Differential Specificities and Simultaneous Occupancy of Human MutSα Nucleotide Binding Sites*

We have examined the permissible nucleotide occupancy states of human MutSα. The MSH2·MSH6 heterodimer binds 1 mol of ADP and 1 mol of adenosine 5′-O-(thiotriphosphate) (ATPγS), with a Kd for each nucleotide of about 1 μm. Anisotropy measurements using BODIPY TR and BODIPY FL fluorescent derivatives of ADP and 5′-adenylyl-β,γ-imidodiphosphate (AMPPNP) also indicate an interaction stoichiometry of 1 mol of ADP and 1 mol of triphosphate analogue per MutSα heterodimer. Di- and triphosphate sites can be simultaneously occupied as judged by sequential filling of the two binding site classes with differentially radiolabeled ADP and ATPγS and by fluorescence resonance energy transfer between BODIPY TR- and BODIPY FL-labeled ADP and AMPPNP. ATP hydrolysis by MutSα is accompanied by a pre-steady-state burst of ADP formation, and analysis of MutSα-bound nucleotide during the first turnover has demonstrated the presence of both ADP and ATP. Simultaneous presence of ADP and a nonhydrolyzable ATP analogue modulates MutSα·heteroduplex interaction in a manner that is distinct from that observed in the presence of ADP or nonhydrolyzable triphosphate alone, and it is unlikely that this effect is due to the presence of a mixed population of binary complexes between MutSα and ADP or a triphosphate analogue. These findings imply that MutSα has two nucleotide binding sites with differential specificities for ADP and ATP and suggest that the ADP·MutSα·ATP ternary complex has an important role in mismatch repair.

In the human cell lines that have been examined, the majority of the MSH2 is associated with MSH6 (11,12,14), and the MutS␣ complex is present in a 6-to 10-fold excess over MutS␤.
Several models for ATPase function in MutS homologue action have been suggested. Two of these invoke ATP-dependent movement of the protein along the helix, which is postulated to link mismatch recognition to activation of downstream events at the strand signal that directs repair (24 -26). Electron microscopic visualization of bacterial MutS⅐heteroduplex complexes has demonstrated the mismatch-and ATP-dependent extrusion of a DNA loop by the bacterial protein (24). Because nonhydrolyzable ATP analogues do not support loop extrusion and because their addition terminates ongoing loop growth, this reaction has been attributed to directional translocation along the helix in a reaction dependent on ATP hydrolysis by the DNA-bound protein. The use of streptavidin end-blocked linear DNAs has also suggested that MutS and MutS␣ leave a mismatch in an ATP-dependent manner, observations that have led to the suggestion that the protein may form a mobile clamp about the helix (22,25,26,29). Although ATP⅐Mg 2ϩ supports the formation of stable complexes on end-blocked heteroduplexes, AMPPNP⅐Mg 2ϩ , 1 ATP␥S⅐Mg 2ϩ , or ATP (no Mg 2ϩ ) do not, suggesting that formation of such intermediates may depend on ATP hydrolysis by heteroduplex-bound MutS␣ (25). The interaction with end-blocked DNA of a mutant form of MutS␣, which supports ATP binding but not hydrolysis, has led to a similar conclusion (27).
A mechanism for MutS/MutS␣ movement along the helix that is independent of ATP hydrolysis by DNA-bound protein has also been proposed (26,29). This molecular switch model posits heteroduplex binding by the MutS/MutS␣⅐ADP complex, mismatch-provoked exchange of ADP for ATP, and free diffusion of the MutS/MutS␣⅐ATP complex along the helix. In this mechanism, ATP hydrolysis occurs after dissociation from the DNA to regenerate the ADP form of the protein. As in the translocation model, this movement is postulated to play an important role in activation of downstream activities that act at the strand signal. Heteroduplex-promoted exchange of ATP for ADP has been demonstrated with human MutS␣ (21,28) and Escherichia coli MutS (29); however, homoduplex DNA also promotes exchange at about 30% of the rate observed with DNA containing a mismatch (28,29). This modest dependence of exchange rate on a mismatch is seemingly insufficient to account for the observed mismatch dependence of activation of downstream activities (Ն20-fold for MutS-and MutL-dependent activation of MutH in the bacterial system (30) and about 10-fold for mismatch-dependent activation of bacterial and human excision systems (31)(32)(33)).
A distinct function for ATP binding and hydrolysis by MutS homologues has also been suggested. In contrast to models invoking ATP-dependent movement of these proteins along the helix, this proposal stipulates that, once mismatch recognition occurs, MutS/MutS␣ remains bound to the mispair during the course of repair (34). In this model, interaction of the mismatch and the strand signal is mediated by DNA bending, and ATP binding by DNA-bound MutS/MutS␣ functions in a kinetic proofreading mechanism that serves to verify mismatch recognition prior to initiation of repair. Although this model has several attractive features, it is inconsistent with the finding that bacterial MutS⅐MutL and human MutS␣⅐MutL␣ complexes leave the mispair by movement along the helix contour (29,35,36) and studies of in vivo assembly of bacterial MutS and MutL foci at mispairs, which suggest that the two proteins migrate from a mismatch in the cell (37).
To further clarify the role of the human MutS␣ ATPase, we have determined the possible states of nucleotide occupancy of the protein. The stoichiometry of MutS␣ interaction with ADP and nonhydrolyzable ATP analogues indicates that the two nucleotide binding sites within the MSH2⅐MSH6 heterodimer display differential specificities for ADP and ATP. These studies also show that di-and triphosphate binding sites can be simultaneously occupied at nucleotide concentrations in the low micromolar range.

EXPERIMENTAL PROCEDURES
Proteins and Oligonucleotides-Human MutS␣ was purified from SF9 cells infected with a baculovirus expressing construct by a modification of the previously described procedure (7,36). Eluates from the Mono Q column (30 -40 mg, about 1 mg/ml) were diluted to a conductivity equivalent to 100 mM KCl and loaded at 2 ml/min onto an 8-ml Mono S column (Amersham Biosciences) equilibrated with 25 mM HEPES-KOH, pH 7.5, 100 mM KCl, 0.1 mM EDTA, and protease inhibitors (1 g/ml each Pefabloc, E64, aprotinin, leupeptin, and pepstatin A). After washing with 40 ml of starting buffer, the column was eluted with a 60-ml gradient of KCl (100 to 370 mM) in 25 mM HEPES-KOH, pH 7.5, 0.1 mM EDTA, and protease inhibitors as above. MutS␣ fractions, which eluted at about 200 mM KCl, were supplemented with 1 mM dithiothreitol, pooled, and dialyzed against 25 mM HEPES⅐KOH, pH 7.5, 0.1 mM EDTA, 1 mM dithiothreitol containing protease inhibitors and 1 M KCl, for 5 h, and then against the same buffer containing 0.2 M KCl for 1 h. After spin concentration at 4°C using a Centriplus YM-30 (Millipore), samples were frozen in liquid N 2 and stored at Ϫ70°C. Concentrations of the protein are expressed as heterodimer equivalents and were determined from the absorbance at 280 nm using an extinction coefficient calculated from the primary amino acid sequences of the MSH2 and MSH6 subunits (38). Nucleotide content of MutS␣ preparations was determined by bioluminescent assay as described previously (25,39). Previous studies have shown that isolated MutS␣ can contain significant levels of bound ADP (25). The dialysis procedure described above results in a significant reduction in the ADP level in MutS␣ isolates. The preparations used in this study contained about 0.07 mol of ADP per mole of heterodimer and were free of detectable ATP.
5Ј-Biotinylated and nonbiotinylated oligonucleotides were obtained from Oligos Etc. (Wilsonville, OR). 31-bp G-T heteroduplex and A⅐T homoduplex substrates were identical to those described previously (24). Appropriate oligonucleotides were annealed at a 1:1 molar ratio in a PerkinElmer Life Sciences Gene Amp 9600 thermocycler by heating to 90°C in 10 mM HEPES-KOH (pH 7.5), 100 mM KCl, followed by cooling to 15°C over 90-min period.
Surface Plasmon Resonance Measurements-MutS␣-DNA interaction was monitored by surface plasmon resonance spectroscopy (SPRS) using a Biacore 2000 optical biosensor system. Streptavidin sensor chips (SA chip, Amersham Biosciences) were prepared according to recommendations of the manufacturer and derivatized with ϳ100 resonance units (RUs) of biotinylated heteroduplex or homoduplex DNA. MutS␣ was injected at a flow rate of 20 l/min in buffer comprised of 10 mM HEPES-KOH (pH 7.5), 1 mM EDTA, 5 mM MgCl 2 , 150 mM NaCl, 0.005% surfactant P20. Experiments were performed at 25°C, and samples were maintained at 4°C prior to injection. Adenine nucleotide effects on association of MutS␣ with DNA were monitored by preincubation of the protein with indicated nucleotides at 4°C for 5 min prior to injection.
Dissociation constants for interaction of MutS␣ and MutS␣⅐nucleotide complexes with heteroduplex DNA were estimated by SPRS by titration of chip-bound DNA with increasing concentrations of MutS␣ or MutS␣⅐nucleotide complexes in the buffer system described above. Maximum binding values obtained after completion of the association phase of the reaction were plotted as a function of MutS␣ concentration, data were fit to a square hyperbola by nonlinear regression analysis, and the K d value was extracted (22). ; ATP␥S, Calbiochem-Novabiochem; or AMPPNP, Calbiochem-Novabiochem) to achieve the desired specific activity. Reactions were incubated for 4 min at 0°C, diluted to 1 ml with room temperature reaction buffer, and immediately filtered through 13-mm, 0.45-m nitrocellulose membranes (Schleicher & Schuell, BA-45), which had been pre-washed with reaction buffer. Filters were dried and radioactivity determined by liquid scintillation counting. For analysis of sequential binding of di-and triphosphates, MutS␣ was first incubated with a near saturating concentration of one nucleotide for 4 min prior to titration with the second nucleotide. Background nucleotide binding was less than 4%, and data shown were corrected for background values.
Human MutS␣ hydrolyzes ATP␥S at a low rate (17), potentially compromising interpretation of filter binding experiments using this nucleotide. To evaluate the significance of ATP␥S hydrolysis, 8 M MutS␣ was incubated under filter binding conditions (0°C in binding buffer) with 8 M [ 35 S]ATP␥S, the reaction was sampled as a function of time, samples were analyzed by chromatography on PEI-cellulose plates (EM Science, Gibbstown, NJ) as described (40), and radioactivity was quantitated using an Amersham Biosciences PhosphorImager. This analysis indicated that hydrolysis under filter-binding conditions occurred at a rate of 1.6% per min. In a second approach 8 M MutS␣ was incubated with 8 M [ 35 S]ATP␥S for 4 and 8 min at 0°C, and MutS␣⅐[ 35 S]ATP␥S complexes were collected on filters. Filter-bound nucleotide was eluted by 2-h incubation in 10 mM EDTA and 0.2% sodium dodecyl sulfate at room temperature, and samples were analyzed by thin layer chromatography as described above. Recovered radioactivity was Ͼ80% ATP␥S for both 4-and 8-min incubations.
In similar experiments, 10 M MutS␣ was incubated at 0°C with 50 M [ 32 P]AMPPNP. No hydrolysis of the ATP analogue was detected after 30 min of incubation. ATP Hydrolysis by MutS␣ and Nucleotide Occupancy under Steadystate Conditions-Reactions (20 l) contained 10 mM HEPES-NaOH, pH 7.5, 150 mM NaCl, 10 M MutS␣, 100 M [␣-32 P]ATP (3000 Ci/mmol, PerkinElmer Life Sciences, diluted to desired specific activity with unlabeled ATP), and 1 mM MgCl 2 . Hydrolysis was initiated by addition of [␣-32 P]ATP and MgCl 2 to reaction buffer containing MutS␣, and incubation was carried out at 0°C. Samples (1 l) were removed as indicated, quenched by addition to 40 l of 0.2% SDS, 10 mM EDTA, and samples were spotted on PEI-cellulose plates, which were processed as described above. Parallel reactions (200 l) were subjected to filter binding analysis to evaluate composition of MutS␣-bound nucleotide. Samples (20 l) were removed at indicated times, diluted to 1 ml with 20 mM HEPES-NaOH, pH 7.5, 150 mM NaCl, and MutS␣-nucleotide complexes collected on nitrocellulose membranes as described above. Filters were immediately transferred to 150 l of 10 mM HEPES-NaOH, pH 7.5, 0.2% SDS, and 10 mM EDTA. After 20 min at room temperature, the nucleotide composition of filter eluates was determined by PEI chromatography. Although only trace levels of [ 32 P]AMP (Յ0.5%) were observed in ATP hydrolytic reactions after 20 min of incubation, 15% of filter-eluted nucleotide was consistently recovered as AMP. Although the origin of this AMP is not clear, it was not observed in control experiments in which ATP hydrolytic reactions were heated at 70°C for 5 min and samples were spotted on nitrocellulose membranes, which were then subjected to the filter elution procedure.
Fluorescence Measurements-Affinities of MutS␣ for the fluorescent nucleotide derivatives BODIPY TR (2Ј or 3Ј)-ADP and BODIPY FL (2Ј or 3Ј)-AMPPNP (Molecular Probes, Inc.) were determined by anisotropy measurement. Reactions contained MutS␣ and nucleotide as indicated in 20 mM HEPES-KOH (pH 7.5), 150 mM KCl, 5 mM MgCl 2 , and 10% (v/v) glycerol. Measurements were performed at 4°C using an SLM-AMINCO 8100 Series 2 spectrofluorometer fitted with calcite polarizers using the L-format. The G factor was determined once for each experiment. BODIPY FL (2Ј or 3Ј)-AMPPNP was excited at 490 nm, and maximal emission was observed at 514 nm. Excitation of BODIPY TR (2Ј or 3Ј)-ADP was at 590 nm with peak emission occurring at 620 nm. Fluorescence anisotropy was calculated using software provided with the instrument.
Because there is significant spectral overlap between the emission spectrum of BODIPY FL (2Ј or 3Ј)-AMPPNP and the excitation spectrum of BODIPY TR (2Ј or 3Ј)-ADP (41), simultaneous binding of the two nucleotides to MutS␣ was evaluated by energy transfer. Energy transfer between BODIPY FL (2Ј or 3Ј)-AMPPNP and the BODIPY TR (2Ј or 3Ј)-ADP acceptor was monitored by exciting at 490 nm and determining the emission at 620 nm. Reaction conditions were as described above for anisotropy measurements using protein and nucleotide concentrations as indicated. Efficiency of energy transfer was determined by the Ratio A method (42, 43) as described previously (41) except that the experimentally determined value for the term ⑀ TR (490 nm)/⑀ TR (590 nm) under the buffer conditions used in this work was 0.048. A value of for the Förster radius, R 0 , which is the distance where energy transfer is 50% efficient, was determined to be 38.1 Å (41). This R 0 value and the measured efficiency of energy transfer were used to estimate the distance between di-and triphosphate binding centers of MutS␣ as described (43).

MutS␣ Binds ADP and Nonhydrolyzable ATP Analogues
with Micromolar Affinities-ATP and nonhydrolyzable ATP analogues reduce the efficiency of heteroduplex binding by human MutS␣ (7,8,21,25), but ADP does not significantly alter the affinity of the protein for a mismatch (25). Human MutS␣ has been shown to bind ATP␥S with an affinity of 0.7 M and to bind ADP (28), although the K d governing the latter interaction has not been established. Using filter binding assay and fluorescent methods, we have examined the interaction of MutS␣ with ADP, ATP␥S, AMPPNP, and the BODIPY fluorescent derivatives of ADP and AMPPNP. As shown in Fig. 1, binding to each of these nucleotides was hyperbolic with respect to MutS␣ concentration. For filter binding assays, K d values and filter retention efficiencies (RE) were obtained by nonlinear least squares fit (44) to Equation 1, where S t is the total MutS␣ concentration (as heterodimer) and A t is the total nucleotide concentration. Fluorescent nucleotide binding data obtained by anisotropy measurement were fit in a similar manner to Equation 2. MutS␣ Has One ADP and One ATP Binding Site per Heterodimer-Because each subunit of the MSH2⅐MSH6 heterodimer contains a Walker nucleotide binding consensus (45)(46)(47), the protein may potentially exist in nine different nucleotide occupancy states (48). To address this issue we have examined the stoichiometry of the interactions of the protein with ADP and nonhydrolyzable ATP analogues. In these experiments a fixed nucleotide concentration was titrated with MutS␣ ( Fig. 2) or a fixed MutS␣ concentration was titrated with nucleotide (Fig. 3). In both types of experiment, the concentration of the fixed component was well above the corresponding K d to ensure that binding was largely stoichiometric. Data were fit to Equation 1 or 2, which assume a single class of binding site. Given the knowledge of the K d and the total concentration of the fixed component, the stoichiometry of binding can be determined. For example, in a titration of a fixed ADP concentration with MutS␣, this analysis yields a value for A t , the apparent ADP concentration at the equivalence point. Division of A t by the actual nucleotide concentration yields the stoichiometry of interaction. As shown in Fig. 2, Equations 1 and 2 provided excellent fits to the binding behavior observed when fixed concentrations of ADP, ATP␥S, BODIPY TR (2Ј or 3Ј)-ADP, or BODIPY FL (2Ј or 3Ј)-AMPPNP were titrated with MutS␣. Good fits to the experimental data were also obtained when MutS␣ was titrated with ADP and ATP␥S (Fig. 3). The stoichiometries of interaction determined by this method are 0.98 -1.13 mol of ADP, 0.96 -1.03 mol of ATP␥S, 1.16 mol of BODIPY TR (2Ј or 3Ј)-ADP, and 0.82 mol of BODIPY FL (2Ј or 3Ј)-AMPPNP per mole of the MSH2⅐MSH6 heterodimer (Figs. 2 and 3). Inasmuch as each subunit of MutS␣ contains one nucleotide binding site (17), these observations show that only half of the expected number of binding sites are available for high affinity occupancy by ADP or nonhydrolyzable ATP analogues. Previous studies have shown that isolates of human MutS␣ can contain significant amounts of bound ADP (25). However, the preparations used in this work contained only 0.07 mol of ADP per mole of MutS␣ and were free of detectable levels of ATP (see "Experimental Procedures"). Consequently, the nucleotide binding stoichiometries observed here can not be due to the fact that one binding site is already occupied in MutS␣ preparations used.
To test the possibility that these binding stoichiometries might be due to differential specificity of the two MutS␣ nucleotide binding sites, the protein was tested for its ability to simultaneously bind ADP and ATP␥S. As shown in Fig. 3, a binding site for ATP␥S was available on the MutS␣ heterodimer after the ADP binding site had been filled, and an ADP binding site was available when the triphosphate site was filled. Thus, MutS␣ heterodimer contains two nucleotide binding sites that display differential specificities for ADP and ATP. Because we have been unable to determine the affinities of MutS␣ for ADP and for ATP␥S under conditions where the other nucleotide binding site is occupied, it is possible that occupancy of the triphosphate site modulates the affinity of the diphosphate site, and vice versa, as has been observed with bacterial MutS (41). However, the data of Fig. 3 imply that binding affinities for the second nucleotide must also be in the low micromolar range.
The two MutS␣ nucleotide binding sites display differential specificities for adenine nucleoside di-and triphosphate, but it is possible that high concentrations can drive ADP into the triphosphate site and vice versa. This question was addressed by titrating 10 M MutS␣ with ADP or ATP␥S. As judged by filter binding assay, occupancy in each case was restricted to a single site at nucleotide concentrations as high as 100 M (Fig.  4). Although these findings indicate that one binding site is highly specific for ADP and the other highly specific for triphosphate, it is important to note that ADP⅐MutS␣⅐ADP and ATP␥S⅐MutS␣⅐ATP␥S complexes might go undetected by this assay. This would be the case, for example, if binding of the protein to the filter leads to release of a weakly bound second nucleotide or if the lifetime of the more weakly bound nucleotide is short relative to filtration time. However, these findings serve to confirm the substantial specificity differences of the MutS␣ nucleotide binding sites.
MutS␣ does not hydrolyze AMPPNP at a detectable rate under filter binding conditions ("Experimental Procedures"), but ATP␥S is hydrolyzed at about 10% the rate of ATP (17). However, filter binding experiments described here were performed at 0°C, conditions under which ATP␥S hydrolysis is exceedingly slow ("Experimental Procedures"). It is also important to note that a positive signal in [ 35 S]ATP␥S binding experiments requires the presence of the ␥-phosphate, and elution of filter-bound complexes formed in the presence of this nucleotide demonstrated more than 80% of the bound label to be ATP␥S ("Experimental Procedures"). Therefore, ATP␥S hydrolysis does not compromise the interpretation of binding experiments using this nucleotide. The experiments described above indicate that the two MutS␣ nucleotide binding sites display differential specificities for ADP and nonhydrolyzable ATP analogues.

Simultaneous Occupancy of the MutS␣ ADP and ATP Binding Sites by Fluorescence Resonance Energy
Transfer-The filter binding studies described demonstrate differential specificities of the two MutS␣ nucleotide binding sites and indicate that the di-and triphosphate sites can be simultaneously occupied. The possibility of simultaneous occupancy was confirmed by application of fluorescence resonance energy transfer. MutS␣ binds BODIPY TR (2Ј or 3Ј)-ADP and BODIPY FL (2Ј or 3Ј)-AMPPNP with high affinity and specificity (Figs. 1 and 2). We have exploited the fact that these two fluorescent derivatives form an energy transfer pair (41) to confirm the idea that the di-and triphosphate binding sites of MutS␣ can be simultaneously occupied. Because energy transfer is proportional to the inverse sixth power of the distance separating the two probes, interaction between the fluorophores is detectable only if the molecules in question are in proximity (49).
Emission spectra of a solution of 5 M BODIPY FL (2Ј or 3Ј)-AMPPNP (donor) and 5 M BODIPY TR (2Ј or 3Ј)-ADP (acceptor) were determined upon donor excitation at 490 nm in the presence and absence of 5 M MutS␣. As shown in Fig. 5  (upper panel), the presence of MutS␣ decreased the emission intensity of the donor at 514 nm and enhanced emission at 620 nm, indicative of sensitized emission of the acceptor, and these effects were abolished by inclusion of excess ATP (Fig. 5, lower  panel). These observations indicate energy transfer between the two fluorophores and imply simultaneous occupancy of MutS␣ di-and triphosphate binding sites. The efficiency of energy transfer in this system is 33%, corresponding to a distance of 46 Å between the fluorophores when bound to MutS␣ in the absence of DNA ("Experimental Procedures"). Because calculation of this distance assumes 100% occupancy of di-and triphosphate, this value is an upper limit for the separation distance of the fluorescent tags on the AMPPNP donor and the MutS␣⅐nucleotide complexes were collected on nitrocellulose membranes over the course of the first 2 min (1.4 half-lives) of hydrolysis, and composition of MutS␣-bound nucleotide was determined. As shown in Fig. 6, the molar ratio of bound ADP:ATP was essentially invariant over this period, with a mean value of 1.61 Ϯ 0.21. Because the burst kinetics observed imply that the rate-limiting step for turnover occurs after hydrolysis, these findings strongly suggest that ADP and ATP are simultaneously bound to a substantial fraction of MutS␣ heterodimers under hydrolytic conditions.
Adenine Nucleotide Effects on MutS␣ Affinity for Heteroduplex DNA-Previous studies have shown that ADP does not significantly affect the affinity of MutS␣ for a mismatch, but the affinity of the protein for a mispair is reduced in the presence of ATP or nonhydrolyzable ATP analogues (7, 8, 17, 19 -21, 23, 25). Because the experiments described here permit prediction of nucleotide occupancy of the free protein, we have evaluated the potential consequences of distinct nucleotide occupancy states on MutS␣-heteroduplex interaction. In these studies, the protein was equilibrated with the indicated nucleotide(s) at 4°C for 5 min prior to analysis of interaction at 25°C with a 31-bp G-T heteroduplex or A⅐T homoduplex by surface plasmon resonance spectroscopy.
As shown in Fig. 7 and Table I, the affinity of MutS␣ for a 31-bp G-T heteroduplex remains constant at 20 Ϯ 6 nM over an ADP concentration range of 15-1000 M, an affinity identical to that observed in the absence of nucleotide ( Fig. 7 and Table I) 122 Ϯ 10 nM for ADP and AMPPNP where di-and triphosphates were present equal concentrations over the range of 100 -1000 M.) We think it unlikely that the intermediate heteroduplex affinities observed in the presence of ADP and a triphosphate analogue are due to production of a mixed population of MutS␣⅐ADP and MutS␣⅐AMPPNP (or MutS␣⅐ATP␥S) binary complexes under these conditions. If this were the case, then binding profiles in the presence of the two nucleotides could be calculated based on the independent binding of MutS␣⅐ADP and MutS␣⅐AMPPNP (or MutS␣⅐ATP␥S) complexes to heteroduplex DNA. As shown in Fig. 5 (E and F), the binding behavior predicted for a mixed population of binary MutS␣⅐nucleotide complexes deviates significantly from experimental binding  Table I. Dashed lines in E and F correspond to expected heteroduplex binding profiles if the potential formation of the ternary ADP⅐MutS␣⅐AMPPNP (or ADP⅐MutS␣⅐ATP␥S) complex is ignored, and nucleotide binding by MutS␣ is restricted to formation of binary MutS␣⅐ADP and MutS␣⅐AMPPNP (or MutS␣⅐ATP␥S) complexes, with these species binding independently and competitively to heteroduplex. These profiles were calculated as follows: In the SPRS method used in these experiments, the total concentration of free MutS␣ and its nucleotide complexes after achievement of equilibrium corresponds to the input concentration of the protein [M] total . If nucleotide binding to MutS␣ is restricted to formation of binary complexes with ADP or a triphosphate analogue, then where [D] is the concentration of ADP, [T] the concentration of AMPPNP (or ATP␥S), and K D and K T are dissociation constants for the MutS␣⅐ADP and the MutS␣⅐AMPPNP/ATP␥S complexes, respectively (Fig. 1). For a model invoking independent and competitive binding of MutS␣, MutS␣⅐ADP, and MutS␣⅐AMPPNP/ATP␥S to heteroduplex, the extent of DNA binding in resonance units is given by Equation 3, where RU max is the maximum observed binding from hyperbolic fit of experimental data, K 1 is the equilibrium constant governing dissociation of the MutS␣⅐ADP complex from heteroduplex DNA (panel A and Table I), K 2 is the equilibrium constant governing dissociation of the MutS␣⅐AMPPNP⅐ATP␥S complex from DNA (panels C and D, and Table I), and K 3 is the dissociation constant of the MutS␣⅐heteroduplex complex ( Table I) 1)) predict that 80% of MutS␣ will be present as the ADP complex, with the remainder largely MutS␣⅐AMPPNP. Because the heteroduplex affinity of the MutS␣⅐ADP complex is 20 times that of MutS␣⅐AMPPNP (Table I), this model predicts that nearly all the heteroduplex binding observed in the presence of the two nucleotides would be due to MutS␣⅐ADP, which is characterized by a mean dissociation constant of 20 Ϯ 6 nM. This value is considerably less than that observed in the presence of both nucleotides (mean value 122 Ϯ 10 nM). Similar arguments apply to the deviation of experimental binding isotherms observed in the presence of ADP and ATP␥S from those predicted by a model that restricts nucleotide occupancy to formation of MutS␣⅐ADP and MutS␣⅐ATP␥S. Whereas difficult to reconcile with a model that invokes a mixed population of binary complexes of MutS␣ with ADP or a triphosphate analogue, the experimental binding curves obtained in the presence of the two nucleotides can be understood in terms of the results described above, which indicate that the ADP⅐MutS␣⅐AMPPNP⅐ATP␥S ternary complex is highly populated in solution when ADP and a nonhydrolyzable triphosphate analogue are both present.

DISCUSSION
Structural analysis of the bacterial MutS⅐heteroduplex complex has indicated intrinsic asymmetry in the dimeric protein, with one subunit preferentially interacting with the mismatch (50,51). In the complex obtained with the Escherichia coli protein, the nucleotide binding site of mismatch recognition subunit is occupied by ADP and the other site is empty (50). Analysis of nucleotide binding by E. coli MutS has demonstrated that this asymmetry is also evident in solution in the absence of DNA: the two nucleotide binding sites within the dimer display differential specificity for ADP or nonhydrolyzable ATP analogues, and di-and triphosphate sites can be simultaneously occupied (41). Furthermore, isolates of hydrolytically defective MutS E694A have been shown to contain 1 mol of bound ATP per dimer equivalent and sub-stoichiometric quantities of ADP (0.1-0.6 mol/mol), suggesting that the ATP⅐MutS⅐ADP complex may occur within the cell (39). The results described here indicate that differential specificity and the potential for simultaneous occupancy are conserved in human MutS␣.
While this report was in preparation, Antony and Hingorani (52) demonstrated that yeast MutS␣ binds only one molecule of ADP but is capable of binding two molecules of ATP␥S with different affinities, although simultaneous occupancy of di-and triphosphate sites was not addressed in these studies. At substoichiometric concentrations, AMPPNP has been shown to potentiate ADP binding by bacterial MutS, but high concentrations of the triphosphate analogue can displace ADP from the protein (41), indicating that, as in the case of yeast MutS␣, bacterial MutS may also be able to bind two molecules of ATP with different affinities. By contrast, we have been unable to detect binding of more than one molecule of ATP␥S to human MutS␣ using a filter assay similar to that employed by Antony and Hingorami with the yeast protein (52). However, as noted under "Results," it is possible that weak binding of a second ATP␥S molecule to the human MutS␣⅐ATP␥S complex might go undetected in our experiments due to potential perturbation of the interaction during collection of the complexes on filters.
As observed previously with bacterial MutS and yeast MutS␣ (40,52), we have shown that ATP hydrolysis by the human protein is accompanied by a pre-steady-state burst of ADP formation, implying that the rate-determining step for turnover occurs after the chemical step. In contrast to results obtained with the yeast protein, where an ADP burst of 1 per MSH2⅐MSH6 heterodimer has been reported (52), the burst described here is sub-stoichiometric, 0.5 mol of ADP per mol of heterodimer. Because the human MutS␣ preparations we have used bind ADP and nonhydrolyzable ATP analogues with a stoichiometry near unity, we think it unlikely that this effect is due to the presence of a significant fraction of inactive protein. However, these results may indicate that the functional form of human MutS␣ is a higher oligomer, as has been suggested for bacterial MutS (53).
MutS homologues are members of the ABC superfamily. Nonequivalence of the nucleotide binding domains of P-glycoprotein (P-gp) and multidrug resistance protein 1 (MRP1), which are also members of this family, has been documented. However, the apparent asymmetry of the nucleotide binding centers within these two classes of drug transporters is thought to reflect distinct modes of action. The two nucleotide binding sites of P-gp are believed to be functionally equivalent, alternately hydrolyzing ATP during the catalytic cycle, and inactivation of one binding site abolishes ATPase activity of the transporter (54,55). By contrast, the nucleotide binding domains of MRP1 do not appear to contribute equally to activity of the protein: inactivation of nucleotide binding domain 2 abolishes drug transport, but inactivation of nucleotide binding domain 1 results in only partial loss of activity (56,57). It is noteworthy in this regard that, like MutS/MutS␣, MRP1 has been shown to bind di-and triphosphate simultaneously (56,57), and, as observed with MRP1, inactivation of the nucleotide binding site within the MSH6 subunit of human MutS␣ results in a more severe ATPase defect than does inactivation of the nucleotide binding center within MSH2 (17). These observations suggest that the functional roles of ATP binding and TABLE I Affinities of MutS␣ for a 31-bp G-T heteroduplex as a function of nucleotide occupancy Binding constants were determined by surface plasmon resonance in the presence of the nucleotides indicated as described under "Experimental Procedures" and the legend to Fig. 7. hydrolysis by MutS␣ may be more similar to those of the MRP1 class of ABC transporters than those exemplified by P-gp.
There is extensive evidence indicating interaction of the ATP hydrolytic centers and the heteroduplex binding site the of MutS homologues. Thus, heteroduplex, but not homoduplex, DNA alters the rate-limiting step for ATP hydrolytic turnover with E. coli MutS and yeast MutS␣ (40,52). Although ADP does not significantly alter the affinity of human MutS␣ for a mispair, the nucleotide enhances the kinetics of heteroduplex association and dissociation (21,25). Moreover, the affinity of MutS homologues for a mismatch is reduced in the presence of ATP or nonhydrolyzable ATP analogues (7,8,(17)(18)(19)(20)(21)(22)(23)25). Although analysis of the effects of ADP and nonhydrolyzable ATP analogues have been useful in defining the modes of interaction of di-and triphosphate forms of MutS/MutS␣ with heteroduplex, the dimer level asymmetry characteristic of the MutS family implies that the two classes of nucleotide binding site within these proteins can be potentially filled in nine different ways (48). We have demonstrated that, in solution, human MutS␣ readily forms a ternary complex with ADP and a nonhydrolyzable ATP analogue, with affinities for each nucleotide being in the low micromolar range. We have also shown that, in the presence of equimolar concentrations of ADP and a nonhydrolyzable ATP analogue, MutS␣ interacts with heteroduplex in a manner distinct from that expected for a mixed population if di-and triphosphate forms of the protein (Fig. 7), an effect we attribute to the high population of ADP⅐MutS␣⅐AMPPNP and ADP⅐MutS␣⅐ATP␥S species in solution. Interestingly, the heteroduplex affinities observed for these ternary complexes are similar to those obtained with MutS␣ under conditions of ongoing ATP hydrolysis (Table I), conditions where the ADP⅐MutS␣⅐ATP complex comprises a significant fraction of the protein in solution (Fig. 6). These findings indicate that the ADP⅐MutS␣⅐ATP complex has a finite lifetime, and hence that this species should be considered as a significant intermediate during ATP hydrolytic turnover by the free protein and, perhaps, during the course of its interaction with DNA.