Structures of cytochrome P450 17A1 with prostate cancer drugs abiraterone and TOK-001.

Cytochrome P450 17A1 (also known as CYP17A1 and cytochrome P450c17) catalyses the biosynthesis of androgens in humans. As prostate cancer cells proliferate in response to androgen steroids, CYP17A1 inhibition is a new strategy to prevent androgen synthesis and treat lethal metastatic castration-resistant prostate cancer, but drug development has been hampered by lack of information regarding the structure of CYP17A1. Here we report X-ray crystal structures of CYP17A1, which were obtained in the presence of either abiraterone, a first-in-class steroidal inhibitor recently approved by the US Food and Drug Administration for late-stage prostate cancer, or TOK-001, an inhibitor that is currently undergoing clinical trials. Both of these inhibitors bind the haem iron, forming a 60° angle above the haem plane and packing against the central I helix with the 3β-OH interacting with aspargine 202 in the F helix. Notably, this binding mode differs substantially from those that are predicted by homology models and from steroids in other cytochrome P450 enzymes with known structures, and some features of this binding mode are more similar to steroid receptors. Whereas the overall structure of CYP17A1 provides a rationale for understanding many mutations that are found in patients with steroidogenic diseases, the active site reveals multiple steric and hydrogen bonding features that will facilitate a better understanding of the enzyme's dual hydroxylase and lyase catalytic capabilities and assist in rational drug design. Specifically, structure-based design is expected to aid development of inhibitors that bind only CYP17A1 and solely inhibit its androgen-generating lyase activity to improve treatment of prostate and other hormone-responsive cancers.

Cytochrome P450 17A1 (CYP17A1, P450c17, EC 1.14.99.9) is a membrane-bound dual-function monooxygenase with a critical role in the synthesis of many human steroid hormones[1]. CYP17A1 17α-hydroxylase activity is required for generation of glucocorticoids like cortisol, while its hydroxylase and 17,20-lyase activities are required for production of androgenic and estrogenic sex steroids (fig S1). CYP17A1 is an important target for the treatment of breast and prostate cancers that proliferate in response to estrogens and androgens[2,3]. In the absence of structural information, CYP17A1 inhibitors have been designed that are thought to bind the cytochrome P450 heme iron[4], but it has been difficult to rationalize or predict other structural features critical for effective, selective CYP17A1 inhibition. In addition, structural information is important to understand 17α-hydroxylase deficiencies and potentially polycystic ovary disease[7]. We determined structures of human CYP17A1 bound to two clinically-relevant CYP17A1 inhibitors (fig. S2). Abiraterone is the active form of a prodrug recently approved by the FDA for metastatic prostate cancer[5,8] and under investigation for breast cancer[9]. TOK-001 is currently in clinical trials for prostate cancer[4].

A truncated, His-tagged version of the human CYP17A1 protein was generated from a synthetic cDNA engineered to remove the single N-terminal transmembrane helix and expressed in E. coli. Resulting CYP17A1 was membrane-bound, so was solubilized with detergent before purification. This CYP17A1 binds abiraterone (fig. 1a) and TOK-001 (not shown) with absorbance decreases at 402 nm and increases at 424 nm, consistent with nitrogen binding to the heme iron (type II interaction) with Kd values of <100 nM (fig. 1a inset). Similar titrations with substrates progesterone (fig. 1b) and pregnenolone (not shown) revealed absorbance decreases at 419 nm and increases at 385 nm, indicative of ligand displacing water from the heme (type I interaction). CYP17A1 binds pregnenolone (Kd <100 nM, not shown) significantly more tightly than progesterone (Kd 229±14 nM; fig. 1b inset). In our hands, full-length enzyme[10] had a similar kcat and 3-fold higher Km (kcat 1.31±0.03 min−1, Km 11.4±0.7 μM) compared to the truncated form (kcat 1.31±0.03 min−1, Km 3.7±0.3 μM). IC50 values for abiraterone (201±1 nM) were lower than for TOK-001 (503.0±1.0 nM) (fig. 1c). Thus, truncated human CYP17A1 is a functional enzyme in terms of ligand binding, catalytic function, and inhibition.

Figure 1

Function of CYP17A1 and inhibition by clinical compounds

a, CYP17A1 titration with abiraterone (10–274 nM) yields progressive shifts in the UV/Vis difference spectrum typical of nitrogen binding to heme iron. b, Similar titration with progesterone (10–1535 nM) indicates water displacement from the heme iron. c, IC50 of abiraterone (circles) and TOK-001 (squares) for progesterone 17 α–hydroxylation.

Both structures with abiraterone (2.6 Å) and TOK-001 (2.4 Å) demonstrate the characteristic cytochrome P450 fold (fig. 2a) and have four very similar protein copies in each asymmetric unit (Table S1). Consistent with spectral binding data, abiraterone and TOK-001 bind with the nitrogen of the C17 pyridine or benzimidazole, respectively, forming a coordinate covalent bond with heme iron (fig. 2b and 2d). The steroid nucleus of these inhibitors rise at a 60° angle above the heme plane, directed between the F and G helices (fig. 2b and 2d), and essentially overlap (fig. 2f). The unsubstituted α face packs flat against the I helix where G301, A302, and adjacent residues form a highly complementary hydrophobic planar surface (fig. 2b). The 3β-OH groups of abiraterone (fig. 2b) and TOK-001 (fig. 2d) hydrogen bond with N202 in the F helix (~2.6 Å and ~2.4 Å, respectively).

Figure 2

CYP17A1 ligand binding

Stick and sphere representations have non-carbon atoms indicated as: blue, N; red, O; rust, Fe; heme, black; abiraterone, grey; TOK-001, cyan. a, CYP17A1/abiraterone structure colored from N-terminus (blue) to C-terminus (red). b, Abiraterone binds (2|Fo|−|Fc| density at 1 σ, blue mesh) at ~60° from heme against helix I (yellow). c, Abiraterone cavity (grey mesh), ~180° rotation from 2b. d, TOK-001 binding (2|Fo|−|Fc| density at 1 σ, blue mesh). e, TOK-001 cavity (grey mesh). f, Overlay of abiraterone (yellow) and TOK-001 (purple) structures with respective voids (mesh). B′ helix removed from panels c-f to view ligands.

Although inhibitors occupy the majority of the enclosed active site, the void extends beyond these ligands in several directions. First, the active site wall nearest the inhibitor β face is not as complementary to the steroid nucleus as for the α face. The C18 and C19 methyl groups project toward a crevice between the B′ helix, the β4 loop, and the loop following the F helix (fig. 2b). Only three side chains of the cavity wall come within 4 Å of C18 or C19. The cavity wall facing the β face of abiraterone or TOK-001 is primarily lined with hydrophobic atoms of A105, S106, A113, F114, I206, L209, V236, and V482 (fig. 2c), but there are two notable exceptions. R239 and D298 extend from the G and I helices, respectively, to orient their basic and acidic termini toward C6. These two polar side chains flank a substantial extension of the active site void adjacent C6. Second, in the abiraterone structure there is additional volume available adjacent to the pyridine ring bordered by V366, A367, I371, and V483 (fig. 2c), which is occupied by benzimidazole in the TOK-001 structure (fig. 2e). Finally, the most substantial active site cavity extension is from the 3β-OH of the inhibitors over the top of helix I and along the underside of helices F and G. This cavity is mostly lined by hydrophobic residues (I198, L243, F300), but its “roof” is bordered by several polar F and G helix residues (Y201, N202, R239, fig. 2c) that interact with, or are located near, waters in this region. The cavity containing TOK-001 is similar but slightly smaller over helix I (fig. 2e).

The single direct hydrogen bond between inhibitors and the protein is part of a larger hydrogen bonding network. In the abiraterone complex this network involves N202, E305, several conserved water molecules, R239, the backbone carbonyl of G297, and in some molecules Y201 (fig. 3). While Y201 is not within hydrogen bonding distance to these waters for molecules A and B, the side chain rotates slightly toward abiraterone in molecules C and D to interact with one or both of the water molecules. TOK-001 has a very similar hydrogen bonding network (fig S3). These interactions are strongly reminiscent of those conserved in the androgen, estrogen, glucocorticoid, mineralocorticoid, and progesterone receptors[11] (fig. 4a,b). In each receptor, the 3β-OH or 3-keto of steroids binds within a deep receptor pocket and forms hydrogen bonds with an arginine, a glutamine/glutamate, and often a conserved water molecule. These interactions are critical for ligand recognition by hormone receptors[12] and may also contribute to CYP17A1 selectivity for pregnenolone, progesterone, and their 17α-hydroxy derivatives. Notably, TOK-001 is both a CYP17A1 inhibitor and androgen receptor antagonist[13] and the similarity of these binding modes is likely the reason for this dual mechanism of action.

Figure 3

Hydrogen bond network with abiraterone

CYP17A1 has a hydrogen bonding network at the top of the active site that interacts with abiraterone. Molecule A/B (yellow) and C/D (green) have slightly different networks with the main difference being the involvement of Y201. Water molecules are indicated by small spheres. C17 and C16 are labeled. Hydrogen bonds indicated by dashed lines with distances indicated.

Figure 4

CYP17A1 compared to the androgen receptor and CYP11A1

a, The hydrogen bonding network near the abiraterone 3β-OH involves N202, R239, and conserved waters. b, The androgen receptor (PDB 3L3X) has a similar hydrogen bond network with R752, Q711, and several waters interacting with the dihydrotestosterone ketone. c, Overlay of CYP17A1 with abiraterone (yellow) and CYP11A1 with 20,22-dihydroxycholesterol (PDB 3NA0, pink) shows dramatically different steroid orientations.

Orientations of the native CYP17A1 substrates are of substantial value in understanding the function of this enzyme. Pregnenolone and progesterone were docked into the CYP17A1/abiraterone structure modeled as the Fe(IV)=O (compound I) catalytic state. Progesterone maintained the N202 hydrogen bond. The distances from C17 and C16 to the catalytic oxygen were 3.7 Å and 3.9 Å, respectively, consistent with the observed 17α-OH (major) and 16α-OH (minor) progesterone metabolites. The pregnenolone C17 atom was 3.6 Å from the compound I oxygen and the 3β-OH hydrogen bonded to N202 (Fig S4). However, the active site topology may be altered in the presence of substrates and this is an important area for further investigation.

CYP17A1 can be compared to three other P450 enzymes involved in steroidogenesis or cholesterol metabolism with reported steroid complex structures: CYP19A1 (aromatase)[14]; CYP11A1 (cholesterol side-chain cleavage enzyme)[15]; and CYP46A1, a cholesterol 24-hydroxylase[16]. Although all four enzymes maintain the canonical cytochrome P450 fold, the other three orient steroids in the opposite direction from CYP17A1. All three have the steroid ligand positioned over the K-L loop directed towards theβ1 sheet as in CYP11A1 (fig. 4c), instead of oriented toward helices F and G.

Over 50 CYP17A1 mutations have been identified, most in patients with 17-hydroxylase deficiencies. The biochemical effects of many clinical mutations can be rationalized by examining the CYP17A1 structure (Fig S5, Table S2). Mutations R96W, R125Q, H373D/N, and R440H/C all alter residues that directly interact with the heme propionates and likely disrupt heme binding, consistent with complete loss of activity[17-22]. Mutations E305G, R347H/C, R358Q, and R449A eliminate only the lyase activity of CYP17A1[23-25]. E305 hydrogen bonds to N202 in the active site, suggesting a role in substrate positioning, while the other residues are on the proximal face of the protein (fig. S5), consistent with their proposed role in cytochrome b5 binding[23,26], which promotes lyase activity. Finally, reduction of the minor 16α-hydroxyprogesterone metabolite is reported for the artificial mutation A105L[27], consistent with its location in the active site facing the β face where the additional bulk may reduce the steroid movement within the active site.

Abiraterone and TOK-001 have several features which make them effective inhibitors of CYP17A1: 1) a heterocyclic nitrogen that coordinates to the heme iron, 2) a planar α face to pack against the I helix, and 3) hydrogen bonding interactions of 3β-OH with conserved polar residues in a hydrogen binding network. These structures provide a model for the binding of substrates and other inhibitors that is very different from binding orientations previously proposed by homology modeling and docking studies and from those demonstrated in other steroid-metabolizing cytochrome P450 enzymes. Perhaps most importantly, the cavity observed is not bilobed as predicted by many modeling studies[28,29]. CYP17A1 interactions with these inhibitors are instead more reminiscent of steroid binding to steroid receptors, which may be the genesis of the TOK-001 dual mechanism of action. Thus, these structures contribute to a better understanding of the function and inhibition of CYP17A1 in a way that should substantially benefit the understanding of enzyme dysfunction in clinical disease and enable structure-based drug design of CYP17A1 inhibitors for treating hormone-responsive cancers, especially prostate cancer.

Methods Summary

A synthetic cDNA for human CYP17A1 was modified to delete residues 2–19, substitute the hydrophilic sequence 20RRCP[23] with 20AKKT[23], and add a C-terminal four histidine tag (fig. S6) before cloning into the pCWori plasmid and overexpression in E. coli JM109 cells. Protein was purified by nickel affinity, cation exchange, and size exclusion chromatography. Abiraterone was synthesized (Methods). Binding affinities were determined using a UV/vis spectral shift assay. Progesterone 17α-hydroxylation was evaluated using HPLC separation and UV detection. For crystallography, inhibitors were included throughout purification. Crystals were grown from CYP17A1 (30 mg/mL) complexed with inhibitor using hanging-drop vapor diffusion to equilibrate against 30% PEG 3350, 0.175 M Tris, pH 8.5, 0.30 M ammonium sulfate, and 3% glycerol. Diffraction data was collected and phased by molecular replacement. Iterative model building and refinement generated the final model. Substrates were docked using Surflex-Dock[30].

Methods

Synthesis and characterization of abiraterone, 17-(3-pyridyl)androsta-5,16-dien-3β-ol

A stirred solution of 17-iodoandrosta-5,16-dien-3β-ol (600 mg, 1.5 mmol) in THF (20mL) in a 100 mL round-bottomed flask was purged with argon. Bis(triphenylphosphine) palladium (II) chloride catalyst (11 mg, 0.016 mmol) was added, followed by diethyl(3-pyridyl)borane (265 mg, 1.8 mmol). To the resultant orange solution, an aqueous solution of sodium carbonate (2M, 5 mL) was added. The flask was fitted with a reflux condenser and the apparatus purged again with argon. The mixture was then heated under reflux (~80 °C) with stirring for 4 days then allowed to cool. The mixture was poured into water and extracted with hot toluene (3x30 mL). The toluene extracts were dried (Na2CO3) and concentrated. Column chromatography was performed with Et2O/toluene (1:2) as the eluent to give abiraterone (350 mg, 66%) as a white crystalline solid: mp 228–230 °C; IR νmax 3307 cm−1 (OH str); 1H NMR δ 1.07 (s, 3, H-19), 1,09 (s, 3, H-18), 3.54 (m, 1, H-3α), 5,41 (dm, 1, J = 5.2 Hz, H-6), 6.01 (m, 1, H-16), 7.24 (dd, 1, pyridyl H-5), 7.66 (dd, 1, pyridyl H-4), 8.47 (dd, 1, pyridyl H-6), 8.63 (d, 1, pyridyl H-2); 13C NMR δ 151.69, 147.92, 147.84, 141.19, 133.68, 132.98, 129.24, 123.03, 121.32, 71.65, 57.56, 50.36, 47.34, 42.32, 37.19, 36.71, 35.26, 31.81, 31.64, 30.45, 20.88, 19.35, 16.59. The HRMS calculated m/z C24H32NO [M+H]+ is 350.2484. The experimental value was 350.2491. Abiraterone was 99% pure by LCMS.

CYP17A1 design, expression, and purification

The human CYP17A1 cDNA was synthesized with codon optimization for E. coli expression (Blue Heron Biotechnology, Bothell, WA). A truncated and His-tagged construct was generated by truncation of the N-terminal transmembrane helix (Δ2–19), substitution of [20]RRCP[23] with [20]AKKT[23], and addition of a C-terminal four-residue histidine tag (fig. S6). N-terminal modifications were designed to increase solubility. This altered cDNA was inserted into the pCWori expression vector and expressed in E. coli JM109 cells.

Cells were grown, harvested, and disrupted as described[10,31]. After centrifugation (5000 × g), CYP17A1 was solubilized with either 4.8 mM Cymal-5 (for crystallography; Affymetrix, Santa Clara, CA) or 2% Emulgen 913 (for assays; Desert Biologicals, Phoenix, AZ), followed by ultracentrifugation (80,000 × g) for 60 minutes. The lysate was loaded onto a NTA-agarose (Qiagen, Valencia, CA) column and purified as reported[31]. Eluted CYP17A1 fractions were pooled, diluted 5-fold with CM buffer (50mM Tris, pH 7.4, 20% glycerol, 100 mM glycine, 1 mM EDTA), and loaded on a HiTrap CM fast flow column (GE Healthcare, Uppsala, Sweden). Protein was eluted in CM buffer with 0.5 M NaCl, concentrated to ~1 mL, and loaded on a Superdex 200 16/60 size exclusion column (GE Healthcare). For crystallography 10 μM abiraterone or TOK-001 (Shanghai Haoyuan Chemexpress Co., Shanghai, China) was added to all buffers.

Protein crystallization, data collection, and structure determination

CYP17A1 crystals were grown using hanging drop vapor diffusion equilibration. CYP17A1 (30 mg/mL) in CM buffer supplemented with 10 μM abiraterone or TOK-001, 2.4 mM Cymal-5, and 0.5 M NaCl was equilibrated against 30% PEG 3350, 0.175 M Tris, pH 8.5, 0.30 M ammonium sulfate, and 3% glycerol at 20 C. Crystals were cryoprotected in 7:1 mother liquor:ethylene glycol and flash cooled in liquid nitrogen. Native data sets were collected at 0.98 Å, 100 K at the Stanford Synchrotron Radiation Laboratory beamline 9–2. Data were processed using Mosflm[32] and Scala[33]. The abiraterone complex was solved by molecular replacement using BALBES[34] with a final search model based on CYP2R1 (PDB 3CZH) and the TOK-001 structure solved using Phaser[33] with the abiraterone structure as a search model. Model building and refinement were accomplished iteratively using COOT[35] and Refmac5[36] in CCP4[33]. Structure validation was performed using WHATCHECK[37] and PROCHECK[38]. Ramachandran plot analysis reveals percent favorable/additional allowed/generously allowed/disallowed residues are 86.3/13.2/0.5/0.0 (abiraterone structure) and 86.2/13.3/0.5/0.0 (TOK-001 structure). X-ray statistics are provided (Table S1). Probe-occupied voids were calculated using VOIDOO[40] (probe radius=1.4 Å; grid mesh=0.4 Å). All figures were prepared using MacPyMOL[39].

Docking

The CYP17A1 active site was defined as described for other cytochromes P450[40] with the addition of an oxygen molecule directly coordinated to the heme to mimic Compound I of the cytochrome P450 catalytic cycle. Substrate coordinates were prepared and energy minimized with SYBYL (Tripos, St. Louis, MO). Charges were assigned using the Gasteiger and Marsili method. Surflex-Dock (Tripos International, St. Louis, Missouri) was used to dock ligands as previously described[40]. The active site was a 10 Å sphere around the heme and pregnenolone. Movement of pregnenolone within the active site was not substantial with the distance of C17 to O=Fe(IV) 4.5 Å, C16 to O=Fe(IV) of 4.5 Å, and a distance from C21 to O=Fe(IV) of 3.0 Å for the lowest energy pose.

Enzyme activity and IC50 determinations

Progesterone 17α-hydroxylation was evaluated using a modified HPLC method with UV-detection[41]. CYP17A1 (50 pmol) and rat NADPH-cytochrome P450 reductase[42] 1:4 were mixed, incubated on ice (20 minutes), and added to buffer (50 mM Tris, pH 7.4 and 5 mM MgCl2) containing progesterone (0 – 50 μM) to a total volume of 500 μL. Phosphatidylcholine (25 μg) was included for side-by-side kinetic comparisons with the full-length enzyme[10]. For IC50 determinations, inhibitors concentrations were 0–1500 nM for abiraterone and 0–3000 nM for TOK-001. After warming (37° C, 3 minutes), reactions were initiated by NADPH addition (20 μL 25 mM), incubated for 10 minutes (37° C), and quenched with 20% trichloroacetic acid (300 μL) and placed on ice. The 17α-hydroxyprogesterone metabolite was identified by UV detection at 248 nm following HPLC separation and coeluted with authentic standards. The HPLC mobile phase was 40% acetonitrile, 60% water with 1% acetic acid and run at 1 mL/min (Phenomenex, Luna 5 μ, C18, 50 × 4.6 mm).

Ligand binding assays

Ligand binding assays based on spectral differences detected upon ligand titration were performed essentially as described[40] except that the CYP17A1 concentration was 0.1 μM, the path length was 5 cm, and the tight binding equation was used.

Functional data was analyzed using Prism (GraphPad Software, La Jolla, CA) and presented as mean ± standard error.