Peripheral Ligand-binding Site in Cytochrome P450 3A4 Located with Fluorescence Resonance Energy Transfer (FRET)*

Background: Cytochrome P450 3A4 (CYP3A4) can bind several substrate molecules simultaneously and exhibits cooperativity. Results: Ligand binding in the active site is preceded by functionally important interactions at a distinct peripheral site. Conclusion: The mechanism of cooperativity involves a ligand-induced allosteric transition. Significance: Allosteric mechanism suggested by our results transforms the view of the grounds and significance of CYP3A4 cooperativity. The mechanisms of ligand binding and allostery in the major human drug-metabolizing enzyme cytochrome P450 3A4 (CYP3A4) were explored with fluorescence resonance energy transfer (FRET) using a laser dye, fluorol-7GA (F7GA), as a model substrate. Incorporation into the enzyme of a thiol-reactive FRET probe, pyrene iodoacetamide, allowed us to monitor the binding by FRET from the pyrene donor to the F7GA acceptor. Cooperativity of the interactions detected by FRET indicates that the enzyme possesses at least two F7GA-binding sites that have different FRET efficiencies and are therefore widely separated. To probe spatial localization of these sites, we studied FRET in a series of mutants bearing pyrene iodoacetamide at different positions, and we measured the distances from each of the sites to the donor. Our results demonstrate the presence of a high affinity binding site at the enzyme periphery. Analysis of the set of measured distances complemented with molecular modeling and docking allowed us to pinpoint the most probable peripheral site. It is located in the vicinity of residues 217–220, similar to the position of the progesterone molecule bound at the distal surface of the CYP3A4 in a prior x-ray crystal structure. Peripheral binding of F7GA causes a substantial spin shift and serves as a prerequisite for the binding in the active site. This is the first indication of functionally important ligand binding outside of the active site in cytochromes P450. The findings strongly suggest that the mechanisms of CYP3A4 cooperativity involve a conformational transition triggered by an allosteric ligand.

The mechanisms of ligand binding and allostery in the major human drug-metabolizing enzyme cytochrome P450 3A4 (CYP3A4) were explored with fluorescence resonance energy transfer (FRET) using a laser dye, fluorol-7GA (F7GA), as a model substrate. Incorporation into the enzyme of a thiol-reactive FRET probe, pyrene iodoacetamide, allowed us to monitor the binding by FRET from the pyrene donor to the F7GA acceptor. Cooperativity of the interactions detected by FRET indicates that the enzyme possesses at least two F7GA-binding sites that have different FRET efficiencies and are therefore widely separated. To probe spatial localization of these sites, we studied FRET in a series of mutants bearing pyrene iodoacetamide at different positions, and we measured the distances from each of the sites to the donor. Our results demonstrate the presence of a high affinity binding site at the enzyme periphery. Analysis of the set of measured distances complemented with molecular modeling and docking allowed us to pinpoint the most probable peripheral site. It is located in the vicinity of residues 217-220, similar to the position of the progesterone molecule bound at the distal surface of the CYP3A4 in a prior x-ray crystal structure. Peripheral binding of F7GA causes a substantial spin shift and serves as a prerequisite for the binding in the active site. This is the first indication of functionally important ligand binding outside of the active site in cytochromes P450. The findings strongly suggest that the mechanisms of CYP3A4 cooperativity involve a conformational transition triggered by an allosteric ligand.
The major human drug-metabolizing enzyme, cytochrome P450 3A4 (CYP3A4), represents the most prominent example of homo-and heterotropic cooperativity among cytochromes P450 (for review see Refs. [1][2][3]. It is now well established that one molecule of CYP3A4 is capable of binding two or more molecules of those substrates that exhibit homotropic cooperativity with the enzyme (4 -10). The initial hypothesis to explain the mechanism of P450 cooperativity was therefore based on an assumption of a loose fit of smaller substrates in a large substrate-binding pocket of the enzyme. It was suggested that efficient catalysis requires the binding of several substrate molecules in one pocket to yield a productive orientation of one of them (11,12). However, this model is insufficient to explain all complex instances of cooperativity in CYP3A4 (for review see Refs. 1,2). According to the current interpretation, the cooperativity in CYP3A4 reflects a true allosteric mechanism. The binding of the first substrate molecule is thought to trigger a structural transition that facilitates the subsequent binding event(s), resulting in a catalytically competent enzyme-substrate complex (13)(14)(15)(16).
The substrate-induced displacement of the spin equilibrium of the heme iron (type I binding) is usually attributed to the formation of the final productive complex of CYP3A4 with several substrate molecules (5, 7-9, 14, 15, 17). However, our recent study (18) of the interactions of 1-pyrenebutanol (1-PB) 2 with CYP3A4 and selected mutants that demonstrate altered cooperativity of steroid hydroxylation (12,19,20) showed that this simple relationship is severely compromised in F213W and F304W mutants. Although these mutants were shown to retain the multisite mechanism of interactions with 1-PB, they exhibit no cooperativity of the 1-PB-induced spin shift. Alteration of cooperativity was shown to result from a displacement of the "spin-shifting" step. Whereas in the wild type the spin shift occurs in the ternary complex only, the mutants exhibit some spin shift upon binding of the first substrate molecule. There-fore, the substrate-induced spin shift in P450 may not result from mere physical displacement of the water ligand of the heme iron by the substrate molecule, as usually thought. Rather, in the case of CYP3A4 and 1-PB, the changes in the position of spin equilibrium result from a substrate-induced change in enzyme conformation that attenuates the hydration and water accessibility of the heme pocket.
This indirect interconnection between substrate binding and the spin state reawakens an earlier hypothesis on the possible role of peripheral substrate binding in P450 enzymes (4,14,15,21,22). It is noteworthy in this regard that the phenylalanine residues 213 and 304, replacement of which with tryptophan results in attenuated cooperativity (12,19,20), participate in the formation of a so-called "phenylalanine cluster" that represents a prominent feature of CYP3A4 (23). The structure of this cluster, which participates in the peripheral binding of progesterone at the distal surface of CYP3A4 (23), is affected substantially by enzyme-substrate interactions in the complex of CYP3A4 with ketoconazole (24). The role of the peripheral substrate binding at the phenylalanine cluster in the mechanisms of CYP3A4 allostery was hypothesized by Williams et al. (23) and advocated by Roberts and Atkins (22). However, direct evidence has been limited.
Further progress in understanding the mechanisms of CYP3A4 allostery and the role and location of peripheral binding site(s) requires development of advanced biophysical techniques. In this study we introduce FRET from the covalently attached donor fluorophore pyrene iodoacetamide (PIA) to the fluorescent ligand fluorol-7GA (F7GA) (25) to dissect the mechanisms of multisite binding in CYP3A4. To achieve sitespecific incorporation of the pyrene fluorophore, we used a series of cysteine-depleted CYP3A4 mutants each bearing only a single modifiable cysteine residue. We demonstrate that the first high affinity substrate binding event takes place outside of the active site. Analysis of the distances from the FRET donor positions to the F7GA-binding sites pinpoints the region of the phenylalanine cluster as the most plausible location of the functionally important peripheral substrate-binding site.
Construction of Cysteine-depleted Mutants of CYP3A4-The single and double cysteine mutants used in this study 3 were based on the cysteine-depleted CYP3A4 construct described earlier (26). In this construct six native cysteines were eliminated through C58T, C64A, C98S, C239S, C377S, and C468A. The single and double cysteine mutants bearing combinations of the above substitutions are termed hereafter as C58, C64, C64/C58, C377/C58, and C468 mutants, in conformity with the positions of the retained cysteine residues. In addition to sitedirected modification of the native cysteine residues, we also made a single cysteine mutant termed C495, in which all native cysteines (excluding Cys 442 ) were eliminated as above.
Expression and Purification of CYP3A4 and Its Mutants-The enzymes were expressed as the His-tagged proteins in Escherichia coli TOPP3 (12). The proteins were purified with a three-column procedure, which includes ion-exchange chromatography on Macro-Prep CM support as described previously (27).
Modification of the Cysteine-depleted Mutants of CYP3A4 with PIA-Modification was performed in 0.1 M HEPES buffer, pH 7.4, containing 1 mM EDTA, 20% glycerol, and 0.2% Igepal CO-630. A 2-3 mM stock solution of PIA in 1:1 (v/v) acetone/ ethanol was added to a 20 M solution of the protein to attain the final concentration of 30 M label. After a 60-min incubation of the mixture with continuous stirring at 25°C, the reaction was stopped with the addition of 1 mM dithiothreitol (DTT). Unreacted label and the detergent were removed by extensive washing with 0.1 M Na-HEPES buffer, pH 7.4, 20% glycerol, 3 mM tris(2-carboxyethyl)phosphine on a small nickelnitrilotriacetic acid-Sepharose column (1 ml of resin per 100 nmol of protein). The modified protein was eluted from the column with 0.15 M histidine-HCl in the same buffer and passed through a column of Bio-Gel P6 (Bio-Rad) equilibrated with the same buffer containing no histidine. The extent of labeling was determined from the absorbance spectrum of the protein using the extinction coefficient of 0.026 M Ϫ1 cm Ϫ1 for the absorbance of PIA at 339 nm (28).
Preparation of CYP-3A4-containing Proteoliposomes-Proteoliposomes were prepared by incorporation of CYP3A4 into liposomes obtained with the octyl glucoside/dialysis technique from a 2:1:0.6 mixture of phosphatidylcholine, phosphatidylethanolamine, and phosphatidic acid as described previously (27). In brief, we used a vortex mixer to prepare a suspension of 10 mg of phospholipids in 2.5 ml of 100 mM Na-HEPES, 150 mM KCl, 0.5 mM EDTA, containing 10% (v/v) glycerol, pH 7.4 (Buffer A), containing 1.54% octyl glucoside. The mixture was incubated for 30 min at room temperature under argon and diluted with Buffer A to the final concentration of octyl glucoside of 0.43%. Liposomes were formed upon dialysis of the above mixture for 72 h at 4°C under constant gentle bubbling of argon gas against three changes of 500 ml of Buffer A, each containing 0.5 g of Bio-Beads SM-2 hydrophobic adsorbent (Bio-Rad). The suspension was then concentrated on 300-kDa cutoff Diaflo membranes (Millipore, Billerica, MA) to a phospholipid concentration of 8 -10 mM and stored at Ϫ80°C under argon. To incorporate cytochrome P450 into pre-formed liposomes, a solution of purified CYP3A4 (100 -150 M) was added to an 8 mM suspension of the liposomes in Buffer A containing 1 mM DTT. The mixture was incubated under an argon atmosphere with continuous stirring for 1 h at room temperature followed by an overnight incubation at 4°C. Separation of unin-3 When terming the mutants as "double" or "single cysteines," we do not count the cysteine residue serving as a heme ligand (Cys 442 ), which is inaccessible to modification and was retained in all our constructs. corporated CYP3A4 was performed by gel filtration on a 1.2 ϫ 40-cm column of Toyopearl HW-75 resin (Tosoh Bioscience GmbH, Stuttgart, Germany). A more detailed description of this technique may be found in our earlier publication (27). Experimental Setup-All titration experiments were carried out at 25°C with continuous stirring in 0.1 M Na-HEPES buffer, pH 7.4, containing 0.6 mg/ml 2-hydroxypropyl-␤-cyclodextrin, 1 mM EDTA, and 1 mM dithiothreitol. 2-Hydroxypropyl-␤-cyclodextrin was included in the media to prevent changes in the fluorescence properties of F7GA due to its aggregation at high concentrations in aqueous solution (25). The F7GA concentration was determined from the absorbance at 440 nm using an extinction coefficient of 14.1 mM Ϫ1 cm Ϫ1 (29). All "titrationby-dilution" and reverse titration experiments were carried out in a 2 ϫ 10-mm ultra-micro fluorescence cell (Hellma GmbH & Co. KG, Müllheim, Germany, product number 105.250). In our regular titrations, we used a 10-mm light path quartz semimicro cell with a stirring compartment (Hellma, product number 109.004F).
Steady-state Fluorescence Measurements-These studies were performed with a combination of either an Edinburgh Instruments FLS920 (Edinburgh Instruments, Edinburgh, Scotland, UK) or a computerized Hitachi F-2000 spectrofluorometer (Hitachi, Ltd., Tokyo, Japan) with an S2000 fiber optic CCD spectrometer (Ocean Optics, Dunedin, FL) as described previously (5). The CCD spectrometer was used to monitor the changes in the transmittance and thus permit an instant correction of fluorescence intensity to compensate for the internal filter effect in reverse titration and dilution experiments. For this purpose, all emission spectra were normalized to the transmittance of a 1-mm slice of the sample (assuming that the source of fluorescence is located in the center of a 2-mm-wide cell) calculated from a measured transmittance spectra. In the studies of FRET from F7GA to the heme, the excitation wavelength was set to 405 nm with a bandwidth of 10 nm. The studies of FRET from PIA to F7GA were performed with excitation at 348 nm and a band with of 10 nm.
Time-resolved Fluorescence Measurements-Studies were performed with an FLS920 fluorometer from Edinburgh Instruments as described previously (25). In the case of titration-bydilution experiments, the amplitude of the decay traces was corrected for the internal filter effect as described above.
The treatment of the decay traces was performed with a technique based on principal component analysis (25). This approach results in a global fit of a series of traces to bi-exponential equations with a common optimized set of decay times, and the amplitudes of the exponential terms vary to reflect the changes in the intensity of fluorescence and the average lifetime of the ligand. The concentration of the enzyme-substrate complex was assessed from the fraction of the fast decay term. The same set of decay traces was used to plot the dependence of the fluorescence intensity (overall amplitude of the decay traces normalized to the F7GA concentration) versus the concentration of the enzyme/substrate mixture.
Fluorescence Dilution Experiments-A 100-l aliquot of enzyme/substrate mixture of desired stoichiometry was placed in a fluorescence quartz micro cell with 2 ϫ 10-mm optical chamber (Hellma GmbH, product number 105.250). Small aliquots of buffer were added to the cell up to the maximal volume of the cell (2000 l). The spectra or fluorescence decay traces were recorded after each addition.
FRET Titration Experiments-The titrations of the PIA-labeled proteins with F7GA were carried out in a 100-l fluorescence micro cell (see above). The excitation wavelength was set to 348 nm, which corresponds to the general minimum in the F7GA excitation and is close to the maximum of PIA excitation. The efficiency of FRET was judged from the decrease in the fluorescence of the pyrene fluorophore.
Data Processing-Absorbance spectra obtained in titration experiments were analyzed by principal component analysis combined with the least square approximation of the principal vectors with a combination of the prototypical spectra of the CYP3A4 low spin, high spin, and type II substrate-bound ferric low spin CYP3A4 P450 states as described previously (5, 30 -32). All data treatment procedures and curve fitting were performed using a 32-bit version of our SPECTRALAB software (31) running under Windows XP TM .
The fitting of the titration curves to an appropriate equation was done with a combination of Marquardt and Nelder-Mead nonlinear least square algorithms as described previously (31). When the interpretation of the results did not require an assumption of multiple binding events (FRET dilution and reverse titration experiments), we used Equation 1 for the equilibrium of bimolecular association (see Ref. 33, page 73, Equation II-53), In the approximation of the results of the titration experiments we used a two-step sequential model of binding as shown in Scheme 1. The calculations were made using the system of equations describing the behavior of this model at comparable concentrations of E and S, as described earlier (5,18,30). This system of nonlinear equations (supplemental Equation S1) along with an outline of the approach used to find its roots may be found in the supplemental material. Tools Used for Molecular Modeling, Ligand Docking, and Structure Analysis-The structural model of CYP3A4 bearing PIA labels at positions 64, 377, 468, and 495 was built on the basis of the x-ray structure of CYP3A4 complex with progesterone (PDB code 1W0E) with the use of the TINKER molecular modeling package (34) version 5.1. Conversion of PDB files into the TINKER-compatible XYZ format and vice versa as well as the assignment of the atom and bond types necessary for the use of the MM3 force field were performed with the OpenBabel chemical toolbox (35). The residue Ser 495 in the 1W0F struc-SCHEME 1 ture was replaced with Cys as in our C495 mutant. The structures of peptide fragments 55-69, 458 -503, and 376 -382 bearing the PIA labels covalently attached to Cys 64 , Cys 377 , Cys 468 , and Cys 495 were optimized with the use of the NEW-TON procedure and the MM3 force field (36), as supplied in TINKER package. The resulting optimized structures of the labeled fragments were then re-inserted into the original structure (PDB code 1W0E).
Search for regions of structure of PIA-modified CYP3A4 that comply with the distance restraints imposed by the set of donor-acceptor distances determined with FRET was done with a custom program written in Delphi TM version 7 (Borland Software Corp., Austin, TX). Molecular docking of F7GA in the proximity of the found sets of residues was performed with Autodock Vina (37). The initial docking was performed within the 20 ϫ 20 ϫ 20-Å boxes built around the geometrical centers of the found sets of residues. The size of the box was then optimized by increasing the dimensions in 1-Å increments in those cases when the optimized positions of the ligand approached within 2 Å of the walls of the box. Similarly, the dimensions were decreased in 1-Å increments if the distances of all of the found positions of the ligand from the walls of the block were larger than 4 Å. The final dimensions of the boxes were 24 ϫ 20 ϫ 10 Å, 24 ϫ 20 ϫ 15 Å, and 23 ϫ 23 ϫ 8 Å for the sets R1-1, R1-2, and R2, respectively. Other manipulations of the molecular models and preparation of the molecular graphics for this publication were done with the use of the UCSF Chimera software (50).
The calculations of the donor-acceptor distances according to the Förster equation (see Equation 13.2 in Ref. 38) were performed with the use of PhotoChemCad software (39). In these calculations, we set the values of the orientation factor ( 2 ) and the refractive index (n) to 0.667 and 1.4, respectively.

RESULTS AND DISCUSSION
Fluorol-7GA as a Model Fluorescent Substrate of CYP3A4-In our search for an appropriate high affinity fluorescent ligand for FRET-based studies of CYP3A4 cooperativity, we identified the laser dye F7GA as an allosteric ligand and potential substrate of CYP3A4. The high intensity of fluorescence and beneficial spectral properties of this dye (Fig. 1) permit its use in a variety of FRET-based techniques. The considerable overlap of the emission band of F7GA with the P450 heme absorbance allows detection of enzyme-substrate interactions by FRET from F7GA to the heme chromophore. Furthermore, F7GA has a perfect resolution of its excitation bands from those of pyrene, a potential FRET donor (Fig. 1). The pyrene/fluorol donor-acceptor pair with its Förster radius of 30 -35 Å provides an excellent means for probing the location of multiple binding sites in the enzyme molecule.
Interactions of F7GA with CYP3A4 Studied by Absorbance Spectroscopy-Binding of F7GA to the wild type CYP3A4 (Fig.  2a) causes a well pronounced spectral transition. Application of principal component analysis to the spectral series obtained in titrations reveals two significant principal components of the F7GA-induced spectral changes in CYP3A4 (Fig. 1, inset). The first component (Fig. 1, solid line) represents mainly a low-tohigh spin shift (type I binding spectrum), whereas the shape of the second principal component (Fig. 1, dotted line) is indicative of type II-like changes.
Resolution of the P450-related spectral changes into two principal components demonstrates that the two types of spectral changes do not take place simultaneously. Rather, the F7GA-induced spin shift takes place in the initial steps of the titration and reveals high affinity and prominent cooperativity (Fig. 2, bottom panels; Table 1), whereas the type II-like transition takes place at higher concentrations of the ligand. Therefore, the absorbance titrations of the wild type CYP3A4 with F7GA reveal at least two sequential binding events with dissimilar spectral signatures and concentration dependence.
As seen in the top panel in Fig. 2b, the mutant F304W exhibits increased amplitudes of the F7GA-induced spectral changes. In contrast to the wild type enzyme, an adequate representation of these changes requires only one principal component (Fig.  2b, inset), which reflects a low-to-high spin transition with some concomitant appearance of the apparent type II complex. Both types of spectral changes thus take place simultaneously in the mutant (Fig. 2b, bottom panel). Fitting of the F7GA-induced changes in the high spin fraction (F h ) of the enzyme with the Hill equation shows no effect of the mutation on the S 50 value. Although decreased in the mutant, the value of the Hill coefficient (n H ) remains Ͼ1, indicative of retained cooperativity (Table 1).
F7GA Binding to PIA-labeled Wild Type CYP3A4 Monitored by FRET from the Label to the Ligand-Labeling of wild type CYP3A4 with pyrene iodoacetamide at a 1:1 molar ratio yielded a CYP3A4-PIA preparation containing Ͼ0.95 M equivalent of the label per mol of the enzyme. Absorbance titration experiments showed that the labeling did not cause any significant changes in the values of S 50 or the amplitude of F7GA-induced spin shift in the enzyme, although there was some decrease in the Hill coefficient ( Table 1).
As shown in Fig. 3a, addition of F7GA to CYP3A4-PIA causes a profound decrease in the pyrene fluorescence concomitant with the appearance of the F7GA emission band (Fig. 3a).  The titration curves (Fig. 3b) reveal prominent cooperativity and may be approximated with the Hill equation, yielding S 50 and n H values of 7.0 Ϯ 0.7 M and 1.7 Ϯ 0.3, respectively (the mean values and the confidence intervals are based on three independent experiments). The cooperativity observed in these experiments indicates that the signal originating from the second bound F7GA molecule predominates in the observed fluorescence changes. Therefore, the amplitude of FRET from PIA to the first bound molecule of the ligand appears to be much lower than the amplitude of FRET resulting from the second binding event.
Resolution of Individual Substrate Binding Events in the Interactions Registered by FRET from F7GA to the Heme-Further insight into the mechanism of interaction required resolution of the parameters of the two individual substrate binding events. These parameters may be then used in mathematical modeling of the binding isotherms obtained through FRET and absorbance titrations to probe their mutual consistency and conformity to the sequential multisite mechanism suggested in our earlier studies (5,18). The two substrate binding events were resolved with titration-by-dilution and reverse titration studies where we monitored FRET from the ligand to the heme, similar to our studies with 1-PB (5, 18). Fig. 4a exemplifies the results of a time-resolved fluorescence dilution experiment with a 1:1 mixture of CYP3A4 with F7GA. This setup is specific to the formation of the 1:1 enzyme-substrate complex (the first binding event), given that interactions obey a sequential mechanism involving binding sites with considerably different affinities to the ligand (5,40).   Titration of the ligand with enzyme (reverse titrations) was also performed. Because the bulk of the measurements are performed at large excess of enzyme, and the affinity revealed in the first binding event is presumed to be higher than in the subsequent step(s), this approach also yields binding constants for complexes with 1:1 stoichiometry. As seen from Fig. 5a, addition of increasing concentrations of CYP3A4 to 0.5 M F7GA results in a decrease in the intensity of fluorescence associated with some displacement of the band to shorter wavelengths. This displacement, which is also seen in the experiments with FRET from PIA to F7GA (Fig. 3), is presumably caused by a translocation of the substrate from solution to a more hydrophobic environment. The K D values obtained in these titrations were fairly close to the values found in 1:1 dilution experiments ( Table 2). 4 In contrast to 1:1 dilution and reverse titration experiments, the series of fluorescence decay traces obtained with dilution at excess ligand must reveal a secondary substrate binding event resulting in the formation of the enzyme with several (at least two) ligand molecules (5). Consistent with the premise of a considerable difference in affinity for F7GA of the two binding sites, the fitting of the dilution curves obtained at excess substrate to Equation 1 yields estimates of K D as high as 11-12 M ( Table 2).
The results obtained with either dilution or reverse titration approaches with the F304W mutant were very close to those found for the wild type enzyme in terms of the dissociation constants and the efficiencies of FRET (Table 2). We may conclude therefore that the interactions of F7GA with either the wild type CYP3A4 or the F304W mutant involve at least two consecutive binding events. In both cases, the dissociation constant for the second binding event is 3-4-fold higher than that of the first. The replacement of Phe 304 with tryptophan has no substantive effect on the multisite mechanism, despite a decrease in cooperativity seen in the absorbance titrations. This situation closely resembles the one depicted in our recent study of the interactions of 1-PB with F304W and other CYP3A4 mutants with attenuated cooperativity (18).

Analysis of the Interactions of F7GA with CYP3A4 Using the K D Values Found in FRET Dilution and Reverse Titration
Studies-The experimentally derived K D1 determined in the 1:1 dilution and reverse titration experiments and K D2 determined in dilution studies at excess substrate may be used to analyze the results of either absorbance spectroscopy or PIA/F7GA 4 It should be noted, however, that both the 1:1 dilution and reverse titration experiments reveal some differences between the parameters determined with the lifetime and the steady-state methods. In particular, the estimate of the FRET efficiency based on the lifetime measurements in the dilution setup is lower than the value from the intensity data, which appears to exceed 100% (Table 2). This discrepancy may indicate that the mechanistic basis of the decrease in the intensity of fluorescence is complex and involves some additional steps that are not accompanied by any changes in the lifetime. Although the difference in the apparent efficiencies of FRET is not seen in the reverse titration, there is a notable difference in the K D values found with the two approaches ( Table 2). FRET titrations in terms of the two-step sequential model of interactions (see Scheme 1). In this analysis we explored the possibility that the substratedependent spin shift, as well as FRET from PIA to F7GA, may take place in both the ES and SES complexes. For this purpose, we used a nonlinear least squares optimization procedure to approximate the experimental results with a combination of the changes in the concentrations of the ES and SES complexes calculated according to a system of equations describing the steady-state behavior of the model (see supplemental Equation  S1) (2). The initial estimates of K D1 and K D2 were set to the mean of the values found in FRET dilution and reverse titration experiments (e.g. 2.6 and 11.85 M, respectively, for the wild type CYP3A4). The set of optimization variables included an adjustable parameter F ES that reflects the fraction of the total signal caused by the formation of the complex ES (A max ES ) as opposed to the ternary complex SES (A max total ) as shown in Equation 2, Results of the fitting of the isotherms of the F7GA-induced spin shift to the sequential binding model (2) are illustrated in Fig. 6. The approximation of the results of PIA/F7GA FRET titration with the same formalism is shown in Fig. 3b (solid line).
In both cases, the satisfactory approximation is possible only with a nonzero value for F ES .
According to our analysis (Table 3), all F7GA-induced displacements of the spin equilibrium of CYP3A4 are exerted upon the formation of the primary complex ES. The subsequent binding of the second substrate molecule does not change the spin state or even result in some reversal of the spin shift (F ES Ն0.5).
In contrast, the amplitude of FRET observed upon the formation of the complex ES amounts to only 19 Ϯ 5% of the amplitude characteristic of the ternary complex (Table 3). Close similarity of the estimates of K D1 and K D2 derived from the approximation of absorbance and PIA/F7GA FRET titration results demonstrate internal consistency of the model (Scheme 1) in depicting the mechanism of interactions revealed by either of the two approaches. Importantly, the difference in the PIA/F7GA FRET efficiencies characteristic of the first and the second bound molecules of the ligand suggests a considerable difference in the distances from PIA to each of the two binding sites. According to these results, the enzyme binds F7GA at two distinct, spatially segregated sites.

Probing the Position of F7GA-binding Sites by FRET Titrations in a Series of Cysteine-depleted Mutants Labeled with PIA-
The uncertain and possibly heterogeneous position of the donor fluorophore in PIA-labeled CYP3A4 prevented any further structural interpretation of FRET results obtained with wild type CYP3A4. Therefore, to localize the two individual F7GA-binding sites, we employed site-directed incorporation of the PIA label into mutants of CYP3A4 bearing modificationaccessible cysteines at positions 64, 377, 468, or 495.
In a previous study (26), we described the design of cysteinedepleted mutants C58/C64, C64, and C58. The yield of C58/ C64 in E. coli was close to that of the wild type enzyme, although the recovery of C64 and C58 mutants was considerably lower. All three mutants were stable and showed no substantial changes in the parameters of interactions with such substrates as 1-PB, ␣-naphthoflavone (ANF), and bromocriptine. We also labeled these mutants with equimolar BADAN and demonstrated that the C64/C58 double mutant bears this thiol-reactive probe uniquely at Cys 64 (26).
In this study, we probed the labeling of these three constructs with PIA. As with BADAN, modification of C64 by PIA is much faster than labeling of C58. Furthermore, the labeled C64 is stable, whereas PIA-labeled C58 is prone to precipitate and convert to the inactive P420 state (data not shown). Therefore, to introduce the PIA label at Cys 64 , we utilized the C58/C64 double cysteine mutant, which gave higher expression.  The same approach was used to introduce the label at Cys 377 through the use of the C377/C58 construct. We also employed the C468 and C495 single cysteine constructs. The yield of expression of C377/C58 and C468 mutants was similar to that of the C58/C64 construct (ϳ200 nmol/liter of culture). The recovery of C495 was lower and did not exceed 70 nmol/liter. All three mutants were stable and showed no effect of the mutation on the parameters of their interactions with 1-PB, ANF, and bromocriptine (data not shown).
The PIA/F7GA FRET titration curves for all four labeled mutants are shown in Fig. 7a. Although the general behavior seen with the PIA-labeled wild type is retained in all four mutants, the amplitude of fluorescence decreases, and values of the Hill coefficient and F ES values differ considerably among the mutants (Table 4).
Before using the FRET efficiencies derived from these experiments to probe the locations of each of the two binding sites in the CYP3A4 molecule, it was necessary to verify that the FRET takes place within one and the same molecule of the enzyme as opposed to different subunits of an oligomer, which CYP3A4 forms in solution. To this end, we studied the interactions of F7GA with PIA-labeled C64/C58 and C468 mutants incorporated into large proteoliposomes with lipid/protein molar ratio of 1000:1. Our recent studies demonstrated that incorporation into these lipid-rich liposomes results in complete monomerization of CYP3A4 enzyme oligomers (27). The results of PIA/ F7GA FRET titrations of CYP3A4 monomers in liposomes (Fig.   7b) were quite similar to those obtained in solution. Both mutants exhibited a distinct cooperativity with Hill coefficients as high as 1.5 Ϯ 0.3 and 1.9 Ϯ 0.5, respectively. As seen from supplemental Table 1S, the values of the S 50 , K D1 , and K D2 were ϳ2 times higher than those observed in solution, whereas the other parameters reveled no considerable effect of the enzyme monomerization in the membrane. The results suggest that the major part of the observed signal derives from FRET between PIA and F7GA fluorophores located in the same enzyme molecule.
Determination of the Distances from the PIA Probes to F7GAbinding Sites-The FRET efficiencies specific to F7GA molecules bound to the high (E 1 ) and low affinity (E 2 ) binding sites (Table 4) were calculated using the known formalism for overall efficiency of FRET in a system with two acceptors and a single donor (41), as described in the supplemental material. The efficiencies were used to estimate the distances of the F7GA molecules from each of the four positions of the donor fluorophore. These calculations require the knowledge of the quantum yield (⌽) of the pyrene fluorophore (the donor), which is very sensitive to the environment and may be decreased by stacking interactions with the neighboring aromatic residues (42).
Experimental determination of the quantum yield (⌽) was based on the use of an ethanol solution of the laser dye coumarin-47 as a reference (⌽ ϭ 0.73 (43)) and involved a correction for quenching of fluorescence due to FRET to the heme. For this correction, we determined the effect of H 2 O 2 -dependent heme   Table 4. All four distances of the PIA probes to the second low affinity binding site (R 2 ) are quite similar (20 -26 Å) and close to the average radius of the CYP3A4 molecule. The location of the labels at different positions in the periphery of the enzyme therefore suggests that the low affinity site is located close to the geometrical center of the enzyme, most probably in the heme binding pocket. In contrast, the distances to the high affinity binding site (R 1 ) differ considerably, which suggest a peripheral location of this site. Initial ligand binding outside of the active site is also supported by the studies of Isin and Guengerich (14,44), who investigated the kinetics of CYP3A4 interactions with its substrates and inhibitors.
Structural Interpretation of the R 1 and R 2 Distances-Further interpretation of the distances found above requires a structural model of PIA-labeled CYP3A4 mutants, which was built on the basis of the structure of CYP3A4 complex with progesterone (PDB entry 1W0F) (23), as described under "Experimental Procedures." The resulting model is illustrated in Fig. 8. The respective PDB file may be found in supplemental material.
The coordinates of pyrene moieties in the above model were used to find the fragments of the CYP3A4 molecule that comply with the distance restraints imposed by the sets of the R 1 and R 2 distances. This was done with a custom program that scanned all non-hydrogen atoms of the side chains in the CYP3A4 structure and calculated their distances to each of the four pyrene moieties (precisely, to the midpoint between the carbon atoms 11 and 12). The root mean square deviation (r.m.s.d.) of these distances from the set of four R 1 or R 2 distances (Table 4) was then calculated. The amino acid residues containing at least one non-hydrogen atom in with r.m.s.d. below the specified threshold were considered as markers of the binding site. The coordinates of the side chain atoms selected in this way were then used to locate the approximate position of the bound F7GA molecule. The residues found with this analysis are shown in Fig. 8a with the backbone colored in magenta or blue (scanning with the set of R 1 values) or green (scanning with the set of R 2 values).
The scanning with the set of R 2 distances and the r.m.s.d. threshold of 4.5 Å identified only four hits. The set of matching residues, which is designated below as "set R1," includes residues Pro 358 , Ile 369 , Leu 483 , and Asn 484 located in the substrate binding pocket of CYP3A4. Therefore, consistent with our initial conclusion, the low affinity F7GA binding event appears to take place in the substrate binding pocket.
The residues that comply with the set of distances R 1 at the threshold of 4 Å diverge into two subsets located on the opposite sides of the molecule. The first subset shown in Fig. 8a, magenta (set R1-1, which includes residues 108, 109, 213, 219, 223, and 238 -241), is located at the distal side of the molecule, close to the phenylalanine cluster (23). The second subset, which is shown in Fig. 8a, blue (set R1-2, residues 424, 426 -437, and 441-445), is located at the proximal surface, close to ␣-helixes K, KЈ, and the meander region.
The interactions of F7GA with CYP3A4 in the proximity of the found sets of residues were modeled with Autodock Vina (37) as described under "Experimental Procedures." The coordinates of the three best poses of the ligand docked in the proximity of the R1-1, R1-2, and R2 sets of residues may be found in the PDB file (supplemental materials). The geometric localization and energy of interactions for the best F7GA positions are characterized in Table 5.
The best position of the F7GA molecule docked in the proximity of the residues of the subset R1-1 is shown in Fig. 8a in magenta. The detailed view of its docking is shown in Fig. 8b. Similar to the peripherally bound molecule of progesterone (23), F7GA molecule packs against the side chains of Phe 219 , Phe 220 , and Ile 238 . The complex is stabilized by a hydrogen bond between the NH group of the ligand and the carbonyl oxygen of Asp 217 . The distances from this position to the four pyrene moieties in PIA-labeled CYP3A4 (Table 5) are in a good agreement with our estimates of R 1 .
The alternative position of the high affinity binding site is located on the opposite proximal side of the molecule. Docking of F7GA in this region suggests two possible positions in the proximity of the set of residues R1-2. Both positions, which are shown in dark and light blue in Fig. 8a, are close to ␣-helices K and KЈ and the meander region. The predicted energies of interactions at these sites are less favorable than at the distal site, although the compliance of the distances is slightly better ( Table 5).
Among the two possible locations of the high affinity binding site, the location at the distal site is considerably more likely. Besides being supported by the x-ray structure of the progesterone complex, the binding at the distal side in the proximity of the phenylalanine cluster (subset R1-1) provides a sound explanation for the effect of site-directed mutagenesis in this region on CYP3A4 cooperativity (12,18,19,45).
Displacement of the spin equilibrium of the heme iron by the first F7GA binding event detected in our experiments implies a role of ligand-induced conformational transitions. This conformational modulation is also consistent with ligand binding at the phenylalanine cluster, which may have a considerable effect on the shape, size, and accessibility of the substrate binding pocket (23,24).
As expected, all poses found by docking of F7GA in the proximity of the residues of set R2 (low affinity binding site) are positioned in the substrate binding pocket. The pose that has the most favorable energy of interactions is shown in Fig. 8a in green. Fig. 8c illustrates the detailed view of the docking. As seen from this figure, the interactions are mediated by residues Arg 105 , Leu 373 , and Ser 119 , which are known to belong to substrate-recognition regions in the CYP3A4 active site (24, 46 -48).
As suggested by the absorbance titrations, the binding of F7GA in the active site does not result in any further low-tohigh spin shift but induces some low amplitude type II spectral changes. These changes are typical for P450 interactions with NH-containing compounds and may be caused by some partial ligation of the heme iron to the nitrogen atom in the butylamino group of F7GA. The fact that the F304W mutation eliminates this spectral transition may indicate altered geometry of a The values given for the docking positions represent the distances from the midpoint between carbon atoms 11 and 12 of the pyrene moiety to the central carbon atom of the F7GA ring system. b Root mean square deviation of the distances calculated from the docking model with the FRET-based estimates. c Free energy of binding was calculated by Autodock Vina. d The designations of the F7GA poses given in parentheses correspond to those used in the PDB file found in the supplemental material.
F7GA binding, which prevents the interactions of the NH group with the heme iron.
Conclusions-Results of our analysis demonstrate that the interactions of CYP3A4 with F7GA may be considered as a sequence of two substrate binding events, where the first molecule of the ligand binds at the periphery of the enzyme. This mechanism of interactions appears to be general and is in good agreement with the results obtained with 1-PB and CYP3A4 mutants with decreased cooperativity (18), as well as the studies of the interactions of CYP3A4 with bromocriptine, testosterone, and ANF (14,22,44).
The most probable position of the peripheral binding site suggested by our results is similar to that observed by x-ray crystallography of either CYP3A4 in the complex with progesterone (23) or CYP2C8 with a peripherally bound palmitic acid (49). Despite these prior observations, there was no direct evidence of any involvement of the peripheral interactions in P450 mechanisms. Our results provide the first indication of a functionally important ligand-binding site outside of the CYP3A4 active site. This finding changes considerably our view of the mechanisms of cooperativity in the enzyme and strongly implicates a conformational transition triggered by a ligand bound to an allosteric site outside of the substrate-binding pocket. Our results raise the intriguing possibility that there may be specific endogenous ligands of this newly discovered peripheral site that play a key role in allosteric regulation of the microsomal drug metabolizing system.