Marked Stepwise Differences within a Common Kinetic Mechanism Characterize TATA-binding Protein Interactions with Two Consensus Promoters*

Binding of the TATA-binding protein (TBP) to promoter DNA bearing the TATA sequence is an obligatory initial step in RNA polymerase II transcription initia-tion. The interactions of Saccharomyces cerevisiae TBP with the E4 (TATATATA) and adenovirus major late (TATAAAAG) promoters have been modeled via global analysis of kinetic and thermodynamic data obtained using fluorescence resonance energy transfer. A linear two-intermediate kinetic mechanism describes the reaction of both of these consensus strong promoters with TBP. Qualitative features common to both interactions include tightly bound TBP-DNA complexes with similar solution geometries, simultaneous DNA binding and bending, and the presence of intermediate TBP-DNA conformers at high mole fraction throughout most of the reaction and at equilibrium. Despite very similar energetic changes overall, the stepwise entropic and enthal-pic compensations along the two pathways differ mark-edly following the initial binding/bending event. Furthermore, TBP-E4 dissociation ensues from both replacement and displacement processes, in contrast to replacement alone for TBP-adenovirus major late promoter. A model is proposed that explicitly correlates these similarities and differences with the sequence-specific structural properties inherent to each promoter. This detailed mechanistic comparison of two strong promoters interacting with TBP provides a foundation for subsequent comparison between consensus and variant promoter sequences reacting with TBP. Assembly of the preinitiation complex, polymerase II recruit-ment, and subsequent gene transcription

Binding of the TATA-binding protein (TBP) to promoter DNA bearing the TATA sequence is an obligatory initial step in RNA polymerase II transcription initiation. The interactions of Saccharomyces cerevisiae TBP with the E4 (TATATATA) and adenovirus major late (TATAAAAG) promoters have been modeled via global analysis of kinetic and thermodynamic data obtained using fluorescence resonance energy transfer. A linear two-intermediate kinetic mechanism describes the reaction of both of these consensus strong promoters with TBP. Qualitative features common to both interactions include tightly bound TBP-DNA complexes with similar solution geometries, simultaneous DNA binding and bending, and the presence of intermediate TBP-DNA conformers at high mole fraction throughout most of the reaction and at equilibrium. Despite very similar energetic changes overall, the stepwise entropic and enthalpic compensations along the two pathways differ markedly following the initial binding/bending event. Furthermore, TBP-E4 dissociation ensues from both replacement and displacement processes, in contrast to replacement alone for TBP-adenovirus major late promoter. A model is proposed that explicitly correlates these similarities and differences with the sequencespecific structural properties inherent to each promoter. This detailed mechanistic comparison of two strong promoters interacting with TBP provides a foundation for subsequent comparison between consensus and variant promoter sequences reacting with TBP.
A strong correlation has been observed between the solution conformation of the TBP-TATA complex and the corresponding functional activity, including overall transcription efficiency and the details of the binary interaction. A very good correspondence is apparent between in vitro and in vivo transcription activity and the DNA solution bend angle in the binary complex (6), consistent with the TBP-TATA structure being a primary determinant of transcription activity. This relationship contrasts with a minimal correspondence of the TBP-DNA complex lifetime to bend angle and to transcription activity (6). The relationships among TATA sequence, DNA solution bend angle, and transcription efficiency are very well described by a two-state model in which the TBP-TATA complex exists in two solution conformations: a transcriptionally inactive form with only slightly bent DNA and a transcriptionally active form with the DNA bent ϳ80° (6). The sequence-dependent probability for the complex to assume the transcriptionally active conformation would then be highly correlated with successful preinitiation complex formation and subsequent transcription.
The interaction between TBP and the consensus adenovirus major late promoter (AdMLP) (TATAAAAG) has been characterized using gel electrophoresis circular permutation analysis (10), DNase I footprinting (11,12), and fluorescence resonance energy transfer (FRET) (13,14). TBP-TATA association reactions proceed at rates significantly slower than diffusion limited (10 -16). FRET studies reveal simultaneous binding and bending of AdMLP by TBP that is well described by a twointermediate process with both intermediate conformers having bent DNA (14). One of these intermediate species is significantly populated and has been proposed as a TBP-DNA complex upon which the preinitiation complex is assembled (14).
The details of the TBP-AdMLP association kinetics, monitored in real time, are highly sensitive to single point mutations in the core TATA sequence (17). Furthermore, dramatic alterations in the solution geometry of a TBP-bound TATA variant (TATAAACG), induced by sequential addition of osmolyte, correspond to equally dramatic, successive changes in the TBP-DNA association binding curve (17). Differences in the mechanistic details of the TBP-promoter interaction must give rise to these kinetic changes and correlate with the sequencedependent binary structures in solution.
Our goal is to further understand the functional consequences of variations in the TATA sequence-dependent solution conformation of the binary complex. A comprehensive kinetic and thermodynamic comparison of two consensus strong promoters reacting with TBP has thus been conducted, providing the foundation for subsequent studies utilizing variant TATA sequences. The interaction of the E4 core promoter sequence (TATATATA) with TBP was chosen for further analysis because previous characterization using DNase I footprinting revealed functional differences relative to the AdMLP sequence (11,12).
Further examination of the TBP-E4 interaction over a range of temperatures and TBP concentrations using FRET in conjunction with steady-state, stopped-flow, and time-resolved measurements yields detailed kinetic and energetic profiles. The interaction of TBP with AdMLP was previously characterized in an analogous manner (14). The complete set of kinetic and thermodynamic data describing that reaction has been further analyzed in this study to provide an unambiguous, direct comparison between the two strong promoters.
The overall interactions of both the E4 and the adenovirus major late promoters with TBP are well described by a linear two-intermediate model (14). This model is the simplest to which the data for each promoter correspond. Important differences in the stepwise progression of these two reactions can be related directly to differences in the sequence-induced structural features of the two promoters, yielding additional insight into the relationships among sequence, structure, and function.
Theory and Determination of TBP-bound E4 End to End Distance Distribution-Detailed discussions of FRET and its application to the present study have been published (6, 13, 14, 18 -24). Very briefly, FRET is the nonradiative transfer of excited state energy from a donor to an acceptor fluorophore, with the efficiency of such transfer exquisitely dependent on the distance between the two dyes. The binding of TBP to TATA-bearing DNA may thus be monitored via changes in the donor emission, because the TBP-induced DNA bending significantly decreases the inter-dye distance. The interaction of E4 with TBP has been monitored herein using FRET in conjunction with steady-state emission, stopped-flow, and time-resolved fluorescence decay measurements.
The solution geometry of TBP-bound E4 was investigated by extracting the probability distributions characterizing the 5Ј end to 3Ј end distances from time-resolved fluorescence decays, as described (6), for the E4 duplex free and TBP-bound. A model-dependent solution bend angle for the bound E4 duplex was then determined as described (6).
Equimolar TBP-DNA Stopped-flow Measurements-Numerous studies have demonstrated that TBP is monomeric under the conditions of our studies (14,(25)(26)(27). However, to test the proposal that TBP dimerization explains the observed TBP-DNA kinetics (15,28), stopped-flow association binding measurements have been conducted using equimolar concentrations of TBP and DNA. A 2 M TBP solution was flowed together with a 2 M AdMLP solution, with T*AdMLP dpx *F comprising 2% of the total DNA (the reliability of using the double-labeled duplex as a trace probe has been demonstrated) (14). Simulations were conducted using the two-intermediate model that has been shown to characterize the TBP-AdMLP interaction together with the previously determined 20°C rate constants (14), with and without an additional step for TBP self-association. The rate constant for TBP dimer 3 2 TBP monomer was assigned the published value of 7.7 ϫ 10 Ϫ4 s Ϫ1 (29). This value and the reported monomer-dimer equilibrium constant, 0.02 M (15), were used to determine the rate constant for the reverse reaction. The experimental and simulated curves were then compared.
The interaction of TBP with AdMLP was characterized previously using fluorescence stopped-flow and steady-state emission measure-ments in conjunction with FRET (13,14). The TBP-E4 interaction has been investigated in an analogous manner. Instrumentation, data acquisition, and data analysis were as described (14) with modifications as noted herein. Briefly, stopped-flow association binding measurements were conducted at 10, 15, 20, 25, and 30°C. Solutions of 40 nM T*E4 dpx *F were flowed together with TBP solutions of 400, 800, or 1600 nM (before mixing). Replicate curves were averaged and very well fit to a biexponential decay model, yielding a set of nine curves, each representing a unique combination of temperature and TBP concentration.
Dissociation Kinetic Measurements and Analysis-To determine the rate of DNA replacement and/or displacement from all conformers of the TBP-E4 complex, dissociation kinetic data were obtained as a function of unlabeled E4 duplex (E4 dpx ) at 15, 20, 25, and 30°C. Briefly, the TBP-T*E4 dpx *F complex was formed using 20 -25.4 nM T*E4 dpx *F and 412 nM TBP, ensuring Ն94% duplex saturation. Dissociation of the complex was initiated by addition of a large excess of unlabeled E4 dpx in one of three concentration ranges: 5.50 -5.56 M, 8.91-9.57 M, or 29.4 -29.6 M. The observed dissociation data were well fit to biexponential decay models as described (14). To ensure a precise comparison, TBP-T*AdMLP dpx *F dissociation kinetic data were collected in an analogous manner using unlabeled AdMLP dpx .
The dependence of the TBP-T*E4 dpx *F dissociation kinetics on unlabeled E4 dpx revealed two processes: passive replacement of labeled by unlabeled duplex, as expected, but also displacement, whereby the unlabeled duplex at very high concentration actively facilitates removal of the labeled duplex. Because only replacement is considered in the overall global analysis, it was necessary to extract that process from the complex dissociation kinetics. Two model-independent procedures were developed for removing the DNA concentration dependence from overall dissociation. In one procedure, the set of three experimental dissociation curves obtained as a function of DNA concentration at a given temperature, defined at discrete time points, was used in a Taylor series expansion about an intermediate DNA concentration to extrapolate to [DNA] ϭ 0. The smoothed experimental data were used to obtain numerical first and second derivatives for the series. An alternate procedure was based on the observation that the two decay constants describing the biphasic dissociation curves were well separated, with fast Ͼ Ͼ slow . fast was thus very nearly the trace of the rate constant matrix and linear in DNA concentration. The value of fast for the pure replacement process was obtained from extrapolation to [DNA] ϭ 0 of a plot of fast versus [DNA]. Because the kinetic curves were normalized, the remaining values needed to describe replacement were slow and the corresponding amplitude change. Both quantities were likewise linear in DNA concentration, although such a correlation is not required. Extrapolation of both values to [DNA] ϭ 0 thus yielded the values needed to construct the pure replacement curves. These curves at 15, 20, 25, and 30°C were subsequently used in TBP-E4 global analysis.
Global Analysis-The correspondence of the ensemble of E4 data to several kinetic mechanisms was explored. These data included nine unique stopped-flow binding and four replacement kinetic curves and six equilibrium binding isotherms. Inclusion of the curves representing equilibrium binding and biphasic replacement rather than the two corresponding constants necessitated modification of the analysis described previously (14). Additional modifications were introduced to ensure appropriate weighting of the kinetic and equilibrium data. Analyses based on the two-, one-, and extended one-intermediate models are described below with changes noted relative to previous studies (14). Finally, to ensure a meaningful comparison, the complete set of AdMLP kinetic and thermodynamic data was reanalyzed in an entirely analogous manner using these modified procedures.
The linear two-intermediate model is as follows.
TBP ϩ DNA L | ; Global analysis yielded values for the six rate constants, k 1 -k 6 , at 30°C, as well as the corresponding six activation enthalpies. The value of ⌬S° for each step was calculated from these quantities. Also determined was the quantum yield of the donor fluorescein in each of the three TBP-bound duplexes relative to that of free T*E4 dpx *F. These values derive from the distance-dependent rate of energy transfer between the two fluorescent labels and are interpreted to reflect the relative extent of DNA bending in each binary complex (6,13,14,17).
Smoothed, normalized representations of each of the nine stoppedflow association and four replacement kinetic curves were first constructed using 41 points/curve (14). Equilibrium binding isotherms were constructed from the experimentally determined K a values using 19 points/curve and scaled from 0 to 0.419, the average overall amplitude change observed in steady-state FRET measurements. These 19 points emphasized the region from 45 to 55% saturation but extended to 98% saturation. Fitting to the shape of the curve rather than to a single K a value allows a priori for heterogeneity in analyses of these equilibria, although heterogeneity was neither detected nor expected for this model.
The theoretical association response functions were determined as described (14). Biphasic replacement response functions, reflecting the approach to equilibrium in the reverse direction, were determined in an analogous manner. The calculated equilibrium binding curves were constructed using K a values derived from the microscopic rate constants associated with the two-intermediate model (Equation 1).
Evaluation of the overall quality of the fit was based on the weighted variance between the observed and theoretical points describing the shapes of the biphasic stopped-flow association ( SF 2 ) and replacement ( R 2 ) curves and the equilibrium binding isotherms ( K 2 ). The weighting factors of 0.041, 0.030, and 0.013 derived from the respective average errors in these data. To obtain global 2 , the three terms were further weighted to reflect the relative information content of each term, as follows.
An ideal global fit to a model then yields global Յ 1, with each term within its experimental error. Error estimates for the optimal parameters were obtained exactly as described (14). The ensemble of TBP-E4 data was initially investigated, in a manner entirely analogous to that just described, using a one-intermediate model.
An equilibrium between two conformations of DNA, one bent and thus predisposed to TBP binding (DNA 1 ), characterizes the model (13,30). Global analysis was as described for the two-intermediate model but with k 3 as a second-order rate constant and the quantum yields of DNA 1 , I 1 , and TBP-DNA FINAL determined relative to a quantum yield of 1 for DNA 2 . Only values of k 1 and k 2 of approximately the same order of magnitude as the other rate constants were deemed acceptable in this analysis. For much larger values of k 1 and k 2 , Equation 5 is equivalent to the one-intermediate model in terms of accommodating the data, with k 3 simply scaled by k 1 /k 2 . The quality of the fit to Equation 5 was assessed as described for the two-intermediate model, with a constraint on the population distribution of DNA 1 and DNA 2 based on the near temperature independence of the fluorescence spectrum of the free duplex. At each iterative step of the global analysis, the corresponding theoretical fluorescein emission, F(T) calc , was determined for each temperature, T, using the following expression.
where QY DNA1 is the quantum yield of free, prebent DNA. F(T) calc values falling outside of the experimentally determined range resulted in penalization of global 2 (Equation 3). Global analyses of the collective TBP-AdMLP data to one-, extended one-, and two-intermediate models were conducted, incorporating these modifications, and the optimal parameters were compared with those obtained previously (14). The analysis included eleven stopped-flow association and four replacement kinetic curves and five equilibrium binding isotherms, the latter scaled to 0.440. In contrast to TBP-E4 dissociation, TBP-AdMLP dissociation was independent of the concentration of unlabeled AdMLP dpx . These latter experimental curves were therefore a direct measure of the replacement process and were used in the global analysis.
FIG. 3. Equilibrium binding isotherm for E4 with TBP at 20°C. The fraction of bound TBP at each titration point is calculated directly from the change in the steady-state emission spectrum. The van't Hoff plots shown in the inset are constructed from the experimentally derived K a values for E4 (OE) and AdMLP (f) and yield values for ⌬H°of 25.1 (Ϯ1.7) and 16.9 (Ϯ2.2) kcal mol Ϫ1 , respectively. These results are in agreement with data obtained by DNase I footprinting over this temperature range for the native AdMLP and E4 promoters (12).

RESULTS
Thermal Stability of TBP-The half-times for inactivation of the free and DNA-bound S. cerevisiae TBP preparation used in these studies (11,13) have been determined by several independent laboratories and methodologies to be ϳ1 and ϳ10 h, respectively, at 30°C under the conditions of these studies (6). 2,3 During the total collection time of 5 min required to obtain a set of five stopped-flow association curves at 30°C, Յ5% of the protein would thus be subject to inactivation, which approximates the variability of replicate curves under a given condition. This conclusion is consistent with the observed superimposition of the first and fifth replicate curves (Fig. 1), the lack of an observed kinetic trend reflecting progressive TBP inactivation, and a total amplitude change identical to that observed in the steady state. The same three observations obtain for replicate curves collected at 10°C over 40 min. These results contrast sharply with those of Jackson-Fisher et al. (28), who reported half-times for TBP inactivation of ϳ4 s and 9 min at 30 and 10°C, respectively.
Equimolar TBP-DNA Binding Studies-That TBP is monomeric under the conditions of this study has been demonstrated by analytical ultracentrifugation (25,26) and biochemical (14) studies. To further establish that the rate constants obtained from stopped-flow binding experiments reflect TBP-DNA binding and not TBP dimer 3 2 TBP monomer , stopped-flow association kinetics were measured with no excess TBP, thus sampling the entire active protein population. These new stopped-flow results were compared with two simulated kinetic curves: one based on the two-intermediate model (Equation 1) and previously determined rate constants (14) and the other based on results obtained using a pull-down assay (15). The experimental kinetic trace was in good agreement with the curve simulated using the two-intermediate model, with the predicted full amplitude decrease of 41 Ϯ 1% observed within ϳ8 s for both the measured and simulated curves. In contrast, the curve simulated from the model including the additional TBP dimermonomer equilibrium differed dramatically from the experimental curve. The latter simulated curve showed an initial rapid phase with an amplitude change of only ϳ3% over 0.5 s, followed by a very slow decay to the full amplitude change, complete only after 50 min, 500 times longer than that observed experimentally. These results show clearly that the kinetic progress curves in this study correspond to protein-DNA association and not a slow, rate-limiting process of dimer dissociation, as has been incorrectly asserted (15,28).
Biphasic kinetics have been observed for TBP-DNA binding in our laboratory using FRET stopped-flow and elsewhere us- values of 1.03, 1.04, and 0.705, respectively, reflect theoretical fits to observed curves within experimental noise and were associated with globally randomized residuals. This close correspondence between the theoretical and experimental data sets is particularly notable considering the additional stringency introduced by noise-free experimental curves and absolute TBP activities.

TABLE I Globally derived TBP-E4 kinetic and thermodynamic parameter values corresponding to the two-intermediate model (Equation 1)
The values shown are for 25°C and 1 M standard state for each step, i, along the reaction pathway. Numbers in parentheses represent the upper and lower error bounds corresponding to the 68% confidence region, determined as described previously (14). k 1 is a second order rate constant with units of M Ϫ1 s Ϫ1 ; k 2 -k 6 are first order rate constants. Quantum yield ratios for I 1 , I 2 , and the final TBP-E4 complex are 0.605 (0.598, 0.612), 0.502 (0.496, 0.529), and 0.550 (0.547, 0.559), respectively. The relative quantum yield reflects the extent of DNA bending (14). Because I 1 and I 2 , as well as the final complex, are present at equilibrium for infinite TBP, the measured   ing a pull-down assay, albeit under different solution conditions. In both cases, the biphasicity is interpreted in terms of linked equilibria: in our work, multiple equilibria associated with the two-intermediate model and, in the latter case, linked equilibria among dimer dissociation, TBP-DNA binding, and TBP inactivation (14,28). However, the contrast between the two data sets is profound; the overall association half-time in our FRET studies is ϳ1 s, whereas that obtained using pulldown assays is 12-20 min, the latter being entirely inconsistent with our experimental findings. Flowing together TBP and DNA with no excess of protein (thus sampling the entire active TBP population) would unambiguously reveal biphasicity deriving from a slow dissociation of dimers with t1 ⁄2 ϭ 20 min.

E4 Distance Distributions, P(R), and the TBP-bound E4 Solution Bend
Angle-Probability distributions describing the 5Ј end to 3Ј end distance distributions for the free and TBP-bound E4 duplex were extracted from measurements of time-resolved fluorescence emission. The mean end to end distances, R, for the free and bound duplex were 54.5 Ϯ 0.1 and 48.0 Ϯ 0.1 Å, respectively. The corresponding values of for the distribution were 6.9 Ϯ 0.2 and 6.1 Ϯ 0.1 Å. A two-kink bending model has been determined previously to best describe the TBP-induced bend for the AdMLP sequence (6). This model yields a solution bend angle for TBP-bound E4 of 71.8 Ϯ 0.2°. This helical bend is very similar to the 76.6 Ϯ 0.2°bend determined for TBPbound AdMLP in solution (6) and to the ϳ 80°bends determined in crystallographic studies for both E4 and AdMLP bound to TBP (31)(32)(33)(34)(35).
Fluorescence Stopped-flow Measurements-The temperature and protein concentration dependence of the TBP-T*E4 dpx *F binding and bending interaction were monitored in real time using fluorescence stopped-flow and FRET. A total of 53 kinetic curves were collected, including replicate curves at each of nine conditions of temperature (10 -30°C) and TBP concentration (200 -800 nM), yielding nine averaged curves. Each of these curves was very well described by biexponential decay, excluding photobleaching (13,14), with an overall change in amplitude of 42.4 Ϯ 0.8%, identical within error to the independently measured steady-state change. The binding of TBP to T*AdMLP dpx *F had been previously measured to obtain 11 analogous stopped-flow progress curves with an amplitude change of 44.0 Ϯ 1.4% (14).
TBP binding to both E4 and AdMLP is biphasic over the experimental temperature range of these studies; the biphasicity increases with decreasing temperature and TBP concentration ( Fig. 2) (14). The association of TBP with the E4 promoter is notably faster and less biphasic than with AdMLP (Fig. 2,  inset). For neither interaction is a pattern based on temperature or TBP concentration apparent.
Equilibrium Binding Constants-Seven equilibrium binding isotherms for the TBP-T*E4 dpx *F interaction were obtained from 10 -30°C, shown for 20°C in Fig. 3. These data were well described by a linear van (Fig. 3, inset) (14). Essentially the same relative affinities are observed with the E4 and AdMLP TATA boxes in the contexts of the native promoters (12).
Dissociation Kinetic Measurements-The time course of the release of labeled DNA from TBP-bound complex was monitored using steady-state emission changes following the addition of large excesses of unlabeled DNA. Three concentrations of the latter were used at each temperature to yield one related set of dissociation curves. All TBP-AdMLP and TBP-E4 dissociation kinetic curves collected at 15, 20, 25, and 30°C were well described by biexponential decay. Notably, TBP-E4 dissociation kinetics, unlike TBP-AdMLP, showed a measurable

FIG. 6. The mole fraction of each species along the TBP-E4 (solid line) reaction pathway at 15°C (A) and 30°C (B) and TBP-AdMLP (dashed line) at 15°C (C) and 30°C (D), generated from the globally derived parameters with [TBP]
‫؍‬ 400 nM. The mole fraction of each species varies with temperature, as shown here, and with [TBP]. The dominance of I 1 over the relevant time interval is apparent at both temperatures. Notably, I 1 is the most prevalent E4 species at equilibrium at 15°C. The general patterns are similar for the TBP-E4 and TBP-AdMLP reaction pathways. dependence on the concentration of unlabeled DNA, revealing contributions to this process from both replacement and displacement.
Inclusion of both replacement and displacement processes in TBP-E4 global analysis would have necessitated additional fitted parameters. The curve corresponding to the replacement process, which is independent of the concentration of unlabeled DNA, was therefore extracted from the set of decays obtained at each temperature using two different procedures (see "Experimental Procedures"). The extracted replacement curves were essentially independent of the method used. The pure replacement curve at 20°C is shown in Fig. 4 together with the set of DNA concentration-dependent dissociation curves from which it was extracted.
The replacement kinetics for both TBP-E4 and TBP-AdMLP are biphasic, although the relative amplitudes of the fast and slow phases differ significantly between the two promoters (Fig. 4, inset). The values of fast for E4 replacement kinetics are approximately two to eight times slower than that for AdMLP. In contrast, the rate constants determined for the dominant slow phases of E4 and AdMLP, 0.0018 Ϯ 0.0002 and 0.0015 Ϯ 0.0004 s Ϫ1 , respectively, are identical within error and are essentially temperature-independent. These values, and their invariance with temperature, are in excellent agreement with those previously reported for the overall dissociation of TBP from E4 (11,12) and AdMLP (12,14).
Global Analysis Using Linear One-and Two-intermediate Models-The collective TBP-AdMLP kinetic and thermodynamic data have been shown unequivocally to be inconsistent with the simplest one-intermediate model (Equation 4) (14). This two-step mechanism likewise could not accommodate the ensemble of TBP-E4 data. Although good correspondence to Equation 4 was obtained using only the four replacement and six equilibrium binding curves, inclusion of the nine stoppedflow binding curves resulted in average weighted residuals approximately four times the experimental error.
Both the TBP-E4 and TBP-AdMLP data were subsequently analyzed using the next simplest, two-intermediate model, Equation 1. The ensemble of TBP-E4 kinetic and energetic data is well described by this mechanism, with global ϭ 0.981 and the weighted average residuals within experimental error for the association, replacement, and equilibrium binding curves (Fig. 5). 4 The optimal kinetic and thermodynamic values corresponding to this model are shown in Table I. As for AdMLP (14), E4 bending occurs simultaneously with TBP binding in the first reaction step, as evidenced by nearly equivalent quantum yield values for the intermediate and final conformers.
Global analysis of collective TBP-AdMLP data using Equation 1 was performed previously with dissociation described using an overall steady-state relaxation expression (14). The more stringent analysis described herein yielded a similar quality fit, with the average weighted residual nearly identical to experiment error. Similar parameter values were also obtained (Table I). The differences were associated primarily with I 2 , the second intermediate, present throughout the reaction at very low concentration and for which parameters derived from both methods of analysis were the least well determined. The defining characteristics of the TBP-AdMLP interaction, such as the thermodynamic profile, are unchanged.
The relative fraction of each species along the association pathway was determined using the microscopic rate constants at low and high temperatures for E4 and AdMLP (Fig. 6). For the latter, the profile was essentially unchanged from that in the initial analysis (14). The species populations vary with TBP concentration and temperature for both the TBP-E4 and TBP-AdMLP reactions, with I 1 present at high concentration for a significant portion of the reaction time for all conditions examined. Notably, I 1 persists at equilibrium for both TBP-promoter complexes, with this conformer even more predominant for E4 than for AdMLP. At equilibrium and 400 nM TBP, I 1 composes 18.2% of all bound E4 species at 30°C and, remarkably, is the dominant bound form at 15°C at 53.1%.
Thermodynamic Profiles for the TBP-E4 and TBP-AdMLP Reaction Pathways-The thermodynamic profiles for TBP-E4  Table I, with (25,26), steady-state fluorescence emission studies (14), and the real time kinetic experiments conducted herein using 1:1 TBP: DNA all confirm that slowly dissociating dimers are not present under our conditions. The thermal stability of TBP under the conditions of these studies has been demonstrated previously and is further confirmed herein.
TBP forms complexes with multiple naturally occurring promoter sequences, with marked preference for four TATA boxes: TATATAA, TATAAAA, TATATATA, and TATAAATA (36). Two such sequences, E4 (TATATATA) and AdMLP (TATAAAAG), are the focus of this study. Extensive real time kinetic and thermodynamic data sets, characterizing the interactions of these two promoters with TBP, have been acquired and the correspondence of these data with numerous models has been explored.
Initially, the data were considered in terms of the simplest, single-step reaction. The clearest evidence of the incompatibility of both TBP-promoter interactions with this mechanism is the multiphasic character of the association and dissociation kinetic curves and the disagreement between the measured equilibrium constants and the ratio of the forward and reverse rate constants (14). The next, more complex one-intermediate model (Equation 4) also failed to accommodate either data set, with average weighted residuals up to six times the experimental error. An extended one-intermediate model (Equation 5), having two conformations of free DNA, accommodated the E4 but not the AdMLP data. Ultimately, the entire ensemble of association, relaxation, and equilibrium binding measures of the AdMLP and E4 interactions with TBP were well accommodated by Equation 1, a linear two-intermediate model.
That these extensive data sets were well described by the same model is not surprising. Both E4 and AdMLP form tightly bound complexes with TBP (36), display high transcription efficiencies (9), have very similar co-crystal structures (31)(32)(33)(34)(35) and have comparable TBP-bound solution structures. The similarity of these structural and functional characteristics is reflected by common qualitative features along the two-intermediate pathway. The values of the relative quantum yields reflect simultaneous DNA binding and bending for both promoters (13,14). Additionally, I 1 is present at high mole fraction throughout both interactions and persists at equilibrium. We therefore propose that for E4, as with AdMLP, both the I 1 and final conformers may bind transcription factor IIB in the subsequent step of RNA polymerase II preinitiation complex assembly (14).
Thermodynamically, the two promoters react with TBP very similarly in the initial binding/bending step, the formation of I 1 from free TBP and DNA (Fig. 7). This initial event proceeds for both sequences through a large energetic barrier together with the largest stepwise increase in entropy. After the first step, however, the thermodynamic profiles of the two reactions are FIG. 8. Raw kinetic data obtained at 15°C for overall TBP-T*E4 dpx *F dissociation using 5.56 M unlabeled DNA (q) and the corresponding pure replacement curve (؋). Theoretical fits derive from analysis of the set of DNA concentration-dependent curves at a given temperature using a dissociation model based on Equation 1, the optimal rate constant values, and displacement from either I 1 (solid line) or TBP-E4 FINAL (dashed line). For the latter model, transformation of the pure replacement curve to capture the fast phase of the observed dissociation requires a rapid rate of displacement from TBP-E4 FINAL . However, because the TBP-E4 FINAL species gives rise to the slow replacement phase of dissociation, its rapid removal eradicates that phase, producing a severe lack of fit. Conversely, a slow rate constant (Ͻk 6 ) for displacement from TBP-E4 FINAL transforms the replacement curve to mimic the slow dissociation phase but results in an equally severe misfit of the fast phase. In contrast, displacement from I 1 simply increases the already rapid rate of I 1 dissociation while retaining the slow dissociation phase deriving from depopulation of the final species. very different. The second step, I 1 3 I 2 , is exothermic and accompanied by a large decrease in entropy for AdMLP. In contrast, this step for E4 is endothermic and has a large increase in entropy. The final step, I 2 3 TBP-DNA FINAL , is thermodynamically similar to the first for AdMLP, highly endothermic and entropically favorable, whereas this step for E4 has only slight endothermic and entropic changes.
Kinetically, the consequences of the stepwise differences are most clearly evidenced by differences in the time evolution of each species along the two pathways (Fig. 6). For E4, I 1 is present at higher mole fraction throughout most of the reaction and at equilibrium under all conditions than for AdMLP. In addition, the mole ratio of I 2 :I 1 is two and a half times greater for E4. Notably, maximum displacement occurs at 15°C, where I 1 is the dominant equilibrium species for E4. These data suggested that the measurable displacement (versus replacement) of E4 by competitor DNA might be attributable to the high relative abundance of the intermediate species.
To identify the predominant species from which displacement occurs, the observed dissociation kinetics were thus compared with theoretical curves incorporating both replacement and displacement. The latter were generated using Equation 1, the six rate constant values at a given temperature, and an additional pathway for displacement from either I 1 or TBP-E4 FINAL . The displacement rate constant that best described the set of three DNA concentration-dependent curves at a given temperature was determined. The model assuming displacement from only I 1 accommodates the dissociation data within error at all temperatures with globally randomized residuals. In contrast, displacement from only TBP-E4 FINAL results in large errors (residuals were one and a half to three times the error) and the systematic deviation of the calculated from the observed dissociation curves. These two theoretical curves are shown in Fig. 8 for 15°C together with the observed dissociation data and the pure replacement curve. This analysis points to a dominant role of the I 1 intermediate in the displacement reaction, although it does not rule out a contribution of the final complex to this process. The facilitated removal of TBP from the E4 promoter by regulatory proteins may similarly proceed from this intermediate conformer rather than from TBP-E4 FINAL . These data afford an independent line of evidence demonstrating the presence of intermediate species at equilibrium. They further reveal a new and novel biological consequence of structural variations in TBP-TATA binary complexes.
That both replacement and displacement processes were observed for TBP-E4 dissociation kinetics suggests caution when deriving a dissociation rate constant directly from a measured "replacement" process. The displacement process is revealed through the use of multiple competitor DNA concentrations in dissociation experiments (37), and its presence or absence is apparent only in retrospect. This concern is of particular importance for measurements aimed at capturing the fast reaction phase. For other protein-DNA complexes with nanomolar affinities, a dependence of dissociation kinetics on competitor DNA concentration may indicate the presence of multiple species at equilibrium.
E4 and AdMLP ultimately form structurally similar binary complexes with TBP. What, then, is the origin of these stepwise similarities and differences? Unique and well characterized structural features are associated with each of these sequences. The uniform deformability of repeating TA base steps, as in the E4 sequence, has been established (38 -45). In contrast, A tracts have been understood to be rigid (31, 46 -48) but with flexibility introduced at the junction between the continuous run of As and mixed sequence DNA (49,50). Very recent bio-physical studies in our laboratory support this view, showing the AdMLP duplex with its partial A 4 tract to be highly flexible, based on the width of the probability distribution of the end to end distance (6). 5 We propose that the structural characteristics inherent to the E4 and AdMLP duplexes give rise to the mechanistic similarities and differences characterizing the two reactions. The 5Ј TATA, identical for E4 and AdMLP, may interact similarly with the protein-binding site in the initial step. These four base pairs would be simultaneously bound, bent and stably anchored via insertion of the 5Ј phenylalanines and formation of most or all of the five hydrogen bonds between the TBP side chains and the bases. Together with loss of solvent at the TBP-DNA interface, these events may account for the large increases in energy and entropy observed for both promoters in step one. The marked differences in the structural features of the downstream half of these two TATA boxes would give rise to the observed mechanistic differences in steps 2 and 3. The large endothermic and entropic increases in step 2 for E4 may derive from ordering of the remaining TA repeats along the TBP saddle, further loss of solvent, and insertion of the second phenylalanine pair. Only slight structural and energetic adjustments in step 3 then yield TBP-E4 FINAL . In this view, the near completion of the E4 interaction in the first two steps is attributable to the unique deformability of repeating TAs.
In contrast, the relatively rigid AdMLP A 4 tract is able to conform to the protein-binding site interface due only to the flexibility at the A 7 -G 8 junction, which allows insertion of the second phenylalanine pair. This protein-DNA structural incompatibility retards the progress of the reaction, delaying until the final step insertion of the 3Ј phenylalanine pair and yielding fully bent DNA but with less perfect complementarity along the interface. The corresponding large changes in energy and entropy for step 3 mimic those of the first step, suggesting similarly significant structural changes.
Consistent with this view are the results of quantitative ⅐ OH footprinting and molecular dynamics simulations for the TBP-AdMLP complex (51). This integrated study reveals polarity in the pattern of contacts along the interface, with the most intimate contacts occurring in the upstream half of the TATA box. In addition, the first half of the TATA sequence displays the highest degree of uniformity in local helix parameters among ⅐ OH footprinting, molecular dynamics simulation, and crystal structure (31,51). Juo et al. (31) also report, from evaluation of crystallography data, that variability in the downstream bases is tolerated, with the upstream end of the TATA box stably anchored to TBP by phenylalanine insertions and hydrogen bonds to the central bases.
In summary, a two-intermediate model with simultaneous DNA binding and bending commonly describes the interactions of two consensus promoters, E4 and AdMLP, with TBP. Significantly different stepwise mechanistic details are proposed to derive explicitly from the sequence-induced structural features of the two TATA boxes. Studies are in progress with C7 (TATA-AACG), a single point mutant of AdMLP, which has a shallow bend angle (ϳ52°) when bound to TBP in solution at 30°C (6) and reduced transcription efficiency (9). TBP-C7 association yields a stopped-flow kinetic curve very different from those for AdMLP and E4, suggesting fundamental differences in the mechanistic details of binary complex formation for consensus and nonconsensus promoters (Fig. 4 in Ref. 17).