Structures of Prostacyclin Synthase and Its Complexes with Substrate Analog and Inhibitor Reveal a Ligand-specific Heme Conformation Change^,

Abstract

Prostacyclin synthase (PGIS) is a cytochrome P450 (P450) enzyme that catalyzes production of prostacyclin from prostaglandin H₂. PGIS is unusual in that it catalyzes an isomerization rather than a monooxygenation, which is typical of P450 enzymes. To understand the structural basis for prostacyclin biosynthesis in greater detail, we have determined the crystal structures of ligand-free, inhibitor (minoxidil)-bound and substrate analog U51605-bound PGIS. These structures demonstrate a stereo-specific substrate binding and suggest features of the enzyme that facilitate isomerization. Unlike most microsomal P450s, where large substrate-induced conformational changes take place at the distal side of the heme, conformational changes in PGIS are observed at the proximal side and in the heme itself. The conserved and extensive heme propionate-protein interactions seen in all other P450s, which are largely absent in the ligand-free PGIS, are recovered upon U51605 binding accompanied by water exclusion from the active site. In contrast, when minoxidil binds, the propionate-protein interactions are not recovered and water molecules are largely retained. These findings suggest that PGIS represents a divergent evolution of the P450 family, in which a heme barrier has evolved to ensure strict binding specificity for prostaglandin H₂, leading to a radical-mediated isomerization with high product fidelity. The U51605-bound structure also provides a view of the substrate entrance and product exit channels.

Prostacyclin (also known as prostaglandin I₂ (PGI₂))⁴ is a potent inhibitor of vasoconstriction, platelet activation, and aggregation (1). Widely known for its vasoprotective activity, PGI₂ is mainly synthesized in the endothelial and smooth muscle cells via an isomerization of prostaglandin H₂ (Fig. 1A), a reaction catalyzed by prostacyclin synthase (PGI₂ synthase; PGIS). PGIS is localized to the endoplasmic reticulum membrane and was assigned to the cytochrome P450 (P450) superfamily as CYP8A1 when its cDNA sequence was determined (2, 3). Heme-containing P450 enzymes are present in most living organisms and form one of the largest protein superfamilies. By acquiring electrons from NAD(P)H and switching between the redox states of heme iron, P450 enzymes typically activate molecular oxygen to catalyze a mono-oxygenation reaction, leading to hydroxylation, epoxidation, peroxidation, or demethylation of chemically divergent substrates (4). Among the hundreds of P450s identified to date, PGIS and its counterpart thromboxane synthase are the only two P450s that metabolize an endoperoxide moiety as their physiological substrate. Moreover, these two enzymes carry out isomerization of endoperoxides without the need for either molecular oxygen or any external electron donors (5).

FIGURE 1. Proposed mechanism for PGI₂ biosynthesis.

A, chemical structures of substrate PGH₂, substrate analog U51605, and inhibitor minoxidil. B, a proposed mechanism for the reaction catalyzed by PGIS (R1, CH₂-(CH₂)₂-COOH; R₂, CH-CHOH-(CH₂)₄-CH₃) (6).

Based on their elegant work, Hecker and Ullrich (6) proposed a now widely accepted catalytic mechanism for PGIS (Fig. 1B), in which an isomerization reaction is triggered by a stereospecific binding of C-11 oxygen of PGH₂ to the heme-iron of PGIS. Following the initial substrate binding, a one-electron transfer from Fe(III) to an O–O bond induces a homolytic cleavage of the endoperoxide, leading to the formation of Fe(IV)-porphyrin and alkoxy radical. Cyclization between the radical and the alkene C-6 takes place to produce a five-member ring and the C-5 radical. The C-5 radical then donates an electron to Fe(IV)-porphyrin to form the C-5 carbocation. Subsequent loss of the C-6 proton and formation of C-5=C-6 double bond yield PGI₂.

To provide a structural understanding on the biosynthesis of PGI₂, we have previously determined the crystal structure of ligand-free human PGIS (hPGIS), which provides the first view of this atypical P450 (7). Consistent with earlier speculation that all P450 structures probably share a similar triangular prism-shaped tertiary architecture (8), the hPGIS structure closely resembles other P450s. However, some notable differences have been recognized. It is generally accepted that structurally divergent regions around the active sites of P450s allow the recognition of a wide variety of substrates of different size, shape, and polarity (9, 10). All of these regions are located at the heme distal side in the F and G helices and F/G loop, which form the roof of the substrate-binding site, as well as the B′ helix, which is at the entrance of the substrate access channel. In accordance with this view, significant differences in the locations of B′, F′, G, and I helices are observed between hPGIS and other P450s. In addition, structural analyses suggest that the most conserved structural elements among P450s are those involved in heme binding, including the invariant proximal cysteine in the “cysteine ligand loop” for heme-iron ligation and two heme-flanking regions for interacting with the A- and D-ring propionates of the heme (11). Through hydrogen bonds and salt bridges, the acidic heme is housed in the otherwise hydrophobic interior of the P450 fold. Interestingly, no interactions were observed between the heme propionates and the hPGIS protein scaffold.

Despite the recognition of these structural features of hPGIS, critical questions regarding the structural basis for PGI₂ biosynthesis and the uniqueness of PGIS as a P450 enzyme remain unanswered. For example, it is not known how PGIS can distinguish between the two chemically equivalent endoperoxide oxygen atoms of substrate PGH₂, with C-11 oxygen being favored over C-9 oxygen as the distal heme ligand (6). In addition, the cyclization reaction is accompanied by the formation of the C-5 radical; it remains a question how this carbon radical donates an electron to Fe(IV)-porphyrin to assist the subsequent catalytic steps. Moreover, given that PGIS favors homolytic cleavage of the peroxide bond of fatty acid hydroperoxides (12, 13), a reaction known to be facilitated in a nonpolar microenvironment, and that both PGH₂ and PGI₂ are hydrophobic and are labile in aqueous solution, will the structure of solvent molecules in the PGIS active site change in the presence of substrate? Finally, does the lack of direct interactions between heme propionates and protein scaffold have potential functional significance? In an effort to gain a better understanding of the structure/function of this atypical P450, we have determined the crystal structures of evolutionarily distant zebrafish PGIS (zPGIS) in the ligand-free and substrate analog U51605-bound forms as well as the inhibitor minoxidil-bound hPGIS (the chemical structures of these compounds are shown in Fig. 1A). These structures provide a framework for deciphering the structural basis for the biosynthesis of PGI₂.

EXPERIMENTAL PROCEDURES

Materials

9,11-Azoprosta-5,13-dienoic acid (U51605) and 15-hydroxy-9,11-(methanoepoxy)prosta-5,13-dienoic acid (U46619) were purchased from Cayman. 1-(o-Chloro-α,α-di-phenyl-benzyl)imidazole (clotrimazole) and 6-(1-piperidyl)-2,4-diaminopyrimidine-3-oxide (minoxidil) were from Sigma. PGH₁ and PGH₂ were synthesized as described previously (12).

Cloning, Expression, and Purification of zPGIS

Total RNA was extracted from a zebrafish obtained from a local pet store, using a ToTALLY RNA^™ kit (Ambion). Reverse transcription-PCR was performed using a Superscript one-step reverse transcription-PCR kit (Invitrogen) with zPGIS-specific primers (forward and reverse primers are nucleotides 514–533 and 1430–1449, respectively, relative to the ATG site; GenBank^™ accession numbers BE558055, BE558156, BE605271, and BE605774), which amplified a fragment lacking the N-terminal segment. The 5′-RACE technique was then employed using zPGIS-specific antisense primers (nucleotides 532–551 for reverse transcription and nucleotides 514 –534 and an adaptor primer for PCR) (14). Subsequently, an additional round of 5′-RACE was required to obtain the translational start site. The 5′-RACE fragments were linked to the 3′-segment by PCR-mediated ligation. The resulting DNA fragment containing the entire coding sequence and ~30 bp of 5′-untranslated region was cloned to the plasmid pBluesrcipt and was sequenced.

The strategy to express the recombinant zPGIS was adopted from that for hPGIS (12). Briefly, the zPGIS cDNA was modified so that the peptide segment MAKKTSS replaced the first 16 amino acid residues, and a 4-histidine tag was added at the C terminus. The resulting cDNA was cloned into a prokaryotic expression vector, pCW, transformed into Escherichia coli BL21(DE3)pLysS, and the recombinant protein was induced by isopropyl 1-thio-β-D-galactopyranoside in the presence of δ-aminolevulinic acid. The purification of zPGIS was essentially as described previously for hPGIS using a nickel affinity column, CM-Sepharose, and gel filtration chromatography (7).

Characterization of the Recombinant zPGIS

The enzymatic product of zPGIS was analyzed by reverse phase HPLC using [1-¹⁴C]PGH₂ as the substrate. Products were extracted using a Sep-Pak C18 cartridge (Waters, Milford, MA) and then resolved by a C₁₈ column following the previous method (15) and detected by on-line liquid scintillation with a β-RAM model 2 β-detector (IN/US System, Tampa, FL). Initial rate determinations were carried out in 200 μl of 20 mM NaP_i, pH 7.4, and 0.2% Lubrol PX by incubating 10 pmol of zPGIS with an appropriate amount of PGH₂ for 20 s at 23 °C. Formation of 6-keto-prostaglandin F_1α (6-keto-PGF_1α), the stable hydration product of PGI₂ (16), was measured by radioimmunoassay (17). Ligand binding affinity was determined by monitoring the absorbance changes of the Soret band as described (12). The zPGIS-CO complex was obtained by treating zPGIS with dithionite under anaerobic conditions and then bubbling CO gas into the solution for a few seconds. The decay rate of the ferrous-CO complex was determined by stopped-flow spectroscopy as described (12).

X-ray Crystallography

After the final gel filtration chromatographic step, fractions containing zPGIS were pooled and concentrated by ultrafiltration to 20 mg/ml. Initial crystallization trials for ligand-free zPGIS were performed with commercially available kits (Hampton Research), using the hanging drop vapor diffusion method. Specifically, 1 μl of concentrated zPGIS solution in gel filtration buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 5 mM β-mercaptoethanol) were mixed with an equal amount of reservoir solution and equilibrated against 450 μl of reservoir solution at 277 K. Conditions that produced small crystals were refined by systematic variation of the precipitant concentration and pH. Diffraction quality crystals were obtained at 277 K using a reservoir solution consisting of 20% polyethylene glycol 3350 plus either 50 mM Tris-HCl (pH 8.0) or 50 mM HEPES (sodium salt; pH 7.5). The crystals of U51605-zPGIS complex were prepared by soaking the ligand-free zPGIS crystals in 2 μl of substitute mother liquor (23% polyethylene glycol 3350, 50 mM HEPES (sodium salt; pH 7.5)) containing 0.25 mM U51605 for 1 week. Prior to x-ray data collection, zPGIS crystals were transferred to a cryostabilization solution (substitute mother liquor plus an additional 20% ethylene glycol) for ~10 s and immediately flash-frozen by plunging into liquid N₂. All diffraction data of zPGIS crystals were collected at NSRRC, Taiwan (beamline BL13B1, ADSC Quantum 315 CCD detector, 100 K) and processed using the HKL2000 program suite (18).

The structure of the ligand-free zPGIS was solved by molecular replacement with the program PHASER (19) by using a polyalanine structure of hPGIS (Protein Data Bank code 2IAG) as the search model. The initial phases derived from molecular replacement solution were improved by density modification with the program RESOLVE (20). The resulting electron density map was of excellent quality, which allowed RESOLVE (21, 22) to build 679 of a total of 950 residues in the asymmetric unit. The programs O (23) and REFMAC5 (24) were then used for rounds of manual model rebuilding and refinement. Two zPGIS molecules (designated chains A and B) are present in the asymmetric unit. No electron density is associated with residues 305–316 of chain A and 155–157, 183–186, 232–243, and 305–316 of chain B, so these regions were omitted from the model. The final model consists of 873 residues (446 and 427 residues in chains A and B, respectively), two heme groups, and 282 water molecules. A total of 95.3 and 4.5% of nonglycine residues have main-chain torsion angles in the most favored and generously allowed regions of Ramachandran plot, respectively.

After deleting the water molecules, ligand-free zPGIS structure was used as a starting point to refine the highly isomorphous crystal structure of U51605-bound zPGIS. Following several cycles of rigid body and restrained maximum likelihood refinement with REFMAC5 (24), the binding sites for U51605 were identified in the resulting omit map. A total of three ligands are bound per zPGIS molecule. As before, the program O (23) and REFMAC5 (24) were then used for rounds of manual model rebuilding and refinement. The final U51605-bound zPGIS model contains 860 residues (with residues 305–316 and 393–394 of chain A and 152–157, 181–190, 211–215, 233–243, and 305–316 of chain B missing), two heme groups, six U51605 molecules, and 223 water molecules.

The inhibitor-hPGIS complex was prepared by adding minoxidil (final concentration 0.1 mM) to a 20 mg/ml hPGIS stock solution. Crystals of the complex were then obtained by co-crystallization, with the technique described previously using ligand-free hPGIS microcrystals as seeds. Diffraction data from a single crystal were collected at 100 K at the NSRRC BL13C1 beamline. Based on the strategy described above for determining the structure of U51605-zPGIS complex, the structure of minoxidil-hPGIS complex was solved using the ligand-free hPIGS structure as a starting model. The final minoxidil-bound hPGIS model contains 939 residues (469 residues in chain A, with residues 320 –329 missing; 470 residues in chain B, with residues 320 –328 missing), two heme groups, two minoxidil molecules, two β-octyl-glucopyranoside molecules, and 765 water molecules. All data collection and refinement statistics are summarized in Table 1.

TABLE 1.

Summary of crystallographic analysis

Parameter	Value
	Ligand-free zPGIS	U51605-bound zPGIS	Minoxidil-bound hPGIS
Space group	P212121	P212121	P21
Unit cell dimensions (a, b, c) (Å)	58.7, 87.9, 190.9	58.5, 88.0, 190.1	68.762, 106.072, 73.90; b = 91.8
Data collection
Wavelength (Å)	1.0000	1.0000	0.97622
Resolution (Å)	30-2.08 (2.15-2.08)	30-2.50 (2.59-2.50)	30-1.62 (1.66-1.62)
Observed reflections	325,736	153,017	499,495
Unique reflections	60,087	32,622	133,648
Completeness (last shell) (%)	99.9 (99.8)	92.3 (94.2)	99.5 (97.5)
Multiplicity	5.4	4.7	3.7
Mean 〈I/σI〉 (last shell)	18.0 (4.3)	12.7 (2.8)	16.1 (2.0)
Rsyma (last shell) (%)	4.6 (21.0)	6.4 (30.2)	4.0 (32.3)
Refinement
Resolution range (Å)	29.91–2.08	27.75–2.48	30–1.62
No. of reflection in working set (test set)	56,958 (3052)	30,957 (1643)	126,835 (6770)
Rcrystb (%)	22.8 (22.5)	21.4 (27.3)	20.3 (24.6)
Rfreeb (%)	26.3 (29.2)	29.2 (35.6)	22.8 (30.0)
r.m.s. deviation from ideal
Bond lengths (Å)	0.008	0.009	0.007
Bond angles (degrees)	1.1	1.4	1.0

^aR_sym = (Σ|Ihkl − 〈I〉|)/(ΣIhkl), where the average intensity 〈I〉 is taken overall symmetry equivalent measurements, and Ihkl is the measured intensity for any given reflection.

^bR_cryst = (Σ||F_o| − k|F_c||)/(Σ|F_o|). R_free = R_cryst for a randomly selected subset (5%) of the data that were not used for minimization of the crystallographic residual.

RESULTS

Cloning and Functional Characterization of zPGIS

The complete coding sequence for zPGIS was isolated using 5′-RACE. The cDNA encodes a 55-kDa protein of 480 amino acid residues (Fig. 2). Sequence alignment reveals that zPGIS shares 45% sequence identity with hPGIS, which consists of 500 amino acid residues (Fig. 2). The primary sequence of zPGIS reveals that the consensus sequence of a highly conserved cysteine ligand loop present in all other P450s, F(G/S)XGX(H/R)XCXG, is somewhat unusual and has the sequence WGTEDNLCPG. This region shares only 5 of 10 residues with the loop of hPGIS (WGAGHNHCLG).

FIGURE 2. Alignment of the primary sequences of human and zebrafish PGIS.

Residues in the putative substrate recognition sites are shaded, and those contacting the heme-ligated U51605 in the zPGIS structure are indicated by asterisks above the sequence. The secondary structural elements are underlined, α helix with a thick line and β sheet with a thin line. The numbering of the residues is shown on the right.

To demonstrate that the cDNA encodes the bona fide PGIS, we adopted a strategy previously used for heterologous expression of hPGIS to obtain the recombinant protein for testing the catalytic activity (12). The zPGIS cDNA was modified by replacing the putative N-terminal transmembrane domain with a hydrophilic sequence, MAKKTSS, and adding a C-terminal His tag. The resulting recombinant zPGIS was purified to electrophoretic homogeneity and was tested for production of 6-keto-PGF_1α, the nonenzymatically hydrolyzed product of PGI₂, using radioimmunoassay and reverse phase HPLC profile analysis. Both assays identified PGI₂ as the enzymatic product. It should be noted that HPLC analysis using [1-¹⁴C]PGH₂ as the substrate yielded only one product peak that co-migrated with the 6-keto-PGF_1α standard (Supplemental Fig. 1A). Furthermore, we used PGH₁ as a substrate to test the enzymatic activity of recombinant protein. PGIS converts PGH₁ to malondialdehyde and 12-hydroxyheptadecadienoic acid, which have absorbance peaks at 268 and 232 nm, respectively (6). Absorption spectra taken after PGH₁ incubated with our recombinant protein displayed the concomitant formation of two absorbance peaks at 268 and 232 nm (Supplemental Fig. 1B). Together, these data indicate that the recombinant protein is the authentic zPGIS.

Purified zPGIS exhibited an absorption spectrum typical of low spin P450s, with a Soret peak at 418 nm and α and β bands at 570 and 537 nm, respectively. The absorption features of zPGIS are indistinguishable from those of hPGIS (12). Steady state kinetic analysis for zPGIS gave K_m and V_max values very close to those for recombinant hPGIS (Table 2). Unlike many P450s, which bind various shapes and sizes of ligands with their nitrogen, carbon, or oxygen atom coordinated to the sixth position of the heme iron, PGIS has only a few known ligands (25). Spectral perturbation titration was conducted to determine the K_d of zPGIS for NaCN (carbon-based ligand), U46619 (oxygen-based ligand and substrate analog), clotrimazole (nitrogen-based ligand), minoxidil (oxygen-based ligand and inhibitor), and U51605 (nitrogen-based ligand and substrate analog). The dissociation constants and ligand-induced shift of Soret peak for zPGIS are comparable with those simultaneously determined for hPGIS (Table 2).

TABLE 2.

Catalytic and ligand-binding properties of recombinant hPGIS and zPGIS

	hPGIS	zPGIS
Parameters of steady-state kineticsa
Km (μM)	13 ± 1	25 ± 5
Vmax, mol of PGI2/min/mol of protein	980 ± 40	1100 ± 100
Optical absorption spectrum
λmax (nm)	418, 537, 570	418, 537, 570
Ligand bindingb, dissociation constant, and Soret peak
NaCN (mM)	8.2 ± 0.7 (432 nm)c	8.9 ± 0.9 (432 nm)
U46619 (μM)	40 ± 11 (410 nm)	37 ± 11 (410 nm)
Clotrimazole (μM)	2.6 ± 0.3 (424 nm)	2.5 ± 0.3 (424 nm)
Minoxidil (μM)	3.4 ± 0.1 (424 nm)	2.5 ± 0.3 (424 nm)
U51605 (μM)	1.8 ± 0.5 (424 nm)	1.9 ± 0.5 (424 nm)

^aReactions were carried out at 23 °C using PGH₂ (2.5–95 μM) as the substrate.

^bDeterminations were conducted at 23 °C in 20 mM NaP_i (pH 7.4) and 0.2% Lubrol PX.

^cPeak of the Soret band upon ligand binding.

We previously reported that recombinant hPGIS in the presence of detergent (0.2% Lubrol PX) has an unstable ferrous-CO complex and decays from a 450 nm Soret peak to a 422 nm Soret peak at a rate of 0.7 s⁻¹ (12). This unique feature is also observed in the zPGIS. When zPGIS is reduced by dithionite, the Soret peak is blue-shifted to 412 nm and diminished in intensity. Upon manual mixing with CO, only the 422 nm Soret peak was observed. The 450 nm ferrous-CO species, monitored by stopped-flow spectroscopy, decayed at a rate of 0.7 s⁻¹. Taken together, these results indicated that although zPGIS and hPGIS share only 45% sequence identity, they have nearly identical functional features regarding the catalytic activity, ligand specificity, and heme environment.

Structure of Ligand-free zPGIS

Consistent with the similarity in catalytic and ligand-binding properties, the structure of zPGIS is essentially identical with that of hPGIS (7). Both enzymes retain the canonical P450 fold of three β-sheets and 12 α-helices (Supplemental Fig. 2). The two structures superimpose with a 1.3 Å root mean square (r.m.s.) deviation over 424 pairs of structurally equivalent Cα atoms. Although the hPGIS is 10 amino acid residues longer at both the F helix and the J/K loop than zPGIS, both regions are located on the protein surface, and only small and localized differences were observed due to this sequence divergence (Supplemental Fig. 2B). Several structural characteristics of hPGIS with potential functional significance were found to be conserved in zPGIS, evident near the active site in particular. Seventeen of twenty-one amino acid residues are identical in the five substrate recognition sites (7) (Fig. 2). The I helix of zPGIS is kinked to a similar extent in the middle, causing its N-terminal half to bend away from the heme plane. In the I helix, the conserved “acid-alcohol” pair important for the P450 hydroxylation reaction is replaced by Gly²⁷⁶–Asn²⁷⁷ in zPGIS (Gly²⁸⁶–Asn²⁸⁷ in hPGIS). The side chain amide of Asn²⁷⁷ points toward the distal pocket with its nitrogen atom 4.6 Å from the heme iron. Trp²⁸² of hPGIS and its counterpart Trp²⁷² of zPGIS are both positioned 8.7 Å from the iron, with the indole ring lying parallel with the heme plane. This tryptophan is thought to serve as a “ceiling” to restrict the binding of most heme ligands. All water molecules present in the active site of ligand-free hPGIS have equivalents in zPGIS, suggesting a conserved network of interactions. Taken together, these findings indicate that the active site structure is highly conserved in PGIS. Another intriguing similarity between the human and zebrafish structures is the lack of well defined interactions between the two heme propionates and protein scaffold. Although the position of the heme propionates is poorly defined in hPGIS, unambiguous electron density is seen for the propionates of zPGIS. In contrast to most P450s in which at least one propionate is on the proximal side of the heme, both the A- and D-ring propionates of zPGIS are positioned on the distal side of the heme. Notably, only one of the propionates of zPGIS forms a salt bridge with the protein scaffold, in contrast to all other structurally characterized P450s, where the propionates are “neutralized” by forming 2–3 hydrogen bonds or salt bridges with the protein matrix.

The main difference between hPGIS and zPGIS lies at the cysteine ligand loop (residues 414–423 for zPGIS). The r.m.s. deviation of overlap of this region (2.1 Å) is higher than the global average (1.3 Å), with the middle four residues (Glu⁴¹⁷, Asp⁴¹⁸, Asn⁴¹⁹, and Leu⁴²⁰) showing the largest deviation (3.4 Å). Although the loops of both proteins are shaped via a hydrogen bond between backbone carbonyl of tryptophan (Trp⁴¹⁴ of zPGIS or Trp⁴³⁴ of hPGIS) and the backbone amide of the invariant cysteine (Cys⁴²¹ of zPGIS or Cys⁴⁴¹ of hPGIS), the loop is stabilized in hPGIS by a hydrogen bond between backbone CO of Gly⁴³⁷ and side chain Nδ of His⁴⁴⁰ but in zPGIS by a bifurcated hydrogen bond from the backbone CO of Thr⁴¹⁶ to the backbone NH of Asp⁴¹⁸ and Asn⁴¹⁹.

Structure of U51605-bound zPGIS

U51605, whose structure resembles PGH₂ closely, differs in having two nitrogen atoms in place of oxygen at C-9 and C-11 and lacking the 15-hydroxy group in the ω-chain (Fig. 1A). The hydroxyl group of PGH₂ at the C-15 appears to be nonessential for catalysis, since 15-keto PGH₂ can also be converted to its PGI₂ derivative by PGIS (6). U51605 was chosen for co-crystallization, because it binds tightly to the enzyme with a K_d of ~1.9 μM and has known inhibitory effects. A U51605-zPGIS complex was prepared by soaking a ligand-free zPGIS crystal in 2 μl of mother liquor containing 0.25 mM of U51605. Three U51605 molecules were identified in each zPGIS-substrate analog complex (Fig. 3); one sits at the active site (Fig. 4A), one sits in the middle of the helices A′, B′, and F′, and strands β1–3 and β1–4 that is also the substrate entrance channel we previously proposed for hPGIS (7) (designated as “Channel 1”), and the third sits between the B′, C, and I helices (designated as “Channel 2”).

FIGURE 3. Structure of the U51605-bound zPGIS.

A, rainbow-colored ribbon representation of U51605-bound zPGIS with the N terminus in blue and the C terminus in red. The heme group is shown as a red stick model. The three U51605 molecules are displayed in stick models embedded within their respective van der Waals surfaces. The heme-ligated U51605 is shown in blue, and the two noncovalently bound U51605 molecules are shown in yellow and green. B, the noncovalently bound U51605 molecules reside in the potential substrate entrance and product exit channels that connect the active site chamber (occupied by the heme-ligated U51605) to protein surface. The channel edges are delineated by the dashed orange lines. For a better view of the channels, the orientation in B was obtained by an ~180° rotation about the vertical axis from the classical view shown in A. All graphic representations were generated using PyMol (available on the World Wide Web).

FIGURE 4. Heme-ligated U51605.

A, a F_o − F_c omit map contoured at 2.8 σ shows unbiased electron density for the heme-ligated U51605. B, close-up view of the zPGIS active site. The key U51605-contacting residues are labeled. The carbon, oxygen, and nitrogen atoms of U51605 are shown in green, red, and blue, respectively. C, structural changes around the active site and Cys ligand loop upon U51605 binding. The ligand-free (blue) and U51605-bound (gold) structures were superimposed over all equivalent Cα atom pairs. The interactions between heme propionates and protein, seen only in the U51605-bound structure, are indicated by dashed lines. The structural superimposition was performed using the least-square fitting algorithm implemented in the CCP4 (24). D, the majority of water molecules (cyan dots) seen in the active site of ligand-free structure (left panel) are displaced upon U51605 binding (right panel). The heme-ligated and the two noncovalently bound U51605 molecules are shown in green and blue, respectively, on the right.

The heme-ligated U51605 fits in a single orientation and is in a predominantly hydrophobic pocket lined by residues Tyr⁹⁷, Ala⁹⁸, Leu¹⁰¹, Trp²⁷², Val²⁷³, Asn²⁷⁷, Ala³³⁵, and Thr³³⁸ (Figs. 2 and 4B). The C-11 nitrogen of U51605 is coordinated to the heme iron with a bond length of 2.2 Å, and the C-9 nitrogen is hydrogen-bonded to the side chain amide nitrogen of Asn²⁷⁷ at a distance of 2.7 Å. The carboxylate group of U51605 bends toward the A-ring propionate and forms a hydrogen bond with the backbone NH of Thr³³⁸. Although the overall structure is unchanged compared with the ligand-free enzyme (r.m.s. deviation of ~0.4 Å over all Cα atoms), significant conformational changes were detected on the proximal side of the heme. The averaged r.m.s. deviation for the Cys ligand loop is ~1 Å, and the largest deviation (~1.3 Å) was observed for residues Glu⁴¹⁷–Cys⁴²¹, which results from a U51605 binding-induced structural change in the heme moiety. Specifically, the porphyrin is pushed toward the proximal side with the A- and D-rings showing the most movement, causing the porphyrin plane to be more flattened when compared with the ligand-free structure (Fig. 4C). The most striking difference is that the two heme propionates, clearly defined by electron density, are pushed downward to the extent that the D-ring propionate is repositioned to the proximal side of the heme and forms a salt bridge with the concomitantly repositioned side chain of Lys¹¹⁹ on the C helix. Moreover, the A-ring propionate forms a salt bridge to the side chain guanidinium of Arg³³⁹ in the β1–4 strand. The side chain amide of Gln⁹⁴ in the β1–5 strand also donates a bifurcated hydrogen bond to the A-ring propionate. Notably, in most P450s, the A-ring propionate forms hydrogen bond(s) or salt bridge(s) with a residue (Arg in most cases) in the β1–4 strand. The β1–5 strand, running through the middle of the A- and D-ring propionates, may provide a residue for hydrogen bonding to either propionate. Additionally, the D-ring propionate forms hydrogen bond(s) or salt bridge(s) with a residue in the C helix. Thus, the mostly absent canonical interactions between heme propionate and protein scaffold at the ligand-free zPGIS recovers these interactions upon substrate analog binding (Fig. 4C). It should be noted that the most conserved residue among those interacting with heme propionates is the second residue N-terminal to the proximal cysteine, presented as Arg or His. As the PGIS heme moves away from the I helix by ~2 Å upon U51605 binding, the cysteine ligand loop is relocated accordingly by ~1 Å. Asn⁴¹⁹, equivalent to the conserved Arg/His, brings about the most prominent conformational change within the cysteine ligand loop, with its side chain rotating by ~90° toward the heme. Further, its side chain amide also rotates by ~90°, positioning seemingly to interact with the D-ring propionate. But the distance is too far (3.4 Å) for a hydrogen bond.

The two noncovalently attached U51605 molecules in Channels 1 and 2 appear to occupy channels leading to the active site. Both molecules adopt an extended conformation such that the α-chain runs antiparallel to the ω-chain, allowing them to fit snugly in the fairly narrow channels (Fig. 3B). As a result, the α-chain in one monomer is exposed to the surface, whereas in the other monomer it is buried inside the protein. Overall, the fact that U51605 adopts a variety of conformations in the Channels 1 and 2 suggests that the substrate analog is given sufficient freedom to adapt its conformation to the interior space of zPGIS. Since both channels connect the active site to protein surface, it is expected that the two peripheral binding pockets represent the substrate entrance/product exit channels.

Comparison of the ligand-free and U51605-bound zPGIS structures also reveals striking difference in water molecules present in the U51605-binding regions (Fig. 4D). The ligand-free zPGIS contains at least 10 tightly bound water molecules at the active site; three are hydrogen-bonded to the backbone carbonyls of either Trp²⁷² or Val²⁷³, one to the side chain amide of Asn²⁷⁷, and one to the A-ring propionate, and five more are in the vicinity of the A-ring propionate. When U51605 binds, most of these water molecules are displaced, and only the two interacting with Trp²⁷² and Val²⁷³ remain intact. Moreover, five water molecules in Channel 1 (watA601, watA627, watA640, watA654, and watA700) are present in the ligand-free structure, whereas only one (equivalent to watA654) is found in the U51605-bound structure. Together, these findings indicate that a large number of water molecules originally seen in the active site and substrate entrance channel of the ligand-free protein are excluded upon U51605 binding, which greatly reduces the polarity of these regions.

Structure of Minoxidil-bound hPGIS

To investigate whether conformational change of heme, particularly those associated with the propionates, is a common phenomenon for ligand binding to PGIS, we determined the crystal structure of hPGIS in complex with the known inhibitor minoxidil. Using the crystallization conditions for ligand-free hPGIS (7), the minoxidil-bound hPGIS crystals were obtained by co-crystallization, and the structure was determined at 1.62 Å resolution. A clear piece of electron density in the heme distal pocket, which corresponds to the bound minoxidil, was easily recognized in the weighted F_o − F_c map (Fig. 5A). Interestingly, it was found that the oxygen of minoxidil is ligated to the heme-iron at a distance of 2.2 Å, in contrast to the previous thought that a primary amine nitrogen may serve as the distal ligand (26). Probably due to the ceiling effect of Trp²⁸² (corresponding to Trp²⁷² of zPGIS; Fig. 2), minoxidil is slanted into the active site at an angle of ~30° to the porphyrin plane (the Fe-O-N angle is 118°; Fig. 5A) and is further stabilized by three hydrogen bonds formed by its two primary amine. The piperidine moiety of minoxidil exhibits two conformations, a boat-shaped conformer in one hPGIS molecule and a chair-shaped conformer in the other.

FIGURE 5. The minoxidil-bound hPGIS.

A, a F_o − F_c omit map contoured at 2.7 σ shows unbiased electron density for minoxidil. B, structural changes around the active site, Cys ligand loop, and B′ helix upon minoxidil binding. The ligand-free (blue) and minoxidil-bound (gold) structures were superimposed over all equivalent Cα atom pairs. The minoxidil-induced stacking between Phe⁹⁶ and His⁴³⁸ is indicated by the dashed lines.

Minoxidil-induced conformational changes in the hPGIS structure are generally very small (an r.m.s. deviation value of ~0.4 Å over all Cα atoms). Similar to the U51605-bound zPGIS structure, minoxidil binding causes little displacement of the I helix. However, significant structural perturbations (>1 Å) in the B′ helix, B/C loop, C′ helix, and the cysteine ligand loop are evident in the presence of minoxidil, possibly resulting from a conformational change in heme (Fig. 5B). Consequently, the heme is pushed (away from the I-helix) toward the proximal side and slightly forward toward the substrate entrance channel with displacements of 1.5–2.5 and 1.6 Å for the pyrrole rings and the proximal thiolate, respectively. The cysteine ligand loop responds by an ~30° rotation of the His⁴³⁸ side chain, allowing it to stack with the benzyl group of Phe⁹⁶, which is located at the N terminus of the B′ helix. Such an π-π interaction repositions the Phe⁹⁶ side chain and probably causes the N terminus of the B′ helix to relax by a quarter of a helical turn, prompting Tyr⁹⁹ to rotate by ~60° toward the heme to the vicinity of the bound minoxidil (at a distance of 4–5 Å).

One of the most significant findings of this study is that PGIS appears to respond differently to different types of bound ligand. Unlike U51605, which restores the interactions between heme propionates and the protein matrix, the positions of pro-pionate relative to the pyrrole rings in the minoxidil-bound structure are unaltered, and no direct propionate-protein matrix interactions are evoked (Figs. 4C and 5B). Instead, the propionates are hydrogen-bonded to a cluster of water molecules upon minoxidil binding. Specifically, six water molecules are present in the minoxidil-bound hPGIS; three are involved in bridging between minoxidil and protein matrix, and three are hydrogen-bonded to the backbone carbonyls of Trp²⁸², Glys⁴⁸², and Phe⁴⁸³. Thus, the heme propionates are anchored by tightly bound water molecules rather than by interacting directly with the protein matrix. It is noteworthy that the ligand-free hPGIS appears to have only one water hydrogen-bonded to the A-ring propionate.

DISCUSSION

Although they share only 45% sequence identity, hPGIS and zPGIS exhibit nearly identical enzymatic activity and three-dimensional structures, making zPGIS a legitimate alternative for structure/functional studies of PGIS. Furthermore, by comparing the structures of the two enzymes in the ligand-free state and in complex with either the substrate analog U51605 or the inhibitor minoxidil, we have identified those amino acid residues important for catalysis as well as rationalized their functional roles.

Toward a Structural Basis for PGI₂ Biosynthesis

The heme-ligated U51605 is stabilized not only by the coordination of its C-11 nitrogen to heme-iron but also by hydrogen bonding to Asn²⁷⁷ and Thr³³⁸ and by making van der Waals contacts with nearby residues (Fig. 4B); C-10 in the cyclopentane ring contacts with the indole of Trp²⁷², and the van der Waals contacts are also found between C-11 and Val²⁷³, between C-19 and Leu¹⁰¹/Ala⁹⁸, and between C-20 and Tyr⁹⁷. This particular spatial arrangement of active site residues provides an explanation for the C-11 oxygen of PGH₂ being favored over C-9 oxygen as the distal ligand. First, the hydrogen bond between Asn²⁷⁷ and C-9 oxygen would be lost if it served as the heme ligand. Second, if this oxygen were rotated into the axial coordination position of iron, the fused ring of PGH₂ would be oriented with the polar C-11 oxygen facing Val²⁷³. Therefore, the C-11 oxygen is stereoselectively coordinated to the heme-iron.

The heme-ligated U51605 structure exhibits a parallel orientation of α- and ω-chains. The α-chain, whose C-6 is 4.2 Å from C-9 nitrogen, lies across the A-ring of the heme, with C-5 located 5.8 Å from the porphyrin plane. This arrangement is similar to the binding model of PGH₂ that we previously proposed (7) and explains in part how PGIS may facilitate PGI₂ biosynthesis via the reaction mechanism proposed by Hecker and Ullrich (6) (Fig. 1B). Specifically, C-5 is found to be close to the porphyrin plane and may even be closer after cyclization, thus favoring the electron donation from the C-5 radical to porphyrin. Another important feature in the U51605-bound zPGIS structure is that most water molecules are excluded from the active site. Because PGH₂ and PGI₂ are hydrophobic and are labile in aqueous solution, it is anticipated that the channeling pathway of PGI₂ biosynthesis should be hydrophobic. Furthermore, the notion that a hydrophobic active site in P450 favors homolytic cleavage of the peroxide bond was demonstrated by mutagenesis studies (27). Indeed, we showed that PGIS favored homolytic cleavage of the peroxide bond of fatty acid hydroperoxides (12, 13). Repulsion of water molecules may thus be a prerequisite in PGI₂ biosynthesis.

Ligand-specific Heme Conformation Change and Its Functional Aspects

The interactions between heme propionate and protein matrix commonly found in other P450s are mostly recovered in the substrate analog-bound PGIS. Experimentally, limited studies were reported regarding the propionate-protein matrix interactions in P450. Mutagenesis studies suggested that these propionate-interacting residues had profound effects in stabilizing the P450 tertiary structure, heme binding, and facilitating the electron transfer (28, 29), but no satisfying conclusion could be reached. However, the theoretical approaches using quantum mechanical/molecular mechanical calculations have been actively pursued recently (30–32). These studies, which aimed at understanding P450cam hydroxylation based on the formation of the Compound I intermediate indicated that hydrogen bonding to the propionates is necessary for stabilizing the intermediate conformation, and the hydrogen bonding increases spin density on the porphyrin. Nonetheless, the Compound II intermediate, which was proposed for the PGIS reaction, has a completely occupied a_2u porphyrin outer orbital, yielding an essentially closed shell porphyrin. A recent quantum mechanical/molecular mechanical calculation for Compound II of P450cam suggested that, although hydrogen bonding is important for the intermediate conformation, the reactivity is not affected much by the presence of hydrogen bonding (33). This prompts us to speculate that the PGIS heme movement is not to facilitate the chemical reaction but to place a physical barrier that limits the binding of most P450 heme ligands. The barrier is shaped to accommodate PGH₂ in such a manner that the substrate is constrained to permit only the PGI₂ formation via an otherwise highly reactive radical reaction. In contrast, thromboxane synthase, having a closely related reaction mechanism as PGIS (6), is not faithful for product formation, converting 50% of PGH₂ to thromboxane A₂ and 50% of PGH₂ to malondialdehyde and 12-hydroxyheptadecatrienoic acid. Notably, thermodynamic computations have shown that two of the four most stable salt bridges in the P450cam are made to the D-ring propionate by Arg¹¹² (in the C helix) and His³⁵⁵ (the second residue N-terminal to the proximal cysteine) (34), supporting the notion that flexibility and the lack of at least one key ionic interaction in the resting PGIS make its heme more loosely anchored at the active site. Although the significance of the propionate-protein interactions is not fully understood at present, our findings should open an avenue to revisit this important subject.

Another intriguing observation regarding the heme conformational change is that the porphyrin becomes more planar upon the binding of substrate analog. The averaged total out-of-plane distortions, calculated by the normal-coordinate structural decomposition method, are 0.67 and 0.56 Å for the ligand-free and U51605-bound structures, respectively, as compared with 0.4–0.5 Å for most P450s (35). This is caused predominantly by “ruffling,” characterized by a rotation of the pyrrole rings about Fe–N bonds. Based on the previous studies that the nonplanar heme is harder to reduce (35) and favors a two-electron oxidation to produce a Compound I intermediate (36), a more planar PGIS heme is probably a favorable conformation for PGI₂ biosynthesis.

In both U51605 and minoxidil-bound structures, the heme group undergoes large ligand-induced conformational changes. This heme movement has never been observed in any other structurally known P450. However, the conformational changes in the heme are different for the two ligands in that U51605 binding recovers the propionate-protein matrix interactions commonly observed in other P450s, but minoxidil binding does not. Further, water molecules are largely retained in the active site upon minoxidil binding but are repelled upon U51605 binding. These findings indicate that the conformational changes are ligand-specific. Our results thus elicit important questions related to the propionate-protein interactions. First, what is the functional role of the propionate-protein interactions in P450s? Will a loosely bound P450 heme lead to a different reaction pathway? Second, what is the driving force(s) that disrupts the substrate-induced propionate-protein matrix interactions in the later steps of the PGI₂-biosynthetic pathway?

Substrate Entrance and Product Exit Channels of PGIS

Given that Channel 2 is tightly closed and Channel 1 is slightly open in the ligand-free structure, it is possible that Channel 1 is for the substrate entrance and Channel 2 is for the product exit. Channel 1, which passes between the F/G loop, the B′ helix, and the β1 sheet, corresponds to Channel 2a assigned by Cojocaru et al. (37) and is thought to be the route for the substrate entrance in P450s. Although bacterial P450s are likely to use the same route for substrate entrance as well as product exit, microsomal P450s, at least for CYP2C5 (38), use a passage between the G and I helices and the B′ helix/BC loop (assigned as Channel 2c) for product exit. This Channel 2c appears to be our Channel 2. Consistent with their molecular dynamics simulations, our structural data implicate that microsomal P450s may use different pathways for substrate entrance and product exit, differing from soluble bacterial P450s.

In conclusion, PGIS differs from other P450s in that a conformational change on the proximal side accompanied by the heme relocation is observed upon ligand binding. It appears that PGIS evolved divergently from other P450s with a flexible proximal side. This unusual characteristic may be achieved in part by the sequence divergence in the cysteine ligand loop where the well conserved His/Arg is replaced by asparagine (Asn⁴³⁹ and Asn⁴¹⁹ in hPGIS and zPGIS, respectively). Lacking the positively charged side chain in asparagine for the ionic interaction with the heme propionate, the cysteine ligand loop of PGIS may thus be endowed with more flexibility. Further, PGIS seems to have evolved to catalyze solely the isomerization of PGH₂ into PGI₂ by placing a barrier on the heme conformation in the P450 structure scaffold in which the direct propionate-protein matrix interactions are observed only when the correct substrate is ligated to the heme. The PGIS structure thus exemplifies another aspect of divergent evolution in the versatile P450 superfamily.

Supplementary Material

supplementary

Supplemental Material can be found at: http://www.jbc.org/cgi/content/full/M707470200/DC1