Conformational Transitions as Determinants of Specificity for the DNA Methyltransferase EcoRI*

Changes in DNA bending and base flipping in a previously characterized specificity-enhanced M.EcoRI DNA adenine methyltransferase mutant suggest a close relationship between precatalytic conformational transitions and specificity (Allan, B. W., Garcia, R., Maegley, K., Mort, J., Wong, D., Lindstrom, W., Beechem, J. M., and Reich, N. O. (1999) J. Biol. Chem. 274, 19269–19275). The direct measurement of the kinetic rate constants for DNA bending, intercalation, and base flipping with cognate and noncognate substrates (GAATTT, GGATTC) of wild type M.EcoRI using fluorescence resonance energy transfer and 2-aminopurine fluorescence studies reveals that DNA bending precedes both intercalation and base flipping, and base flipping precedes intercalation. Destabilization of these intermediates provides a molecular basis for understanding how conformational transitions contribute to specificity. The 3500- and 23,000-fold decreases in sequence specificity for noncognate sites GAATTT and GGATTC are accounted for largely by an ∼2500-fold increase in the reverse rate constants for intercalation and base flipping, respectively. Thus, a predominant contribution to specificity is a partitioning of enzyme intermediates away from the Michaelis complex prior to catalysis. Our results provide a basis for understanding enzyme specificity and, in particular, sequence-specific DNA modification. Because many DNA methyltransferases and DNA repair enzymes induce similar DNA distortions, these results are likely to be broadly relevant.

DNA-modifying enzymes have evolved a delicate balance between sequence specificity and efficient catalysis (1,2). Indeed, the lack of our understanding of the underlying causes for specificity may account in part for the general lack of success in re-engineering such enzymes, despite relative successes with other classes of enzymes (3)(4)(5). Many DNA-modifying enzymes have specificity constants that approach the predicted values for diffusion-controlled reactions (k cat /K m ϳ 10 8 -10 9 M Ϫ1 s Ϫ1 ) (6,7). DNA-modifying enzymes have attained such efficiency in part because of facilitated diffusion mechanisms (8). Remarkably, such highly efficient enzymes also show significant discrimination between cognate and noncognate sequences.
These enzymes frequently undergo significant conformational changes upon binding their cognate sequence, a clear example of an induced fit mechanism. The resultant intermediates provide a theoretical basis for modulating specificity through the concept of kinetic proofreading, as proposed for several DNA polymerases (9). These checkpoints can directly impact specificity by modifying the partitioning of reaction intermediates prior to catalysis differentially between cognate and noncognate substrates (9 -11).
S-Adenosylmethionine-dependent bacterial DNA methyltransferases modify bases at cytosine (C-5 and N-4) and adenine (N-6) and play important roles in many cellular pathways, including mismatch repair, restriction modification, and gene regulation (12). A detailed structural and mechanistic characterization of DNA methyltransferases is also motivated by recent demonstrations that some bacterial and mammalian enzymes are viable candidates for antibiotic and cancer drug development, respectively (13)(14)(15)(16). The bacterial DNA adenine methyltransferase EcoRI forms part of a restriction/modification system and modifies the N-6 position of the second adenine in the sequence GAATTC. The schematic and kinetic mechanism of M.EcoRI presented in Scheme 1 represent the chemical and conformational transitions observed during the methylation of cognate DNA. M.EcoRI uses facilitated diffusion to locate and methylate its cognate DNA to protect the bacterial genome from the complement restriction endonuclease (8,17). Upon location of the target sequence, M.EcoRI induces a 50°bend in the DNA sequence and flips the target adenine out 180°to an extrahelical position to facilitate catalysis. Solution-based FRET 2 studies were used to demonstrate that along with bending the DNA, M.EcoRI also expands the duplex DNA structure presumably through an intercalative mechanism (18), and intercalation and bending occur nearly simultaneously with DNA binding at 25°C (18).
The specificity constant of M.EcoRI (k cat /K m , ϳ10 7 -10 9 M Ϫ1 s Ϫ1 ) for cognate DNA is near the rate of diffusion (8,17,19,20). Specificity constants for M.EcoRI toward several noncognate sites are decreased 5-fold (A6, GAATCC), 3500-fold (A4, GAATTT), and 23,000-fold (A3, GGATTC), which involve relatively minor changes in DNA affinity (19). The molecular basis of this discrimination remains unclear, as is the sequence discrimination for any DNA methyltransferase. We recently described an intriguing bending-impaired M.EcoRI mutant (H235N) whose ability to flip its target adenine is decreased 2000-fold compared with the WT enzyme, whereas the cognate modification was left largely intact (21). Moreover, the mutant is at least 1000-fold more discriminating than the WT enzyme (21). The molecular basis of how changes in precatalytic conformational transitions can result in such a profound increase in specificity is the focus of this study. We find that much of the discrimination of the WT enzyme against noncognate DNA occurs from the selective destabilization of precatalytic intermediates. Our results are presented in the context of kinetic proofreading concepts, which provide a quantitative framework for understanding sequence-specific DNA modification.

MATERIALS AND METHODS
Enzyme Expression and Purification-M.EcoRI was purified from MM294 Escherichia coli cells with the plasmid pXRI as described previously (18). 6 liters of culture were induced with 1 mM isopropyl 1-thio-␤-D-galactopyranoside at an A 600 0.4 for 3 h. Cells were sonicated in extraction buffer (200 mM NaCl, 6.5 mM K 2 HPO 4 , 3.5 mM KH 2 PO 4 , 1 mM EDTA, and protease inhibitor mixture (Sigma)). The lysate was centrifuged (30 min at 4°C); the supernatant was loaded onto a 20-ml phosphocellulose (Whatman) column, and the protein was eluted with a salt gradient between 200 and 800 mM NaCl. Fractions containing M.EcoRI were loaded onto a 10-ml hydroxyapatite (Bio-Rad) column equilibrated with extraction buffer. The protein was eluted with a gradient between 6.5 mM and 1 M KH 2 PO 4 and dialyzed overnight in extraction buffer containing 10% glycerol. The dialyzed protein was loaded onto a Bio-Rex (Bio-Rad) column, and the protein was eluted with a gradient of 200 and 800 mM NaCl. Protein from this procedure was typically greater than 95% pure by densitometry. Enzymes were stored at Ϫ80°C with 10% (v/v) glycerol, 0.3-0.4 m NaCl, 10 mM potassium phosphate, pH 7.8, 1 mM EDTA, and 7.5 mM ␤-mercaptoethanol.
Equilibrium Anisotropy-Anisotropy experiments were carried out on a Fluoromax-2 fluorimeter (ISA SPEX) equipped with an L-format autopolarizer. Equilibrium measurements were performed by monitoring the change in anisotropy of 20 nM singly labeled (Alexa-488) duplex DNA as increasing enzyme was titrated into the reaction mix. An excitation of 488 nm and emission of 515 nm with slit widths of 8 nm were used for data collection. Anisotropy was calculated using the following equation; R ϭ I ʈ Ϫ I Ќ /I ʈ Ϫ 2I Ќ (22). Dissociation constants were determined by fitting data to the following quadratic formula: Equilibrium FRET-Equilibrium fluorescence experiments were performed as described previously (18). All measurements were collected with a PerkinElmer Life Sciences LS50B luminescence spectrometer at 22°C. Excitation at 485 nm was used to excite Alexa-488, and an emission spectrum from 500 to 650 nm was collected. Enzyme at a final concentration of 200 nM was mixed with 20 nM DNA and 1 M sinefungin in reaction buffer (50 mM Tris, pH 8.0, 5 mM EDTA, 100 mM NaCl, and 1 mM dithiothreitol). The S-adenosylmethionine analogue sinefungin (Sigma) was used for all equilibrium and transient experiments. SCHEME 1. Kinetic mechanism of M.EcoRI. E⅐DNA* indicates a bent enzyme DNA complex. E⅐DNA** indicates a base-flipped complex. E⅐DNA*** indicates an intercalated complex. DNA-CH3 indicates that the N-6 of the adenine is methylated. FRET was used to measure bending and intercalation. Flipping was measured by 2-aminopurine fluorescence. The chemistry step is measurable by single turnover experiments. Brackets indicate that flipping and intercalation may be occurring simultaneously due to the biphasic signal for 2AP flipping. Schematics are provided to facilitate conceptualization of the various intermediates in the reaction mechanism.
Stopped-flow FRET-Stopped-flow measurements were collected on an Applied Photophysics SX.18MV stopped-flow reaction analyzer outfitted with a single channel emission photomultiplier tube. An excitation wavelength of 485 nm and an emission cut-off filter of 515 nm were used for data collection. Coupled DNA was rapidly mixed with M.EcoRI preincubated with sinefungin in reaction buffer. The change of donor fluorophore emission was used to detect bending and intercalation due to the greater signal change for the donor fluorophore during FRET with the acceptor fluorophore (18). DNA concentrations were 20 nM for cognate DNA and 100 nM for noncognate DNA substrates. Enzyme and cofactor concentrations were varied from 100 to 2000 nM. Error measurements are from at least two experiments with the DNA and enzyme mixes freshly prepared each time.
Stopped-flow Flipping-Transient base flipping data for cognate and noncognate substrates were measured in a stoppedflow apparatus by monitoring the increase in fluorescence of 2AP as it transitions from base-stacked to extrahelical in the presence of enzyme. 2AP fluorescence is roughly 14-fold greater when extrahelical, and has been validated previously as a probe for the measurement of base flipping kinetics (23)(24)(25)(26). An excitation wavelength of 310 nm and an emission cut-off filter of 320 nm were used for data collection. 100 nM 2AP containing DNA was rapidly mixed with M.EcoRI preincubated with sinefungin in reaction buffer, monitoring the emission above 320 nm. Enzyme and cofactor concentrations varied between 500 and 2000 nM. Error measurements are from at least two experiments with the DNA and enzyme mixes freshly prepared each time.
Reverse Kinetics-Stopped-flow measurements were collected with excitation wavelengths and cut-off filters as described for the FRET and 2AP studies. The rate of extrahelical base restacking was measured by monitoring the dissociation of the preformed 2AP-containing DNA-enzyme complex. The loss of fluorescence is attributed to the restacking of the target base. A final concentration of 100 nM 2AP DNA, 100 nM enzyme, and 6 M sinefungin were mixed with 5 M unlabeled competitor DNA. The unbending/unintercalation data were collected by monitoring the change in FRET as a 25-50-fold excess unlabeled cognate DNA was rapidly mixed into the system. All forward and reverse stop flow data were fit to either a single or double exponential equation using the program Sigma Plot.
Data Analysis and Simulation-Kinetic simulations to model bending and intercalation data were performed using Scientist (Micromath Inc., Salt Lake City, UT) and KinTekSim (KinTek Corp., Austin, TX). Global fitting analysis of 4°C cognate bending and intercalation data were performed using a two intermediate reaction mechanism. Goodness-of-fit statistics were used to evaluate the fit of the data to the reaction mechanism, yielding an R 2 value of 0.99. KinTekSim modeling was done for binding, bending, and intercalation as shown in Scheme 1. Second-and first-order rate constants for the cognate DNA simulation were as follows: k on ϭ 3.0 ϫ 10 7 M Ϫ1 s Ϫ1 , k bend ϭ 100 s Ϫ1 , k intercalate ϭ 10 s Ϫ1 , and k unintercalate ϭ 0.02 s Ϫ1 . Simulations for noncognate substrates were performed by increasing the unintercalation rate constant to 2000-and 4000-fold.

Transient DNA Bending Is Observable with Cognate and
Noncognate Substrates-FRET between Alexa-488 and TAMRA coupled to the 5Ј ends of 14-mer oligonucleotides was previously used to examine the bending transition of the target DNA by M.EcoRI (18). Because cognate DNA bound by M.EcoRI is bent ϳ50°as determined by both atomic force microscopy and gel shifting methods (21,27), an increase in energy transfer was anticipated. The opposite changes were observed, consistent with a decreased energy transfer upon protein binding (18) (Fig. 1A). Further analysis revealed the decreased FRET was not dye-dependent, and the anticipated increase in FRET does occur using an enzyme that recognizes the same DNA substrate but only induces DNA bending (18,28). Because of the robustness of the decreased FRET M.EcoRI-DNA data, an intercalative mechanism was proposed (18). Recent protein/DNA cocrystal structures of T4DAM and E. coli DAM reveal several residues intercalating into the DNA (29,30), supporting the concept of intercalation for methyltransferases.
Using a FRET-based assay (18), we examined equilibrium and pre-equilibrium kinetics of binding, bending, and intercalation for the cognate sequence with the target adenine in the complement strand methylated (cognate bottom methylated) (GAATTC) and 3 noncognate sequences as follows: A6 (GAATCC), A4 (GAATTT), and A3 (GGATTC). The equilibrium FRET for cognate DNA previously demonstrated a decrease in FRET between the donor and acceptor fluorophores on the DNA. Noncognate A6 also has a decrease in FRET upon M.EcoRI binding similar to the cognate signal, whereas noncognate substrates A3 and A4 have no observable change in FRET (Fig. 1, A-C). To ensure that all analysis on DNA bending, base flipping, and intercalation are of a saturated enzyme substrate complex, the affinity of the enzyme for each substrate was determined. Enzyme K D DNA values were determined by equilibrium anisotropy (data not shown) and are within error of the previously determined K D DNA values obtained by gel mobility shift assays (19). Enzyme concentrations used in each experiment were more than 3-fold higher than the DNA dissociation constant (K D DNA ) to ensure that the enzyme was bound to the substrate.
The lack of detectable equilibrium FRET changes with both A4 and A3 may be due to compensatory signal decreases and increases resulting from changes in bending and intercalation, respectively. This was further investigated by measuring the pre-equilibrium kinetics of the FRET-based bending/intercalation experiments on a stopped-flow apparatus. Cognate and noncognate (A6, A4, and A3) substrates with dyes on both 5Ј ends, at a concentration of 100 nM, were rapidly mixed with 500  nM M.EcoRI-sinefungin while monitoring the donor signal. We previously demonstrated that M.EcoRI binding to cognate substrate at 25°C involves a protein concentration-dependent increase in the donor signal consistent with a decrease in FRET ( Fig. 2A) (18). Most surprisingly, substrates A4 and A3 show a very rapid decrease in donor signal (130 and 240 s Ϫ1 , respectively), followed by a slower increase (37 and 22 s Ϫ1 , respectively) (Fig. 2, C and D). Noncognate A6 has a very rapid decrease in donor signal (rate constants are unattainable), followed by an increase in donor signal with a k obs rate constant of 27 s Ϫ1 (Fig. 2B).
Based on classical FRET theory (18,22), we suggest that the initial decrease in the donor signal (increased FRET) is caused by the bending transition (18,28), whereas the increase in donor signal (decreased FRET) (18) is caused by intercalation. The assignment of DNA bending to the decrease in the donor signal is further validated in the following section when the decrease in the donor signal is observed for the cognate DNA with the temperature lowered to 4°C (Fig. 3A). Further support for the assignment of the bending transition to the initial decrease in donor fluorescence is provided by the observed concentration dependence of this phase in the fluorescence data (Fig. 3A). Finally, the cocrystal structure of the endonuclease R.EcoRV bound to the cognate DNA sequence for M.EcoRI GAATTC reveals a significant degree of DNA bending; R.EcoRV binding to this same DNA shows the anticipated decreased donor signal (increased FRET) (28). Taken together, these observations support our assignment that the initial increase in FRET for M.EcoRI is due to DNA bending.
The k obs rate constant for the increase in donor signal with the rapid mixing of 20 nM cognate DNA with 500 nM M.EcoRI is 10 s Ϫ1 (18). Under similar conditions the exponential fits for the donor signal increases of A4 and A3 (37 and 22 s Ϫ1 , respectively) are faster than the cognate increase in donor signal (Fig.  2, A, C and D). The faster rate constants for A4 and A3 arise from a faster approach to equilibrium for these noncognates versus cognates.
DNA Bending Precedes Intercalation with the Cognate Site-Our previous results showed that cognate DNA binding, bending, and intercalation are nearly simultaneous (18). Furthermore, based on the biphasic FRET signal observed with the noncognate substrates, tentatively assigned to rapid bending followed by slower intercalation (Fig. 2, B-D), we sought to deconvolute these events with cognate DNA and M.EcoRI at 4°C. Upon rapidly mixing M.EcoRI with cognate DNA, an initial decrease in the donor signal followed by a slower increase was observed. The small, but reproducible, decrease in signal occurs in the initial 25 ms of data collection (Fig. 3A). As stated above, the increase in FRET is most reasonably assigned to the anticipated changes resulting from a bending transition (28,31,32). This is followed by a decrease in FRET that has been assigned to the intercalation of the DNA by the enzyme (18). Measurement of the rate constant for intercalation with varying M.EcoRI concentrations (100, 300, 500, 700, and 1400 nM) and constant cognate DNA (20 nM) reveals that at high concentrations the rate of intercalation (donor signal increase) does not increase linearly with enzyme concentration (Fig. 3B). The mild concentration dependence of the intercalation data suggests that intercalation is only partially second order (Fig. 3B) (33). Furthermore, because the bending transition remains concentration-dependent, this step continues to be simultaneous with DNA binding.
The intercalation signal for cognate DNA is the only observable signal at 22°C (Fig. 2A), and it was previously shown that the rate of intercalation increases linearly with increasing enzyme concentration at 22°C (18). The concentration dependence for intercalation of cognate DNA at 22°C suggests that both bending and intercalation are nearly simultaneous with binding (18). The new signal at 4°C (Fig. 3A), which we assign to bending for cognate DNA, demonstrates that bending precedes intercalation. We therefore focused subsequent bending and intercalation studies of noncognate substrates at 4°C (see below). The decrease in the signal amplitudes for the intercalation phase with the noncognates A4 and A3 in comparison to cognate DNA (Fig. 2, C and D versus A) suggests a significantly altered protein/DNA interaction.
The Intercalation Rate Constant for Noncognate DNA Is First Order-The transient rates of bending and intercalation for noncognate DNAs A4 and A3 were examined at 4°C by the rapid mixing of 100 nM DNA with 500, 700, 900, and 1100 nM M.EcoRI (Fig. 4, A and B, A4 and A3). There is very little change in the observed rates of bending and almost no change in the observed intercalation rates for both substrates with increasing enzyme concentration (Fig. 4, inset), which suggests that both bending and intercalation are no longer second order steps with A4 and A3 substrates. The noncognate rates observed for each curve were calculated by plotting the two signal directions as separate single exponentials (33,34). It is important to note that the level of amplitude change in the two phases is different for noncognate compared with the cognate DNA (Fig. 3A versus Fig. 4, A and B), i.e. the intercalation phase rises much higher than the bending phase for the cognate DNA than the noncognate DNA. Several scenarios could account for this. 1) Most of the bending transition for cognate DNA occurs within the dead time of the stopped-flow measurement, giving the appearance of a greater intercalation amplitude. 2) The number of residues intercalating the noncognate substrate is less than the cognate substrate, decreasing the expansion of the DNA during intercalation and ultimately decreasing the amplitude of the signal change during intercalation. 3) The reverse rate constant for the intercalation step increases for the noncognate substrate, reducing the observable intercalated intermediate population.
Base Flipping Precedes Intercalation-The use of 2AP as a base flipping probe has been well documented for both methyltransferases and repair enzymes (23)(24)(25)(26)35). The base flipping rate constants for cognate and noncognate substrates were measured under pre-equilibrium conditions. The equilibrium signal of 2AP-containing A4 and A3 DNA bound by M.EcoRI is less than 35% of the cognate 2AP signal (data not shown). Based on the observed 2AP equilibrium data, pre-equilibrium experiments were designed to reveal tractable changes during stopped-flow experiments to measure the base flipping and restacking rate constants. The observed rate of base flipping was measured at different concentrations and time frames. An initial increase in 2AP fluorescence is observed in the first 100 ms of the flipping experiment (Fig. 5A). There is an observable concentration dependence of the first phase of flipping, followed by a concentration-independent phase (Fig. 5A). The biphasic signal for 2AP flipping was previously assigned to flipping, followed by an isomerization event for the target base (26,35).
To assign a temporal order for bending and base flipping, we examined the enzyme-induced cognate base flipping at 4°C by monitoring 2AP base flipping, and we compared this to the FRET signal for cognate bending (only observed at 4°C). 2APcontaining DNA was rapidly mixed with M.EcoRI preincubated with sinefungin (Fig. 5B). A short ϳ4-ms lag is observed before the increase of 2AP fluorescence (Fig. 5B), which is consistent with previous observations (26). This lag persists even with increasing concentrations of 500, 1000, 1500, and 2000 nM M.EcoRI. Closer inspection of the cognate FRET data at 4°C (Fig. 3A) reveals that no lag is observed, and the bending event is near completion within this time frame. Furthermore, under similar conditions to the FRET experiments, the transient flipping experiment takes more than 200 ms to reach equilibrium. These data clearly show that the DNA bending event precedes base flipping, as suggested previously (21,36). Also, the observed rate of intercalation from the FRET signal plateaus with increasing enzyme concentration (Fig. 3B). This plateau suggests that the intercalation step is a first order process. Furthermore, because the observed rate of base flipping does not plateau under similar conditions (Fig. 5C), base flipping must precede intercalation. The concentration profile (Fig. 5C) demonstrates that the event that we and others have assigned to base flipping (23,25,35) is concentration-dependent with a k on ϭ 3.6 ϫ 10 7 M Ϫ1 s Ϫ1 , which is similar to the previously reported k on ϭ 2.1 ϫ 10 7 M Ϫ1 s Ϫ1 as determined by anisotropy (18,36) and FRET (18). The intercept that approximates k off ϭ 40 s Ϫ1 is very close to the restacking transition of Ͼ50 s Ϫ1 previously calculated from global fitting (36). Because we observe concentration-dependent behavior for both bending and base flipping, we suggest that the k obs rate of base flipping would eventually plateau if a high enough enzyme concentration were achievable.
Unbending and Restacking Rate Constants Increase for Noncognate DNA-The reverse rate constants for unbending and restacking were examined by trapping experiments. A preformed complex of DNA-M.EcoRI-sinefungin was rapidly mixed with 25-50-fold excess unlabeled DNA at 22°C. The change in fluorescence for either the doubly labeled DNA or the 2AP-containing DNA was attributed to unbending and restacking, respectively. The rate constants of unbending for noncognates A6, A4, and A3 were all extremely fast compared with cognate (Table 1; Fig. 6, A-C). The A6 noncognate causes a decrease in the donor signal (increased FRET), opposite the signal observed for the forward rate (Fig. 2B), as would be predicted. The change in the donor signal for A4 and A3 noncognates increases, suggesting that A4 and A3 are only unbending (Fig. 2, C and D).
Restacking rate constants were obtained through similar trapping experiments used to monitor unintercalation and unbending, except that the DNA substrates used for the preformed complex contained 2AP in substitution for the target adenine and did not have fluorophores coupled to the 5Ј ends. The equilibrium 2AP signals upon binding by M.EcoRI with all three noncognate substrates were uniformly small, and A3 could not be reproducibly obtained; hence, restacking rate constants were limited to the cognate and noncognates A6 and A4 (Fig. 7, A-C).

DISCUSSION
The hallmarks of enzyme function, specificity, rate enhancements,  and regulation demand a molecular understanding to provide a basis for their rational redesign (37)(38)(39). The redesign of enzyme specificity (40 -44) reflects a deep understanding of enzyme function, leading to the potential practical application of such powerful catalysts. DNA methyltransferases provide an excellent platform for the study of mechanisms leading to enzyme specificity. They are challenged with the daunting task of modifying a single base within a particular recognition element, which itself is embedded in a large amount of competing substrate. They do so by exploiting large scale conformational changes both within the enzymes and the target DNA, and the bacterial and mammalian enzymes form part of the critical epigenetic pathways essential to bacterial pathogenesis, human development, and tumorigenesis. The redesign of sequencespecific DNA methyltransferases to modify unique positions within the genome may enable in vivo site-specific DNA modification and gene regulation for basic research and biomedical applications (45,46). M.EcoRI-induced changes in DNA conformation, including bending, flipping, and intercalation, provide a compelling basis for regulating sequence discrimination (10,18,47). Early evidence for this was provided by a bending-impaired H235N M.EcoRI whose decreased apparent base flipping constant also resulted in a dramatic increase in the substrate discrimination of the mutant (21). We previously hypothesized that these transitions may play important roles in determining discrimination of the WT enzyme against noncognate methylation (21).
Observable Bending/Decrease in Intercalation Amplitude-Examination of reaction intermediates and conformational changes prior to methylation by multiple fluorescence techniques presented here have allowed us to better understand the role of conformational changes in the reaction mechanism of M.EcoRI. Furthermore, these studies address fundamental questions pertaining to DNA-modifying enzymes, specificity, and transitions between intermediates on a reaction pathway. Our studies further probe the widely held view that structurally accessible intermediates define the specificity of an enzyme (2,34).
FRET-based experiments with noncognate substrates were performed to examine the rates and amplitude changes of DNA bending and intercalation. In accordance with the limited change in the specificity of M.EcoRI toward the noncognate A6, we see limited changes in the equilibrium (Fig. 1B) and preequilibrium (Fig. 2B) bending and intercalation, compared with the cognate site. The lack of any change in the A4 and A3 equilibrium FRET (Fig. 1C) is further supported by the kinetic preequilibrium FRET data (Fig. 2, C and D), which shows the intercalation signal reaching an equilibrium equivalent to the start of the bending phase, suggesting that the population of intermediates between the bent and intercalated states of the enzyme with A4 and A3 differ from the populated intermediates of bent and intercalated A6 and cognate with enzyme.
Temporal Order Is Bending First, Followed by Flipping and Intercalation-The use of 2-aminopurine as a probe for base flipping kinetics of methyltransferases has been well documented (23,25,26,35). There are minimal changes in M.EcoRI substrate affinity when the target base is replaced with 2-amin- opurine, further validating its use as a relevant probe for studying base flipping (26). Also, the appendage of FRET fluorophores to the substrate DNA has not affected any observable catalytic parameters of M.EcoRI (18). The lack of any gross perturbation to the enzyme-substrate complex by the fluorophores further allows us to compare data collected using the two fluorescence techniques.
To provide temporal order to the steps of bending, flipping, and intercalation, we compared our FRETbased experiments at 4°C (Fig. 3A) with our base flipping experiments at 4°C (Fig. 5B). The observation of a lag in the flipping signal in the initial 4 ms is consistent throughout the concentration profile (Fig. 5B) and has been reported previously (36). The bending signal (Fig. 3A) is near completion within 10 ms, and no lag is observed, indicating that bending precedes base flipping. The curvature in the concentration profile of the intercalation phase at 4°C observed by FRET (Fig.  3B) and the lack of any curvature in the concentration profile for base flipping at 4°C (Fig. 5C) demonstrates that intercalation follows base flipping (33). However, base flipping rates are anticipated to eventually plateau with increasing enzyme concentrations because base flipping also follows bending. The assignment of the temporal order for the intermediates by collecting rate constants for observable DNA bending, base flipping, and intercalation at temperatures of 22 and 4°C also reveals that a significant portion of the energy barriers for these enzyme-induced conformational changes is entropic.
Reverse Rate Constants Dictate Specificity-The kinetic rate constants for the reversal of the bending, intercalation, and flipping transitions were determined by trapping experiments. The total change in amplitude for the FRET signal observed for A4 and A3 (unbending/unintercalation) (Fig.  6, C and D) is the same as the amplitude for the FRET signal observed with the forward step of bending (Fig. 2, C and D). Furthermore, the changes occur in the opposite direction (Fig. 6, C and D) of those observed in forward experiments (Fig. 2, C and D). The partitioning of the enzyme/DNA intermediates back toward the unbent intermediate and which involves no intercalation is favored with these noncognate substrates. Fitting the trapping dotted line indicates modeling with a 4000-fold increase in the unintercalation rate constant. The 4000-fold increase (dotted line) was added to show that an even greater increase in unintercalation will increase the approach to equilibrium and reduce the total amplitude change for the intercalation signal. Note that the concentration of DNA for this simulation has increased to 100 nM to normalize the amplitude changes between cognate and noncognate substrates. D, an illustrative energy diagram of reaction intermediates for cognate and noncognate substrates. Solid line indicates cognate DNA; dashed line indicates noncognate DNA; TS chem indicates transition state to the methylation step. Energy wells for base-flipped and intercalated DNA are raised for the noncognate DNA substrates relative to the cognate DNA substrate because the reverse rate constants change significantly with noncognate DNA (Figs. 6 and 7), with minor changes in the forward rate constants (Fig. 2). Only differences between the cognate and noncognate substrates at the base-flipped and intercalated intermediates are revealed in the energy diagram as these are the only intermediates definitively addressed by the data from this work. No other intermediates are changed as we have no evidence to suggest otherwise. data to single exponentials reveals that the reverse rate constants for A4 and A3 are 2500-and 1900-fold faster than cognate, respectively (Table 2). We reconfirmed the reverse rate kinetics by performing the same trapping experiment but using the 2AP-containing substrates. Indeed, the A4 noncognate substrate produces a restacking rate constant 2500-fold greater than the cognate substrate ( Fig. 7; Table 2). Interestingly, both restacking (Table 2) and unbending (Table 1) transitions for A4 show similar noncognate/cognate enhancements. Also, the increase in the reverse rate constant for A4 versus cognate is very close to the reported 3500-fold difference in specificity toward the cognate substrate, which suggests that enzyme specificity may arise from the destabilized enzyme-noncognate complex and thus a more rapid partition away from the catalytic complex. The inability to collect A3 base flipping and restacking data may be caused by additional instability of the base-flipped-enzyme complex, thereby showing the greatest difference in specificity from the cognate substrate of 23,000-fold. Reverse kinetic steps have been shown to control enzyme specificity (28,39,48), and the specificity can be dominated by a single step in the overall mechanism (48,49 (Fig. 8C).
Simulation and Conclusion-To obtain rate constants for the transitions between the intermediates of bending and intercalation, the bending and intercalation data for the cognate substrate obtained at 4°C (Fig. 3) was fit globally to a two intermediate reaction mechanism using the program Scientist. The goodness-of-fit statistics obtained when solving for four different rate constants (Fig. 8A) yields an R 2 value of 0.99, further supporting our assignment of the two FRET phases to bending followed by intercalation. The obtained rate constants were used to simulate the bending and intercalation phases with 20 nM DNA and 300, 500, and 700 nM enzyme (Fig. 8A). The calculated rate constants yield simulated FRET data that mimic the observed FRET data in shape and amplitude (Fig. 8A versus  Fig. 3A).
Global fitting analysis was also applied to the base flipping data obtained at 4°C (Fig. 5B). Use of a two intermediate reaction mechanism allowed for the incorporation of an intermediate prior to the flipping step, which produces a lag in both the raw flipping data (Fig. 5B) and the simulated flipping data (Fig.  8B). Unfortunately, the lag in Fig. 8B is poorly defined due to large errors for the rate constant of the reverse step for the first intermediate. The calculated rate constants for flipping (Fig.  8B, k 3 ϭ 550 s Ϫ1 ) versus the calculated rate constant for intercalation (Fig. 8A, k 3 ϭ 97 s Ϫ1 ) further supports our assignment of base flipping preceding intercalation.
Prior work, based on modeling experiments with the computer program KinTekSim, assigned a lower limit of 10 s Ϫ1 for the rate of bending (18). Expansion of this KinTekSim modeling exercise in combination with using the obtained rates from the global fit as lower limits (lower limits because these data are collected at 4°C) allows us to further probe the influence of reverse rates on the population of the intermediates of bent and intercalated DNA. Fig. 8C shows the simulated FRET data for the cognate substrate (solid line). When the rate constants for unintercalation are increased 2000-fold (Fig. 8C, dashed line; as seen for noncognate substrates A3 and A4) and 4000-fold (dotted line; doubling of observed noncognate rate constants for perspective), the k obs for the approach to equilibrium increases, and the total amplitude change for the intercalation phase decreases (Fig. 8C). We suggest that the noncognate substrates destabilize the intercalated intermediate, thereby populating the bent intermediate to a greater extent. The change in the population distribution for the noncognate versus cognate substrates tracks very closely with the changes in M.EcoRI sequence discrimination. Thus, M.EcoRI specificity is predominantly dictated by the reverse rate constants of the intercalation step ( Table 1, Fig. 6, and Fig. 8D). However, because the forward bending and intercalation k obs rates with A4 and A3 do not increase linearly with increasing enzyme concentration, the forward transitions may have also been impacted with the noncognate substrates, thereby contributing to the discrimination of the enzyme. The observed temporal order of bending, flipping, and then intercalation, along with increased unintercalation for the noncognate substrates suggests that by intercalating residues the enzyme stabilizes a base-flipped complex. In support of this idea we observe increased restacking rate constants for noncognate substrates that closely match the unintercalating rate constants. The caveat to this argument is that flipping and restacking data are obtained with a base analogue, which may skew the rate constants for flipping, yielding the mechanistic step order of flipping preceding intercalation.
Our proposal that DNA sequence-dependent modulations of the enzyme's partitioning away from the intermediate prior to methylation (Fig. 8D) is consistent with M.EcoRI using a facilitated diffusion mechanism to locate its specific site (8,17). Noncognate sites do not cause the enzyme to dissociate from the DNA, rather stabilizing enzyme/DNA intermediates that occur prior to the catalytically relevant complex, involving a bent, flipped, and intercalated conformation. This form of kinetic proofreading allows for the efficient selection of the cognate site over the noncognate sites and involves conformational checkpoints that contribute to the specificity of the enzyme prior to catalysis.