Catalytically Relevant Electrostatic Interactions of Cytochrome P450c17 (CYP17A1) and Cytochrome b5*

Background: Cytochrome b5 (b5) stimulates the 17,20-lyase activity of cytochrome P450c17 (CYP17A1). Results: A carboxylate group in b5 residues Glu-48/Glu-49 is required for 17,20-lyase activity, and cross-linked peptides Lys-88(CYP17A1)-Glu-61(b5) and Lys-347(CYP17A1R347K)Glu-42(b5) involve adjacent residues. Conclusion: These data support a molecular model of a CYP17A1-b5 complex. Significance: The CYP17A1-b5 interaction site might be a drug target to inhibit androgen synthesis. Two acidic residues, Glu-48 and Glu-49, of cytochrome b5 (b5) are essential for stimulating the 17,20-lyase activity of cytochrome P450c17 (CYP17A1). Substitution of Ala, Gly, Cys, or Gln for these two glutamic acid residues abrogated all capacity to stimulate 17,20-lyase activity. Mutations E49D and E48D/E49D retained 23 and 38% of wild-type activity, respectively. Using the zero-length cross-linker ethyl-3-(3-dimethylaminopropyl)carbodiimide, we obtained cross-linked heterodimers of b5 and CYP17A1, wild-type, or mutations R347K and R358K. In sharp contrast, the b5 double mutation E48G/E49G did not form cross-linked complexes with wild-type CYP17A1. Mass spectrometric analysis of the CYP17A1-b5 complexes identified two cross-linked peptide pairs as follows: CYP17A1-WT: 84EVLIKK89-b5: 53EQAGGDATENFEDVGHSTDAR73 and CYP17A1-R347K: 341TPTISDKNR349-b5: 40FLEEHPGGEEVLR52. Using these two sites of interaction and Glu-48/Glu-49 in b5 as constraints, protein docking calculations based on the crystal structures of the two proteins yielded a structural model of the CYP17A1-b5 complex. The appositional surfaces include Lys-88, Arg-347, and Arg-358/Arg-449 of CYP17A1, which interact with Glu-61, Glu-42, and Glu-48/Glu-49 of b5, respectively. Our data reveal the structural basis of the electrostatic interactions between these two proteins, which is critical for 17,20-lyase activity and androgen biosynthesis.

associated with several diseases, including polycystic ovary syndrome, Cushing syndrome, and prostate cancer. Traditionally, disorders of excessive androgen production have been treated with gonadotropin suppression to prevent testicular testosterone synthesis or, more recently, with abiraterone acetate, a potent CYP17A1 inhibitor (2). Abiraterone treatment, however, has significant potential endocrine side effects and toxicities, including hyperkalemia and hypertension, due to the accumulation of precursor steroids upstream of the 17-hydroxylase reaction (3). Consequently, an unmet clinical need is a selective inhibitor of the 17,20-lyase reaction, which will lower testosterone production without requiring concomitant glucocorticoid therapy or disturbing drug metabolism.
Although the 17-hydroxylase activity of CYP17A1 is relatively insensitive to physiologic changes or assay conditions, the regulation of the 17,20-lyase activity is complex. A high abundance of the flavoprotein P450 oxidoreductase (POR), the presence of the 17-kDa heme protein cytochrome b 5 (b 5 ), and serine phosphorylation all augment 17,20-lyase activity with little influence on 17-hydroxylase activity (4). In particular, a molar equivalent of b 5 increases the rate of the 17,20-lyase reaction 10-fold, via an allosteric mechanism that does not require electron transfer (5). A specific allosteric binding site on b 5 , which includes the acidic residues Glu-48, Glu-49, and possibly Arg-52, appears to mediate this stimulation of 17,20-lyase activity (6 -8). Nonspecific interactions of the hydrophobic membrane-spanning regions of b 5 and CYP17A1 are also required to optimize 17,20-lyase activity (9). Furthermore, the basic residues Lys-88, Lys-89, Arg-347, Arg-358, and Arg-449 of human CYP17A1 are critical for the enzyme's b 5 -dependent acyl-carbon cleavage activities (8,10,11). Studies of CYP17A1 mutations from patients with isolated 17,20-lyase deficiency (10,12) demonstrate that alterations in the interaction of CYP17A1 with its redox partner proteins POR and b 5 form the biochemical basis for this selective enzyme defect. POR mutation G539R also causes a form of isolated 17,20-lyase deficiency (13), as do the missense and nonsense mutations in the CYB5A gene encoding b 5 (14,15). These observations emphasize that specific interactions of CYP17A1 with POR and b 5 are critical for maximal 17,20-lyase activity and thus androgen production.
Previous studies have generated a model in which b 5 interacts with P450 or P450-POR complexes via at least two surfaces, which incorporate the acidic residues Glu-48 -Glu-49 for CYP17A1 and Asp-58 -Asp-65 for CYP3A4, CYP2E1, and CYP2C19 (7,16). For multiple conformations of CYP3A4, the contribution of b 5 residues Glu-48 -Glu-49 is minor and varies with phospholipid. Therefore, it might be possible to identify compounds, which block the electrostatic interactions of b 5 with CYP17A1 and thus eliminate androgen production without disturbing 17-hydroxylase activity or drug metabolism. Here, we describe the use of site-directed mutagenesis and cross-linking coupled with mass spectrometry to delineate protein-protein interactions of CYP17A1, POR, and b 5 .
Plasmids-The expression plasmids were generous gifts obtained from the following investigators: modified human CYP17A1-G 3 H 6 in pCW and N-27-human POR-G 3 H 6 in pET22 from Professor Walter L. Miller (University of California, San Francisco) and human b 5 in pLW01-b 5 H 4 from Professor Lucy Waskell (University of Michigan).
Site-directed Mutagenesis-Human b 5 mutations were generated following site-directed mutagenesis methods described previously (6). Briefly, plasmid DNA encoding the wild-type b 5 cDNA was amplified by using 0.25 units of Pfx and 0.5 units of Taq DNA polymerases and a set of primers containing the appropriate base substitution as listed in Table 1. Reactions conditions were as follows: denaturing at 95°C for 15 s, annealing at 62.5°C for 20 s, and extension at 68°C for 3 min 30 s. After 30 cycles of amplification, the resulting mutated plasmids were selected by digestion with DpnI, and 5-l PCR products were transformed in DH5␣-competent cells. The CYP17A1-R347K plasmid was created in the same manner as the b 5 mutations, and the CYP17A1-R358K plasmid was prepared by overlap extension PCR, using the primers listed in Table 1. Constructs were sequenced to ensure accurate mutagenesis.
Expression and Purification of Recombinant Proteins in Escherichia coli-Modified human CYP17A1 was expressed with GroEL/ES chaperones (pGro7 plasmid) in E. coli JM109 cells as described (17). Modified human POR was expressed in E. coli C41(DE3) cells (OverExpress, Lucigen, Middleton, WI) and purified according to the previously published procedure (18). For both cases, 1 liter of Terrific Broth (supplemented with 0.5 mM 5-aminolevulinic acid for CYP17A1) and appropriate antibiotics were inoculated with 20 ml of an overnight pre-culture. The cells were grown at 37°C with shaking at 250 rpm until the A 600 reached 0.8 -1.2 AU, at which time the culture was induced with 0.4 mM isopropyl ␤-D-thiogalactopyranoside and grown for 48 h at 25-27°C. After cell lysis with a French press, the recombinant proteins were solubilized using mild nondenaturing detergents as follows: 0.5% Nonidet P-40 for CYP17A1 and 0.2% cholate, 0.2% Triton X-100 for POR (18). After centrifugation at 100,000 ϫ g for 45 min, the supernatant was mixed with nickel-nitrilotriacetic acid affinity resin and purified in a single step upon elution with 300 mM imidazole, followed by buffer exchange using PD-10 columns. For cross-linking experiments, higher purity CYP17A1-protein preparations were obtained by further affinity purification with His Select resin (0.5 ml) and Sartorius TM Vivapure-S spin columns (Mini H) according to the manufacturer's operating instructions. Purified CYP17A1 preparations showed a specific content of 8 -12 nmol of P450/mg of protein with 3-10% P420. The protocol for expression and reconstitution of recombinant human b 5 was based on the procedure of Mulrooney and Waskell (19). Yeast microsomes containing native human CYP17A1 and POR and control microsomes without human P450 enzymes were prepared from strain YiV(B) transformed with V60-c17 or native V60 plasmids, respectively, as described (20). Reconstituted Enzyme Assays-In a 2-ml polypropylene tube, purified human CYP17A1 was mixed with 4-fold molar excess of POR, various amounts of b 5 , and 20 g of control yeast microsomes as a phospholipid source in less than a 10-l volume and incubated for 5 min. The reaction mixture was then diluted to 0.2 ml with 50 mM potassium phosphate buffer (pH 7.4), and the substrates progesterone (10 M, for 17-hydroxylase activity) or 17-hydroxypregnelonone (5 M, for 17,20-lyase activity) with 80,000 cpm tracer steroid in methanol (2% of incubation volume) were added. Experiments with microsomes containing native CYP17A1 (5 pmol) and POR from transformed yeast were preincubated with b 5 variants (10 pmol) at room temperature for 5 min before adding substrate. The resulting mixture was preincubated at 37°C for 3 min before adding NADPH (1 mM) and incubating at 37°C for another 20 min. The reaction mixture was extracted with 1 ml of dichloromethane, and the organic phase was dried under a nitrogen flow.
Chromatography, Data Acquisition, and Determination of Kinetic Constants-Reaction products were analyzed using an Agilent 1260 Infinity HPLC system with a UV detector and ␤-RAM4 in-line scintillation counter (LabLogic, Brandon, FL). Extracted steroid products were dissolved in 20 l of methanol, and 5-l injections were resolved with a 50 ϫ 2.1 mm, 2.6 m, C 8 Kinetex column (Phenomenex, Torrance, CA), equipped with a guard column at a flow rate of 0.4 ml/min. A methanol/ water linear gradient was used as follows: 27% methanol from 0 to 0.5 min; 39% to 16 min; 44% to 20 min; 60% to 22 min; 71% to 30 min; 75% to 30.5 min, and 27% to 33 min. Products were identified by retention times of external standards chromatographed at the beginning and ends of the experiments. The flow rate of the scintillation mixture (Bio-SafeII, Research Products International, Mount Prospect, IL) was 1.2 ml/min, and the data were processed with Laura4 software (LabLogic) and graphed with GraphPad Prism 6 (GraphPad Software, San Diego).
Spectroscopic Methods-UV-visible spectra were recorded on a Shimadzu UV-2600 spectrophotometer with UV-Probe software using a quartz cuvette with a path length of 10 mm.
Cross-linking and Immunoblot Experiments-Purified human CYP17A1 (50 pmol), b 5 (WT or mutations), and phospholipids (PE/PS ϭ 1:1) were reconstituted at a molar ratio of 1:15: 100. After incubation for 20 min at 25°C, freshly prepared EDC was added to a final concentration of 2 mM in a total volume of 20 l containing 50 mM potassium phosphate (pH 7.0). The mixture was shaken gently at room temperature for 2 h, followed by adding an equal volume of Laemmli sample buffer and boiling for 5 min. Control incubations without EDC were conducted in parallel. The cross-linked proteins were resolved with SDS-PAGE on 7.5% gels, and the bands corresponding to the heterodimeric complex of CYP17A1 and b 5 were excised and submitted to LC/MS analysis at MSBioworks (Ann Arbor, MI). For immunoblot analyses, the SDS-polyacrylamide gels were electroblotted onto PVDF membrane. The membranes were blocked for 1 h at room temperature in PBS containing 5% skim milk plus 0.1% Tween 20 and were incubated overnight at 4°C with mouse anti-His tag antibody, followed by incubation with secondary antibody for 1 h at room temperature. The blots were incubated for 1 min with ECL reagent and exposed to photographic film.
LC/MS Analysis-The excised gels containing cross-linked proteins were processed by in-gel digestion using a robot (Pro-Gest, DigiLab) and Promega sequencing-grade trypsin. The digested sample was analyzed by nano-LC/MS with a Waters NanoAcquity HPLC system interfaced to a ThermoFisher Q Exactive tandem mass spectrometer. Peptides were loaded on a trapping column and resolved on a 75-m analytical column at 350 nl/min; both columns were packed with Jupiter Proteo resin (Phenomenex). The mass spectrometer was operated in data-dependent mode, with MS and tandem MS performed in the Orbitrap at 70,000 and 17,500 full width at half-maximum resolutions, respectively. The 15 most abundant ions were selected for tandem MS.
Data Processing-Data were searched using a local copy of Mascot, and the DAT files were parsed into scaffold software for validation, filtering, and creating a nonredundant list for each sample. Cross-linked products were identified by analyzing MS data with the software StavroX 3.2.10 (21) and Sqid-Xlink 1.0 (22). MS data containing all tandem MS data for each precursor ion are loaded as a standard Mascot generic file (mgf) for StavroX or as dta files for Sqid-Xlink. The parameters are as follows: K miscleavage ϭ 3; R miscleavage ϭ 1; fixed modification ϭ carbamidomethyl; variable modification ϭ oxidation, acetyl (protein N terminus), pyro-Glu (N-terminal Glu), and deamidation (NQ). CrossLinker parameters are as follows: EDC mass ϭ Ϫ18.01; composition ϭ ϪH 2 O; precursor mass deviation ϭ 50.0 ppm; fragment mass deviation ϭ 20.0 ppm; lower mass limit ϭ 200.0 Da; upper mass limit ϭ 5000.0 Da; and S/N ratio ϭ 2.0.
Three-dimensional Model Analysis-A three-dimensional model of CYP17A1-b 5 heterodimer was predicted using the GRAMM-X (23) and HADDOCK (24 -26) software tools, which are rigid-body protein-protein docking programs, to present the molecular interactions between CYP17A1 and b 5 . Each of the programs returned the 10 most probable models (clusters) out of thousands of candidates based on their geometry, hydrophobicity, and electrostatic complementarity. The first-ranked model in cluster 1 of HADDOCK and the top model in GRAMM-X exhibited highly similar interacting modes; therefore, the structure from HADDOCK was selected as the final model for further analysis. Chain A of human b 5 molecule (PDB code 2I96, no accompanying journal publication as of August 2014) was set as ligand molecule and chain A of human CYP17A1 (PDB code 3RUK published by DeVore and Scott (27) as receptor molecule. As constraints, the two intermolecular cross-links identified below as well as amino acids identified from mutagenesis studies, Lys-88, Arg-347, and Arg-358 of CYP17A1 and Glu-42, Glu-48, Glu-49, and Glu-61 of b 5 , were used during all calculations to predict the interaction of these two target proteins. Molecular graphic figures were generated using PyMOL. Predictions of binding affinity changes upon mutation were calculated by BeAtMuSiC version 1.0 (28), available on line, and molecular interface analysis was performed using PDBePISA.

Site-directed Mutagenesis of b 5 Residues
Glu-48 and Glu-49 -Previous work has identified the acidic residues Glu-48 and Glu-49 as critical b 5 residues for stimulation of 17,20-lyase activity (6,7). To determine the steric and electrostatic requirements of these key residues, a series of b 5 mutations at positions 48 and 49 (GG, GA, AG, GC, CG, GD, DG, GQ, QG, and DD) were generated, expressed, and purified as described previously (7). The side-chain groups of these mutations vary in size, shape, charge, hydrophobicity, chemistry, and capacity to serve as salt bridges or hydrogen bond donors. All mutations displayed absorption spectra comparable with that of the wildtype b 5 in the reduced and oxidized state (Fig. 1), indicating structural and electronic integrity.
Each b 5 mutation was tested for its capacity to stimulate the 17,20-lyase activity of CYP17A1 ( Fig. 2A). When yeast micro-somes containing CYP17A1 and POR were incubated with b 5 and 17-hydroxypregnenolone, the presence of wild-type b 5 increased the rate of product formation 11-fold. The activities of b 5 mutations GG, GA, AG, GC, CG, DG, GQ, and QG were totally abolished, and product formation was reduced to that observed without b 5 added. The GD and DD mutations partially restored the capacity to stimulate 17,20-lyase activity at 23 and 38% that of wild-type b 5 , respectively. These results were replicated in a reconstituted system containing purified CYP17A1, POR, b 5 , and control yeast microsomes as lipid source (Fig. 2B). These data indicate that at least one carboxylate group is an essential feature of this interaction site and that even conservative aspartate substitutions are poorly tolerated at b 5 residues Glu-48 and Glu-49.
Arg-to-Lys Mutations of Key CYP17A1 Residues-CYP17A1 mutations that remove the full positive charge at arginine 347 or 358 selectively disrupt 17,20-lyase activity yet preserve 17-hydroxylase activity (11). We reasoned that mutation of these arginines to the basic residue lysine might preserve interactions with POR and b 5 , particularly the carboxylate-mediated interaction with b 5 . CYP17A1 mutations R347K and R358K were strategically constructed in an effort to maintain 17,20lyase activity, yet render residues 347 and 358 vulnerable to  DECEMBER 5, 2014 • VOLUME 289 • NUMBER 49 chemical reaction with carbodiimide cross-linking agents. The resulting R347K and R358K mutations exhibited similar 17-hydroxylase (Fig. 2C) and b 5 -stimulated 17,20-lyase activity as wild-type CYP17A1 (Fig. 2D). These mutations were then employed with wild-type CYP17A1 in cross-linking experiments with b 5 .

CYP17A1-Cytochrome b 5 Interaction Surface
Cross-linking CYP17A1 with b 5 -To identify CYP17A1 residues in close proximity to carboxylate groups of b 5 during complex formation, mixtures of CYP17A1 and b 5 were incubated with the water-soluble zero-length cross-linker EDC, which covalently links lysine amino groups to the carboxylate groups of aspartic or glutamic acid. CYP17A1 mutations R347K and R358K were employed to specifically target these key residues, and b 5 mutation E48G/E49G was used as a negative control. When incubations of b 5 and CYP17A1 were treated with EDC, an ϳ70-kDa species was formed, consistent with a CYP17A1-b 5 heterodimer (complex, Fig. 3A). A 70-kDa species also appeared in incubations containing the CYP17A1 mutations R347K and R358K (Fig. 3A); however, complex formation was abolished or markedly reduced for incubations containing the b 5 mutation E48G/E49G (Fig. 3B). These results suggest that poor complex formation is the primary reason why b 5 mutation E48G/E49G fails to stimulate 17,20-lyase activity. A small amount of complex formation between CYP17A1 mutation R358K and b 5 mutation E48G/E49G (Fig. 3B) appears to be a nonspecific cross-linked product, because this b 5 mutation does not stimulate the 17,20-lyase activity of CYP17A1 mutation R358K (data not shown). Addition of substrate to these experiments enhanced formation of the putative CYP17A1-b 5 complex in the order 17-hydroxypregnenolone ϭ pregnenolone Ͼ progesterone Ͼ no substrate (Fig. 3C).
Mass Spectrometry Analysis of Protein Cross-links-Following SDS-PAGE and Denville blue staining of the cross-linked species (Fig. 4A), peptide mixtures generated from in-gel trypsin digestions were separated by nano-HPLC and analyzed by mass spectrometry. These digests contained peptides from both CYP17A1 and b 5 , as well as inter-and intramolecular cross-linked products. To identify cross-linked peptides involving the two proteins from this intricate mixture, the experimentally obtained monoisotopic masses were compared with calculated masses of peptides and cross-linked products derived from the StavroX (21) and squid-Xlink (22) software packages as described under "Experimental Procedures." Fig. 4B shows a representative high resolution fragmentation spectrum of cross-linked peptides between wild-type CYP17A1 and b 5 : CYP17A1-WT ( 84 EVLIKK 89 )-b 5 ( 53 EQAGGDATENFED-VGHSTDAR 73 ). The underlined residues, Lys-88 of CYP17A1 and Glu-61 of b 5 , have formed an inter-protein amide bond via the action of EDC. Cross-linking of CYP17A1 mutation R358K and b 5 with EDC afforded the same Lys-88(CYP17A1)-Glu-61(b 5 ) peptide (Fig. 4, C and D), and no additional cross-linked species were identified.

CYP17A1-Cytochrome b 5 Interaction Surface
performed, in which CYP17A1, POR, and b 5 were treated with EDC, and one band corresponding to the ternary complex was detected by Coomassie Blue staining (data not shown). LC/MS revealed peptides derived from CYP17A1, POR, and b 5 in the complex with 51-60% sequence coverage, including the crosslinked peptide Lys-88(CYP17A1)-Glu-61(b 5 ). Because of lower mass accuracy of the data, however, other candidate crosslinked peptides could not be characterized with confidence. Table 2 summarizes the two intermolecular cross-link locations identified with StavroX and Squid-Xlink analyses.
Model of b 5 Interaction with CYP17A1-Using distance constraints derived from the cross-linking data and b 5 residues Glu-48 and Glu-49 as essential components, low resolution structural models were calculated using HADDOCK (24 -26) rigid body energy minimization and semi-flexible and water refinement (Fig. 6A). The x-ray crystal structure of human CYP17A1 with bound abiraterone (PDB code 3RUK) reported by DeVore and Scott (27) was used as the receptor molecule. Like other P450s, the core of CYP17A1 contains helices D, E, I, and L, which compose a four-helix bundle that carries the triangular prism-shaped structure of the protein molecule. The long I helix (purple in Fig. 6, B and C) runs over the distal surface of the heme. The proximal side, where the heme iron is ligated to the thiolate of a conserved cysteine, is a highly conserved region where interactions with redox-partner proteins occur. In the human b 5 molecule (PDB code 2I96), the prosthetic heme group is sandwiched in a cleft between two helix-loophelix motifs (Fig. 6B), and a three-stranded ␤-sheet ties together these helices. The structure of the CYP17A1-b 5 complex generated from HADDOCK shows that seven key amino acids, including Lys-88, Arg-347, and Arg-358 in CYP17A1 and Glu-42, Glu-48, Glu-49, and Glu-61 in b 5 , are all located at the interface of the complex. This model incorporates the available evidence and features direct electrostatic and functional interactions between these charge residues. In addition to the two cross-links identified, the major contribution to the interaction enthalpy between CYP17A1 and b 5 arises from the two electrostatic interactions between positively charged Arg-347 (JЈ helix) and Arg-358 (K helix) side chains of CYP17A1 and negatively charged Glu-48 and Glu-49 (␣3 helix) carboxyl groups of b 5 . Cationic Arg-358 on CYP17A1 interacts with both Glu-48 and Glu-49 of b 5 , whereas anionic Glu-48 on b 5 interacts with both Arg-347 and Arg-358 of CYP17A1.
Surface Contacts Analysis by PISA-To analyze the surface contacts of the CYP17A1-b 5 complex, we used the protein interfaces, surfaces, and assemblies service PDBePISA at the European Bioinformatics Institute. PISA yielded the interface area of 1126.0 Å 2 for the interface between CYP17A1 and b 5 in the complex. The interaction surface shows good shape complementarity with numerous electrostatic interactions, including 20 potential hydrogen bonds and 23 potential salt bridges (Table 3 and Fig. 6D), plus hydrophobic interactions. Interfacing residues identified within the CYP17A1-b 5 complex, including Lys-83/Lys-88/Lys-91 (B helix), Lys-136 (C helix), Arg-347 (JЈ helix), Arg-358 (K helix), and Arg-449 (L helix) in CYP17A1 and Glu-42/Glu-43 (turn 2), Glu-48/Glu-49 (␣3 helix), and Glu-61/Asp-65 (␣4 helix) in b 5 , were all consistent with the results from cross-linking and mutagenesis experiments. As evidenced in Fig. 2, A and B, replacement of b 5 residues Glu-48, which interacts with Arg-347, Arg-358, and Arg-449 of CYP17A1 or Glu-49, which interacts with Arg-358 and Arg-449 of CYP17A1, with smaller or neutral amino acids resulted in a marked loss of catalytic activity. Prediction of changes in protein-protein binding affinity from these b 5 mutations was analyzed by BeAtMuSiC version 1.0. The difference in binding energies (⌬G bind ) calculated for wild-type b 5 and these mutations affords the change in binding energy (⌬⌬G bind ), which provides an estimate of the relative binding affinities. A substantial decrease in the binding affinity (⌬⌬G bind ϭ 3.5 kcal/ mol) was observed with the inactive CYP17A1-b 5 mutation E48G/E49G complex. Smaller yet significant decreases in binding energy were calculated for the partially active CYP17A1-b 5

DISCUSSION
The 17,20-lyase activity of CYP17A1 is strongly regulated via interactions with the redox partners POR and b 5 . Given the difficulties encountered when using traditional structural biology methods to study such multiprotein complexes, elucidating the architecture of these assemblies and their multiple interactions remains a challenging task. To overcome these difficulties, our study combined data from mutagenesis, chemical cross-linking, mass spectrometry, and computational modeling to provide an internally consistent description of CYP17A1 interactions with its allosteric modulator b 5 . First, we found

TABLE 2 Cross-linked peptide pairs between CYP17A1 and cytochrome b 5
Filtering cutoff is false discovery rate of Ͻ0.05, and a spectral count of Ͼ2.

CYP17A1
Cytochrome b DECEMBER 5, 2014 • VOLUME 289 • NUMBER 49 that only some b 5 mutations containing at least one acidic residue at positions 48 and 49 stimulate 17,20-lyase activity, affirming the stringent requirement for a carboxylate group in the ␣3 helix of b 5 for this function. Second, we showed that b 5 forms specific cross-links with CYP17A1, including mutations R347K and R358K, and that b 5 mutation E48G/E49G disrupts this cross-linking. EDC, a zero-length cross-linking reagent, only reacts with amino and carboxylate groups in close proximity, which allows a very precise localization of protein interaction sites. Third, we identified two cross-linked peptides, which clustered in two adjacent regions for each protein. Finally, from the distance constraints imposed by the cross-linking and mutagenesis data, we used computational methods and existing x-ray crystal structures to generate a low resolution model of the protein complex. These results represent the first observation of CYP17A1 and b 5 cross-links and reveal the closest contacts between CYP17A1 and b 5 , which involve Arg-347 and Arg-358 of CYP17A1 with Glu-48 and Glu-49 of b 5 at distances of 1.7-3.9 Å. All four CYP17A1 residues (Lys-88, Arg-347, Arg-358, and Arg-449) and three b 5 residues (Glu-48,

CYP17A1-Cytochrome b 5 Interaction Surface
Glu-49, and Arg-52) predicted by mutagenesis studies to be interacting with each other are on the interface of our model (Fig. 6D, Table 3). Surprisingly, even though mutagenesis and NMR studies strongly implicate b 5 residues Glu-48 and Glu-49 as critical for stimulating 17,20-lyase activity, we did not identify these residues engaged in cross-linked amide bonds from CYP17A1. Instead, acidic residues adjacent to the b 5 ␣3 helix containing Glu-48 and Glu-49 formed cross-links with CYP17A1 residues. Similarly, genetic and biochemical evidence confirm the importance of CYP17A1 residues Arg-347 and Arg-358 for 17,20lyase activity and interactions with POR and b 5 ; however, EDC does not react with the guanidinium group of arginine and thus cannot be used to identify interacting residues on b 5 . We therefore employed CYP17A1 mutation R347K to trap residue 347 in a cross-link with Glu-42 of b 5 . In contrast, mutant CYP17A1 residue Lys-358 was not found in cross-linked peptides with b 5 in our experiments. It is possible that other cross-linked peptides containing mutant CYP17A1 residues Lys-347 or Lys-358 and/or b 5 residues Glu-48 or Glu-49 were present in our exper-iments but were not identified for several possible reasons. The most likely reason is that cross-linked products involving Lys-358 are not amenable to detection under the mass spectrometric conditions and computational methods employed, due to either insolubility, resistance to trypsin digestion, poor ionization, or subsequent further chemical modification. In addition, the side chain of Lys-358 appears to be buried deep in the interaction surface and might not be accessible to the EDC reagent in the solvent or might rapidly form an intramolecular crosslink. A zero-length cross-linking agent such as EDC identifies only a very limited number of interacting residues due to strict steric and chemical requirements. This analysis underscores the limitations of cross-linking experiments alone and emphasizes the need to combine these results with mutagenesis, modeling, and other structural studies. Nevertheless, these experiments yielded sufficient results to construct a structural model, which explains the available biochemical data.
The sites of cross-linking identified in this study are all located on the proximal surface of CYP17A1 near the heme, in the conserved redox partner-binding site. Lys-88 of CYP17A1, which cross-links to Glu-61 of b 5 , is located at the end of helix B. The CYP17A1 mutation K88A shows a 30 -40% diminished affinity for b 5 (29,30), and CYP17A1 mutation K89N exhibits a selective 3-fold reduction in 17,20-lyase activity (31). In fact, Lys-88 of CYP17A1 corresponds to Lys-96 of CYP3A4, and the latter has been shown to interact with the same residue on helix 4 of b 5 (Glu-61, human b 5 numbering) (32). When Arg-347 of CYP17A1 located on helix JЈ was mutated to lysine, a cross-link was readily formed with Glu-42 of the b 5 molecule (Fig. 5A). This cross-link specifies the orientation of the ␣2-loop-␣3 segment of b 5 , including key residues Glu-42, Glu-48, and Glu-49, to align with the proximal concave surface of CYP17A1 and is critical for determining the topological features of the CYP17A1-b 5 complex. Further support for this model derives from site-directed mutagenesis studies, where substitution of either Arg-347, Arg-358, or Arg-449 to alanine in human CYP17A1 prevented binding of b 5 and eliminates 17,20-lyase activity without impairing 17-hydroxylase activity (11).
The interaction sites of b 5 with CYP2B4 (33, 34), CYP2E1 (35), CYP3A4 (32), and CYP17A1 (this study and Refs. 6, 7) identified by cross-linking, NMR mapping, and site-directed mutagenesis, are summarized in Table 4. Electrostatic attractions appear to dominate the orientation between each P450 and b 5 to generate pre-collisional encounter complexes. Multiple sequence alignments among these P450s has shown that there are a significant number of favorable contacts between amino acid residues at different regions of the structure. For CYP2B4, a total of seven side chains participates in b 5 interactions, and these residues are all located in the C-helix on the proximal surface of the protein, except for Lys-433, which is located in the ␤-bulge (33). All of the five residues of b 5 that interact with CYP2B4 are located on the ␣4-loop-␣5 segment, which are distinct from the residues that interact with CYP17A1. However, CYP2E1 uses Lys-428 and Lys-434 to interact with Glu-58 and Glu-61 of b 5 (35). Even though Lys-434 of CYP2E1 corresponds to Lys-433 in CYP2B4, these residues do not interact with the same residue of b 5 . CYP3A4 interactions with b 5 appear to be more complex than for these other P450s. Lys-421 of CYP3A4 and Arg-347/Arg-358 of CYP17A1 were found to interact with the same acidic residues (Glu-42, Glu-48, and Glu-49) of b 5 . In addition, CYP3A4 also uses another two cationic residues, Lys-96 (corresponding to Lys-88 of CYP17A1) and Lys-127 (corresponding to Arg-122 of CYP2B4), to interact with Glu-61 of b 5 (32), based on crosslinking data. One limitation to comparing our results with other EDC cross-linking studies of P450s with b 5 is our use of lysine mutagenesis to target arginine residues implicated in interactions with b 5 . Therefore, we cannot exclude that additional arginine residues of these P450s participate in b 5 binding, but these interactions have not been identified. It is apparent that the b 5 binding profiles of CYP2B4 are relatively similar to that of CYP2E1 but remarkably different from that of CYP17A1, yet the promiscuous CYP3A4 enzyme shows a combination all of these interactions with b 5 . Our data are consistent with the recent NMR study of Estrada et al. (8), in which acidic residues Glu-48 and Glu-49 of b 5 and Arg-347, Arg-358, and Arg-449 of CYP17A1 are all critical in protein-protein binding. Evidence from chemical shift mapping also indicates that b 5 and POR compete for a binding surface on CYP17A1 and that the presence of substrate influences b 5 -induced conformational changes in CYP17A1 (8). In our hands, cross-linking was most efficient in assays reconstituted with phospholipids (PE/PS ϭ 1:1) and substrates, and the formation of the CYP17A1-b 5 complex increased in the following order: no substrate Ͻ progesterone Ͻ pregnenolone ϭ 17-hydroxypregnenolone (Fig. 3C). Minor discrepancies might result from the different experimental conditions used, including protein concentrations and phospholipid presence, forms of CYP17A1 employed (amino-terminal truncated and soluble versus modified and detergent-solubilized), and other factors. Nevertheless, the ability of substrate binding to affect CYP17A1-b 5 complex formation is consistent in both studies.
In summary, we have described the electrostatic interactions of CYP17A1 with b 5 , using several methods to generate an internally consistent model. This work is an important step in elucidating the structural features responsible for the 17,20lyase reaction and P450 redox partner binding in general. The 17,20-lyase activity of CYP17A1 is the only pathway to androgen biosynthesis known and is an important target for drug discovery. Abiraterone, galeterone, and related steroid azoles (2,36,37) are potent CYP17A1 inhibitors that markedly suppress androgen synthesis, which is desirable for treating prostate cancer (37) and androgen excess disorders such as 21-hydroxylase deficiency (38). These tight-binding heme-directed inhibitors, however, have the undesirable consequence of inhibiting 17-hydroxylase activity as well, which leads to the accumulation of upstream steroids that cause hypertension (39,40) as occurs in genetic 17-hydroxylase deficiency (41), unless combined with glucocorticoid therapy. Our current results might be exploited to develop selective inhibitors of the 17,20lyase reaction, by targeting this CYP17A1 interaction with b 5 , if such compounds do not disrupt reduction by POR. Proteinprotein interactions are traditionally viewed as difficult targets for chemical intervention, as these interacting surfaces often lack the deeper pockets and clefts found in enzyme-active sites. Nevertheless, there is a growing interest in this approach and success using fragment-based methods to target protein interfaces at "hot spots" or allosteric sites that alter protein conformations (42). Given the immense burden of androgen-dependent diseases, further studies to define and to target the CYP17A1-b 5 interaction are likely to yield clinically important results.