Structure of human cortisol-producing cytochrome P450 11B1 bound to the breast cancer drug fadrozole provides insights for drug design

Human cytochrome P450 11B1 (CYP11B1) is responsible for the final step generating the steroid hormone cortisol, which controls stress and immune responses and glucose homeostasis. CYP11B1 is a promising drug target to manage Cushing's disease, a disorder arising from excessive cortisol production. However, the design of selective inhibitors has been hampered because structural information for CYP11B1 is unavailable and the enzyme has high amino acid sequence identity (93%) to a closely related enzyme, the aldosterone-producing CYP11B2. Here we report the X-ray crystal structure of human CYP11B1 (at 2.1 Å resolution) in complex with fadrozole, a racemic compound normally used to treat breast cancer by inhibiting estrogen-producing CYP19A1. Comparison of fadrozole-bound CYP11B1 with fadrozole-bound CYP11B2 revealed that despite conservation of the active-site residues, the overall structures and active sites had structural rearrangements consistent with distinct protein functions and inhibition. Whereas fadrozole binds to both CYP11B enzymes by coordinating the heme iron, CYP11B2 binds to the R enantiomer of fadrozole, and CYP11B1 binds to the S enantiomer, each with distinct orientations and interactions. These results provide insights into the cross-reactivity of drugs across multiple steroidogenic cytochrome P450 enzymes, provide a structural basis for understanding human steroidogenesis, and pave the way for the design of more selective inhibitors of each human CYP11B enzyme.

CYP11B1 and CYP11B2 perform the final steps in biosynthesis of the glucocorticoid cortisol and mineralocorticoid aldosterone, respectively, whereas CYP19A1 produces estrogenic sex steroids (Fig. S1). Alterations in any of these steroidogenic P450 enzymes can cause human disease. The major human glucocorticoid cortisol regulates several catabolic and signaling pathways, including glucose homeostasis and stress and immune responses (1)(2)(3). Excess CYP11B1 cortisol production results in Cushing's disease with suppressed immune response, obesity, muscular weakness, hypertension, depression, and infertility (4 -6). Aldosterone produced by CYP11B2 regulates water and electrolyte homeostasis (7). Overproduction, called primary hyperaldosteronism, causes high blood pressure, one of the most common cardiovascular diseases (8,9). Estrogens control female development and fertility but can promote breast cancer proliferation, and selective inhibitors of CYP19A1 treat estrogen-responsive breast cancer (10). Selective inhibitors of CYP11B1 or CYP11B2 are of similar clinical interest (11)(12)(13)(14)(15)(16)(17) for Cushing's disease and hypertension. The development of such inhibitors has been impeded by the 93% amino acid sequence identity between them and their similar steroid-binding functionality. Although structures of membrane proteins are difficult to obtain, several are available of CYP11B2 with its physiological substrate 11-deoxycorticosterone and two nonselective inhibitors. However, there are no CYP11B1 structures to identify differences that can be exploited for clinical drug design or understand differences in enzymatic activity. Although CYP11B1 homology models were generated from the CYP11B2 structures (18 -20), these highlight similarities more than differences. Thus, the structural basis for distinctions in CYP11B enzyme function is unknown, impeding drug design.

CYP11B1 characterization
To obtain structural data for human CYP11B1, the enzyme was engineered to omit the N-terminal signaling sequence, recombinantly expressed, and solubilized and purified with detergent. Titrations of CYP11B1 and similarly produced CYP11B2 (21) with their respective substrates 11-deoxycortisol and 11-deoxycorticosterone (Fig. S2) revealed a blue shift in absorbance associated with substrate binding in the active site displacing the water normally on the heme iron (type I interaction). When reconstituted with NADPH and adrenodoxin reductase and adrenodoxin to deliver electrons, both CYP11B1  S3) and CYP11B2 (21) produce their respective cortisol and aldosterone products. Thus, both recombinant enzymes are fully functional. CYP11B1 and CYP11B2 both bound the racemic breast cancer drug fadrozole with a red shift in absorbance, consistent with fadrozole nitrogen coordinating the heme iron, replacing the water (type II interaction; Fig. S2). For comparison, fadrozole is reported to bind its intended CYP19A1 target with similar spectral changes and a K d of 16 nM, compared with 9 nM for CYP11B1 and 370 nM for CYP11B2, respectively (21). Thus, although fadrozole is used clinically as breast cancer drug inhibiting CYP19A1, it can have off-target effects on both CYP11B enzymes but serves as a starting point for more selective inhibitors.

Structure of CYP11B1 bound to the inhibitor fadrozole
A 2.1 Å X-ray structure of CYP11B1 co-crystallized with the nonselective fadrozole inhibitor (PDB entry 6M7X) reveals the characteristic cytochrome P450 fold (Fig. 1A). Consistent with binding studies, fadrozole binds in the CYP11B1 active site with a nitrogen of the imidazole ring coordinated directly to the heme iron, with the fused ring system oriented almost perpendicularly to the plane of the heme and the benzonitrile directed toward ␤1 (Fig. 1). Ligand nitrogen binding to the iron is known to provide significant affinity for P450 enzymes, but because the heme is conserved across P450 enzymes, this often contributes to significant nonspecificity, as is the case here, and results in undesirable side effects. Residues surrounding the active site can modify the affinity for individual P450 enzymes. In CYP11B1 (Fig. 1B), the nitrile of fadrozole hydrogen bonds to water, forming a network. Atoms within hydrogen-bonding distance of this water include the backbone nitrogen and carbonyl of Glu-383 in ␤ 1-4 , a side-chain nitrogen of Arg-384, and a heme propionate. Thus, this water (and its interactions with fadrozole) is highly stabilized. The roof of the active site opposite the heme is composed of largely hydrophobic residues compatible with hydrophobic regions of fadrozole.

Comparisons between CYP11B1 and CYP11B2 structures with the inhibitor fadrozole
Remarkably, comparison of the CYP11B2-fadrozole structure with the new CYP11B1-fadrozole complex revealed opposite preferences for the two fadrozole enantiomers. Although both CYP11B1 and CYP11B2 were crystallized with the racemic fadrozole used clinically for breast cancer (22), the density clearly supports CYP11B1 binding of the S enantiomer and CYP11B2 binding of the R enantiomer (21) (Fig. 2A). The R enantiomer of a similar compound tested preclinically as an inhibitor of CYP11B2 also binds CYP11B2 in a similar orientation (23). Although the imidazopyridine moiety is similarly oriented, binding the heme iron in both structures ( Fig. 2A), the benzonitrile projects toward ␤ 1-4 in CYP11B1 and along the central I helix in CYP11B2, where it interacts instead with Arg-120 (Fig. 2B). The basis for preferred binding of distinct enantiomers by CYP11B enzymes is based on unique active-site architectures. Despite all active site residues being identical in both CYP11B enzymes, large hydrophobic residues Trp-116, Phe-231, Trp-260, Phe-381, Phe-487, and Ile-488 forming the active site roof differ substantially in placement and orientation. Whereas Trp-116, Trp-260, Ala-313 are spread apart to accommodate the benzonitrile of (R)-fadrozole in CYP11B2, they are closer to each other in CYP11B1. On the other side of the active site, differential positioning of Phe-381 and Phe-487 permits CYP11B1 to accommodate the benzonitrile of (S)fadrozole. Previous computational studies suggested CYP11B1 preference for (S)-fadrozole, but this was based on a predicted orientation with the benzonitrile moiety directed toward the I helix (24), highlighting the necessity of experimental structures, even when structures of close homologs are available.
Many changes in active-site residue position are due to repositioning of their secondary structure elements. At the overall structural level, CYP11B1 and CYP11B2 are very similar on the proximal side of the heme, but major shifts occur in helices on the distal portions of the enzyme, including helices F-G and In both, the enzyme is shown in rainbow colors from blue at the N terminus to red at the C terminus. The heme is shown as black sticks with the iron as a rust sphere. The inhibitor fadrozole (structure inset) is shown in cyan sticks. In B, water is shown as a small red sphere with atoms within hydrogen bonding distance indicated by black dashed lines.

Cytochrome P450 11B1 structure
AЈ-A ( Fig. 2A). These elements compose the active site "roof" opposite the heme and are thought to be involved in substrate access to the otherwise buried active site. Thus, residue substitutions outside the active site ( Fig. S5) impact overall protein conformation and active site interactions (Fig. 3A).
Another notable difference in the CYP11B active sites is the conformation of the central I helix that forms one "wall" of the active site ( Fig. 1, yellow to yellow/green). Although this helix often has a slight interruption in the typical hydrogen-bonding pattern as it crosses over the heme, this region is much less helical in CYP11B1, having a break of five amino acid residues compared with two in CYP11B2 (Fig. 3B). A shift of this nonhelical region toward the active site center may reduce (R)fadrozole binding. Notably, the identity of residue 320 immediately following this nonhelical region is crucial for the 18-oxidase function of CYP11B2 (25) and may be important in differential binding and orientation of the respective substrates. Notably, introducing CYP11B1 residues into the CYP11B2 helices H (G288S) and I (L301P, E302D, and A320V) increased 11␤-hydroxylation efficiency for 11-deoxycortisol (25)(26)(27). Whereas residues 288, 301, and 302 are far from the active site and cause little difference at the backbone level, the larger CYP11B1 Val-320 correlates with different rotamers of the flanking Phe-193 and Phe-321 and shifts in the ␤ 4 system and beyond (Fig. 3B).
Structural revelation of CYP11B disparate preferences for fadrozole enantiomers is consistent with functional results. (R)-Fadrozole, also known as FAD286, has 5-fold selectivity for CYP11B2 over CYP11B1 in rat adrenal cortical preparations (28). As such, (R)-fadrozole was a promising compound in preclinical studies for hypertension caused by hyperaldosteronism. In contrast, the S enantiomer is reported to be the active CYP19A1-inhibiting agent in breast cancer treatment (29) and is clearly preferred by CYP11B1 (22). A fluorinated version of (R)-fadrozole called LCI699 or osilodrostat is the first CYP11B1-inhibiting compound in clinical trials for the treatment of excess cortisol in Cushing's disease. A multicenter study revealed that whereas LCI699 did normalize CYP11B1produced cortisol levels as desired, a simultaneous decrease in aldosterone was observed, likely partially due to simultaneous inhibition of CYP11B2 (30). However, a longer phase II trial did not note blood pressure changes, underscoring the complexities with poorly selective agents. Defining CYP11B1 preference for the S conformation paves the way for more selective drug design for the treatment of Cushing's disease.
The substrates of CYP11B1 and CYP11B2 are 11-deoxycortisol and 11-deoxycorticosterone, respectively. Whereas CYP11B1 is an efficient 11␤-hydroxylase (Fig. S3), CYP11B2 performs 11␤-hydroxylation but is much better at subsequent C18 hydroxylation and 18-oxidation to generate aldosterone, although the latter reaction is poor (Fig. S4). The structure of CYP11B2 with 11-deoxycorticosterone reveals the steroidal core oriented lengthwise over the heme, with C18 4.02 Å and C11 4.23 Å from the iron. Comparison with the CYP11B1 structure reveals that, if unaltered, the side chain of Phe-130 would sterically hinder binding of 11-deoxycorticosterone or 11-deoxycortisol with the orientation shown in CYP11B2, pushing the steroidal core away from it. As a result, C11 would likely be the site of oxidation closest to the heme iron, consistent with its function. It has been suggested that CYP11B2 is poor at overall aldosterone production because the multiple intermediates may dissociate from CYP11B2 via an open channel observed from the active site to the protein surface (21). However, CYP11B1 also has a similar channel (Fig. 4) yet is very efficient at 11␤-hydroxylation. This does not rule out differences in the conformational flexibility of the two proteins, but this idea should be investigated further, specifically with substrates and redox partner proteins that could modulate conformational changes. Regardless, the current structure also provides a context for evaluating naturally occurring CYP11B1 mutations that occur throughout the enzyme and cause disease in patients (19).

Conclusions
Overall, the CYP11B1 structure revealed crucial structural distinctions from CYP11B2 that underlie inhibitor binding and enzyme function and help rationalize the impact of diseaseassociated mutations while yielding valuable guidance in more

Protein expression and purification
CYP11B1-A synthetic, codon-optimized cDNA for CYP11B1 expression was generated (Blue Heron Biotech, Bothell, WA). This construct coded for a truncated protein with MATK prior to the 27th amino acid of the WT sequence and a His 4 tag at the C terminus. This gene was cloned into the pCWori vector, and the resulting construct was transformed into the E. coli strain DH5␣ already containing the pGro7 vector encoding the GroEL/ES chaperone. The construct was later modified to code for the sequence MAKKTSS prior to the 31st amino acid of the WT sequence (Fig. S5). A single colony from a lysogeny broth (LB) agar plate containing 100 g/ml carbenicillin for pCW selection and 20 g/ml chloramphenicol for pGro7 selection was used to initiate a 30-ml LB starter grown at 37°C with the same antibiotics. For expression, 800 ml of terrific broth (TB) with 2ϫ potassium phosphate buffer and respective antibiotics in a 2.8-liter Fernbach flask were inoculated with 8 ml of overnight culture and incubated at 37°C and 220 rpm until A 600 of 0.4 -0.6 was reached. CYP11B1 expression was then induced with 1 mM isopropyl ␤-D-1-thiogalactopyranoside (IPTG) and expression of the GroEL/ES chaperone with 4 mg/ml arabinose.

Cytochrome P450 11B1 structure
At this point, cultures were supplemented with 1 mM ␦-aminolevulinic acid heme precursor. Protein expression was carried out at 27°C at 190 rpm for 48 h, after which bacterial pellets were collected by centrifugation at 6,000 rpm and 4°C for 15 min. The cell pellet was resuspended in lysis buffer (50 mM potassium phosphate buffer (pH 7.4), 20% (v/v) glycerol, 500 mM sodium acetate, 1.5% (w/v) sodium cholate, 1.5% (v/v) Tween 20, 0.1 mM PMSF, 0.1 mM DTT), and the cells were disrupted by passage through a French press with one pass at 16,000 p.s.i. The cell debris was removed by ultracentrifugation at 35,000 rpm for 50 min at 4°C. The resulting supernatant was loaded on a Ni-NTA column (Ni/NTA Superflow, Qiagen) equilibrated with 3 CV of equilibration buffer (50 mM potassium phosphate buffer (pH 7. (v/v) glycerol, and 500 mM NaCl. The resulting purified protein was flash-frozen and stored at Ϫ80°C until further use. Purity was examined using SDS-PAGE analysis, and the ratio of absorbance of the Soret peak versus the total protein absorbance was determined at 280 nm. The R Z for pure CYP11B1 saturated with fadrozole is usually ϳ1.1.
Adrenodoxin reductase-The pET17b plasmid with human adrenodoxin reductase cDNA containing a C-terminal 6-residue His tag was kindly provided by Dr. Richard Auchus. The expression vector was co-transformed with the pGro7 vector in C41(DE3), and transformed cells were grown on an LB agar plate containing 100 g/ml carbenicillin for pET17b selection and 20 g/ml chloramphenicol for pGro7 selection. A single colony from the agar plate was used to grow a 30-ml starter culture with respective antibiotics in LB medium at 37°C for 5 h. For expression cultures, 1 liter of TB medium in 2.8-liter Fernbach flasks was inoculated with 25 ml of starter culture and grown at 37°C and 250 rpm. At an A 600 of 0.5, expression of adrenodoxin reductase was induced with 100 mg/liter IPTG and the chaperones with 3 g/liter arabinose. The cells were also supplemented with 0.1 mg/liter riboflavin 5-phosphate to support FAD synthesis. Cultures were incubated for 40 h at 20°C and 140 rpm. After collecting the cells at 6,000 rpm at 4°C for 15 min, the cell pellet was resuspended in lysis buffer (50 mM potassium phosphate buffer, pH 7.4, 300 mM NaCl, 20% (v/v) glycerol, 10 M FAD, 0.1 mM PMSF) and disrupted by a French press with one pass at 16,000 p.s.i. The cell debris was removed by ultracentrifugation at 35,000 rpm for 50 min at 4°C, and the supernatant was loaded on a Ni-NTA column, which was previously equilibrated with 3 CV of lysis buffer. The column was washed with 6 CV of lysis buffer, and reductase eluted with a histidine gradient from 0 to 80 mM, during which the main reductase fraction eluted at ϳ40 mM histidine. The eluted protein was concentrated in a 50-kDa centrifugal device for subsequent size-exclusion chromatography using 50 mM potassium phosphate, pH 7.4, 300 mM NaCl, and 20% (v/v) glycerol as a buffer. Eluting adrenodoxin reductase was identified by an absorbance at 450 and 280 nm. The respective fractions were pooled and concentrated, and the buffer was exchanged to storage buffer consisting of 50 mM potassium phosphate buffer (pH 7.4) and 20% (v/v) glycerol. The protein was flash-frozen and stored at Ϫ80°C. Purity was analyzed by SDS-PAGE analysis as well as spectral analysis of the ratio of the total protein absorbance at 270 nm versus the FAD absorbance maximum at 450 nm. At a value of 7.8, the protein is considered to be pure (31). Adrenodoxin reductase concentration was determined using the extinction coefficient at the absorbance maximum at 450 nm (⑀ ϭ 11.3 mM Ϫ1 cm Ϫ1 ) according to the Beer-Lambert law. Using this method, adrenodoxin reductase could be obtained with a final yield of 160 nmol/liter expression culture.
Adrenodoxin-A pLW plasmid containing coding for human adrenodoxin with a six-residue N-terminal histidine tag was kindly provided by Dr. Richard Auchus. The expression vector was transformed into C41(DE3) cells, and cells were plated on LB agar containing 100 g/ml carbenicillin for plasmid selection. A starter culture was grown from a single colony in 30 ml of LB medium supplemented with 100 g/ml carbenicillin at 37°C overnight.
For expression, 1 liter of TB medium in a 2.8-liter Fernbach flask was inoculated with 10 ml of starter culture and grown at 37°C and 250 rpm. At A 600 of 0.4 -0.6, adrenodoxin expression was induced with 1 mM IPTG, after which cultures were incubated at 28°C for an additional 20 h.
After cell collection by centrifugation at 6,000 rpm at 4°C for 15 min, the cells were resuspended in lysis buffer (50 mM potassium phosphate buffer, pH 7.4, 300 mM NaCl, 5% (v/v) glycerol, 0.1 mM PMSF) and disrupted by a French press with one pass at 16,000 p.s.i. After ultracentrifugation at 35,000 rpm for 50 min at 4°C, the supernatant was loaded on a 30-ml Ni-NTA column (Qiagen) equilibrated with lysis buffer. The column was washed with 8 CV of 50 mM potassium phosphate buffer, pH 7.4, 300 mM NaCl, 5% (v/v) glycerol, 4 mM histidine, and 0.1 mM PMSF, Cytochrome P450 11B1 structure and the adrenodoxin was eluted with 5 CV of the same buffer containing 80 mM histidine. Fractions containing adrenodoxin were identified by a higher absorbance at 414 nm, pooled, concentrated in a 10-kDa centrifugal device, and then diluted 5-fold with binding buffer (50 mM potassium phosphate buffer, pH 7.4, 5% (v/v) glycerol, and 0.1 mM PMSF) to decrease the salt concentration. For anion-exchange chromatography, two connected 1-ml HiTrap Q columns (FF) were equilibrated with 10 CV binding buffer. The protein was loaded, washed with 5 CV of binding buffer and then with 20 CV of the same buffer containing 60 mM NaCl, and eluted with 15 CV of binding buffer increased to 300 mM NaCl. Fractions exhibiting an absorbance at 414 nm and 280 nm were concentrated, and the buffer was exchanged to 50 mM potassium phosphate buffer, pH 7.4, with 5% (v/v) glycerol at Ϫ80°C. Purity was verified by SDS-PAGE analysis as well as the analysis of the absorbance spectrum. The absorbance ratio of the iron-sulfur cluster at 414 nm versus the total protein absorbance at 276 nm of 0.8 was considered to reflect a pure adrenodoxin sample. Adrenodoxin concentration was determined using the extinction coefficient at the absorbance maximum at 414 nm (⑀ ϭ 11 mM Ϫ1 cm Ϫ1 ) according to the Beer-Lambert law (32). Using this method, adrenodoxin could be obtained with a final yield of 2,000 nmol/liter expression culture.

Crystallization and structural determination
Initial crystals were grown from 0.5 l of fadrozole-saturated (Soret maximum at 423.5 nm) purified CYP11B1 (50 mM potassium phosphate buffer (pH 7.4), 20% (v/v) glycerol, 500 mM NaCl; 30 mg/ml determined using the Soret absorbance maximum and ⑀ ϭ 100 mM Ϫ1 cm Ϫ1 ) mixed with 0.5 l of 0.1 M imidazole, pH 7.4, and 12% (v/v) PEG 20,000 in a sitting drop using the Crystal Gryphon Robot (Art Robbins Instruments, Sunnyvale, CA). These drops were equilibrated against 65 l of the same imidazole solution at 20°C. After 3 weeks, small crystal plates could be detected. A seed stock was prepared using 1 l of these crystals mixed in 50 l of the imidazole solution and stored at Ϫ80°C until use. One l of this seed stock was mixed with 1 l of 40 mg/ml purified CYP11B1 for hanging-drop vapor-diffusion experiments against the same imidazole solution (500 l) at 20°C. Diffraction data were collected on beamline 12-2 of the Stanford Synchrotron Radiation Lightsource using Blu-Ice and processed using XDS (33) ( Table 1). The CYP11B1 structure was solved by molecular replacement using Phaser (34) and a search model consisting of CYP11B2 in complex with fadrozole (PDB entry 4FDH (21)). Model building and refinement were performed iteratively with Coot (35) and Phenix (36), respectively ( Table 1). The ligand fadrozole was generated using elBOW in Phenix (36). Figures were generated with PyMOL (37). Atomic coordinates and structure factors for the reported CYP11B1 crystal structures have been deposited with the Protein Data Bank under accession code 6M7X.