Cooperativity in Oxidation Reactions Catalyzed by Cytochrome P450 1A2

Rabbit liver cytochrome P450 (P450) 1A2 was found to catalyze the 5,6-epoxidation of α-naphthoflavone (αNF), 1-hydroxylation of pyrene, and the subsequent 6-, 8-, and other hydroxylations of 1-hydroxy (OH) pyrene. Plots of steady-state rates of product formation versus substrate concentration were hyperbolic for αNF epoxidation but highly cooperative (Hill n coefficients of 2-4) for pyrene and 1-OH pyrene hydroxylation. When any of the three substrates (αNF, pyrene, 1-OH pyrene) were mixed with ferric P450 1A2 using stopped-flow methods, the changes in the heme Soret spectra were relatively slow and multiphasic. Changes in the fluorescence of all of the substrates were much faster, consistent with rapid initial binding to P450 1A2 in a manner that does not change the heme spectrum. For binding of pyrene to ferrous P450 1A2, the course of the spectra revealed sequential changes in opposite directions, consistent with P450 1A2 being involved in a series of transitions to explain the kinetic multiphasicity as opposed to multiple, slowly interconverting populations of enzyme undergoing the same event at different rates. Models of rabbit P450 1A2 based on a published crystal structure of a human P450 1A2-αNF complex show active site space for only one αNF or for two pyrenes. The spectral changes observed for binding and hydroxylation of pyrene and 1-OH pyrene could be fit to a kinetic model in which hydroxylation occurs only when two substrates are bound. Elements of this mechanism may be relevant to other cases of P450 cooperativity.

P450 4 enzymes are some of the most versatile catalysts found in nature (1). These hemoproteins mainly catalyze mixed function oxidations, using the general cycle shown in Fig. 1, but can use similar chemistry at the heme iron prosthetic group to do reductions, rearrangements of unstable oxygenated chemicals, and several other reactions (3). Collectively, the thousands of reactions catalyzed by the mammalian P450s can be categorized as using sterols, fatty acids, fat-soluble vitamins, and xenobiotic chemicals as substrates (4).
Most reactions catalyzed by P450s can be described with hyperbolic, classical Michaelis-Menten kinetics. However, examples of heterotropic cooperativity (i.e. one ligand enhancing the rate of oxidation of another (5)) were reported in the 1970s (6 -9), and in the early 1990s reports of homotropic cooperativity (e.g. sigmoidal and other atypical v versus S plots (5) began to appear (10,11). Examples of in vivo evidence for heterotropic cooperativity are known (12,13). Homotropic cooperativity is difficult to demonstrate in vivo, but the process can be documented in isolated hepatocytes (14).
Explanations and hypotheses have been presented for (human) P450 3A4, perhaps the most commonly reported example of P450 cooperativity, and also other P450s for which cooperative behavior has been reported, particularly (human) P450s 2C8, 2C9, 2B6, and 1A2 (4,(15)(16)(17)(18). Most of these P450s that have been crystallized have large cavities for ligand binding (19 -21). A full discussion of the field is beyond the scope of this introduction and the reader is referred to some of the more recent reviews (4,17,18). Some of the major features of the more common proposals include the following: (i) a classic allosteric reaction with a second ligand-binding site controlling binding to the catalytic substrate-binding site (22,23); (ii) cooccupancy of the cognate substrate-binding site with multiple ligands, leading to a more productive rearrangement of the substrate for catalysis because of either steric packing or more specific protein-ligand interactions (11,24,25); (iii) selective populations of P450 (in slow equilibrium) that interact in different ways (26); (iv) multimeric states of P450 (27) and possibly interactions with accessory proteins (28) that are involved; and (v) elements of an induced fit process, in which the protein adapts a conformation as a result of an initial interaction with a ligand (29). Some of these possibilities are not mutually exclusive. Deficiencies in this area are the small extent of cooperativity seen in many cases (i.e. low n values in Hill plots), the often limited number of experimental data points, and the over interpretation of models based on limited data.
P450 1A2 is an important enzyme, not so much in regard to physiological function per se but because of its role in the oxidation of many drugs (30) and carcinogens, particularly aryl and heterocyclic amines (31). A recent crystal structure of the protein bound to the inhibitor ␣NF provides insight into how this protein binds substrates (32). The active site is somewhat larger than necessary for the ligand ␣NF to fit, but the accessible volume is considerably less than that available in other mammalian P450s, e.g. 2C8, 2C9, and 3A4 (19 -21). P450 1A2 also exhibits cooperativity in some reactions, at least with regard to homotropic behavior. In the case of O-dealkylation of 1-alkoxy-4nitrobenzenes, rabbit P450 1A2 showed patterns that could be described mathematically as negative cooperativity (33). However, a more appropriate fit of the results involves two sets of k cat and K m values. Evidence that these phenomena are the result of concurrent occupancy of the P450 (1A2) by two ligands comes from a study with two ligands, a 1-alkoxy-4-nitrobenzene substrate and the iron ligand 1,4-phenylene diisocyanide, which elicits a heme spectral perturbation opposite to that of the "type I" substrate, i.e. a "type II" change involving an increase in A 420 and decrease in A 390 . A key finding was that the extrapolated K d values for P450 1A2 binding of 1-alkoxy-4nitrobenzenes were perturbed when measured in the presence of 1,4-phenylene diisocyanide and vice versa (33). However, these studies do not reveal details of the mode of interaction.
One approach we have utilized to study the cooperativity of P450 3A4 is time-resolved analysis of events in ligand binding, i.e. kinetic studies (34,35). These studies have revealed complex, multiphasic, and slow events that are considered relevant to issues in cooperativity, and evidence has been presented that these events have their basis in sequential as opposed to parallel events (34,35). These complex events are not seen in the binding of ligands to "simpler" mammalian P450s, e.g. P450 2A6 (36), or bacterial P450 101A1 (37) or P450 105D5 (38). We studied rabbit P450 1A2 because its heme spectral properties permit a number of studies that cannot be done with the human ortholog, which is isolated exclusively in the high spin iron state (33,39). We found a very strong positive cooperativity pattern for the oxidation of pyrene, unprecedented in P450 literature. Pyrene also proved to be a very useful ligand because of its fluorescence properties. Pre-steady-state kinetic analyses were done using UV-visible, fluorescence, and CD spectroscopy, and comparisons were made with ␣NF (a substrate that does not show apparent cooperativity). The availability of the crystal structure of a human P450 1A2-␣NF complex provided a basis for modeling ligands in rabbit P450 1A2, which may be relevant to cooperative phenomena.
Enzymes-P450 1A2 was purified from liver microsomes prepared from ␤-naphthoflavone-treated New Zealand White rabbits using a two-step ion-exchange chromatography procedure (40), as modified (39). The preparations used in the work contained Ͻ13% cytochrome P420, an inactive form. NADPH-P450 reductase (rat) was expressed in Escherichia coli and purified as described elsewhere (41).
Measurement of Enzyme Activity-Substrates were incubated with a reconstituted enzyme system that generally included 0.25 M P450 1A2, 0.50 M NADPH-P450 reductase, 23 M L-␣-1,2-dilauroyl-sn-glycero-3-phosphocholine (dispersed by sonication before use), and the substrate of interest, generally in a reaction volume of 1.5 ml. All incubations were in 100 mM potassium phosphate buffer (pH 7.4) (the enzyme precipitates at lower ionic strength). Stock concentrations of substrates were prepared in CH 3 OH, and the final concentration of CH 3 OH in reactions was Յ1% (v/v). The reported solubility of pyrene in H 2 O is 0.65 M (42). The solubility under the experimental conditions used here is probably higher, with a 1% (v/v) concentration of organic solvent (and dispersion in some cases with the phospholipid vesicles and possibly the proteins). Catalytic studies with polycyclic aromatic hydrocarbons, including pyrene and particularly benzo[a]pyrene, have often been done at relatively high concentrations (50 -80 M (43-46)). However, we generally limited the concentration in the spectroscopic experiments to a low micromolar range to avoid artifacts.
Incubations (at 37°C) were initiated by the addition of an NADPH-generating system (45) and generally run for 2 min (pyrene), 1.5 min (1-OH pyrene), or 20 min (␣NF). Reactions were stopped by the addition of 2.0 ml of CH 2 Cl 2 and mixed with a vortex device, and 1.5 ml of the organic layer was removed and dried under an N 2 stream for HPLC analysis.
Pyrene and 1-OH pyrene reaction products were dissolved in CH 3 OH and separated by HPLC using the following program at a flow rate of 1.0 ml min Ϫ1 (solvent A, 90% 10 mM NH 4 CH 3 CO 2 buffer (pH 6.8), 10% CH 3 OH (v/v); solvent B, 100% CH 3 OH): 0 -20 min, linear gradient from 45 to 80% B; 20 -21 min, linear gradient to 100% B; 21-25 min, 100% B (isocratic). Separation of products was monitored using a McPherson FL-750 BX fluorescence detector (Chelmsford, MA), with excitation at 240 nm and emission monitored using a 380 nm long pass filter, as well as by UV scanning (UV3000 detector, see above, in series) from 230 to 400 nm (49). After the relevant spectra had been accumulated, individual wavelength traces were used subsequently.
Benzo[a]pyrene hydroxylation assays, in which mainly 3-hydroxylation is measured, were done using the basic fluorescence extraction method of Nebert and Gelboin (43), with slight modification (45). 3-OH benzo[a]pyrene (Midwest Research Laboratories, Kansas City, MO) was used as a standard, with standardization of the stock solution using an extinction coefficient from Raha (50).
Characterization of Di-OH Pyrene Products-An enzyme system of 0.25 M P450 1A2, 0.50 M NADPH-P450 reductase, phospholipid (see above), and 50 M 1-OH pyrene (total volume 285 ml) was incubated for 10 min under the same conditions as described above, and the products were extracted. HPLC conditions were altered slightly; the following program was used with a flow rate of 1.0 ml min Ϫ1 (solvent A, 90% H 2 O, 10% CH 3 OH (v/v); solvent B, 100% CH 3 OH); 0 -30 min, linear gradient from 45 to 70% B). Seven sequential HPLC injections (20 l each) were made, and the four product fractions (peaks 1-4) were collected, based on monitoring by UV absorbance at 275 nm ( Fig. 2A). The fractions from each injection were combined, extracted with CH 2 Cl 2 , and dried using a rotary evaporator and then under an N 2 stream. The four fractions were further dried in vacuo, using a vacuum pump, in a dessicator with P 2 O 5 for 5 h at room temperature. Each sample was then redissolved in d 6 -acetone and transferred to a 2.5-mm NMR tube under an argon atmosphere for NMR spectroscopy analysis, where one-dimensional, COSY, HSQC, and HMBC 1 H ( 13 C) NMR spectra were obtained. Peaks 1 and 4 were then spiked with 60 nmol of CH 3 CO 2 H as an internal standard in d 6 -acetone; 300 nmol of CH 3 CO 2 H was added to peaks 2 and 3. The integration of the resulting singlet at 1.96 ppm was set at H ϭ 3 to allow quantitation of the di-OH pyrenes. The fluorescence emission spectrum of each of the fractions was recorded in the NMR solvent (excitation at 240 nm). The fractions were then dried and redissolved in CH 3 OH, and the absorbance spectra (200 -600 nm) were recorded.
Anaerobic Experiments-Samples were deaerated in all-glass cuvettes or tonometers using a vacuum/argon manifold system described earlier (51) and modified (52). Contents of tonometers were transferred into the drive syringes of the stoppedflow spectrophotometer as described (52). Anaerobic titrations were done in a glass cuvette/titrator assembly using techniques described previously (53).
Spectroscopy-UV-visible spectra were recorded with either a Cary14/OLIS, an Aminco DW2/OLIS, or a Hewlett-Packard 8452A/OLIS diode array spectrophotometer (On-Line Instrument Systems, Bogart, GA). Steady-state spectral titrations of ferric P450 1A2 with ␣NF and pyrene were carried out as described previously (54). In the case of the 1-OH pyrene titrations, a different procedure was used because of the inherent absorbance of the ligand in the near-UV region. Tandem (Yankeelov) cuvettes were used. At each concentration of 1-OH pyrene, each of two cuvettes contained 1.0 ml of 2.0 M P450 1A2 in one chamber and twice the desired final concentration of 1-OH pyrene in the other side. A base-line spectrum was recorded in the OLIS-Cary 14 spectrophotometer (mean of duplicate scans). The contents of the two chambers of the sample cuvette were then mixed; the reference cuvette was not mixed. The resulting difference spectrum was recorded (mean of duplicate scans).
Fluorescence spectra were recorded with either an OLIS DM-45 (On-Line Instrument Systems) or a Fluorolog-3 spectrofluorometer in the 111 configuration (JY Horiba, Edison, NJ). Steady-state fluorescence spectra were acquired at room temperature, with slit widths corresponding to a 1-3 nm band pass, depending on fluorophore concentration. Either CH 3 OH or luminescence grade propanol was used to dissolve ligands. For the steady-state titration experiments, either the emission (at a fixed wavelength) or the excitation (at a fixed emission wavelength) scans were recorded.
CD spectra were recorded in 1-mm cuvettes with a J-810 instrument (Jasco, Easton, MD) in the Vanderbilt Center for Structural Biology. In the case of the CD scans, P450 1A2 was incubated with substrate for 15 min before taking measurements; kinetic scans were initiated as soon as substrate was added. The absorbance in all CD studies was Ͻ0.20. CD output, in millidegrees, was converted to standard mean residue ellipticity, [] (in degrees cm 2 dmol Ϫ1 ), using the relationship shown in Equation 1, where signal is the CD output in millidegrees, c is the protein concentration in millimolar, n is the number of amino acid residues, and l is the cell path length in centimeters. Stopped-flow absorbance and fluorescence kinetic measurements were made using an OLIS RSM-1000 instrument (On-Line Instrument Systems) as described in detail previously (35). Absorbance (UV-visible) measurements were made in the rapid-scanning mode (scanning 10 3 spectra s Ϫ1 ); kinetic parame-ters were derived from traces of absorbance (versus time) at individual wavelengths and fit to single, bi-, or tri-exponential equations using the manufacturer's software. Kinetic fluorescence measurements were made with the OLIS RSM-1000 instrument using long pass or band pass filters (Oriel, Stratford, CT) attached to the photomultiplier tube and utilized a direct kinetic mode, with fitting of data as described for absorbance. In most cases, 1.24-mm slits (8 nm bandwidth) were used. Data are presented as either "scatter" plots or "connected" signals versus time. Generally, four to eight individual stopped-flow experiments were averaged before subsequent data analyses.
Mass and NMR Spectrometry-Liquid chromatography-MS-MS spectra were acquired using an Acquity UPLC system (Waters, Milford, MA) with the same octadecylsilane (C 18 ) column (6.2 mm ϫ 80 mm, 3 m) as described (see above) connected to a Finnigan LTQ mass spectrometer (ThermoFisher Scientific, Waltham, MA) and operated in the electrospray ionization negative ion mode. Reaction products were redissolved in CH 3 OH and kept at 4°C prior to injection with an autosampler, whereas the column was kept at ambient temperature. In addition to an ion scan from m/z 100 to 500, two selective ion monitoring scans at m/z 217 and 233 were taken, which triggered MS-MS fragmentation events. Product ion MS-MS scans were acquired from m/z 100 to 300. The following liquid chromatography conditions were used, with a 375-l min Ϫ1 flow rate (solvent A, 90% 10 mM NH 4 CH 3 CO 2 buffer (pH 6.8), 10% CH 3 OH (v/v); solvent B, 100% CH 3 OH): 0 -30 min, linear gradient from 45 to 100% B. Electrospray ionization conditions were as follows: source voltage 4 kV, source current 100 A, auxiliary gas flow rate setting 20, sweep gas flow rate setting 5, sheath gas flow setting 34, capillary voltage Ϫ26 V, capillary temperature 350°C, tube lens voltage Ϫ138 V. MS-MS conditions were as follows: normalized collision energy 35%, activation Q setting 0.250, and activation time 30 ms. NMR experiments were acquired using either a 14.0-tesla Bruker magnet (Bruker, Billerica, MA) equipped with a Bruker AV-II console operating at 600.13 MHz in the case of peaks 2, 3, and 4 or an 11.7 tesla Bruker magnet equipped with a Bruker DRX console operating at 500.13 MHz in the case of peak 1. All spectra were acquired in 2.5-mm NMR tubes using a Bruker 5-mm TCI cryogenically cooled NMR probe. Chemical shifts were referenced internally to acetone (7.1 ppm), which also served as the 2 H lock solvent. For onedimensional 1 H NMR, typical experimental conditions included 32 K data points, 13 ppm sweep width, a recycle delay of 1.5 s, and 128 -512 scans, depending on sample concentration. For two-dimensional 1 H-1 H COSY spectra, experimental conditions included 2048 ϫ 512 data matrix, 13 ppm sweep width, recycle delay of 1.5 s, and 32 scans per increment. The data were processed using a squared sinebell window function, symmetrized, and displayed in magnitude mode. Multiplicity-edited HSQC experiments were acquired using a 1024 ϫ 256 data matrix, a J(C-H) value of 145 Hz, which resulted in a multiplicity selection delay of 34 ms, a recycle delay of 1.5 s, and 24 scans per increment, along with GARP decoupling on 13 C during the acquisition time (150 ms). The data were processed using a p/2 shifted squared sine window function and displayed with CH/CH 3 signals phased positive and CH 2 signals phased negative. J 1 (C-H) filtered HMBC experiments were acquired using a 2048 ϫ 256 data matrix, a J(C-H) value of 9 Hz for detection of long range couplings resulting in an evolution delay of 55 ms, J 1 (C-H) filter delay of 145 Hz (34 ms) for the suppression of one-bond couplings, a recycle delay of 1.5 s, and 96 scans per increment. The HMBC data were processed using a p/2 shifted squared sine window function and displayed in magnitude mode.
Analysis of Kinetic and Binding Data-As indicated above, pre-steady-state data were initially fit to exponential equations in the OLIS software. In some cases binding rate measurements were made at several ligand concentrations and fit to plots of rate versus ligand concentration to estimate k on and k off values. For complex fits, the program DynaFit (55) was used to fit data to possible mechanisms.
Steady-state kinetic data (rate versus substrate concentration) were fit to hyperbolic or sigmoidal plots, in the latter case developed as a Hill plot of the form shown in Equation where S denotes substrate, and n is an exponent used for fitting but has no direct physical meaning. Fitting was done with the program GraphPad Prism (GraphPad, San Diego, CA). The DynaFit program was also utilized with sigmoidal steady-state kinetics. Some of the ligand binding data were fit to the quad- A the maximum change, E the enzyme concentration, and X the ligand concentration (the form used in the program was Y ϭ Homology Modeling and Docking-A recently published crystal structure of a human P450 1A2-␣NF complex (32) was used as the template to develop a homology model of rabbit P450 1A2 using the SWISS-MODEL automated homology modeling tool (56). Coordinate files for molecules of pyrene were generated using the Dundee PRODRG2 server (57) and manually docked into the binding cavity (replacing ␣NF) using Turbo Frodo. 5 Simulated annealing was then performed in CNS Solve 1.1 using the input file model anneal (59,60). The heme group was fixed during simulated annealing. The energy-minimized models were not adjusted following annealing.

Characterization of Oxidation Products of ␣NF and Pyrene-
Although ␣NF is generally considered as an inhibitor of P450 family 1 enzymes (6, 61), we found that it was slowly oxidized by rabbit P450 1A2. The only product detected using HPLC was the 5,6-epoxide (partly recovered as the 5,6-dihydrodiol (44)), which was identified by t R and UV comparison with the product 5 System is from Ref. 58. isolated from reactions with P450 3A4 (previously established by NMR methods) (25). 6 Oxidation of another fluorescent substrate, pyrene, yielded five (fluorescent) products (Fig. 2, A and B). 1-OH pyrene was identified by co-chromatography, and its identical UV and mass spectra were compared with commercial material. The time course of the products (Fig. 2C) suggested that peaks 1-3 might be products of 1-OH pyrene, and an experiment using 1-OH pyrene as substrate confirmed this hypothesis (Fig. 2D).
MS established that the other four products were di-OH pyrenes (supplemental Fig. S1). A reaction with a limited pyrene concentration indicated a rise-fall relationship for 1-OH pyrene, supporting its role as an intermediate product (Fig. 2C). 1-OH pyrene was shown to be converted to all of these ( Fig. 2, B and D). UV and fluorescence spectra were also recorded (supplemental Figs. S2 and S3). The sites of hydroxylation ( Fig. 3) were established using a series of proton NMR experiments (supplemental Fig. S4). The last of the four eluted peaks (peak 4) was obtained in trace amounts, contaminated with peak 3, and the structure was not identified. The compounds in peaks 2 and 3 were readily identified by their NMR patterns, which were both simplified because of the symmetry in the molecules. Furthermore, the NMR and UV spectra of peaks 2 and 3 matched those of 1,6 and 1,8 di-OH pyrene reported in the literature (62,63).
Identification of peak 1 as 1,5 di-OH pyrene should be considered tentative. Peak integration of the one-dimensional 1 H NMR (supplemental Fig. S4A) shows seven protons in the aromatic region, not including the hydroxyl protons. However, all nonsymmetric di-OH pyrenes should have eight protons, so the absent proton is of concern because the MS data clearly revealed a molecular ion corresponding to a di-OH pyrene (supplemental Fig. S1A). These spectral data eliminate a symmetrical di-substituted pyrene (i.e. 1,3, 1,6, or 1,8 di-OH pyrene). Absence of a singlet and the presence of a triplet in the one-dimensional 1 H NMR spectrum preclude 1,2 and 1,7 di-OH pyrene, respectively. Furthermore, 1,4, 1,9, and 1,10 di-OH pyrene were tentatively eliminated based on HSQC spectra analysis and correlation information provided by COSY spectra. Thus peak 1 is tentatively assigned as 1,5 di-OH pyrene, with some caveats regarding missing HMBC correlations and the absence of one of the expected eight protons in the aromatic region.   MARCH 14, 2008 • VOLUME 283 • NUMBER 11

JOURNAL OF BIOLOGICAL CHEMISTRY 7297
Steady-state Kinetics of P450 1A2 Reactions-A plot of the rate of 5,6-epoxidation versus ␣NF concentration yielded a typical hyperbolic plot (Fig. 4A). In contrast to the plots of v versus S for ␣NF (Fig. 4A) and many other rabbit P450 1A2 substrates (64 -66), the plots for pyrene 1-hydroxylation were consistently very sigmoidal (Fig. 4B). When the data were fit to a Hill expression (Equation 2) the apparent n values were Ն3. When the oxidations of 1-OH pyrene to the di-OH products were analyzed, the v versus S plots were sigmoidal (Fig. 4C), with n values of 2-3.
In other studies with (recombinant) human P450 1A2, the same oxidations of ␣NF and pyrene were observed, but the rates were ϳ5-fold slower. The plot was also sigmoidal for pyrene 1-hydroxylation (supplemental Fig. S5).
Steady-state Binding of Ligands to Ferric P450 1A2-Rabbit P450 1A2, as isolated, contains a mixture of low and high spin heme iron (39,67). Binding of some substrates further shifts the equilibrium to the high spin form (39), a so-called type I shift (54). (Recombinant human P450 1A2 is isolated nearly completely in the high spin form and is not amenable to such analysis (33,39).) The preparation used in this work was 77% low spin, as established by second-derivative analysis of the ferric Soret peak (68, 69) (supplemental Fig. S6).
Titration of P450 1A2 with either ␣NF (supplemental Fig.  S7), pyrene, or 1-OH pyrene yielded such spectral changes ( Pre-steady-state Kinetics of Ligand Binding to Ferric P450 1A2, Absorbance-Pre-steady-state kinetic analysis was used to address the question of whether the spin-state change (Fig. 5) is the first event observed upon binding of substrates and other ligands. Preliminary analyses with equimolar concentrations of P450 1A2 and pyrene yielded a type I change (A 390 increase, A 424 decrease), fit to an estimated second order rate of ϳ4 ϫ 10 5 M Ϫ1 s Ϫ1 (Fig. 6A). Binding of the known substrate 1-isopropoxy-4-nitrobenzene (33), under pseudo-first order conditions, yielded a trace that could be best fit to a bi-exponential equation with rates of 2.9 and 0.08 s Ϫ1 (Fig. 6B). Binding of ␣NF under pseudo-first order conditions yielded a plot that fit a bi-exponential equation with rates of 9.8 and 1.2 s Ϫ1 (Fig. 6C). 7 Collectively the rates of these changes are rather slow to describe initial binding of ligands to proteins (70). Furthermore, when the ␣NF experiment (Fig. 6C) was repeated with the same reagents and the fluorescence quenching of ␣NF was observed, the plot was biphasic with the fast phase accounting for most of the decrease, at a rate of 50 s Ϫ1 , Ͼ5-fold faster than the absorbance change (see more detailed analyses, see below). These results imply that the heme Soret changes are secondary to faster processes that occur when P450 1A2 binds ligands. The binding of 1-OH pyrene (20 M) also yielded biphasic kinetics with rates (5.7 and 0.23 s Ϫ1 ) similar to those observed using ␣NF and 1-isopropoxy-4-nitrobenzene (Fig. 6D).

Steady-state Interaction of Pyrene with Ferric P450 1A2 and
Excimer Formation-Steady-state fluorescence emission spectra of pyrene, in solution with or without P450 1A2, were acquired by optical pumping at either 275 or 338 nm, two excitation wavelengths especially effective in generating pyrene excimer emission at ϳ 466 nm. For excimer emission to occur, the two pyrenes must be in close proximity and base-stacked (71,72). P450 1A2 quenched the fluorescence of pyrene in a concentration-dependent manner (Fig.  7A). With an assumption that one molecule of pyrene is bound to each P450 1A2, a quadratic fit yielded an estimated K d of 56 Ϯ 1 nM (cf. Fig.  5C). If two pyrenes are bound per P450 under these conditions, then the expression for K d is more complex, but nevertheless the affinity of P450 1A2 for pyrene is strong, and the fluorescence is highly quenched.
A titration of P450 1A2 with pyrene, the opposite of the approach used in Fig. 7A, revealed further features (Fig. 7B). In this experiment, fluorescence emission was collected at either 370 or 485 nm (recording excitation scans), corresponding to monomer and excimer fluorescence, respectively, and the intensities of the fluorescence excitation bands at ϳ 335 nm from either monomer or pyrene were quantified. The ratio of fluorescence emission at 370 to 485 nm is related to the pyrene monomer to excimer ratio (m/e) (73), as well as to the amount of quenching (Fig. 7B). Because of the quenching of pyrene fluorescence by P450 1A2, it is useful to use this ratio instead of the actual fluorescence values. In the absence of P450 the pyrene dimerizes with increasing concentrations when titrated into the buffer solution. In the titration with P450 1A2, at lower concentrations of pyrene, the pyrene (monomer) fluorescence appears to be preferentially quenched. The amount of fluorescent monomer increases with increasing pyrene concentration. After addition of ϳ0.6 pyrene/P450 1A2, the ratio of monomer/excimer decreases, indicative of an increase in the contribution of a dimer of pyrene (Fig. 7B).
Pre-steady-state Kinetics of Ligand Binding to Ferric P450 1A2, Fluorescence-Preliminary results indicated that the quenching of the fluorescence of ligands occurred faster than the changes in the heme spectrum (see above). Accordingly the kinetics of these processes were analyzed in more detail.   MARCH 14, 2008 • VOLUME 283 • NUMBER 11

JOURNAL OF BIOLOGICAL CHEMISTRY 7299
The quenching of ␣NF fluorescence seen upon mixing with P450 1A2 fit a bi-exponential plot (Fig. 8A). As the ␣NF concentration was increased, a second order rate constant could be estimated to be 6.5 ϫ 10 6 M Ϫ1 s Ϫ1 (results not presented). 8 This rate constant is much faster (Ͼ10-fold) than observed for the UV-visible changes (Fig. 6C).
The reaction of pyrene with P450 1A2 showed a very rapid decrease in the monomer fluorescence (Fig. 8B). The rate of 70 s Ϫ1 at a pyrene concentration of 2 M suggests a second order rate constant of ϳ10 6 -10 7 M Ϫ1 s Ϫ1 , similar to that estimated for ␣NF (Fig. 8A).
As the monomer fluorescence decreased, a small increase in the apparent excimer (dimer) fluorescence occurred and then this slowly decreased (Fig. 8C). One interpretation of this time course is a sequence of pyrene monomer binding to protein, the formation of a dimer, and then the movement of the dimer to interact with protein residues to decrease the excimer fluorescence. Alternatively, the protein may be binding a pre-formed dimer from solution, and the fluorescence properties may be changing as it moves within the protein. We favor the latter explanation because at higher concentrations of pyrene (10 -20 M), an initial decrease in Ͼ455 nm emission (to ϳ1 s) was followed by an increase in this fluorescence and then a decrease (5-10-s period).
The binding of 1-OH pyrene to P450 1A2 was also biphasic (Fig. 8D), with the fast phase estimated at ϳ 85 s Ϫ1 using a 1-OH pyrene concentration of 15 M (slower phase ϳ3 s Ϫ1 ). Analysis of the rates of the fast phase as a function of the 1-OH pyrene concentration yielded an apparent k on of ϳ3 ϫ 10 6 M Ϫ1 s Ϫ1 (plot not presented). This rate is much faster than that of the Soret absorbance change (ϳ3 s Ϫ1 at 15 M; Fig. 6D), which corresponds approximately to the slower fluorescence phase (Fig. 8D). To our knowledge, 1-OH pyrene does not form excimers (at concentrations relevant to these experiments), and none of the steady-state fluorescence spectra suggested such a phenomenon.
Interactions of Pyrene with Ferrous P450 1A2-Although the catalytic cycle for P450 1A2 is usually depicted in the general manner shown in Fig. 1, the point can be made that the events should not necessarily be considered in a linear, sequential manner (36). P450 1A2 is reduced at the same rate in the absence or presence of substrates (52), and the binding and release of substrate at the ferrous level (and possibly others) is possible and probably happens with many P450-substrate combinations (36).
Pyrene binds to ferrous P450 1A2 and produces a change in the heme spectrum (Fig. 9A). Anaerobic titration yielded an apparent K d value of 2.3 (Ϯ0.8) M (Fig. 9B). Kinetic analysis of the binding showed a rapid increase in A 426 (Fig. 9, C and D), followed by a slower bi-exponential decrease in A 426 (Fig. 9, E and F).
The kinetic course of the reaction is of significance, in that it distinguishes between multiple populations of enzyme (in slow equilibrium or aggregational states) participating in the same reaction at different rates, in contrast to an apparent single population of enzyme undergoing a sequential series of reactions with a substrate. The pattern observed here with sequential trends in opposite directions is clearly more consonant with the latter view of multiple sequential reactions.
CD Measurements of Interaction of Ligands with Ferric P450 1A2-The addition of any of four ligands (pyrene, 1-OH pyrene, ␣NF, or 1-isopropoxy-4-nitrobenzene) to P450 1A2 produced a significant change (7-29%, depending on the ligand) in the CD spectra (UV region, ϳ220 nm). The change may be due in part to a decrease in the ␣-helicity of the protein, as judged by the changes at 220 nm, although such changes proved too rapid to be measured by CD stopped-flow techniques (74, 75) (data not presented).
Docking of Ligands in a Model of P450 1A2-The human P450 1A2 crystal structure (32) allows a homology model of the rabbit enzyme to be generated using the Swiss Model program. The feasibility of binding of multiple ligands to the rabbit P450 1A2 model active site was assessed by manually docking the ligands and then performing simulated annealing of the pro- 8 The binding model is probably more complex than a simple two-state (bound/unbound) situation, in that the apparent k off (20 s Ϫ1 ) and k on yield the ratio K d ϭ k off (20 s Ϫ1 )/k on (6.5 ϫ 10 6 M Ϫ1 s Ϫ1 ) of ϳ3 M, which is not consistent with the low K d value estimated from the steady-state absorbance results (supplemental Fig. S7). tein-ligand complex. Superimposing the human P450 1A2 crystal structure (Protein Data Bank code 2HI4) containing the heme group and ␣NF with the rabbit P450 1A2 model places the ␣NF moiety directly in the active site of the homology model (32) (Fig. 10A). The homology model is almost identical to the experimentally derived structure (root mean square deviation 0.097), which is not surprising given the sequence similarity of the two enzymes (75% (77)). It is of interest to note that the only site of ␣NF that is oxidized (5,6 bond; Figs. 3 and 4A) is positioned furthest from the heme iron in both the published human P450 1A2 structure (32) and our own model (Fig. 10A), and presumably this is not a catalytically competent configuration. Two pyrene molecules were docked in the active site of the homology model. Although two pyrene molecules can clearly fit in the active site of the rabbit homology model, stacking of the two molecules does not appear to be favorable. Rather, an end-on-end conformation, with some overlap, is less perturbing to the homology model used here (Fig. 10B). By trying several different starting conformations, a model was obtained in which two pyrene molecules stack in excimer fashion (i.e. completely stacked), but this orientation causes some disruption of the I-helix and may not represent a viable complex. More likely, the formation of any fluorescent excimer may be transient in nature, which would be consistent with the relatively weak fluorescence signal (Figs. 7B and 8C). The overall binding interactions observed in the human P450 1A2-␣NF co-crystal structure are largely maintained in the model containing two pyrene molecules, with van der Waals interactions comprising most of the enzyme-substrate contacts (Fig. 10C). Other notable similarities include the potential for orthogonal and parallel aro-matic interactions between the pyrenes and the phenylalanine side chains of residues 125 and 226, which are essentially identical to those reported previously with ␣NF (32).
The docking studies (Fig. 10, B and C) suggested that two benzo-[a]pyrene molecules could readily fit into the P450 1A2 active site. An assay of benzo[a]pyrene hydroxylation, using a fluorimetric assay (which detects mainly 3-hydroxylation (44,45,78)) showed highly cooperative behavior (Fig. 11). The low rate is similar to that reported previously with this enzyme at single benzo[a]pyrene concentrations (78 -80).
Simulation of Binding and Steadystate Hydroxylation Kinetics-A variety of spectral methods was used to monitor interactions of P450 1A2 in this work. The interactions with pyrene are of particular interest, and we developed a kinetic model that could explain the UV-visible changes that occur upon binding (Figs. 5A and 6A). Kinetic model building was also done with hydroxylation rates. Although a Hill expression (Equation 2) can be used to fit the steady-state kinetic data, it does not provide a mechanistic explanation for the behavior. The goal was a model, having a minimal number of steps, that could adequately fit three diverse sets of data: (i) pre-steady-state absorbance changes observed upon the binding of pyrene to P450 1A2 (Fig. 6A); (ii) the sigmoidal v versus S plots observed for pyrene 1-hydroxylation (Fig. 4B); and (iii) the temporal pattern of pyrene depletion, 1-OH pyrene formation, and depletion, and formation of di-OH pyrenes (experiment of Fig. 2C done with a higher concentration of pyrene). The model also requires a low K d value for the first pyrene binding event (Fig. 5A) and a reasonably low K d value for 1-OH pyrene binding (Fig. 5B).
All modeling was done with the program DynaFit (55). Initial modeling was done with the pyrene 1-hydroxylation v versus S data set (Fig. 4B). The strongly sigmoidal nature of the points presented a challenge, and a reasonable solution was a model with two substrates with only the dimeric complex (ESS) being catalytically active. The phenomena are reminiscent of the case of a bacterial NO reductase, which was concluded to only, reduce when two (NO) molecules are in the active site (81). The behavior of the enzyme could be described by Equation 3, and a quadratic expression (5) as shown in Equation 4, which reduces to Equation 5, when K 1 Ͼ Ͼ K 2 (81). Thus, one solution to the modeling of the v versus S points is to use a high K d (ϳ8 M) for binding of the first pyrene molecule, followed by a very low K d for the addition of the second ligand, i.e. Ͻ1 M, to produce the sharp rise in the rate following the initial binding event. Such a model has elements of classical positive cooperativity (82). However, this model is inconsistent with the observed tight initial binding of a single pyrene (Fig. 5,  A and C). Another potential solution is the use of a 3-pyrene model, with tight binding of the first pyrene, loose binding of a second, and a tight binding of a third pyrene, with only the  (52), with an NADPH-generating system (45). The base line was set to zero throughout the wavelength range with ferrous P450 1A2, and changes were set against this base line. The arrows show the direction of changes with increasing pyrene concentration. B, absorbance changes from A fit to a quadratic equation with K d ϭ 2.3 Ϯ 0.8 M. C, P450 1A2 (5 M) was mixed with 1 M NADPH-P450 reductase (and 75 M L-␣-dilauroyl-sn-glycero-3-phosphocholine) and reduced anaerobically with an NADPH-generating system. In the experiments shown, the final concentration of pyrene (after mixing) was 20 M. Difference spectra were collected every 1 ms and are shown for 5-ms intervals up to 41 ms. D, ⌬A 426 data were fit to a biexponential expression (absorbance increase, then decrease) with k 1 ϭ 106 Ϯ 11 s Ϫ1 and k 2 ϭ 1.0 Ϯ 0.3 s Ϫ1 . E, latter phases of reaction after mixing. The first spectrum shown was collected at 80 ms, and subsequent spectra were collected every 2.4 s. F, data from E were fit to a biexponential decrease in A 426 with k 2 (corresponding to second part in B) ϭ 0.66 Ϯ 0.11 s Ϫ1 and k 3 ϭ 0.041 Ϯ 0.002 s Ϫ1 .
trimer serving as substrate. Such a model gave reasonable sigmoidal fits (results not presented) but seems inconsistent with the spatial docking models (Fig. 10), although conceivably the third pyrene could reside outside of the substrate region (i.e. on the periphery of the protein, see below). However, in the absence of any direct physical evidence in its favor, a 3-pyrene model should be considered too speculative at this point.
A model could also be readily developed with two pyrenes, with the first pyrene binding tightly and the second more loosely, and this model generated sigmoidal plots (Fig. 12). The model was considered for its ability to also fit the presteady-state binding data. A "k on " rate constant of 4 ϫ 10 6 M Ϫ1 s Ϫ1 was used, based upon the fluorescence quenching work (e.g. Fig. 8B). 9 Using a typical extinction coefficient for the ⌬A 390 -A 420 observed upon substrate binding to a P450 (6.5 ϫ 10 4 M Ϫ1 cm Ϫ1 ) (36), the absorbance trace could be fit, but only if an equilibration event was added following the binding of the second pyrene (this model fit even with equal initial concentrations (2 M) of enzyme and pyrene, as well as with excess pyrene). The importance of adding an equilibration event (Fig. 12A) was seen in the comparison of Fig. 12B with the same model without such a step (results not presented). In the model the initial complex produces no absorbance change but the ESS and SES (rearranged) complexes do.
When the ESS % SES equilibrium step was added to the initial mechanism developed for the v versus S data (see above), it did not change the character of the sigmoidal plot (Fig. 12C), although the rate of the hydroxylation of 1-OH pyrene was reduced (k 6 in Fig. 12A). Both the v versus S (Fig. 12C) and ⌬A 390 -A 420 versus time (Fig. 12B) fits showed some sensitivity to the rate of binding of pyrene in the models, although rates of 2-4 ϫ 10 6 M Ϫ1 s Ϫ1 could be used. A model with only one substrate was unsatisfactory in that hyperbolic fits were always produced regardless of the rate constants (results not shown). Interestingly, the sigmoidal character of the plots required two distinct K d values (k Ϫ1 /k 1 and k Ϫ2 /k 2 ) for 9 We used an apparent binding rate constant of 4 ϫ 10 6 M Ϫ1 s Ϫ1 in our kinetic modeling and analysis, which is based on typical rates of interaction of P450s and small ligands measured by fluorescence quenching (Fig. 8). We have used such rate constants previously in analysis with P450 3A4 (34,35).
For further discussion of "on" rates for proteins, see Fersht (70). The rate constant is not intended to be a diffusion-controlled limit because, as discussed (70), only a fraction of initial hits between a protein and a ligand are productive even in the sense of quenching the fluorescence of a ligand, because of electrostatics and geometry. The rate constants of the order of ϳ5 ϫ 10 6 M Ϫ1 s Ϫ1 are consistent not only with the P450 1A2 (Fig. 8) and other P450 binding (34, 35) estimated by fluorescence interaction but also some heme Soret changes measured with P450s that appear to have simple binding phenomena (36, 37).  pyrene (S) binding but was not dependent on the conversion of 1-OH pyrene (P) to di-OH pyrenes (Q).
The final set of boundaries to the model was fitting of the temporal pattern of changing pyrene, 1-OH pyrene, and di-OH pyrene concentrations in solution starting only with pyrene as substrate (Fig. 2C). The fits were not unreasonable, although the steady-state level of 1-OH pyrene calculated from the model was only about one-half the measured concentration (Fig. 12D). Raising k 4 in the model (Fig. 12A) improved this difference but also produced a rate that was too fast in the v versus S plots (Fig. 12C).
The final model (Fig. 12A) is presented, along with the fits to each set of data (Fig. 12, B-D). Scripts and other details of the programs are presented in supplemental Figs. S8, S9, and S10).
The low K d values (spectroscopic binding) of pyrene (Fig. 5, A and C) and 1-OH pyrene (Fig. 5, B and D) are consistent with the model (Fig. 12A), assuming that the presence of a pyrene does not affect the binding of 1-OH pyrene (ESP versus EP complexes).

DISCUSSION
Spectroscopy-Several spectroscopic methods were utilized in the kinetic studies, and these reveal different features of the system. CD spectroscopy provided some evidence for conformational changes, possibly in ␣-helicity, upon ligand binding, and the time frame appeared too rapid to be captured by stopped-flow techniques (results not shown). The changes in the fluorescence of ligands (Figs. 7 and 8) are valuable in that they provide evidence for an interaction of the ligands (pyrene, 1-OH pyrene, and ␣NF) with P450 1A2 prior to the UV-visible changes associated with the heme iron spin-state transition, as in the case with P450 3A4 and its ligands (34,35). The UVvisible changes are also useful regarding the multiphasic behavior and provide insight into the relative rates of individual steps, following the initial Soret-invisible kinetic event. The UV-visible spectral changes seen in the binding of pyrene with ferrous P450 1A2 (Fig. 9) are of particular relevance in that the multiphasic changes are in a series of opposite directions and distinguish a sequential reaction sequence from parallel reactions of two slowly interconverting enzyme populations.
Our work leaves some uncertainty about the course of events related to pyrene excimers. The situation is complex because excimers form in solution (as well as proposed in the active site, which are difficult to distinguish spectroscopically) and binding to P450 1A2 could quench the fluorescence of monomers or excimers. The course of the titration in Fig. 7B can be interpreted in the context of two pyrenes being inserted sequentially into the active site (Figs. 10 and 13), partially overlapping ( Fig.  10B) to produce the transient excimer fluorescence (Fig. 8C), and then moving (Fig. 13) to decrease the excimer fluorescence. Alternatively, the kinetic course of events (Fig. 8, B and C) could be explained by binding of a dimer from solution, movement into the active site, and then changes in the fluorescence as a function of the interactions with individual amino acids. This scenario may not seem as consistent with the titration results (Fig. 7). However, some other interpretations of the fluorescence data cannot be ruled out at this time because of the equilibria between monomer and excimer populations in both free solution and the protein (i.e. Fig. 3) and the lack of knowledge about the fluorescent properties of the bound species in each form. One issue is the concentration dependence of the mechanism, i.e. at higher pyrene concentrations a preformed dimer may bind whereas at lower concentrations a dimer could form in the active site. These possibilities would yield similar results in most of the kinetic modeling.
Overall, the kinetic results clearly indicate a rapid ligand binding process revealed by the fluorescence studies (Fig. 8) and at least a biphasic process following that, observable with UVvisible methods. The kinetics of the events appear to vary with individual ligands (Fig. 8). The course of events in the interaction of pyrene with ferrous P450 (Fig. 9) is indicative of a sequential process, which is proposed to be operative in the other systems.
Kinetics and Modeling-The kinetic fitting work (Fig. 12) is consistent with the UV-visible results for pyrene binding (Fig.  6A), the cooperative v versus S pattern for pyrene 1-hydroxylation (Fig. 4B), and the time course of conversion of pyrene to 1-OH pyrene and di-OH pyrenes (Fig. 2, C and D). We should emphasize that this is a minimum kinetic model, and the true course of the binding interaction may involve more steps, e.g. additional conformational changes or rapid steps following slow ones that may not be revealed. The scheme presented in Fig. 12A is adequate to describe the course of the UV-visible changes seen in the Soret spectra (Fig. 12B). Attempts to include a conformational change following binding of the first ligand were not successful, but if a step were fast enough this step would not be seen. A schematic model, presented in Fig.  13, does have a step following initial binding, in which the ligand translocates into the active site. This step seems obvious but could either be fast or could be coupled with the binding of the second ligand. Two paths diverge in Fig. 13. With a larger substrate, e.g. ␣NF, only one molecule fits, and conformational changes occur after site occupation (Fig. 13, step 3a). With the smaller substrate pyrene (or benzo[a]pyrene), a second molecule enters (Fig. 13, step 3b), and finally conformational changes occur with this dimer-occupied P450 (step 4).
The kinetic model presented in Fig. 12A is intended to be a minimal model, i.e. the simplest model that can be used to rationalize a set of data (83). The condition was used that a single model had to be capable of explaining the following: (i) the biphasic kinetics of pyrene binding to P450 1A2 (Fig. 6C); (ii) the sigmoidal v versus S plot for 1-hydroxylation of pyrene (Fig. 4B); and (iii) the steady-state time course of oxidation of pyrene to 1-OH pyrene and then the di-OH pyrenes (Fig. 2, C  and D), plus be consistent with: (iv) the rate of interaction of pyrene with P450 (ϳ4 ϫ 10 6 M Ϫ1 s Ϫ1 , Fig. 8A); (v) the tight unimolecular binding of pyrene to P450 (Fig. 5, A and C); and (vi) tight binding of 1-OH pyrene to P450 (Fig. 5, B and D). The 6-step model presented in Fig. 12A satisfies these criteria reasonably well. The model may contain additional steps, and a kinetic model cannot necessarily exclude all alternative possibilities. The sigmoidal nature of the fit to the v versus S data in Fig. 12C is not ideal, but nevertheless the fitted line (from DynaFit) gives a Hill plot with an n value of 1.81 (Ϯ0.01) and S 50 value of 9.9 (Ϯ0.01) M. A shortcoming of the modeling is the prediction of the experimentally measured steady-state level of 1-OH pyrene (Fig. 12D). The latter could be adjusted by increasing k 4 but this change rendered v too fast in Fig. 12C (v versus S). In the modeling, we used the ESS complex as the FIGURE 13. Scheme depicting proposed events in ligand binding to ferric P450 1A2. See text (under "Discussion") for more discussion.
Step 1, ligand L first interacts with P450 1A2 at a peripheral site, and the binding quenches the fluorescence of ligand (Fig. 8).
Step 2, L is translocated to the interior of the protein, where it can interact with the heme and produce the difference spectrum (Fig. 6).
Step 3a, if there is only space available for one molecule of L in the active site (e.g. ␣NF), a conformational change in the P450 occurs.
Step 3b, If L is small enough for two molecules (of L) to occupy the active site (e.g. pyrene or 1-OH pyrene), a second molecule of L can enter the active site (via the same path as in steps 1 and 2).
Step 4, conformational change of the P450 following occupancy with the second molecule of L.
catalytically competent one, instead of SES (Fig. 12A). In principle, either or both of these complexes could be catalytic, and the mathematical solutions should be equivalent, although k 3 , k -3 , and possibly other rate constants would need adjustment.
One deficiency of our knowledge is that we do not have information as to whether a pyrene is still present in the P450 1A2⅐1-OH pyrene complex that forms di-OH pyrenes (Q in the modeling, see Fig. 12A). The modeling was done with the pyrene (S) still present, although a model with the pyrene removed should give similar results although having one more step (also, in the model of Fig. 12A, the dissociation of di-OH pyrene (Q) was ignored in that Q and ESQ were considered together in Fig. 12D). In additional modeling (not shown), the mechanism of Fig. 12A could be readily adapted to fit the hydroxylation of 1-OH pyrene (v versus S data, see Fig. 4C) by setting S to represent 1-OH pyrene, P to represent di-OH pyrenes, and eliminating k 6 . We also approached the issue of whether a heteromeric P450⅐pyrene⅐1-OH pyrene complex could explain the observed kinetics or if a P450 1A2⅐1-OH pyrene complex was required. Modeling showed that a P450⅐pyrene⅐1-OH pyrene complex is catalytically competent to form di-OH pyrenes and a dimeric (1-OH pyrene) 2 ⅐P450 complex yielded fits to di-OH pyrene formation (Fig. 12D) that were far too slow (Ͼ20-fold, results not shown).
In summarizing the kinetic modeling, the major features that were cited for this model are as follows: (i) two pyrene molecules bound; (ii) catalytic activity only with two bound pyrenes; (iii) K d,2 Ͼ Ͼ K d,1 ; and (iv) an equilibration step occurring after binding of the second pyrene. Other features are probably less critical. As mentioned earlier, we could fit the sigmoidal v versus S data set with mathematical equations either having K d,1 Ͼ K d,2 (81) or K d,1 Ͻ K d,2 (Fig. 12A). Only the latter is consistent with the spectral binding results (Fig. 5, A and C), and the kinetics of the change in absorbance (Fig. 12B) fit well even at a low pyrene concentration. The same conclusion about the relative affinities of P450 3A4 for binding two testosterone molecules (K d,1 Ͻ K d,2 )was reached by Roberts et al. (84) using different approaches.
Structural Models-Analysis of the structural models indicates that the active site of rabbit or human P450 1A2 is only large enough for a single ␣NF molecule, but two pyrenes can be readily accommodated (Fig. 10). The ␣NF configuration is clearly a noncatalytic one, in that the only site of oxidation (5,6 bond) is the most distant from the iron atom (32). With pyrene, the exact orientations of the two molecules docked in the active site are uncertain. In Fig. 10 (B and C), a slight overlap could easily be accommodated without distortion of any side chains. With some movement of the I-helix, a complete pyrene-pyrene stack could be achieved in the models. Whether the fluorescence spectra represent a slightly overlapped pyrene pair or a low population of highly overlapped pyrenes is not known in the absence of more data.
At this point we are unable to explain the site of hydroxylation (C-1) of pyrene (as opposed to C-2 or -4), in that docking in the model is uncertain. Another point to be made is that the information available is not sufficient to make predictions of sites of oxidation of other substrates for P450 1A2. As pointed out above, even when a crystal structure is available the site most likely to be oxidized may not be predicted (32).
The hyperbolic nature of the kinetics of oxidation of ␣NF and the cooperativity of pyrene and 1-OH pyrene oxidations may be rationalized in terms of the size of the P450 1A2 active site, which holds only one ␣NF but two of the other molecules (Fig.  10). However, our ability to predict the cooperative behavior of other ligands is still limited. One can ask the question of why phenacetin O-deethylation is apparently not cooperative (at least for human P450 1A2) (85). Why are the homotropic cooperativity patterns of pyrene and the 1-alkoxy-4-nitrobenzenes rather opposite in their appearance (Fig. 4B) (33)? One possible explanation may be that the results are related to the flat planar character of the polycyclic aromatic hydrocarbons, in that pyrene and benzo[a]pyrene both showed homotropic cooperativity (Figs. 2B and 11).
Cooperativity-Although pyrene 1-hydroxylation is highly cooperative with rabbit (Fig. 4B) and human (supplemental Fig.  S5) P450 1A2, the related enzymes (human) P450 1A1 and 1B1 have been reported to exhibit classical hyperbolic behavior (46). P450 3A4-catalyzed pyrene 1-hydroxylation has been reported to show a cooperative behavior of a type opposite to that here (Fig. 4B), i.e. described with a low K m /high K m fit (not sigmoidal) (28), and the patterns were altered in the presence of cytochrome b 5 . Benzo[a]pyrene, which contains one more aromatic ring than pyrene, has been studied extensively in the context of chemical carcinogenesis (86). The docking studies (Fig. 10, B and C) suggested that two benzo[a]pyrene molecules could readily fit into the P450 1A2 active site. A study of benzo-[a]pyrene hydroxylation, using a fluorimetric assay (which detects mainly 3-hydroxylation (45)) showed cooperative behavior (Fig. 11). Surprisingly, examination of the literature did not yield any plots of benzo[a]pyrene oxidation as a function of substrate concentration by rabbit or other P450 1A2 enzymes.
The question can be raised as to how much the results of this cooperativity work with P450 1A2 extend to other P450s that have been shown to have weaker cooperative behavior (e.g. P450s 3A4, 2C9, and 2B6). Interestingly, the active site of P450 3A4 is known to be very large (20,21,29). Interactions among multiple ligands have been invoked as a mechanism for cooperativity with P450 3A4 (11), and stronger fluorescence evidence for the presence of pyrene excimers in the protein has been presented (73). However, in the 1-hydroxylation of pyrene by P450 3A4 only weak homotropic cooperativity was observed, with an apparent Hill coefficient (n) of ϳ1.7 in some cases, i.e. with cytochrome b 5 (28,87). One possibility is that some cases of positive and negative cooperativity in the P450s can be explained by models such as that described in Fig. 12A, and the variation in the degree of cooperativity can be understood in terms of the extent of the differences between the k cat and K d (or at least K m ) values for activity with substrate monomer and dimer, e.g. in this case a 100-fold difference between K d,1 and K d,2 in the model yielded an apparent n value of 1.8 in the Hill plot (see above).
Another point to be made is that reports of weak cooperativity must be viewed cautiously. In cases of weak cooperativity, the results are heavily dependent upon the accurate measure-ment of experiments done using low concentrations of substrate (or ligand). These measurements are the most sensitive to error. Another issue is that points obtained with low substrate concentrations are sensitive to depletion of substrate, product inhibition, etc., and in our own experience, artifactual plots can easily be obtained under conditions of too prolonged incubation. In the literature in this field it is not uncommon to find plots with few or even no points at less than one-half maximal activity (or signal) and subsequent deconvolution of limited data to complex models with three ligands etc.
Conclusions-We report what appears to be the most cooperative of P450 reactions characterized to date. The results are consistent with a kinetic model for selective catalytic activity for two substrate molecules. Structural models are consistent with the hypothesis that hyperbolic oxidation of ␣NF is related to single occupancy of the active site as opposed to the highly cooperative kinetics seen with two substrate molecules in the active site. The kinetics of interactions of the enzyme with several ligands are multiphasic, with slow steps, and the kinetics of interaction with pyrene can be fit to a two-ligand kinetic model. Some of the findings with this highly cooperative P450 system may extend, with modification, to other cooperative P450s.