Effect of DNA Binding on Geminate CO Recombination Kinetics in CO-sensing Transcription Factor CooA*

Background: CooA proteins are CO-sensing transcription factors. Results: DNA binding to CooA-CO speeds up geminate rebinding of CO. Conclusion: DNA binding reduces heme heterogeneity and CO rebinding barrier. This along with distal pocket trapping maintains the “on” state long enough for transcription to take place. Significance: This work provides a deeper understanding of the allosteric transition in CooA proteins. Carbon monoxide oxidation activator (CooA) proteins are heme-based CO-sensing transcription factors. Here we study the ultrafast dynamics of geminate CO rebinding in two CooA homologues, Rhodospirillum rubrum (RrCooA) and Carboxydothermus hydrogenoformans (ChCooA). The effects of DNA binding and the truncation of the DNA-binding domain on the CO geminate recombination kinetics were specifically investigated. The CO rebinding kinetics in these CooA complexes take place on ultrafast time scales but remain non-exponential over many decades in time. We show that this non-exponential kinetic response is due to a quenched enthalpic barrier distribution resulting from a distribution of heme geometries that is frozen or slowly evolving on the time scale of CO rebinding. We also show that, upon CO binding, the distal pocket of the heme in the CooA proteins relaxes to form a very efficient hydrophobic trap for CO. DNA binding further tightens the narrow distal pocket and slightly weakens the iron-proximal histidine bond. Comparison of the CO rebinding kinetics of RrCooA, truncated RrCooA, and DNA-bound RrCooA proteins reveals that the uncomplexed and inherently flexible DNA-binding domain adds additional structural heterogeneity to the heme doming coordinate. When CooA forms a complex with DNA, the flexibility of the DNA-binding domain decreases, and the distribution of the conformations available in the heme domain becomes restricted. The kinetic studies also offer insights into how the architecture of the heme environment can tune entropic barriers in order to control the geminate recombination of CO in heme proteins, whereas spin selection rules play a minor or non-existent role.

Carbon monoxide oxidation activator (CooA) proteins are heme-based CO-sensing transcription factors. Here we study the ultrafast dynamics of geminate CO rebinding in two CooA homologues, Rhodospirillum rubrum (RrCooA) and Carboxydothermus hydrogenoformans (ChCooA). The effects of DNA binding and the truncation of the DNA-binding domain on the CO geminate recombination kinetics were specifically investigated. The CO rebinding kinetics in these CooA complexes take place on ultrafast time scales but remain non-exponential over many decades in time. We show that this non-exponential kinetic response is due to a quenched enthalpic barrier distribution resulting from a distribution of heme geometries that is frozen or slowly evolving on the time scale of CO rebinding. We also show that, upon CO binding, the distal pocket of the heme in the CooA proteins relaxes to form a very efficient hydrophobic trap for CO. DNA binding further tightens the narrow distal pocket and slightly weakens the iron-proximal histidine bond. Comparison of the CO rebinding kinetics of RrCooA, truncated RrCooA, and DNA-bound RrCooA proteins reveals that the uncomplexed and inherently flexible DNA-binding domain adds additional structural heterogeneity to the heme doming coordinate. When CooA forms a complex with DNA, the flexibility of the DNA-binding domain decreases, and the distribution of the conformations available in the heme domain becomes restricted. The kinetic studies also offer insights into how the architecture of the heme environment can tune entropic barriers in order to control the geminate recombination of CO in heme proteins, whereas spin selection rules play a minor or non-existent role.
Heme proteins are well known to perform a variety of key biological functions, such as diatomic ligand transport and storage, electron transfer, and catalysis. A rapidly growing family of heme proteins, which act as sensors and regulators, has also recently emerged. These sensing and regulatory proteins perform a variety of biological functions by selectively responding to changes in the level of diatomic messenger molecules, such as CO, O 2 , or NO. When the messenger molecule binds to the sensory domain of the protein, it changes the activity of the protein, typically by altering the heme ligation and eventually inducing a global conformational change (1,2). One prototype of heme protein sensors is the carbon monoxide oxidation activator (CooA). CooA is a prokaryotic CO-sensing transcription factor that regulates the expression of the genes necessary for the host bacterium to grow, using CO as the sole energy and carbon source (1).
CooA from Rhodosprillum rubrum (RrCooA) 3 was the first heme protein discovered that has a clear biological role in CO sensing (3)(4)(5). When bound to CO, it becomes activated and then initiates the transcription of a series of genes encoding the CO oxidation system in R. rubrum (5). It belongs to the catabolite activator protein (CAP) family of transcription regulators (1). RrCooA is a homodimeric protein, where each monomer consists of an N-terminal effector-binding domain, enclosing a b-type heme, and a C-terminal DNA-binding domain, containing a classical helix-turn-helix motif (6). The two domains are connected via a long helix (the C-helix) that serves as the dimerization interface (6). In the inactive form of RrCooA, the heme is six-coordinate (6). The proximal side of the heme is ligated by His-77, and the distal side is ligated by the N-terminal Pro-2 of the opposite subunit of the dimer. The N-terminal Pro-2 ligand is displaced by CO binding, which activates the protein for DNA binding (2,6).
CooA from the thermophilic bacterium Carboxydothermus hydrogenoforms (ChCooA) is one of the CooA homologues that have been found and functionally characterized following the analysis of 207 bacterial genomes (7). The heme environment of ChCooA is similar to that of RrCooA in both its ferrous inactive state and CO-bound active state (8,9). His-82 (analogous to His-77 in RrCooA) and the N-terminal amino group are the two axial ligands to the heme in the ferrous state of ChCooA (8,9). The N-terminal amino group is also displaced upon CO binding in this system (8,9).
Comparison of the crystal structure of the "off" ferrous RrCooA with that of CAP in the "on" state ( Fig. 1, A and B) provided the first glimpse of the range of global motions triggered by effector binding to each protein (6). This comparison indicated that the DNA-binding domain undergoes a very large reorientation relative to the effector-binding domain upon activation (6). The recently resolved structure of the apo-CAP supports this conclusion; it revealed that in the inactive apo-CAP, the two recognition helices are not solvent-exposed, although the two DNA-binding domains show multiple orientations relative to the effector-binding domain (10). Large conformational changes are needed to expose and position the two recognition helices in the correct orientation required for DNA binding (10). Resonance Raman investigations of RrCooA and several key mutants have led to a plausible mechanistic model explaining various aspects of CooA activation by CO binding (11)(12)(13)(14). This model suggests that, upon CO binding, the conformational changes that rearrange the DNA-binding domain into the proper orientation for specific DNA binding are triggered by the sliding of the heme into an adjacent cavity along with a displacement of the C-helix toward the opposite heme. Recently, the crystal structure of the CO-bound form of a variant of ChCooA (LL-ChCooA) has been solved (9) (Fig. 1C). In this structure, one of the monomers is without its heme, but the other one contains a CO-bound heme and shows heme sliding and C-helix displacement similar to those predicted by resonance Raman and UV resonance Raman studies (11)(12)(13)(14). The structure also shows an unexpected movement of the N-terminal region, displaced by CO, so that it is repositioned between the heme and the DNA-binding domains. The repositioned N-terminal region serves as a bridge between the two domains, which might be crucial to reorient the two DNA recognition F helices to the correct position for specific DNA binding. Nevertheless, because the structures of wild-type CooACO and its DNA-bound form have not yet been solved, many of the atomic level structural details for CO-mediated allosteric transition in the CooA family of proteins are still not fully understood.
The geminate CO recombination kinetics to RrCooA at room temperature have been reported previously (15,16), revealing that CO rebinds to the heme on a sub-nanosecond time scale with a high geminate yield (15,16). Time-resolved resonance Raman spectroscopy investigations (13) have shown that the geminate rebinding kinetics of CO to RrCooA speeds up upon DNA binding. In this study, we investigate the ultrafast dynamics of CO rebinding to different CooA complexes in order to better understand the activation mechanism of this very important class of proteins. We measured the effects of  1G6N) show substantial differences in the orientation of the DNA-binding domains relative to the effector-binding domains. Bottom, x-ray structure of LL-ChCooACO (C; 2HKX) showing monomer A with CO-bound heme in an active conformation and monomer B without heme. The N-terminal tail in the active monomer is docked between the DNA binding and the heme binding domains. The N-terminal regions are highlighted in yellow. The configuration of the distal pocket in LL-ChCooACO is shown in D. The Leu-121 residue in yellow is from the monomer A, and the other residues in green are from the monomer B. Gly-122 is also part of the distal pocket but is not shown here for clarity.
DNA binding as well as the removal of the DNA-binding domain on the geminate CO recombination kinetics in CooA. We observe that a distributed set of rates rather than a single exponential rate is a common kinetic property displayed by all CooA complexes. We show evidence that this is the consequence of a quenched (or slowly varying on the time scale of CO rebinding) distribution of heme conformations. We hypothesize that upon DNA binding, the inherently flexible DNA-binding domain becomes more ordered, as reflected by the observed reduction of the structural heterogeneity in the heme-binding domain. Finally, we discuss how the architecture of the heme environment reshapes the energy landscape of the protein by modulating the enthalpic and entropic barriers, allowing the protein to control the geminate recombination of CO to the heme. Heme proteins are evidently designed to ensure efficient geminate rebinding if the diatomic ligand has a functional purpose, such as CO binding to CooA. On the other hand, proteins can evolve to impede geminate rebinding if the diatomic ligand is adventitious or poisonous, as is the case for CO binding to myoglobin and hemoglobin.

EXPERIMENTAL PROCEDURES
Sample Preparation-The purifications of the full-length RrCooA (17) and truncated RrCooA (18) have been described previously. ChCooA, containing a His tag, was expressed and purified as described (19). Ferric CooA proteins were diluted in 50 mM Tris-HCl buffer, pH 8.0, and 6 mM CaCl 2 , containing 150 mM KCl for RrCooA or 500 mM KCl for truncated RrCooA and ChCooA. All solutions were continuously degassed with argon for at least 30 min. Ferrous samples were obtained by adding an excess of sodium dithionite under anaerobic conditions. Then an appropriate amount of dithiothreitol (DTT) was added to the ferrous sample to obtain a final concentration of 5 mM DTT. CO adducts were prepared by flushing CO gas over the surface of the ferrous samples for about 30 min. DNA-bound CooA samples were prepared as follows. Two strands of 26-mer oligonucleotides containing P cooF (5Ј-ATAACTGTCATCTG-GCCGACAGACGG-3Ј and 5Ј-CCGTCTGTCGGCCAGAT-GACAGTTAT-3Ј) were purchased from Eurofins MWG Operon (Atlanta, GA). These strands were heated at 80°C for 5 min and then annealed at room temperature for ϳ3 h to form double-stranded DNA. An appropriate amount of DNA was added to the ferric sample, and then the CO complex was prepared as reported previously (13). The CO-CooA-DNA samples were then incubated at room temperature for 30 min. CooACO complexes in the presence of the nonspecific fragment of DNA gt11 (5Ј-TTGACACCAGACCAACTGGTA-ATG-3Ј and 5Ј-CATTACCAGTTGGTCTGGTGTCAA-3Ј) were prepared following the same procedure. In all cases, the final concentration of the protein was 50 M, and that of DNA was 200 M. For each experiment, the sample was loaded into a 1-mm path length cell, and the absorption spectra were recorded before and after the experiment using a spectrophotometer (U-4100 Hitachi). All measurements were carried out at room temperature (T ϭ 295 K).
Experimental Setup-The temporal kinetic traces were measured using a two-color pump probe instrument that has been described in detail elsewhere (20,21). The instrument involves two Ti:sapphire regenerative amplifiers, both operating at 190 kHz, which are seeded by two synchronized Ti:sapphire oscillators: a master (Mira P, 76 MHz, 2.5 ps (full width at half-maximum)) and a slave (Mira F, 76 MHz, 100 fs (full width at half-maximum)). The output of the femtosecond regenerative amplifier (Coherent REGA9000) was focused in a 500-m BBO crystal to generate the second harmonic at 403 nm. This beam was chopped at a rate of 2 kHz and used to pump the sample. The amplified picosecond output of a home-built tunable picosecond regenerative amplifier was doubled using a 2-mm BBO crystal. This beam was tuned to 438 nm and used as a probe. The two beams were then collinearly combined and focused in the sample using an achromatic lens with a 4-inch focal length. To refresh the sample, the two collinear beams were moved to scan a chosen area of the sample using an xy scanner. The time delay between the pump and probe pulses can be controlled electronically using a phase shifter in the synchrolock system (continuous scan from 0 to 16 ns). The timing jitter between the pump and probe pulses was ϳ800 fs. The polarizations of the pump and the probe beam were set to the magic angle (ϳ54.7°) to eliminate rotational relaxation artifacts. After the sample, the pump was blocked using a long pass cut-off filter, and the transmitted probe light was detected by a photodiode connected to a lock-in amplifier. The CO dissociation rate constants (k off ) were determined by rapidly mixing RrCooACO (or RrCooACO ϩ DNA) with an equal volume of 10 mM potassium ferricyanide, and the time evolution of the absorption change at 422 nm was recorded by using a spectrophotometer (U-4100 Hitachi) in time scan mode.

RESULTS
Rebinding Kinetics of CO in RrCooA Complexes-Fluorescence anisotropy-based DNA-binding assays have shown that RrCooA binds to its target P cooF promoter sequence with high affinity, only in the presence of CO (11,19). Fig. 2 shows the CO geminate recombination kinetics to both RrCooA and trun- cated RrCooA, lacking the DNA-binding domain (18), at room temperature. Note that the data are presented on a logarithmic scale. The figure also displays the CO rebinding kinetics to RrCooA in the presence of a P cooF DNA fragment (26 bp) as well as in the presence of a nonspecific piece of DNA ( gt11). The gt11 has a length (24 bp) similar to that of P cooF and is used as a nonspecific DNA control (for the sequence, see "Experimental Procedures").
As can be seen in Fig. 2, the CO rebinding kinetics in RrCooA are non-exponential and occur on a subnanosecond time scale with high geminate yield. This agrees with previous measurements (15,16). In the presence of the P cooF sequence, the kinetics are faster by a factor of 2-3, and the fraction of the photodissociated CO that escapes into the solvent is reduced from about 4% to less than 2%. As can be seen in Fig. 2, the CO geminate recombination kinetics in RrCooA are unchanged upon the addition of the gt11. Thus, the changes in the kinetic response, observed in the presence of the P cooF DNA fragment, reflect conformation changes around the heme induced by specific DNA binding. These changes are not simply due to the presence of DNA in solution. Another striking feature in Fig. 2 is that the CO geminate recombination kinetics in truncated RrCooA are very similar to those in the RrCooA-P cooF complex. This suggests that removal of the DNA-binding domain induces conformational changes in the heme domain of RrCooACO that are similar to those induced by DNA binding.
Rebinding Kinetics of CO in ChCooA Complexes-ChCooACO also binds to the P cooF DNA sequence with high affinity (19). Fig. 3 displays the CO rebinding kinetics to ChCooA in the absence and presence of the P cooF sequence at two different salt concentrations. Here again, the kinetic trace of CO rebinding to ChCooA in the presence of the nonspecific DNA fragment ( gt11) is presented as a control. The plots in Fig. 3 show that the geminate recombination of CO to ChCooA is also non-exponential. More than 98% of the photodissociated CO molecules have rebound within 10 ns. The kinetic response at longer times speeds up in the presence of target DNA, making the kinetic profile appear more homogeneous. As seen in the figure, this effect was found to be independent of salt concentration.
Geminate Rebinding Kinetics of NO and Endogenous Ligand in ChCooA-A recent study has found that, surprisingly, NO binding to ChCooA results in a six-coordinate Fe(II)-NO species that binds the P cooF DNA sequence very tightly at room temperature (19). In Fig. 4, we present the NO geminate recombination kinetics to ferrous ChCooA in the absence of DNA. Fig. 4 also displays the rebinding kinetics following photolysis of the internal ligand (the N-terminal amino group) in ferrous ChCooA. The rebinding of both of these ligands can be fit well (down to survival populations of 1% or less) by using a biexponential kinetic response function. Earlier work on NO binding kinetics to heme proteins and model systems has established that, unlike CO, NO rebinding does not depend on the heme "doming" coordinate and typically displays biexponential rather than broadly heterogeneous kinetics (22). The time constants of NO rebinding to ChCooA are 5.6 ps (91%) and 23 ps (9%). These values are similar to the ones previously measured in other heme-based sensor proteins, such as FixL and Dos (23,24). The fact that NO rebinds to ChCooA on the picosecond time scale with high efficiency means that it can pose a serious competition against CO. However, at higher (physiological) temperatures, it appears that the heme iron-His-82 bond breaks when NO binds to ferrous CooA, thereby forming an inactive five-coordinate NO-bound CooA species (19,25).
The N-terminal amino group rebinding can also be fit with biphasic exponential kinetics, and the following time constants and amplitudes are extracted: 8.7 ps (85%) and 38 ps (15%). These values are similar to those found for endogenous ligand rebinding in other six-coordinate heme proteins (26 -28). The slower phase (38 ps in the present case) is only observed in proteins where functional ligand exchange takes place (27,28). Thus, this phase could represent the rebinding of the endogenous ligand from an "open" conformation, which must be pres-  ent to allow ligand exchange between the N-terminal amino group and the signaling CO molecule. Generally, the open conformation is thought to facilitate the replacement of the endogenous heme ligand by an external physiological signaling ligand (28). Therefore, the thermal dissociation of the N-terminal amino group and a transition to the open conformation, which allows the CO to bind, are probably key steps in initiating ChCooA activation.
Based on the crystal structure of LL-ChCooACO, Borjigin et al. (9) have proposed that the N terminus, displaced by CO binding, might play an important role in CooA activation. Indeed, the crystal structure (Fig. 1C) shows that, upon CO binding, the released N terminus is expelled from the distal heme pocket and resides between the DNA-binding and the heme-binding domains. This occurs through the formation of several new hydrogen bonds and hydrophobic contacts (9). This suggests that the released N-terminal tail acts as a "hook" that helps to stabilize the on state of the protein by holding the DNA-binding domain in the correct orientation for DNA binding to the protein. On the other hand, all of the photodissociated N-terminal amino group population in ferrous ChCooA rebinds within the first 200 ps (Fig. 4). This indicates that, upon thermal dissociation, the N-terminal amino group is unlikely to leave the distal pocket and allow protein activation unless it is displaced by an external competitor ligand, such as CO or NO.

DISCUSSION
Kinetic Model-In order to fit the kinetic data of CO rebinding to the different CooA complexes investigated here, we used a distributed linear coupling model that has been described in detail elsewhere (29,30) (see also the supplemental material). This is a straightforward extension of models often used to fit electron transfer (Marcus theory) but without the implicit averaging over the time and spatial variations of the Stokes shift within the ensemble. The fits using this model extend over 3 decades in time, as shown as solid lines in Figs. 2 and 3. The distributed coupling model (29,30) is based on a quenched distribution of heme reorganization energies, so that the total activation enthalpy, H, for CO binding to an individual heme within the ensemble can be well approximated by two terms, where H p represents the "proximal barrier" due to the heme distortion from its quasiplanar configuration and a is a quantitative measure of the iron out-of-plane equilibrium position associated with the heme doming coordinate. The quantity K is an effective force constant representing all linear restoring forces involved in bringing the iron-porphyrin-ligand system back into its planar transition state configuration. H 0 represents the remaining (a-independent) contributions to the enthalpic barrier, which contains energies involving ligand docking sites and steric constraints associated with the distal pocket. For the sake of simplicity, we assume here that the distal barrier H 0 is constant. However, on the time scale of pico-to nanoseconds, ligand binding can be coupled to the structural relaxation of the protein, which gives rise to time dependence in both the enthalpic and entropic barriers. The ensemble distribution of the iron out-of-plane displacements in the photolyzed ferrous state is taken to be a Gaussian with a mean value (a 0 ) representing the average out-of-plane displacement and a variance ( a ) describing the width of the distribution. The quadratic relationship between H and a in Equation 1 leads to an asymmetric distribution of proximal barrier heights, H p , that can be expressed as follows.
The survival population of the five-coordinate photoproduct at time t after photolysis is then given by Equation 4, The quantity k 0 in Equation 5 is the Arrhenius prefactor, and k B is Boltzmann's constant. The algebraic transformations performed using Equation 4 to generate a simplified fitting function are given in the supplemental material. We note that the data analysis is facilitated in situations where the geminate rebinding amplitude is very large because the distribution of the observed geminate rates, [k g ], coincides with the distribution of the fundamental rates for ligand rebinding, [k BA ] (i.e. the escape rate is much smaller than the rebinding rate). In all CooA samples investigated here, the geminate amplitude for CO rebinding is very high (Ͼ96%). The fits uniquely determine the distribution function g(H p ) and the constant k 1 at room temperature. Detailed studies of the rebinding kinetics as a function of temperature, which extract k 0 and H 0 using Equation 5, will be presented elsewhere. Because k 1 was observed to be temperature-independent (see supplemental material), we find that H 0 is ϳ0 and k 0 is ϳ50 GHz (see Table 1). The values of a 0 and a depend upon K, the effective force constant. This parameter can be approximated by using the frequency of the heme doming mode observed near 40 -50 cm Ϫ1 (31).
Our recent investigation of the low frequency mode activity in ChCooACO shows that the doming mode frequency in this protein is ϳ50 cm Ϫ1 (32), which results in K ϭ 22 N/m. The fundamental CO kinetic and heme distribution parameters for the different CooA proteins are presented in Table 1. For comparison, Table 1 also includes the kinetic parameters for CO binding to myoglobin and protoheme (FePPIX) with and without 2-methyl imidazole (2MeIm). For the latter proteins, the doming frequency is ϳ40 cm Ϫ1 , and therefore the force constant is somewhat lower (K ϳ14 N/m). In order to gauge the sensitivity of the distribution parameters to the value of K, we also used the lower force constant (14 N/m) in fitting the CooA data. When this was done, the values of a 0 and a were systematically increased by a relatively small amount: ϳ0.065 and ϳ0.018 Å, respectively.
Quenched Disorder at Room Temperature-We want to emphasize here that we tried several methods to fit the data: 1) a two-exponential fit that has been used previously to fit the CO rebinding kinetics of RrCooA over a shorter time range (t Ͻ 2 ns) (15,16); 2) a stretched exponential fit that is often used when ligand rebinding is coupled to protein structural relaxation, resulting in a time-dependent transition state barrier height (49); 3) a power law, which was originally used to introduce the concept of protein conformational substates, by a phenomenological fit of the temperature-dependent CO rebinding kinetics in myoglobin below the solvent glass transition (33)(34)(35); and finally 4) the distributed coupling model, which quantitatively accounts for the distribution of heme rebinding barriers (29,30).
In supplemental Fig. S1, we compare the quality of the fits obtained by applying the different fitting methods to the CO rebinding kinetics in the DNA-bound RrCooA, as an example. As can be seen from supplemental Fig. S1, both the two-exponential and the stretched exponential fits fail to describe the long time tail of the kinetic trace, whereas the power law function and the distributed coupling model fit the data very well over the whole time range. This clearly shows that the nonexponential behavior of the CO rebinding arises from an underlying barrier distribution that can be best described by either the power law function or the distributed coupling model. Therefore, in the time widow investigated here, t Ͻ 15 ns, even at room temperature, the photodissociated CO molecules face a quenched distribution of rebinding barriers resulting from a frozen or slowly evolving distribution of heme and protein conformations. This demonstrates that the averaging time for the heme rebinding barrier, driven by the heme conformational fluctuations, is longer than 15 ns.
DNA Binding Tightens Heme Distal Pocket-Analysis of the temperature-dependent CO rebinding kinetics of ChCooA (see supplemental Fig. S3), RrCooA, and truncated RrCooA 4 has shown that the distal enthalpic barrier, H 0 , is zero in all of these proteins (see Table 1). Therefore, the rate k 1 defined in Equation 5 is equal to the prefactor k 0 , which (among other things) describes the entropic contribution to the Gibbs free energy barrier for CO binding. The values of k 0 for the DNA-bound RrCooACO and truncated RrCooACO are larger than those for RrCooACO ( Table 1). The increased prefactor means that the ratio of the number of transition states to the number of dissociated states is increasing. This can occur by the reduction of the distal pocket volume and, thus, by the number of states available to the photodissociated CO. Therefore, it is suggested that DNA binding or truncation of the DNA-binding domain leads to a reduction in the volume of the distal pocket.
This view agrees with UV resonance Raman results that monitor the vibrational modes associated with Trp-110 in the C-helix (13,14). Comparison of the Trp-110 bands of the inactive ferrous RrCooA with those of RrCooACO shows that these bands are enhanced upon CO binding (13,14). This enhancement of the Trp-110 UV resonance Raman signal was attributed to the sliding of the heme into a narrow cavity and the movement of the C-helix toward the opposite heme, burying the indole side chain of Trp-110 and further narrowing the distal pocket (13,14). The Trp-110 signal of RrCooACO is increased further upon DNA binding (13) or by truncation of the DNA-binding domain (14). This led to the conclusion that both DNA binding (13) and the deletion of the DNA-binding domain (14) induce further displacement of the C-helix in a similar way. Our results are consistent with the Trp-110 UV resonance Raman data and independently demonstrate that the additional displacement of the C-helix leads to a reduction in the volume of the distal pocket in both the truncated and DNAbound RrCooACO.
Uchida et al. (36) have shown that DNA binding to RrCooA slightly slows down the CO bimolecular rebinding. This suggests that, in addition to tightening the distal pocket and blocking CO escape, the conformational changes of the protein induced by DNA binding also increase the barrier for CO entry to the heme pocket.
DNA Binding Weakens Iron Histidine Bond- Fig. 5 displays the proximal activation enthalpy probability distribution, g(H p ), for CO rebinding to the different CooA complexes studied here. The distributions can be classified by the average enthalpic barrier (͗H p ͘ ϭ 1 ⁄ 2Ka 0 2 ) needed to be overcome in order to move the iron into the heme plane (the values of ͗H p ͘ are listed in Table 1). ͗H p ͘ is reduced upon DNA binding or by truncation of the DNA-binding domain. Vibrational coherence spectroscopy measurements (32) suggest that the effective force constant for all of these protein complexes remains the same (K ϳ22 N/m), so we find that the value of the average out-of-plane displacement of the transient five coordinate heme is smaller in DNA-bound and truncated RrCooACO compared with that of RrCooACO. This is consistent with the notion that the proximal bond to the heme is slightly weakened, either upon DNA binding or by the removal of the DNA-binding domain. These kinetic findings are in agreement with resonance Raman results, where backbonding correlation plots of the Fe-C and C-O stretching modes show that the Fe-His bond is weaker in RrCooA com-  pared with that of myoglobin and that the Fe-His bond is further weakened when DNA binds to the protein (12). Fig. 5 is that the distribution g(H p ) is narrower in the DNA-bound RrCooACO and the truncated RrCooA when compared with that of the full-length RrCooACO. The widths of the distributions, a , can be found in Table 1, where a ϳ20% reduction in a takes place. This means that the kinetic heterogeneity, due to variations of the heme conformations, in RrCooA is significantly reduced either by DNA binding or by the removal of the DNA-binding domain. Even the intrinsically more homogeneous thermophilic ChCooA displayed a ϳ10% reduction in its barrier distribution width upon DNA binding.

DNA Binding Reduces Structural Heterogeneity in Hemebinding Domain-An important conclusion from
To explain these results, we propose that the inherent flexibility of the DNA-binding domain, especially at the hinge region that connects the two domains, adds additional structural heterogeneity to the heme-binding domain. This additional heterogeneity arises from feedback that propagates from the DNA-binding domain to the heme-binding domain, probably along the same pathway as the transduction signal. The distribution of the heme geometries is tuned by the protein matrix mainly via the Fe-His link. The fact that the distribution function g(H p ) is almost the same for both DNA-bound RrCooA and the truncated RrCooA suggests that this additional heterogeneity is abolished when RrCooA binds to its target DNA. Thus, we conclude that DNA binding enhances the rigidity of the DNA-binding domain and locks it into a conformation with narrower distribution that is reflected by the reduction of structural disorder at the heme-binding site.
A disordered DNA-binding domain is a common feature revealed in the x-ray structures of many transcription factors (6,9,10,37). The B factor values of the DNA-binding domains are significantly higher than those of the heme-binding domains in the structures of ferrous RrCooA (6) and LL-ChCooACO (9). Also, the recent structure of the wild-type apo-CAP has revealed that the DNA-binding domain adopts multiple orien-tations relative to the nucleotide-binding domain, indicating that the inactive form exists in equilibrium of several different conformations (10). NMR and isothermal titration calorimetry investigations of cAMP binding to the dimeric CAP have shown that although the binding of the first cAMP has a minimal effect on the fast dynamics of CAP, binding of the second cAMP quenches protein motions and drastically limits the range of its thermal fluctuations (38). Therefore, CAP activation by cAMP results in a dramatic decrease of the conformational entropy of the protein.
Akiyama et al. (39). have used a small angle x-ray scattering analysis to characterize the structure of RrCooA in solution. Their results suggest that the solution shape of the inactive ferrous RrCooA is very similar to that of CO-bound RrCooA. They hypothesized that the hinge region connecting the hemebinding domain to the DNA-binding domain is kinked in the inactive ferrous CooA so that the two domains are positioned close to each other. This is in contrast with the crystal structure of the ferrous RrCooA that shows the C-helix of one monomer completely extended, whereas the other is kinked (6). Kuchinskas et al. (18) explained this apparent discrepancy by suggesting that the off state may be characterized by an ensemble of DNA-binding domain orientations, whereas the DNA-bound on state has only one orientation. In this regard, the crystal structure of the inactive RrCooA can be considered as one of the several solution structures sampled by small angle x-ray scattering. The results presented here are consistent with the presence of heterogeneity in the absence of DNA binding, which is evidently not observed in the x-ray structure.
Simple Model for Allosteric Transition in CooA-Models for the allosteric transition in transcription factors are generally deduced by comparing the available x-ray structures of the apo (unliganded) and the holo (liganded) proteins. Such models usually suggest that binding of the messenger molecule shifts the protein from one discrete conformation that does not bind DNA to another discrete conformation that has high affinity for DNA binding, or vice versa.
Combining our kinetic data with the results of the studies discussed above (6,10,38,39), we propose a three-state model for the allosteric transition in CooA. This model is based on the observed levels of kinetic heterogeneity and extends the twostate model suggested by Kuchinskas et al. (18). In this model ( Fig. 6 and Scheme 1), CooA is in three possible equilibrium states, each with a different protein conformational distribution: 1) the off state, which has the broadest distribution, with the DNA-binding domain highly disordered; 2) the on state, with CO bound, which has a narrower (or intermediate) distribution of conformations, which are organized for DNA binding; and 3) the final "locked" DNA-bound state, where the DNA-binding domain is locked in place by the binding process. The off state is considered as a broad ensemble of conformations composed of different orientations of the DNA-binding domains, whereas the CO-bound on state is assumed to have a narrower distribution of conformers oriented for DNA binding. The DNA binding shifts the equilibrium toward the final locked set of conformations with the narrowest distribution. In this view, CooACO can be seen as an intermediate state, where the conformations are optimized and the heterogeneity is nar-

Effect of DNA Binding on CO Rebinding Kinetics in CooA
rowed to enhance DNA binding. The kinetic data (Figs. 2, 3, and 5) clearly distinguish between the CooACO and the DNAbound CooACO protein state distributions. We note that, although the small angle x-ray scattering results (39) reflect very similar mean structures in solution for both the inactive and the active forms of CooA, the conformational fluctuations around the mean structure are expected to be different in the two states. Thus, the reduction in heterogeneity, due to successive binding of CO and DNA, must reduce the conformational entropy of the protein. This demonstrates that conformational entropy plays an important role in the free energy of CooA activation.
Effect of DNA Binding on CO Dissociation Kinetics-Puranik et al. (40) have measured the bimolecular dissociation rate of CO in RrCooA. They found that the dissociation kinetics are not monophasic, and they were able to fit their data using two exponentials with equal amplitudes. The two extracted rates were 0.02-0.04 and 0.002 s Ϫ1 (40). To explain this, they hypothesized that RrCooACO could have two slowly interconverting conformations (i.e. open and closed). The faster dissociation rate was assigned to the open conformation, and the slower rate was assigned to either the rate of the interconversion between the two conformers or to the escape of CO from the closed conformation. The two conformations (open and closed) could not be revealed by spectroscopic studies (40). However, from the analysis of the dissociation rate data, Puranik et al. (40) hypothesized that only the closed CooACO population binds to the target DNA, whereas the other half of the population in the open conformation does not bind DNA.
However, it should be noted that the non-exponential behavior of the CO off-rate does not necessarily mean that the protein has two distinct open and closed conformations. Another possibility, which would explain the non-exponential CO dissociation rate, arises if conversion of the protein from the on state to the inactive off state takes place on the same time scale as the CO escape from the protein. In such a case, relaxation effects associated with a broad and slowly evolving protein distribution can lead to a non-exponential kinetic response for the CO dissociation.
We measured the bimolecular dissociation rate of CO in RrCooA, with and without the target DNA. As can be seen in Fig. 7, DNA binding slows down CO dissociation kinetics. We fitted the kinetic data using two exponentials as well as a stretched exponential function. Both methods fit the CO dissociation kinetics in absence of DNA with good fidelity. However, in the presence of the target DNA, both methods fail to fit the early time data (t Ͻ 10 s). The fitting parameters are given in supplemental Tables S1 and S2.
The results of the two-exponential fit indicate that both rates (fast and slow) slow down upon DNA binding. This contrasts with the two-state model proposed by Puranik et al. (40). Within this model, the fast rate should not depend on the presence or the absence of DNA because it represents CO dissociation from CooA molecules that are in an open conformation, which does not bind DNA.
Thus, we conclude that there are two plausible scenarios that can explain the non-exponential behavior of CO dissociation in CooA. In the first, the CO escape is occurring on the same time scale as the protein interconversion from its on state to its off state. In the second, there are allosteric interactions between the two subunits of CooA. When the CO molecule dissociates from one subunit, it triggers an allosteric signal that propagates FIGURE 6. A, proposed model for the allosteric transition in the CooA family of transcription factors. CooA exists in three equilibrium states. In the off state, the N terminus of one monomer is coordinated to the heme of the other monomer, and the protein has a broad distribution of conformations, characterized by different DNA-binding domain (DBD) orientations. In the on state, CO displaces the N terminus. The released N terminus positions itself between the heme-binding domain and the DBD and acts as a "hook" that helps to stabilize the on state by holding the DBD and narrowing the distribution to an ensemble of conformers that are optimized for DNA binding. In the final locked DNA-bound state, the DBD is locked in place with the binding process, resulting in a very narrow distribution. B, one-dimensional schematic diagrams of the energy landscape of the protein along the generalized domain orientation (DO) coordinate that describes the DBD orientation relative to the heme binding domain, in the three states. The allowed range of the domain orientation, as governed by the available thermal energy, becomes systematically more limited, from left to right, in the ferrous off state, the CO-bound on state, and finally the DNA-bound locked state. SCHEME 1. Three-state model for the allosteric transition in CooA. The solid lines are fits to the data, using a two-exponential function (red) and a stretched exponential (green). The fitting parameters are given in supplemental Tables S1 and S2.
to the second subunit, inducing a conformational change that delays the escape of its CO molecule.
Enhanced Structural Rigidity of ChCooA-The distribution function g(H p ) of the thermophilic ChCooACO is narrower compared with that of RrCooACO. This suggests that the overall structure of ChCooA is less flexible than that of RrCooA at room temperature. The enhanced structural rigidity of ChCooA can also explain the following differences between the two proteins. 1) Unlike ChCooA, RrCooA undergoes proximal ligand switch from His-77 to Cys-75 upon oxidation (2). 2) NO binds to ferrous ChCooA and forms a six-coordinate species, whereas NO binding to RrCooA breaks the proximal iron-histidine bond, thus forming a five-coordinate NO species (19). Both differences 1 and 2 suggest that the iron-histidine bond is stronger in ChCooA. 3) The prefactor k 0 (Table 1) is the same in ChCooACO and DNA-bound ChCooACO, whereas it increases in RrCooACO upon DNA binding. We interpret this to mean that, upon DNA binding, the volume of the distal pocket is reduced in RrCooACO but not in ChCooACO. This is consistent with the idea that the distal heme pocket in ChCooA is more stable and harder to distort than it is in RrCooA.
Hydrophobic Trap for CO-The transduction signal initiated by CO binding to the heme accomplishes two important tasks. The first involves the conformational changes required for the rearrangement of the DNA-binding domain, ensuring a proper set of configurations that allow specific DNA recognition and binding. The second is the formation of a tight hydrophobic trap for the bound CO so that the system remains in the on configuration long enough for transcription to take place. Heme sliding and C-helix displacement are key steps in both the rearrangement of the DNA-binding domain and the trapping of the CO molecule (11,13). In addition, DNA binding further tightens the already narrowed distal pocket and completely closes the way for CO escape. In the DNA-bound state, thermal dissociation of CO cannot deactivate the protein because the CO will only geminately rebind to the heme, precluding a reversal of structural rearrangements that might lead to protein deactivation and dissociation of the DNA.
As shown in Fig. 1D, the distal pocket of ChCooACO is exclusively composed of non-polar residues. This also helps to discriminate against O 2 and other potential polar ligands. Neither ChCooA nor RrCooA can form a stable complex with O 2 . In heme proteins, where O 2 is the functional ligand, the distal pocket is more polar, and the oxy complex is stabilized by hydrogen bonding of O 2 to nearby polar residues. For example, O 2 forms a stable complex by hydrogen bonding to His-64 in myoglobin (41) and to Arg-220 in FixL (42). The tight distal heme pocket also precludes the activation of CooA by potential bulky ligands, such as imidazole (43). The formation of a tight trap for the functional ligand, which guarantees its fast geminate recombination with very high efficiency, is also seen in other heme-based sensor proteins, such as FixL (42). NP4 (21,44), a nitric oxide carrier protein, also forms a hydrophobic trap around NO in the low pH environment of the saliva in the blood feeding insect Rhodnius prolixus.
The physiological importance of trapping CO in RrCooA can best be illustrated by considering the opposite limit (I g Ͻ Ͻ 1), where only a small amount of CO rebinds geminately and the majority escapes into the solvent, as is the case for Mb (45). The bimolecular rebinding rate for CO in RrCooA is 14 M Ϫ1 s Ϫ1 (14), and the rebinding of the displaced endogenous ligand Pro-2 is biphasic, with rates of 4000 and 84 s Ϫ1 (40). If we take the concentration of CO dissolved in the environment of the bacterium R. rubrum to be ϳ1 M, we find that CO binds to RrCooA with a rate of 14 s Ϫ1 , which is much smaller than either of the Pro-2 rebinding rates. This means that, when there is no hydrophobic trap and the geminate yield is small, CO thermal dissociation from the heme leads to escape into the solution and little chance of reactivation by CO due to the fast Pro-2 binding. Thus, if significant CO escape to the solvent is allowed, it means that there is a high probability that the protein will be deactivated and that the transcription process will be disrupted.
To be more quantitative, the two-exponential fit of CO dissociation in RrCooA ( Fig. 7 and supplemental Table S1) gives k off of ϳ0.04 and 0.006 s Ϫ1 , but a distribution of rates within this interval is an equivalent possibility. This means that, on average, CO stays bound to RrCooA for ϳ25-150 s. By using a simple three-state model, we find that k off is related to the thermal Fe-CO bond dissociation rate, k AB , by the equation, k off ϭ (1 Ϫ I g )k AB (45), where I g is the geminate amplitude. If we take k off ϭ 0.02 as an average rate for CO dissociation, then we find that k AB ϭ 0.02/(1 -0.96) ϭ 0.5 s Ϫ1 . Taking this as a measure of the thermal Fe-CO bond breaking rate in CooA, the overall CO residence time when the geminate amplitude is small only approaches ϳ2 s. This is not long enough for RrCooA to undergo a conformational change from the inactive state to the active state, search for and bind to its promoter sequence, and then recruit RNA polymerase to start transcription. In fact, Elf et al. (46) found that in Escherichia coli, the search time for the Lac repressor to find its target DNA is in the range of 65-360 s. In a recent paper about real-time observation of transcription initiation and elongation on an endogenous yeast gene, the authors found that the gene firing rate is directly dictated by the search time of the transcription factor for its target DNA (47). They estimated this search time to be about 52 Ϯ 8 s. (47). Therefore, the ϳ25-150 s residence time found for CooA, when CO is trapped in the distal pocket, greatly enhances the chance of RrCooA to find its target DNA, bind to it, and initiate transcription.
Finally, we should delineate the difference between CO trapping and the affinity of CooA for CO. Puranik et al. (40) have shown that the endogenous ligand in CooA acts as a self-inhibitor for CO binding. Indeed, the affinity of RrCooA for CO is greatly decreased when Pro-2 is ligated (40). The affinity of the Pro-2-bound, six-coordinate RrCooA for CO is (ϳ0.5 M Ϫ1 ), whereas the CO affinity of the five-coordinate intermediate is (1600 M Ϫ1 ) (40). This means that when CO binds to RrCooA and displaces Pro-2 (causing Pro-2 to dock outside the distal pocket with return rates of 4000 s Ϫ1 and 84 s Ϫ1 ), the CO can remain trapped in the narrow distal pocket and continuously rebind to the heme for ϳ25-150 s. However, if a CO molecule dissociates and escapes into the solvent, it is difficult, especially at physiological CO concentrations, for another CO molecule to displace Pro-2 and bind to the heme. This is why the trapping action of the distal pocket is so essential to the function of the protein. The low affinity of CooA for CO caused by endogenous ligand inhibition guarantees that CooA activation will not occur until CO concentration in the environment reaches the micromolar level (40). This prevents the bacterium from wasting valuable energy on expressing an array of CO-metabolizing proteins in response to traces of CO.
Role of Entropy in Geminate Recombination-The prefactor, k 0 , for CO rebinding in the CooA proteins investigated here is found to be Ն0.5 ϫ 10 11 s Ϫ1 . Such values of k 0 are comparable with those found in heme model compounds FePPIX-CO (30) (see Table 1) and MP8-CO (48) in glycerol water mixtures, and they are approximately 2 orders of magnitude larger than the prefactor for CO binding to myoglobin (k 0 ϳ10 9 s Ϫ1 ) (34,35,49). This suggests that the entropic contribution to the free energy barrier plays a major role in controlling the CO rebinding rate to the heme and that it can be tuned over a wide range by adjusting the architecture of the protein matrix. Indeed, the Arrhenius prefactor can be approximated as follows (30), where 0 is an attempt frequency, and S † Ϫ S 0 is the entropic barrier. In order to account for the large difference between the prefactor for CO binding in CooA and myoglobin, one has to assume that either the number of accessible states for the protein-ligand system in the unbound state (⍀ 0 ) or in the rebinding transition state (⍀ † ) or in both are significantly different in these two proteins. As shown in the crystal structure of LL-ChCooACO (9), there is a cluster of hydrophobic residues that surround the bound CO and form a tight pocket on the distal side of the heme (Fig. 1D). We suggest that, upon photodissociation, these residues greatly restrain the translational and rotational motion of the CO molecule. This, along with the short time scale for rebinding, significantly reduces the entropic barrier. In contrast, the photolyzed CO in myoglobin is found to lie approximately in the plane of the heme at a distance of 4 Å from the iron (50,51). Therefore, the CO in myoglobin must sample a much larger number of accessible states in order to find the correct upright transition state orientation from which it can rebind. A recent study of CO binding to truncated hemoglobin HbO from Mycobacterium tuberculosis has shown that CO rebinds predominantly within the first nanosecond with a high efficiency (52). Molecular dynamics of CO motion after its dissociation from the heme has revealed that the rotational motion of CO in this protein is constrained around the heme normal, compared with that of CO in myoglobin (52).
As can be seen from Table 1, the distal heme pocket of myoglobin has been engineered to present a significant distal enthalpic barrier for CO rebinding (H 0 ϭ 7 kJ/mol). In contrast, the distal barrier (second term in Equation 1) is zero in the protoheme and all of the CooA complexes. This means that the heme environment in CooA has evolved to eliminate the distal pocket enthalpic barriers that retard CO binding, such as ligand docking sites and steric constraints. The only barrier seen by CO when binding to CooA is the intrinsic barrier presented by the domed heme (first term in Equation 1).
In supplemental Fig. S2, we compare the rebinding kinetics of CO to RrCooA and DNA bound RrCooA with those of CO and NO binding to myoglobin. NO rebinds geminately to Mb on the same time scale as CO binding to CooA complexes. However, the geminate recombination of CO to Mb occurs on a much slower time scale (microsecond) with a very low geminate yield. NO rebinds predominantly to the ferrous heme in Mb and other heme proteins with a rate of ϳ10 11 s Ϫ1 (21,22,24). This rate does not depend on temperature (22). This means that the enthalpic barrier for NO binding is zero, and the prefactor is k 0 ϳ10 11 s Ϫ1 in contrast to the prefactor for CO binding to Mb, which is ϳ10 9 s Ϫ1 . It has been suggested that this huge difference in the prefactors for NO and CO binding to Mb is due to the spin selection rules (53,54). NO binding to the heme iron has a spin allowed channel (⌬S ϭ 1), whereas CO binding to the ferrous heme is a spin forbidden process (⌬S ϭ 2). However, Table 1 shows that the prefactor for CO binding to CooA complexes and heme model compounds is also of the order ϳ10 11 s Ϫ1 . This clearly demonstrates that spin selection rules do not play a dominant role in controlling the kinetics of ligand geminate recombination to heme systems. Studies of heme protein samples in very high magnetic fields also demonstrate that spin selection rules play a nearly undetectable role in ligand rebinding kinetics (55).
It is interesting to note that in Mb, which has H 0 ϭ 7 kJ/mol, the entropic contribution to the free energy barrier for CO binding is also much larger than that in CooA complexes, where H 0 ϭ 0. One possible explanation for this observation is that in addition to the reduction of the volume of the distal pocket in CooA, the entropic barrier might also have a synergistic dependence on the total enthalpic barrier presented by the protein to the ligand. In CooA, the activation enthalpy for CO binding is much smaller than that in Mb, which greatly speeds up CO rebinding and may not allow enough time for the photolyzed CooACO to fully explore all of its available system states, ⍀ 0 . In such situations (i.e. when the rebinding time scale is less than the entropy production time scale), the "nonequilibrium" entropic barrier for CO rebinding will be significantly decreased (30).
One consequence of the non-equilibrium reduced entropy barrier hypothesis is that one would expect the prefactor (which contains the entropy barrier) to also be distributed. This occurs quite naturally because the proximal enthalpic barrier is distributed. The photolyzed proteins with smaller activation enthalpy will rebind quickly and will have less time available for "entropy production." They will therefore rebind with a lower entropic barrier and an increased prefactor, k 0 . Photolyzed proteins with larger enthalpic barriers will have retarded binding so that the photolyzed system has more time to engage in "entropy production." This will increase their entropic barriers and reduce k 0 . If CO rebinding in CooA is fast enough to fall within the non-equilibrium entropy limit, the value of k 0 extracted from the simple model presented here would represent an average of the k 0 distribution. Clearly, a more formal model that includes "nonequilibrium" entropy production is needed in order to fully appreciate and quantify the role of entropic factors in controlling the chemical reactions of biomolecules on ultrafast time scales.