relaxed selectivity to repairsuch lesions. Alternatively,7,24 vsr may be requiredto maintain important DNA stretches related todcm, e.g. Chi sequences (GCTGGTGG) and REP(CC(G/T)GA and TC(C/A)GG) motifs, which are

important in recombination and gene regulation,respectively. Finally, vsr is known to act on U:Gmismatches in vitro18 and in vivo.6 It has been pro-posed that the protein might supplement the uracilglycosylase-mediated repair process and aid therepair of dU lesions.6 Again, relaxed selectivitywould be advantageous in such a role.

Acknowledgements

This work was supported by the UK BBSRC, the UKMRC, the EU and the Wellcome Trust. D.P.T. was aBBSRC-supported PhD student. We thank PaulineHeslop for expert technical assistance, and Dr RenosSavva (Birkbeck College, London) and Professor Laur-ence Pearl (Institute of Cancer Research, London) forsupplying the plasmids that overexpressed the vsr endo-nuclease.

References

1. Lindahl, T. & Nyberg, B. (1974). Heat-induceddeamination of cytosine residues in DNA. Biochemis-try, 13, 3405-3410.

2. Duncan, B. K. & Miller, J. (1980). Mutagenic deami-nation of cytosine residues in DNA. Nature, 287,560-561.

3. Lieb, M. (1983). Speci®c mismatch correction in bac-teriophage lambda crosses by very short patchrepair. Mol. Gen. Genet. 191, 118-125.

4. Lieb, M. (1985). Recombination of the l repressorgene: evidence that very short patch (VSP) correc-tion restores a speci®c sequence. Mol. Gen. Genet.199, 465-470.

5. Zell, R. & Fritz, H. (1987). DNA mismatch repair inEscherichia coli counteracting the hydrolytic deamina-tion of 5-methylcytosine residues. EMBO J. 6, 1809-1815.

6. Gabbara, S., Wyszynski, M. & Bhagwat, A. S. (1994).A DNA repair process in Escherichia coli correctsU:G and T:G mismatches to C:G at sites of cytosinemethylation. Mol. Gen. Genet. 243, 244-248.

7. Lieb, M. & Bhagwat, A. S. (1996). Very short patchrepair: reducing the cost of cytosine methylation.Mol. Microbiol. 20, 467-473.

8. Hennecke, F., Kolmar, H., Brundl, K. & Fritz, H. J.(1991). The vsr gene product of E. coli K-12 is astrand- and sequence-speci®c DNA mismatch endo-nuclease. Nature, 353, 776-778.

9. Tsutakawa, S. E., Muto, T., Kawate, T., Jingami, H.,Kunishima, N., Ariyoshi, M. et al. (1999a). Crystalstructure and biochemistry of VSR protein. Mol. Cell,3, 621-628.

10. Dzidic, S. & Radman, M. (1989). Genetic require-ments for hyper-recombination of very short patchrepair: involvement of Escherichia coli DNA polymer-ase I. Mol. Gen. Genet. 217, 254-256.

11. Marinus, M. G. (1973). Location of DNA methyl-ation genes on the Escherichia coli K-12 genetic map.Mol. Gen. Genet. 127, 47-55.

12. Sohail, A., Lieb, M., Dar, M. & Bhagwat, A. S.(1990). A gene required for very short patch repairin Escherichia coli is adjacent to the DNA cytosinemethylase gene. J. Bacteriol. 172, 4214-4221.

508 DNA Recognition by the E. coli vsr Protein

13. Dar, M. E. & Bhagwat, A. S. (1993). Mechanism ofexpression of DNA repair gene vsr, an Escherichiacoli gene that overlaps the DNA cytosine methylasegene. Mol. Microbiol. 9, 823-833.

14. Urieli-Shoval, S., Gruenbaum, Y. & Razin, A. (1983).Sequence and substrate speci®city of isolated DNAmethylases from Escherichia coli C. J. Bacteriol. 153,274-280.

15. Marinus, M. G. (1984). Methylation of prokaryoticDNA. In DNA Methylation (Razin, A., Cedar, H. &Riggs, A. D., eds), pp. 81-109, Springer-Verlag,Heidelberg, Germany.

16. Bandaru, B., Gopal, J. & Bhagwat, A. S. (1996).Overproduction of DNA cytosine methyltransferasescauses methylation and C to T mutations at non-canonical sites. J. Biol. Chem. 271, 7851-7859.

17. Choi, S. H. & Leach, J. E. (1994). Identi®cation of theXorII methyltransferase gene and a vsr homologfrom Xanthomonas oryzae pv. oryzae. Mol. Gen. Genet.244, 383-390.

18. Turner, D. P. & Connolly, B. A. (2000). Interaction ofthe E. coli G:T-mismatch endonuclease (vsr protein)with oligonucleotides containing its target sequence.J. Mol. Biol. 304, 765-778.

19. Drotschmann, K., Aronshtam, A., Fritz, H. J. &Marinus, M. G. (1998). The Escherichia coli MutLprotein stimulates binding of VSR and MutS toheteroduplex DNA. Nucl. Acids Res. 26, 948-953.

20. Mansour, C. A., Doiron, K. M. J. & Cupples, C. G.(2001). Characterisation of functional interactionsamong the Escherichia coli mismatch repair proteinsusing a bacterial two-hybrid assay. Mutat. Res. 485,331-338.

21. Jones, M., Wagner, R. & Radman, M. (1987).Mismatch repair of deaminated 5-methylcytosine.J. Mol. Biol. 193, 155-159.

22. Lieb, M. (1987). Bacterial genes Mut L, Mut S, anddcm participate in repair of mismatches at 5-methyl-cytosine sites. J. Bacteriol. 169, 5241-5246.

23. GlaÈsner, W., Merkl, R., Schellenberger, V. & Fritz,H. J. (1995). Substrate preferences of VSR DNAmismatch endonuclease and their consequences forthe evolution of the Escherichia coli K-12 genome.J. Mol. Biol. 245, 1-7.

24. Lieb, M. & Rehmat, S. (1995). Very short patchrepair of T:G mismatches in vivo: importance ofcontext and accessory proteins. J. Bacteriol. 177, 660-666.

25. Tsutakawa, S. E., Jingami, H. & Morikawa, K.(1999b). Recognition of a TG mismatch: the crystalstructure of very short patch repair endonuclease incomplex with a DNA duplex. Cell, 99, 615-623.

26. Pingoud, A. & Jeltsch, A. (1997). Recognition andcleavage of DNA by type-II restriction endonu-cleases. Eur. J. Biochem. 246, 1-22.

27. Lesser, D. R., Kurpiewski, M. R. & Jen-Jacobson, L.(1990). The energetic basis of speci®city in the EcoRIendonuclease-DNA interaction. Science, 250, 776-786.

28. Jen-Jacobson, L. (1995). Structural-perturbationapproaches to thermodynamics of site-speci®c pro-tein-DNA interactions: energetics of macromolecules.Methods Enzymol. 259, 305-344.

29. Jen-Jacobson, L. (1997). Protein-DNA recognitioncomplexes: conservation of structure and bindingenergy in the transition state. Biopolymers, 44, 153-180.

30. Thielking, V., Alves, J., Fliess, A., Maass, G. &Pingoud, A. (1990). Accuracy of the EcoRI restrictionendonuclease: binding and cleavage studies witholigodeoxynucleotides containing degenerate recog-nition sequences. Biochemistry, 29, 4682-4691.

31. Taylor, J. & Halford, S. E. (1989). Discriminationbetween DNA sequences by the EcoRV endonu-clease. Biochemistry, 28, 6198-6207.

32. Alves, J., Selent, U. & Wolfes, H. (1995). Accuracy ofthe EcoRV restriction endonuclease: binding andcleavage studies with oligodeoxynucleotide sub-strates containing degenerate recognition sequences.Biochemistry, 34, 11191-11197.

33. Engler, L. E., Welch, K. K. & Jen-Jacobson, J. (1997).Speci®c binding by EcoRV endonuclease to its recog-nition site GATATC. J. Mol. Biol. 269, 82-101.

34. Lieb, M., Allen, E. & Read, D. (1986). Very shortpatch DNA mismatch repair in phage lambda:repair sites and length of repair tracts. Genetics, 114,1041-1060.

Edited by J. Karn

(Received 1 March 2001; received in revised form 22 May 2001; accepted 24 May 2001)