Identification of the Interactions between Cytochrome P450 2E1 and Cytochrome b5 by Mass Spectrometry and Site-directed Mutagenesis*

The reaction cycles of cytochrome P450s (P450) require input of two electrons. Electrostatic interactions are considered important driving forces in the association of P450s with their redox partners, which in turn facilitates the transfer of the two electrons. In this study, the cross-linking reagent, 1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride (EDC), was used to covalently link cytochrome P450 2E1 (CYP2E1) with cytochrome b5 (b5) through the formation of specific amide bonds between complementary charged residue pairs. Cross-linked peptides in the resulting protein complex were distinguished from non-cross-linked peptides using an 18O-labeling method on the basis that cross-linked peptides incorporate twice as many 18O atoms as non-cross-linked peptides during proteolysis conducted in 18O-water. Subsequent tandem mass spectrometric (MS/MS) analysis of the selected cross-linked peptide candidates led to the identification of two intermolecular cross-links, Lys428(CYP2E1)-Asp53(b5) and Lys434(CYP2E1)-Glu56(b5), which provides the first direct experimental evidence for the interacting orientations of a microsomal P450 and its redox partner. The biological importance of the two ion pairs for the CYP2E1-b5 interaction, and the stimulatory effect of b5, was confirmed by site-directed mutagenesis. Based on the characterized cross-links, a CYP2E1-b5 complex model was constructed, leading to improved insights into the protein interaction. The described method is potentially useful for mapping the interactions of various P450 isoforms and their redox partners, because the method is relatively rapid and sensitive, and is capable of suggesting not only protein interacting regions, but also interacting orientations.

Cytochrome P450s (P450s) 2 are a superfamily of b-type hemoproteins responsible for the metabolism of a wide variety of exogenous compounds such as drugs and carcinogens, and endogenous compounds such as prostaglandins and steroids (1). P450 reactions require input of two electrons (supplemental Fig. S1) (1,2). The efficiency of electron transfer is one of the key determinants of the reaction kinetics. In microsomal systems, NADPH is the ultimate source of the two electrons, and NADPH-dependent cytochrome P450 oxidoreductase (P450 reductase) together with cytochrome b 5 (b 5 ) facilitates the electron transfer. Knowledge of the interactions between P450s and their redox partners is fundamental to a complete understanding of the mechanisms of P450 reactions.
The interactions between P450s and b 5 have drawn much attention because of variable effects b 5 has on different P450 isoforms and P450 reactions. It has been shown that b 5 may stimulate, inhibit or have no effects on P450-catalyzed reactions depending on the particular isoform of P450 and the substrate of the reaction (3)(4)(5). However, there is no consensus on whether b 5 transfers electrons to P450, or causes an allosteric effect on P450, or whether both mechanisms are simultaneously operative (6 -10). In addition, the mechanism of substrate-and P450 isoform dependency is unknown (5,11).
CYP2E1 is a P450 isoform whose reactions are highly stimulated in the presence of b 5 . For example, we previously found that b 5 stimulates CYP2E1-catalyzed oxidation of acetaminophen to its toxic metabolite, N-acetyl-p-benzoquinone imine (NAPQI), by 25-fold (12). For other CYP2E1-catalyzed reactions, such as aniline p-hydroxylation and 7-ethoxycoumarin O-deethylation, b 5 stimulates the reactions by 270-fold and 67-fold, respectively (5,11). In contrast to its stimulating effects on CYP3A4, CYP3A5, CYP2C19, CYP2B6, and CYP2C8, apo-b 5 is unable to replace holo-b 5 in stimulating CYP2E1-catalyzed reactions (11,12). The requirement for the heme group increases the probability that b 5 stimulates CYP2E1-catalyzed reactions by facilitating electron transfer rather than by only causing a positive allosteric effect. Because intermolecular complex formation immediately precedes electron trans-* This work was supported by National Institutes of Health Grant GM32165 fer (13), the identification of the protein interacting regions and orientations in the CYP2E1-b 5 complex is an important goal in understanding the effects of b 5 on CYP2E1-catalyzed reactions.
It has been shown, by site-directed mutagenesis and chemical modification, that several cationic residues on the proximal face of P450, and several anionic residues on b 5 surrounding the solvent-exposed heme edge, are important for the functional interaction between the two proteins (4,14,15). However, the protein interacting surfaces have not been fully characterized, and the protein interacting orientations have never been determined. One complex model of b 5 and CYP101, a microbial P450, was proposed by Sligar and co-workers (16) based on visual optimization of the intermolecular electrostatic interactions and minimization of the distance between the redox centers of the two proteins. However, the protein interacting orientations in this model have not been substantiated by experiments. In this study, a complex model of b 5 and CYP2E1, a microsomal P450, is proposed based on two chemical crosslinks characterized using mass spectrometry.
Because electrostatic interactions are considered the major driving forces for the P450-b 5 interaction (4,14,15), a watersoluble cross-linking reagent, 1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride (EDC), was chosen to covalently link CYP2E1 with b 5 . EDC generates "zero-length" cross-links (amide bonds) between basic (Lys) and acidic (Asp or Glu) residues that come into very close proximity (supplemental Fig. S2) (17). The short length of EDC cross-links generally leads to specific intra-and intermolecular linkages without sampling multiple protein orientations (17). EDC was previously used to form a cross-link between CYP2B4 and b 5 (18). However, this complex was not structurally characterized.
To identify the cross-links in the CYP2E1-b 5 complex, the complex was digested and the generated peptides were isotopically labeled. It has been noted that during trypsin-catalyzed proteolysis, two oxygen atoms from solvent are incorporated into the ␣-carboxyl group of a peptide C-terminally ending with a lysine or an arginine residue (19). When the proteolysis is conducted in fully enriched 18 O-water, the generated peptides are labeled with two 18 O atoms at their C termini ( Fig. 1) (19,20). Recently, 18 O-labeling has been used to identify cross-linked peptides in a digest mixture (21,22). Through the comparison of peptide ions generated from two digestions, one conducted in 16 Owater and the other in 18 O-water, cross-linked peptides can be distinguished from non-cross-linked peptides by virtue of incorporating more than two 18 O atoms during proteolysis. The structures of the selected cross-linked peptide candidates are subsequently characterized by tandem mass spectrometric (MS/MS) analysis.
Here we describe the characterization of two cross-links in the CYP2E1-b 5 complex using 18 O-labeling and mass spectrometry. The importance of the identified ion pairs in the interacting proteins was confirmed by site-directed mutagenesis. Finally, a model that sheds light on the CYP2E1-b 5 interaction was constructed.
Cloning of Histidine-tagged Recombinant Human CYP2E1-The vector construct pCWhum3A4(His) 6 (10) was the source of the pCWoriϩ expression vector for cloning the histidinetagged CYP2E1 expression vector construct. The cDNA of human CYP2E1 (24) was used as the template for PCR. Amplifications were performed using Platinum Pfx DNA polymerase. Reactions were assembled and heated to 94°C for 3 min, followed by 30 cycles of denaturation at 94°C for 45 s, annealing at 56°C for 45 s, and extension at 72°C for 1 min. Cycling was followed by a final extension at 72°C for 5 min. DNA from the  JULY 21, 2006 • VOLUME 281 • NUMBER 29 reactions was purified using a Qiagen PCR purification kit and digested with NdeI and SalI. The digests were then electrophoresed on a 1.1% agarose gel and the CYP2E1 amplicons of expected size were gel-purified using a Qiagen Geneclean kit. The pCWhum3A4(His) 6 plasmid was digested with NdeI and SalI to enable gel-purification of the linearized plasmid from which the CYP3A4 insert had been removed. The vector was treated with calf intestinal phosphatase and the CYP2E1 fragment was ligated to the vector to generate the histidine-tagged CYP2E1 expression vector construct. Ligation reactions were used to transform DH5␣ cells. Positive clones containing the desired inserts were verified initially by colony PCR and restriction analysis, and finally by DNA sequencing using the dideoxy chain termination method.

Identification of Interactions between CYP2E1 and b 5
Site-directed Mutagenesis of CYP2E1-The cloned plasmid pCWhum2E1(His) 6 was used as the template for amplification reactions with the QuikChange mutagenesis kit (Stratagene, La Jolla, CA) according to the manufacturer's protocol. Oligonucleotide primers used in the generation of K428A and K434A single mutants plasmids were as follows (mismatches indicated by the underlined bases): K428A forward, 5Ј-GTGACTATTTCGCGC-CATTTTCCAC-3Ј; K428A reverse, 5Ј-GTGGAAAATGGCGC-GAAATAGTCAC-3Ј. K434A forward, 5Ј-AAGCCATTTTCCA-CAGGAGCACGAGTGTGTGCTGGAGAA-3Ј; K434A reverse, 5Ј-TTCTCCAGCACACACTCGTGCTCCTGTGGAAAATG-GCTT-3Ј. A double mutant of pCWhum2E1(His) 6 containing both the K428A and the K434A replacements was constructed using the K434A single mutant plasmid and the K428A forward and reverse primers. DpnI-digested DNA was transformed into XL1-Blue cells, and DNA from several of the resulting colonies was isolated. The cDNA sequence was analyzed for the presence of the desired mutations and the absence of extraneous mutations (University of Washington Sequencing Facility).
Protein Expression and Purification-A single isolated colony of histidine-tagged CYP2E1 was used to inoculate 10 ml of LB-ampicillin media, which was cultivated with shaking at 37°C overnight and then diluted 1:100 in Terrific Broth containing 100 mg of ampicillin liter Ϫ1 , 1.0 mM thiamine, and trace elements (25). The cultures were shaken (180 rpm) at 37°C until the A 600 reached 0.4, after which isopropyl-1-thio-␤-Dgalactopyranoside (1.0 mM) and ␦-ALA (0.5 mM) were added, and the cultures were shaken (160 rpm) at 28°C for 36 h. Cells were harvested by centrifugation at 5000 ϫ g (4°C, 15 min), resuspended in storage buffer (50 mM KP i , pH 7.4, 20% glycerol, and 0.5 mM EDTA), pooled into 50-ml Falcon tubes, and recentrifuged at 4000 ϫ g (4°C, 30 min). The supernatant was discarded, and the cell pellets were resuspended in resuspension buffer (100 mM Tris-HCl, pH 7.4, 20% glycerol, with the addition of 1 ml of protease inhibitor mixture per liter of the initial culture volume). After the addition of lysozyme (5 mg/liter), the culture was stirred at 4°C for 1 h. Cells were homogenized and spun at 150,000 ϫ g. The pellets were resuspended in the resuspension buffer by homogenization and stirred at 4°C for 1 h after the addition of 1% Emulgen 911. After centrifugation at 150,000 ϫ g for 25 min, imidazole was added to the red/orange supernatant to a final concentration of 20 mM. The supernatant was applied to a Ni-NTA agarose column that had been pre-equilibrated with 15 column volumes of equilibrium buffer (50 mM KP i , pH 7.4, 20% glycerol, 0.5 M KCl, 0.05% sodium cholate, 50 M ␣-NF, and protease inhibitors). The column was washed with 20 column volumes of washing buffer A (50 mM KP i , pH 7.4, 20% glycerol, 40 mM imidazole, 0.05% sodium cholate, 0.02 mM DTT, and protease inhibitors). CYP2E1 enzyme was eluted with elution buffer A (50 mM KP i , pH 7.4, 20% glycerol, 350 mM imidazole, and 0.02% sodium cholate). The eluted fractions were dialyzed against dialysis buffer A (100 mM KP i , pH 7.4, 20% glycerol, 0.5 mM EDTA, and 0.1 mM DTT). The dialysate was applied slowly (20 ml/hr) to a hydroxyapatite column (1.5 ϫ 3 cm) and washed with 5 column volumes of washing buffer B (50 mM KP i , pH 7.4, 20% glycerol, 0.5 mM EDTA, and 0.1 mM DTT). The CYP2E1 enzyme was eluted with elution buffer B (300 mM KP i , pH 7.4, 20% glycerol) and dialyzed against dialysis buffer B (50 mM KP i , pH 7.4, 20% glycerol).
Rat b 5 expression plasmid was kindly provided by Dr. Ronald W. Estabrook and expressed in BL21-DE3 cells using previously described conditions (23). Expression and purification of P450 reductase was accomplished as previously described (26).
Cross-linking Reactions-All enzymes used for cross-linking reactions were dialyzed against dialysis buffer B. CYP2E1, b 5 (human b 5 or rat b 5 ) and DLPC were reconstituted with molar ratio of 1:1:500. The solution was gently stirred for 10 min and held at room temperature for 2 h. EDC was added to 8 mM final concentration from a 100 mM stock. The reaction was allowed to proceed at room temperature for 2 h.
Proteolytic Digestions-For in-gel proteolysis, the cross-linking reaction was quenched by the addition of an equal volume of 2 ϫ SDS loading buffer containing 200 mM DTT. SDS-PAGE was performed, and the protein gels were stained by either Coomassie Blue or copper II chloride dihydrate. Copper staining was carried out by soaking the gel in 300 mM CuCl 2 . When the desired degree of opacity was reached, the staining solution was removed, and the gel was kept in water. The band of interest was excised and washed three times with destaining buffer (25 mM Tris-HCl, 192 mM glycine, pH 8.3). After destaining, the gel piece was washed with 500 l of 100 mM ammonium bicarbonate buffer (pH 8.5) for 10 min, dehydrated in 500 l of acetonitrile at room temperature for 15 min, and dried in a Speed-Vac for 20 min. Subsequently, the gel was rehydrated with 300 l of 100 mM ammonium bicarbonate buffer containing 10 mM DTT, and incubated at 56°C for 50 min to reduce the disulfides. Then the gel piece was washed with 500 l of 100 mM ammonium bicarbonate buffer (pH 8.5) for 10 min, dehydrated in 500 l of acetonitrile at room temperature for 15 min, and dried in the SpeedVac for 20 min. To alkylate cysteine residues, the gel piece was rehydrated with 200 l of 100 mM ammonium bicarbonate buffer containing 60 mM iodoacetamide and incubated in the dark at room temperature for 50 min. The gel piece was cut into two halves and both pieces were then washed with 500 l of 100 mM ammonium bicarbonate buffer (pH 8.5) for 10 min, dehydrated in 500 l of acetonitrile at room temperature for 15 min and dried in the SpeedVac. The dried gel pieces were rehydrated with two digestion solutions (50 mM ammonium bicarbonate, pH 8.5, with sequencing grade trypsin at an enzyme:substrate ratio of 1:25 (w/w), prepared with 16 O-and 18 O-water respectively. The digestion was allowed to proceed at 37°C for 24 h, and the reaction was quenched with 0.1% trifluoroacetic acid.
For post-proteolysis (a proteolysis step conducted in 16 Owater followed by a post-digest labeling step carried out in 18 Owater), subsequent to trypsin digestion in 16 O-water, the peptide mixture was dried completely in the SpeedVac. Digestion solution prepared with 18 O-water was then added, and the oxygen exchange reaction was allowed to proceed at 37°for 24 h. The reaction was quenched with 0.1% trifluoroacetic acid.
For in-solution proteolysis, the cross-linking reaction was quenched by the removal of EDC through dialysis against dialysis buffer B. Glycerol was removed by a second dialysis against dialysis buffer C (50 mM KP i , pH 7.4). The sample was dried completely in the SpeedVac and resuspended in 6 M urea, 100 mM Tris-Base, pH 8.0, to yield a protein concentration ϳ2 mg/ml. 50 l of the protein sample was transferred to another microcentrifuge tube and reduced by adding 10 l of 100 mM Tris-Base, pH 8.0, containing 100 mM DTT. The reaction was carried out at room temperature for 1 h. Subsequent alkylation reactions were initiated by adding 30 l of 100 mM Tris-Base, pH 8.0, containing 500 mM iodoacetamide. The reactions were allowed to proceed in the dark at room temperature for 1 h. Samples were subsequently diluted with 50 mM ammonium bicarbonate and centrifuged using an Ultrafree-4 Centrifugal Filter Unit (Millipore, Billerica, MA). The dilution and centrifugation steps were repeated three times. Samples were then split into two equal aliquots, which were dried in the SpeedVac. The two dried peptide samples were reconstituted in two digestion solutions (prepared with 16 O-and 18 O-water respectively). The digestion was allowed to proceed at 37°C for 24 h, and the reaction was quenched with 0.1% trifluoroacetic acid.
ESI-QTOF MS Analysis-Protein mass spectra were recorded on an API-US quadrupole/time-of-flight (QTOF) mass spectrometer (Micromass, Manchester, UK). Protein samples were injected on a 300 m i.d. ϫ 5 cm perfusion column, packed with 20 m POROS R2 particles (PerSeptive Biosystems, Framingham, MA), operated at a flow rate of 20 l/min and interfaced on-line with the QTOF mass spectrometer. Instrument parameters were as follows: source temperature, 100°C; N 2 drying gas, 50 liters/hr; electrospray voltage, 3.8 kV; and cone voltage, 60 V. Data acquisition was carried out from m/z 800 -2400 using a 2.4 s scanning time. The gradient elution profile was set as follows: 5% solvent B for 2 min and 5-90% solvent B over the next 5 min. (solvent A, 5% acetonitrile, 0.1% trifluoroacetic acid; solvent B, 95% acetonitrile, 0.1% trifluoroacetic acid).
Peptide digests were analyzed using the QTOF mass spectrometer equipped with a CapLC system (Waters, Milford, MA). The stream select module was configured with a 5 mm ϫ 300 m i.d. trap column packed with 5-m C 4 particles (LC Packings, San Francisco, CA) connected by a ZU1XC metallic union (Valco, Houston, TX) to a 20 cm ϫ 75 m i.d. nanoscale analytical column packed in-house with 5-m Jupiter C18 particles (Phenomenex, Torrance, CA) using the method described by Kennedy and Jorgenson (27). Peptide samples were injected onto the trap column at 10 l/min, cleaned-up and back-flushed to the analytical column at 0.5 l/min using gradient elution. Binary gradients of 5-60% solvent B were generated over 30 min, followed by 60% B for 5 min and 60 -90% B for 5 min (solvent A, 3.3% acetonitrile, 1.7% 2-propyl alcohol, and 0.1% trifluoroacetic acid; solvent B, 63.3% acetonitrile, 31.7% 2-propyl alcohol, and 0.1% trifluoroacetic acid). QTOF parameters were set as follows: electrospray potential, 3.6 kV; cone voltage, 35 V; and source temperature, 100°C. The instrument was operated at a mass resolving power of 6,000. For MS/MS, the scan time was set to 2 s, the precursor isolation width set to 4 Da, and the collision energy set to 25-45 eV according to the m/z of the precursor and the charge state.
IT-FT-ICR MS Analysis-Peptide digests were analyzed by electrospray ionization in the positive ion mode on a hybrid ion trap-Fourier transform ion cyclotron resonance mass spectrometer (Thermo Electron Corp., San Jose, CA). Nanoflow HPLC was performed using a similar approach to that described by Yi et al. (28). The electrospray voltage was applied via a liquid junction using a platinum wire inserted into a micro-tee union (Upchurch Scientific, Oak Harbor, WA). Ion source conditions were as follows: ESI voltage, 1.4 kV; capillary temperature, 200°C; capillary voltage, 44 V; and tube lens voltage, 180 V. All other voltages were optimized using a tuning solution composed of caffeine (Sigma), MRFA (Bachem, King of Prussia, PA) and Ultramark 1621 (Lancaster Synthesis, Windham, NH). Injection waveforms for the LTQ-FT ion trap and ICR cell were kept on for all acquisitions. ICR resolution was set to 50,000 (m/z 400). ICR ion populations in the ICR cell were held at 1e6 and 5e5 for MS and MS/MS, respectively. For MS/MS, the precursor isolation width was set to 10 Da and the collision energy set to 40 and 55% for quintuply and quadruply charged precursor ions, respectively.
Acetaminophen (APAP) Oxidation Catalyzed by CYP2E1-Assays of APAP oxidation metabolites were performed with slight modifications to a previously described procedure (12). Briefly, purified CYP2E1 was reconstituted with recombinant rat P450 reductase, purified human cytochrome b 5 , DLPC, DOPC, and DLPS at a molar ratio of 1:2:2:250:250: 250, respectively. The reconstituted complex was dialyzed twice against 1000 volumes of 50 mM potassium phosphate buffer, pH 7.4, for 4 h at 4°C to remove residual glycerol, which was in the storage buffer. The incubation mixture containing 0.1 M CYP2E1, selected concentrations of APAP (0.15, 0.3, 0.6, 1.2, 2.4, 4.8, 9.6, 15, and 19.2 mM), 10 mM GSH, and 50 mM potassium phosphate buffer (pH 7.4) was preincubated for 3 min at 37°C. The reaction was initiated by the addition of an NADPH-generating system (0.4 mM NADP ϩ , 10 mM glucose 6-phosphate, and 0.4 units of glucose-6-phosphate dehydrogenase). The final incubation volume was 100 l. The assay was conducted at 37°C for 10 min and terminated by addition of 10 l of 33% sulfosalicylic acid (w/v). APAP metabolite 3-glutathione-S-yl-APAP (GS-APAP) was quantified on an Agilent 1100 series system. A Hewlett-Packard 1049A electrochemical detector was connected to the UV detector in tandem and was set at a constant voltage of 0.6 V. Separations were performed on a 3.5-m Zorbax SB-C18 column (4.6 mm ϫ 15 cm, Agilent). The isocratic mobile phase was composed of 25 mM ammonium phosphate buffer (pH 5.3) containing 10% methanol. A solvent flow rate of 1.0 ml/min was used. Nonlinear regression analysis was performed using the software GraphPad Prism 3.0 (GraphPad Software Inc., San Diego, CA) with the Michaelis-Menten equation fitted to the kinetic data.
p-Nitrophenol (PNP) Hydroxylation Catalyzed by CYP2E1-Reconstituted CYP2E1 was prepared as described above. The incubation mixture containing 30 pmol of CYP2E1, 500 M PNP, and 50 mM potassium phosphate buffer (pH 7.4) was preincubated for 3 min at 37°C. The reaction was initiated by adding NADPH (1 mM final concentration). In all cases the final incubation volume was 250 l. The assay was conducted at 37°C for 10 min and terminated by the addition of 8 l of trifluoroacetic acid. The formation of p-nitrocatechol was quantified using the same HPLC column and system as used for the APAP oxidation assay. The electrochemical detector was set at a constant voltage of 0.7 V. The isocratic mobile phase was as previously described (29) and was composed of 40 mM sodium phosphate buffer (pH 2.6) containing 1 mM heptane sulfonic acid, 80 M sodium EDTA, and 20% methanol. A solvent flow rate of 1.0 ml/min was used. Determination

Characterization of Protein Molecular Masses-Molecular
masses of the investigated proteins were measured using an ESI-QTOF mass spectrometer. The measured values matched the predicted values (supplemental Figs. S3, S4, and S5), which provided quality assurance for the protein samples used in the following cross-linking experiments.
Cross-linking Reactions-First, the cross-linking reaction was carried out with equimolar CYP2E1 and human b 5 . Based on the measurement of the protein molecular masses, the molecular mass of a 1:1 complex (CYP2E1-human b 5 ) is ϳ70 kDa, and the molecular mass of a 1:2 complex (CYP2E1-(human b 5 ) 2 ) is ϳ85 kDa. Both complexes were absent in the blank sample, a mixture of equimolar CYP2E1 and human b 5 without the addition of the cross-linking reagent (Fig. 2, A and B, lane 3). Treatment of the mixture with EDC resulted in the formation of both complexes (Fig. 2, A and B, lane 2) in the Coomassie Blue-stained gel. The yields of the 1:1 and 1:2 complexes were ϳ35 and 10% with respect to CYP2E1 ( Fig. 2A). The 1:2 complex was visible after the protein gel was destained for 2 h ( Fig. 2A), and was invisible after 12 h (Fig.  2B), whereas the 1:1 complex remained visible even after the protein gel was destained for 24 h. Subsequent increases in the amount of human b 5 to two, three and four times that of CYP2E1 did not generate additional detectable complexes in the protein gel. Therefore, the generation of cross-linked species with one molecule of CYP2E1 binding to more than two molecules of human b 5 is unfavorable. CYP2E1 was found to interact with rat b 5 in the same way as it did with human b 5 .
In-gel proteolysis and subsequent MS/MS analysis were conducted with all gel bands containing the complexes. The presence of both CYP2E1 and b 5 polypeptide chains in the complexes was verified by data base searching with MASCOT (Matrix Science, London, UK).  Isotopic Labeling and Mass Spectrometric Analysis of the Peptides-To achieve maximal incorporation of 18 O atoms in the peptides, in-gel proteolysis, post-proteolysis, and in-solution proteolysis were conducted as described under "Experimental Procedures." Extensive incorporation of 18 O atoms was observed only through in-solution proteolysis (supplemental Fig. S6), which resulted in complete incorporation of 18 O atoms in more than 95% of the peptides. Data analysis of the peptide ions generated from the in-solution proteolysis facilitated an extensive search of cross-linked peptide candidates. By comparing peptide masses in the 16 O-digest and the 18 O-digest, peptides with mass increases of more than 4 Da, that is, incorporation of more than two 18 O atoms, were selected as crosslinked peptide candidates.
Masses of the 16 O-and 18 O-labeled peptides were then analyzed by ESI-QTOF mass spectrometry. Under electrospray ionization (ESI) conditions, peptide ions exist in multiple charge states. The values of monoisotopic peak shifts represent different peptide mass increases depending on the charge states of the ions. The QTOF mass spectrometer provided sufficient resolution to resolve the isotopic distribution of different charge states so that monoisotopic shifts were observed unambiguously.
Using DetectShift, a customized software program (30), fourteen peptide ions were selected as candidate ions that incorporated more than two 18 O atoms. Out of the fourteen candidate ions, six were targeted for subsequent MS/MS analysis because the masses of the six ions matched three intermolecular cross-linked peptides ( Table 1). The remaining eight ions were possibly generated from intramolecular cross-linked peptides or complex cross-linked peptides containing multiple cross-linked residue pairs. Such complex cross-linked peptides are not characterized in most current mass spectrometric studies because of the difficulties in predicting peptide masses and in assigning fragment ions. Here, we only characterize intermolecular cross-linked peptides containing one cross-linked residue pair and two linear peptides.
Characterization of the Intermolecular Cross-linked Peptide Candidates Using ESI-QTOF MS-The six ions representing three intermolecular cross-linked peptide candidates (Table 1) first underwent MS/MS characterization using the QTOF mass spectrometer.
The first cross-linked peptide candidate (Table 1), with its triply charged ion at m/z 1071.19 and quadruply charged ion at m/z 803.65, incorporated three 18 O atoms as a result of 18 O-labeling (supplemental Fig. S7). The measured mass of this peptide matched the calculated mass of an intermolecular cross-linked peptide (human b 5 : M OXIDIZED AHHHHM OXIDIZED AEQS-DEAVK)-(CYP2E1: NYGMGKQGNESR). However, the recorded MS/MS spectrum (supplemental Fig. S7) did not confirm the predicted peptide sequence. The possible reason for the generation of the false positive is that, although the two precursor ions were not generated from a cross-linked peptide under investigation, their masses happened to match the mass of the intermolecular crosslinked peptide.
The second cross-linked peptide candidate (Table 1), with its quadruply charged ion at m/z 946. 45 Fig. S8) of the quadruply charged precursor ion confirmed the peptide sequence. However, the cross-linked sites were not identified because only a limited number of terminal fragment ions were obtained. The MS/MS spectrum of the quintuply charged precursor ion did not provide more information to locate the cross-linked sites.
The third cross-linked peptide candidate (Table 1), with its quadruply charged ion at m/z 907.43 and quintuply charged ion at m/z 726.14, incorporated three 18 O atoms as a result of 18 Olabeling (Fig. 4). The measured mass of this peptide matched the calculated mass of an intermolecular cross-linked peptide (human b 5 : EQAGGDATENFEDVGHSTDAR)-(CYP2E1: YSDYFKPFSTGK). Both precursor ions were targeted for MS/MS analysis. Unfortunately, their MS/MS spectra were not observable using the ESI-QTOF mass spectrometer.
Characterization of the Intermolecular Cross-linked Peptide Candidates Using IT-FT-ICR MS-To further characterize the two intermolecular cross-linked peptides, the protein complex digests were subsequently analyzed on a hybrid ion trap-Fourier transform ion cyclotron resonance mass spectrometer (IT-FT-ICR MS).
For the second cross-linked peptide candidate, the measured masses of the quadruply charged ion at m/z 946.4325 (Ϯ0.9 ppm) and the quintuply charged ion at m/z 757.3460 (Ϯ0.8 ppm) (supplemental Fig. S9) precisely matched the calculated mass of the intermolecular cross-linked peptide (human b 5 : EQAGGDATENFEDVGHSTDAR)-(CYP2E1: YSDYFKPFST-GKR), further confirming the existence of the peptide.
Both precursor ions were targeted for MS/MS analysis. Peptide fragmentation was induced by collision-induced dissociation (CID). Under CID conditions, b-and y-type ions are predominant fragment ions (31), and these ions were used for peptide structure characterization. Fragment ions resulting from possible losses of H 2 O, NH 3 , CO, or CO 2 , were included in the fragment ion assignments. The nomenclature of the fragment ions follows that proposed recently (32). MS/MS spectrum assignments were performed using AssignXLink, a customized software program (30). The assignments of the MS/MS spectra of both the quadruply charged precursor ion at m/z 946.4325 (Fig. 5, Table 2) and the quintuply charged precursor ion at m/z 757.3460 (Fig. 6, Table 3) revealed the same cross-link, Glu 56 (b 5 )-Lys 434 (CYP2E1).
Site-directed Mutagenesis Evidence for the Importance of the MSidentified Ion Pairs in the CYP2E1-b 5 Interaction-To confirm the importance of the two ion pairs identified in the CYP2E1-b 5 interaction, Lys 428 and Lys 434 on CYP2E1 were mutated to alanine residues independently, and simultaneously. The equilibrium dissociation constants of b 5 with wild-type CYP2E1, and with the CYP2E1 mutants were measured to assess the interactions, and thus the importance of the two lysine residues. In addition, a comparison of the catalytic activity of wild-type CYP2E1 to the catalytic activities of the CYP2E1 mutants in the presence of b 5 was made to assess the stimulatory effect of b 5 .
Because the stimulatory effect of b 5 is substrate-dependent, two probe substrates of CYP2E1, APAP and PNP, were selected to  Table 2 for the MS/MS assignments.
provide a more complete assessment of the importance of the two ion pairs in the CYP2E1-b 5 interaction. APAP oxidation showed Michaelis-Menten kinetics both in the presence and absence of b 5 . However, PNP hydroxylation exhibited substrate inhibition in the presence of b 5 , and nonsaturatable enzyme kinetics in the absence of b 5 , which agrees with previous reports (33,34). Therefore, the catalytic efficiency (V max / K m ) of APAP oxidation, and the rate of PNP hydroxylation at a concentration of PNP (500 M) that saturates CYP2E1, were used to assess the interactions of b 5 with CYP2E1 wild type and the mutants.
The most commonly used method to measure the equilibrium dissociation constants of P450 and its redox partners is to measure the spin state changes of the P450 by the addition of the redox protein. However, the addition of b 5 did not alter the spin state of CYP2E1 significantly. Therefore, the apparent equilibrium dissociation constant of the CYP2E1-b 5 interaction (Table 6) was determined by measuring the rate of CYP2E1catalyzed APAP oxidation at constant concentrations of CYP2E1 and P450 reductase, and various concentrations of b 5 .
The catalytic activities of the wild-type CYP2E1, K428A single mutant, K434A single mutant, and K428A/K434A double mutant are shown in Table 6. In the absence of b 5 , the activities of the three mutants showed no significant differences from wild-type CYP2E1, indicating no significant changes of overall protein structure or local conformation in the three mutants. The addition of b 5 to wild-type CYP2E1 increased the catalytic efficiency of APAP oxidation by 65-fold, and the rate of PNP hydroxylation by 8-fold. In comparison, the addition of b 5 to the single mutants K428A and K434A stimulated the reactions to a lesser extent, indicating the involvement of the two lysine residues in the CYP2E1-b 5 interaction, and the requirement of such interactions for maximal stimulation by b 5 . The substitution of alanine for both lysine residues in the K428A/K434A double mutant further reduced the stimulatory effect of b 5 , which indicates that the two ion pairs are located on the same interacting protein surface, and work coordinately in the interaction. Lys 434 may play a more important role, because the K434A mutant decreased the b 5 stimulatory effect more than the K428A mutant. For APAP oxidation, the stimulatory effect of b 5 on the K428A mutant (50-fold) was 77% of that of wildtype CYP2E1 (65-fold), whereas the stimulatory effect of b 5 on the K434A mutant (21-fold) was only 32% of that of wild-type CYP2E1 (65-fold). The substitution of both lysine residues by alanine residues did not totally block, but further reduced the stimulatory effect of b 5 on the metabolism of the two substrates. This indicates that although the two interacting sites are important, additional interactions are also involved.
The measured apparent equilibrium dissociation constants (K d ) satisfactorily complement the catalytic parameters. The K d values of the mutants K428A, K434A, and K428A/K434A are 2-fold, 3-fold, and 13-fold of the K d value of wild-type CYP2E1, indicating the importance of the two lysine residues, especially Lys 434 , for the CYP2E1-b 5 interaction. The K d values suggest that the two lysine residues work synergistically for the CYP2E1-b 5 interaction, because the interruption of both ion pairs reduces the protein affinity by 13-fold, which is more than the effect of the interruption of each ion pair combined.
Construction of the Complex Model-A high resolution x-ray crystallographic structure of bovine b 5 (PDB CYO, Ref. 35) and a homology model of human CYP2E1 (36) constructed on the basis of a rabbit CYP2C5 x-ray crystallographic structure (PDB DT6, Ref. 37) were used to construct a model of the CYP2E1-b 5 complex. The docking was accomplished using two software programs, the "O" (38) and the Molecular Operating Environment (Chemical Computing Group Inc.). The distance between the two residues in each identified cross-link was minimized. Some side chains of the residues on the protein interacting surfaces were reoriented using the "O" program to avoid overlap of the side chains. The resulting complex model was energy-minimized using Engh & Huber energy parameters in the Molecular Operating Environment program. In the complex, the interacting surface on CYP2E1 consists of the meander region, the ␤-bulge, the C-helix, the L-helix, and the JЈ-helix (Fig. 9A), whereas the interacting surface on b 5 contains the ␣3 helix, the ␣4 helix, the loop region between the two helices, and one propionate group of the b 5 heme (Fig. 9A). The ␣3-helix of b 5 is perpendicular to the L-helix and parallel to the JЈ-helix of CYP2E1, with the start of the ␣3-helix proximate to the start of the L-helix. The loop region between the ␣3and ␣4-helices of b 5 apposes the meander region of CYP2E1, and the end of the ␣4-helix of b 5 is close to the ␤-bulge of CYP2E1. One propionate group of b 5 heme protrudes toward the end of the C-helix of CYP2E1. The CYP2E1 and b 5 hemes are nearly perpendicular, and the shortest distance between aromatic heme atoms is ϳ14 Å. Indicated by the model, hydrophobic interactions are not a major driving force in the association of CYP2E1 and b 5 because no correspondingly positioned hydrophobic patches were observed on the protein interacting surfaces. In contrast, the model reveals several well positioned intermolecular ion pairs besides the two pairs characterized by chemical cross-linking and mass spectrometry. Table 7 shows all the proposed intermolecular electrostatic interactions. By rotating the b 5 molecule in the complex 180 degrees to the right, the distribution of the charged residues (Table 7) on the protein interacting surfaces can be observed (Fig. 9B). In addition to the electrostatic interactions, seven intermolecular H-bonds (supplemental Table S1) were observed using the FindHBond function in the Chimera program. These H-bonds may further stabilize the complex.

DISCUSSION
Chemical cross-linking in combination with mass spectrometry has developed into a powerful method for mapping low resolution three-dimensional protein structures, and for investigating molecular interfaces in protein complexes. Among several strategies (39) developed to identify crosslinked peptides in a digest mixture, 18 O-labeling (21,22) is an attractive one because it is suitable for all cross-linking reactions and, once optimized for maximal incorporation, is easy to conduct.
Compared with 18 O-labeling, the commonly used method (40 -43) of selecting cross-linked candidates based on peptide mass matching shows two disadvantages. First, many false signals are generated when deconvoluted masses are acquired from ESI MS data using the MaxEnt 3 function in MassLynx software (Micromass, Manchester, UK). When such deconvoluted masses are used for peptide mass matching, false crosslinked candidates are selected and true candidates potentially missed. Furthermore, it is time-consuming to assign the deconvoluted masses to the corresponding ions in the spectrum because of the multiple charge distribution of the ions. Second, certain proteolytic rules have to be designated to generate a  Table 3 for the MS/MS assignments.

TABLE 3 IT-FT-ICR MS/MS assignments for the precursor ion [M ؉ 5H] 5؉ ‫؍‬ 757.3460
The MS/MS spectrum is shown in Fig. 6. Ions marked with subscript ␣ are from human b 5 peptide Glu 48 -Arg 68 , and ions with subscript ␤ are from CYP2E1 peptide Tyr 423 -Arg 435 . Nomenclature of the fragment ions follows that proposed recently (31). predicted cross-linked peptide data base with a reasonable size. Therefore, peptides generated from abnormal cleavages are excluded from the data base. In contrast, in 18 O-labeling experiments, the accurately measured monoisotopic and isotopic signals, instead of the software-generated deconvoluted masses, are analyzed, and cross-linked candidates caused by any type of proteolysis are selected. In our study, mass shifts of peptides serve as a filter to select cross-linked peptide candidates that incorporate more than two 18 O atoms. To select as many candidates as possible, experimental conditions were optimized for maximal isotopic incor-poration. Both proteolysis and postproteolysis conditions have been used to incorporate 18 O atoms from 18 O-water into peptides (44). However, we found that proteolysis resulted in more complete isotopic incorporation than post-proteolysis in comparative experiments with both the CYP2E1-b 5 complex and bovine serum albumin. In addition, conducting 18 O-labeling during ingel proteolysis resulted in lower isotope incorporation than during insolution proteolysis. For in-gel proteolysis, complete dryness of gel pieces is difficult to achieve, even though the gel pieces are dried for a prolonged period of time (ϳ4 h) before reconstitution in digestion solutions prepared with 18 O-water. The retained 16 O-water in the gel pieces can cause lower than maximal incorporation of 18 O atoms. Complete incorporation of 18 O atoms was observed for ϳ60% of peptides obtained from in-gel proteolysis. In comparison, complete incorporation was observed for Ͼ95% of peptides obtained from insolution proteolysis, which facilitated an extensive search for the cross-linked peptide candidates.

Labeling number Measured Calculated Error Assignments
Subsequent MS/MS analysis of the three selected candidates led to the identification of two intermolecular cross-linked peptides, whose structures were unambiguously characterized by IT-FT-ICR MS. IT-FT-ICR MS provides two major advantages over other available MS techniques in peptide structure elucidation. First, in IT-FT-ICR MS, less abundant precursor ions can accumulate to a desired ion population through the use of an automated gain control in the linear ion trap cell (45). Precursor ions of cross-linked peptides are generally less abundant than those of non-cross-linked peptides because of their lower stoichiometry. Moreover, MS data acquired from ESI mass spectrometers yield a wide charge distribution of peptide ions, further reducing the abundance of individual precursor ions. Without a precursor ion accumulation function, the QTOF MS/MS did not generate enough fragment ions (supplemental Fig. S8) to locate the cross-linked sites, and the fragmentation of our third crosslinked peptide (Table 1) Table 4 for the MS/MS assignments.  Table 5 for the MS/MS assignments.

Identification of Interactions between CYP2E1 and b 5
high resolution and mass accuracy of IT-FT-ICR MS make it possible to accurately assign a large number of fragment ions with multiple charge states (Tables 2-5), thereby identifying the cross-linked sites.
The importance of the two ion pairs, Lys 428 (CYP2E1)-Asp 53 (b 5 ) and Lys 434 (CYP2E1)-Glu 56 (b 5 ), identified by mass spectrometry in the CYP2E1-b 5 interaction was confirmed by site-directed mutagenesis, and by the measurement of protein apparent equilibrium dissociation constants (K d ) and catalytic activities. The   other intermolecular interactions, are structurally responsible for the stimulatory effect of b 5 on CYP2E1 oxidation rates.
In agreement with these results, a model constructed on the basis of the two identified "zero-length" cross-links, suggests nine ion pairs (Table 7) and seven H-bonds (supplemental Table S1) as part of the CYP2E1-b 5 interacting surface. Compared with Lys 428 , Lys 434 appears to play a more significant role because the K d value of the K434A mutant is higher than that of the K428A mutant, and because there is less stimulatory effect in the K434A mutant. This result agrees with the model, as well, in that among all proposed ion pairs, Lys 434 is involved in two ion pair interactions, Lys 434 -(CYP2E1)-Glu 56 (b 5 ) and Lys 434 -(CYP2E1)-Asp 60 (b 5 ), whereas, Lys 428 is involved only in one interaction, Lys 428 (CYP2E1)-Asp 53 (b 5 ) ( Table 7).
In the study, two ion pairs, Lys 428 (CYP2E1)-Asp 53 (b 5 ) and Lys 434 (CYP2E1)-Glu 56 (b 5 ), were identified by mass spectrometry and seven more ion pairs were proposed by the constructed model (Table 7). Because EDC covalently links basic (Lys) and acidic (Asp or Glu) residues that come into proximity, a question arises why three of the proposed ion pairs containing lysine residues, Lys 342 (CYP2E1)-Glu 43 (b 5 ), Lys 422 (CYP2E1)-Glu 48 (b 5 ), and Lys 434 (CYP2E1)-Asp 60 (b 5 ), were not identified by mass spectrometry. There are four possible reasons. First, the competition for ion pair formation, caused by the proximity of one charged residue to several others with FIGURE 9. CYP2E1-b 5 complex model. CYP2E1 is colored yellow with its heme group colored black; b 5 is colored gray with its heme group colored green. The interacting residues on CYP2E1 and b 5 are colored blue and red, respectively. A, interacting regions on CYP2E1 and b 5 are colored brown and purple, respectively. Protein regions far away from the interacting surfaces are truncated. B, molecule of b 5 in the complex is rotated 180 degrees to the right to display the interacting surfaces on CYP2E1 and b 5 completely. Nitrogen atoms of the side chains of the positively charged interacting residues on CYP2E1 are colored blue; oxygen atoms of the side chains of the negatively charged interacting residues on b 5 are colored red. Atom radii are decreased to 1.2 Å so that CYP2E1 heme can be seen.

Identification of Interactions between CYP2E1 and b 5
opposite charge, may decrease the abundance of each ion pair, therefore preventing it from being identified by mass spectrometry. For example, Lys 422 (CYP2E1)-Glu 48 (b 5 ), an ion pair predicted by the model, was not identified by mass spectrometry as a crosslink. An examination of the model reveals that Glu 48 (b 5 ) interacts with two other residues, Arg 344 (CYP2E1) and Arg 444 -(CYP2E1), whereas Lys 422 (CYP2E1) forms an intramolecular ion pair with Asp 341 (CYP2E1). All the above interactions decrease the effective local charges of Lys 422 (CYP2E1) and Glu 48 (b 5 ), thus reducing the abundance of the ion pair between the two residues.
In agreement with our experimental results, the model shows that the two ion pairs identified by mass spectrometry are more specific because of less local distraction. Second, peptide yields from proteolysis vary, and the ability of peptides to form multiply charged molecular ions under the same ESI conditions differ. As a result, peptides with low yields, and those that do not efficiently generate ions, may not be observed. Third, the actual intermolecular atomic distances of the three undetected ion pairs may deviate from the predicted values and may be beyond the distance range of EDCmediated cross-linking reactions, because neither protein flexibility nor protein local conformational changes induced by proteinprotein interactions can be predicted by a rigid complex model (46). Fourth, lacking an x-ray crystallographic structure of CYP2E1, a homology model was used in this study to construct the complex model. The possible deviation of the atomic coordinates in the homology model leads to a less accurate prediction of the intermolecular atomic distances.
In agreement with previous proposals, our model ( Fig. 9) constructed on the basis of the two cross-links shows that electrostatic interactions (Table 7) are the main stabilizing forces for the protein-protein interaction and contribute to the proper relative orientations of the prosthetic groups, which may result in a change of dielectric constant and facilitate the subsequent electron-transfer process. For all the proposed electrostatic interactions (Table 7), the negatively charged residues are contributed by b 5 . These residues are located on the surface region where the b 5 heme group protrudes toward the solvent. The positively charged residues contributed by CYP2E1 are distributed across the proximal face of CYP2E1 where the buried CYP2E1 heme group comes closest to the solvent.
The surface region on b 5 has previously been reported for its interactions with several redox proteins, such as P450, NADH-b 5 reductase, cytochrome c, NADPH-P450 reductase and stearyl coenzyme A desaturase (4). The negatively charged residues on this surface region are conserved across species (3).
In our study, the analysis of the complexes of CYP2E1-rat b 5 and CYP2E1-human b 5 identified the same cross-links, suggesting that the evolutionary conservation of the negatively charged residues is related to their functions in protein-protein interactions.
Previous studies, using geometric fit algorithms for protein docking, have shown three surface regions on CYP2B4 that can possibly bind to b 5 (47). The first surface region is located on the distal face of CYP2B4 and includes residues in the A-helix, the E*-helix, the F-helix, the I-helix, and the ␤4-sheet. The second region is located on a face perpendicular to the heme plane of CYP2B4 and includes residues in the BЈ-helix. The third region is located on the proximal face of CYP2B4 and includes residues in the C-C*-helices, the ␤5-sheet, the K-helix, the meander region, the ␤-bulge and the L-helix. Studies with CYP2B4 and several other P450 isoforms, which used site-directed mutagenesis, synthetic peptide binding and chemical modification, suggest important roles for several basic residues on the third surface region in b 5 binding (3,4,14,48). In the CYP2E1-b 5 interaction, most interacting residues on CYP2E1 (Table 7), such as Lys 422 and Lys 428 in the meander region, Lys 434 in the ␤-bulge region, Arg 126 in the C-helix, and Arg 444 in the L-helix, are located on the third surface region. However, different from the third surface region, the CYP2E1-b 5 model shows the involvement of the JЈ-helix of CYP2E1, instead of the ␤5-sheet, for b 5 binding.
Compared with the CYP101-b 5 complex model constructed in 1989 (16), our model shows the same interacting surface on b 5 and some overlapped regions on P450. The main difference in the two models is that the B-helix of CYP101 is part of the b 5 binding region, whereas the JЈ-helix of CYP2E1 is part of the b 5 binding region. Furthermore, the protein docking orientations in the two models differ by about 90 degrees. In the CYP101-b 5 model, the ␣3-helix of b 5 is close to the B-helix, the meander region and the ␤-bulge of CYP101; the ␣4-helix of b 5 is in the proximity of the C-helix of CYP101; and the b 5 heme propionate protrudes toward the L-helix of CYP101. In the CYP2E1-b 5 model, the ␣3-helix of b 5 is close to the L-and the JЈ-helices of CYP2E1; the ␣4-helix of b 5 apposes the meander region and the ␤-bulge of CYP2E1; and the b 5 heme propionate protrudes toward the end of the C-helix of CYP2E1. Because of the structural differences in P450 isoforms, it remains unclear whether the protein interacting orientations are P450 isoform-dependent.
The generation of a CYP2E1-(b 5 ) 2 complex (Fig. 2) suggests more than one b 5 binding region on CYP2E1. As mentioned, three surface regions on CYP2B4 were proposed for their possible involvement in b 5 binding (16). The homology model of CYP2E1 shows three similar surface regions. Other than the predominant interaction (Fig. 9), b 5 may approach the other surface regions, yielding the CYP2E1-(b 5 ) 2 complex. In addition, it has been shown that there are at least two positively charged regions on P450, one for the binding of b 5 , and the other for the binding of P450 reductase (18,49,50). In the absence of P450 reductase, the second molecule of b 5 probably occupies the reductase binding region. However, the association of CYP2E1 to the second molecule of b 5 appears to be much less favorable because the CYP2E1-(b 5 ) 2 complex band was very weak and disappeared in the protein gel destained for more than 12 h (Fig. 2). It is possible that the less favorable interaction between CYP2E1 and b 5 does not provide as many stabilizing forces as the primary interaction does. Ion pairs in the less favorable protein complex were not identified by mass spectrometry because of their low abundance.
In conclusion, two intermolecular ion pairs in the CYP2E1-b 5 complex were characterized using chemical cross-linking, isotopic labeling and mass spectrometry, and the biological importance of the ion pairs was confirmed by site-directed mutagenesis. This study reveals the protein interacting surfaces and provides the first direct evidence to support protein orientations in a P450-b 5 complex. Inasmuch as some P450 isoforms (e.g. CYP1A2 and CYP2D6) have a more positively charged surface and are not affected by b 5 , additional studies are underway to assess the specificity of this cross-linking method.