Crystal structure of human indoleamine 2,3-dioxygenase: Catalytic mechanism of O₂ incorporation by a heme-containing dioxygenase

Overall Structure of IDO.

The overall structure (Fig. 1A) shows that IDO is folded into two distinct domains (small and large). The large domain is an all-helical domain and is comprised of 13 α-helices and two 3₁₀ helices. Four long helices (G, I, Q, and S) in the large domain run parallel to the heme plane and interact with the neighboring helix by hydrophobic interactions. Helix Q provides an endogenous ligand (H346 imidazole) for the heme iron at the fifth coordination position (proximal side) (Fig. 1B). The heme-binding pocket is created mainly by these four helices and other helices (K-L and N). The side chains of helices K-L and N also contribute to heme–protein interactions and connect the two domains. The small domain and a long loop (residues 250–267) connecting the two domains above the sixth-coordination site of the heme (distal side) cover the top of the heme pocket. The small domain is comprised of six α-helices, two short β-sheets, and three 3₁₀ helices. Contact between these two domains is very extensive with a buried surface area of 3,100 Ų. The interface is formed by a combination of hydrophobic interactions, salt bridges, or H bonds mediated by the side chains of amino acid residues. The folding of each domain of IDO was not detected among other known protein structures in the similarity-searching database dali (15).

The entrance for the ligand or substrate entrance toward the pocket is indicated from the molecular surface representation shown in Fig. 2. It was not possible to construct an atomic model for the region between R and S helices (residues 360–380) because of the disordered electron density. This region is presumably a flexible loop outside the heme pocket.

Heme Environment.

The proximal side of the heme is occupied by only side chains from the large domain. Fig. 3 shows the heme with the surrounding residues in the PI-bound form. The heme 6-propionate contacts to the water molecule (wa2) and R343 in the proximal side (Fig. 3A). wa2 also interacts with wat1 and L388 carbonyl. Previous site-directed mutagenesis studies (16) have suggested that D274 is a key residue in the IDO catalytic reaction. However, D274 in the present structure is located in the proximal side, creating a salt bridge with R343, which is in contact with the 6-propionate. Therefore, the large decrease in enzymatic activity upon the mutation of D274 is likely derived from the structural instability and failure to maintain the heme position. The proximal ligand H346 is H-bonded to a water molecule (wa1) (Fig. 3A). The environment of the heme proximal side suggests that the H346 imidazole of IDO is partially anionic by the 6-propionate and L388 carbonyl with the mediation of two waters (wa1 and wa2). It is consistent with the previously reported resonance Raman spectroscopic data (17) in which the Fe-His stretching frequency (ν_Fe-His) was observed at 231 cm⁻¹. This value is in contrast to the highly anionic character of the proximal His imidazole in peroxidases (248 cm⁻¹) and the neutral one in myoglobin (220 cm⁻¹). The 7-propionate of the heme is pointed upward from the heme plane and interacts with the hydroxyl group of S263.

The distal heme pocket, which is a coordination site of the heme iron for the sixth external ligand (PI and CN⁻), is comprised of a combination of the small domain, the large domain, and the loop connecting the two domains (Fig. 1). Because molecular oxygen (O₂) binds at this site, the structural characteristics of the distal heme pocket should be responsible for the dioxygenase reaction catalyzed by IDO. His residue has been predicted to be a distal and catalytic residue on the basis of EPR spectroscopic data (18). However, the structures in complex with PI and CN⁻ reveal that His is not present in the heme distal side, and that neither polar residues nor water molecules are available to interact with the iron-bound ligand (Fig. 3B). Whereas PI binds to the iron as a noncompetitive inhibitor in the ferric and ferrous forms, the small ligand CN⁻ and l-Trp markedly enhance the affinity of each other for the ferric form. Although we cannot describe the ligand geometry of the CN⁻-bound form in detail at 3.4-Å resolution, the N and C^β atoms of A264 in the connecting loop are estimated to be located within 3 Å from the CN⁻. In a structural comparison between the PI and the CN⁻-bound forms (Fig. 4), the connecting loop exhibits a large conformational change in the main chain with a 1.3-Å displacement, whereas rms deviation values for all common C^α atoms is 0.2 Å. The highly conserved sequences (AGGSAG for residues 260–265) in this loop appear to provide flexibility. We suggest that the ligand and/or substrate binding induce a conformational change in the connecting loop.

In the PI-bound form, Phe-163 interacts with the phenyl group of PI in the π-π stacking. Ser-167, which is located 3.7 Å above the 3-methyl group of the heme, is also one of the closest residues to the iron. At the back of the pocket, the side chains of Phe-164, Val-130, and Cys-129 also contribute to the wall but are far from the iron (>10 Å).

In the heme distal pocket of both crystal forms, we observed clear density (Fig. 5) for two molecules of 2-(N-cyclohexylamino)ethane sulfonic acid (CHES), a component of the crystallization buffer. The cyclohexan-ring of CHES-1 is very close to the ligand (4.0 Å) and iron (5.6 Å). The side chain of F226 and the alkyl chain of R231 are in contact with the cyclohexan ring of CHES-1 at distances of 3.8 and 4.2 Å, respectively (see Figs. 3 and 5). G261 and G262 in the connecting loop also interact with the ethanesulfonic group of CHES-1. CHES-2 is adjacent (4–5 Å apart) and antiparallel to CHES-1. The amino group of each CHES interacts with 7-propionate of the heme. Despite these interactions, the enzymatic activity is not inhibited upon addition of CHES (>50 mM). It seems reasonable to consider that the high concentration of CHES binds to the putative substrate-binding pocket.

Site-Directed Mutagenesis.

To identify the important amino acid residue for the catalytic reaction, we replaced the residues that make up the wall of the distal pocket to Ala by site-directed mutagenesis. As shown in Table 1, the mutations for polar amino acids (S167A, C129A, and S263A) retain the dioxygenase activity. The reason the activity of the mutant of S263A is reduced to 15% is that the side chain of S263 appears to interact with the heme 7-propionate to stabilize the heme position. In contrast, the mutants of F226A, F227A, and R231A have drastically reduced the dioxygenase activity. This result supports the hypothesis that the CHES-1 molecule mimics the substrate for IDO in the crystal. It is possible that F226 and R231 are directly involved in substrate recognition by hydrophobic interactions. The side chain of F227 interacts with the guanidium group of R231 by cation–π interactions in the present structures, which suggests an indirect contribution of F227 to substrate recognition by stabilizing the conformation of R231. It is noteworthy that the mutation of F163A does not affect the activity and other parameters, whereas F163 forms a π-π stacking interaction with PI in the PI-bound form.

Reaction Mechanism.

A range of substrates and inhibitors are known for IDO. It suggests that the inhibitory effect of methylation (1-methyl d-Trp) or substitutions of the indole N with O or S (furan or thiophene analogs) are caused by the lack of H at the 1-position of the indole derivatives. In addition, the electron-donating group at C-5 or C-6 of l-Trp has been known to be a good substrate. Based on these analyses, several reaction mechanisms have been proposed [reviewed by Sono et al. (11)]. Regarding the step after the formation of the ternary complex (IDO Fe²⁺-O₂:substrate), two types of schemes have been proposed (Fig. 6). The earlier proposed mechanisms involve the protein base-assisted abstraction of proton from the indole NH group (Fig. 6, scheme B). Terentis et al. (17) recently proposed a mechanism involving proton abstraction by the iron-bound O₂, followed by the electrophilic addition of the dioxygen-attacking double bond between C-2 and C-3 of the pyrrole ring (Fig. 6, scheme A). Because the present crystal structure and mutational analysis demonstrate that no polar/charged protein side chains act as a catalytic base, we conclude that proton abstraction by iron-bound dioxygen (Fig. 6, scheme A) is the most plausible event for the reaction mechanism of IDO. The interaction of an indole NH group with the proximal oxygen atom (= oxygen atom bonded to iron) (Fig. 6, 2) would lead to the rearrangement of electrons of the substrate and weaken the FeO₂ bond. It then induces the electrophilic addition of a terminal oxygen atom of the bound dioxygen to the relatively electron-rich C-3 of indole to form the intermediate Fe²⁺:3-hydroperoxyindolenin complex (Fig. 6, 3). It has been speculated that the product N-formyl Kyn is converted by the intermediate dioxetane (Fig. 6, 4) (17).

In the Fe²⁺-CO:L-Trp and Fe³⁺-OH⁻:L-Trp complex forms of IDO, close contact of the substrate to the iron-bound ligand has been implicated by resonance Raman spectroscopy (17). It should also be noted that, in the case of the O₂-bound form of the bacterial hemoglobin (19), the interaction between the proximal oxygen and the side chain of the distal Tyr residue weakens the FeO₂ bond, suggesting that, in addition to the moderately strong FeN_His bond of IDO, the interaction of the proximal oxygen atom with the NH of the indole ring might also be important for the dioxygen-releasing mechanism from heme iron without or before the cleavage of the OO bond.

Interestingly, the mutations of F226A, F227A, and R231A do not affect the K_d values (Table 1). Nevertheless, they dramatically reduce activity. These mutants suggest that a necessary condition for the reaction is the proper geometry between the substrate and dioxygen for proton abstraction, which is brought about by the strict complementarities between the indole ring of the substrate and a hydrophobic moiety of the protein. It is also suggested that strict complementarities are not required for l-Trp binding, which may reflect the broader substrate specificity of IDO. The structural differences between indoleamine derivatives may be absorbed by the flexibility of the connecting loop.

Comparison of IDO with Heme-Containing Protein.

The mechanism for the reaction of IDO proposed above can be comparable to that for heme monooxygenases (20). The P450 family consists of typical monoxygenase enzymes that contain heme-iron in their active site for dioxygen activation (21). In the case of P450, as well as nitric oxide synthases and heme oxygenase, the electropositive patches on the proximal surface are important for the binding of its redox partner (22). In the distal side of P450, hydroxyl or carboxyl groups in the conserved Thr or acidic residues play a crucial role in the formation of a specific H-bonding network to deliver the protons from solvent water to the active site during the catalytic reaction of P450. In peroxidase, the presence of a strong H bond between the proximal His and an Asp leads to anionic character in the proximal His. The distal Arg polarizes the OO bond and distal His acts as an acid-base catalyst for the heterolytic cleavage of the OO bond. In the case of myoglobin, the absence of the distal Arg and H bond in proximal His is indicative of the reversible binding of O₂. In contrast, IDO has a symmetric charge distribution on the molecular surface, and no suitable region for the interaction with the possible reductase can be identified in the proximal surface (Fig. 2). Furthermore, the significant polar amino acid or water molecules are absent from the active site of IDO, suggesting that the H-bonding network connecting the active site to the solvent region is not present in the distal pocket of the IDO structure. These structural characteristics around the heme pocket of IDO are in good agreement with the proposed catalytic reaction, in that it does not require either protons or electrons supply.

The nonheme Fe²⁺-containing extradiol dioxygenases cleave the CC bond adjacent to the -OH of catecholic substrates by activating dioxygen and incorporating both oxygen atoms. The crystal structures of the substrate-bound form indicate that the two -OH groups of the substrate, together with three protein side chains and water, bind directly to Fe²⁺ (13, 23). In the case of naphthalene dioxygenase, the dioxygen binds side-on with both oxygen atoms coordinated to the nonheme iron (12). Thus, the reaction mechanism for nonheme dioxygenases is quite different from that of IDO, which is allowed by the flexible coordination geometry and the accessible space around the nonheme iron.

Protein Expression, Purification, and Kinetics.

The coding region for human IDO was cloned into the pET-15b vector (Novagen). The wild-type and mutant IDO were expressed in transformed Escherichia coli BL21(DE3) cells and purified by using a Ni affinity column and anion exchange chromatography. The assays for the wild-type and mutant IDO were performed as described (24), except that the buffer with 0.1 M potassium phosphate buffer and 0.1 M NaCl (pH 8.0) was used at 37C°. The rate of product formation was determined from the slope of the initial increase in absorbance at 480 nm derived from Kyn as a function of time. K_d values of l-Trp for ferric, ferric CN⁻, and ferrous forms of IDO were determined by monitoring the peak height of the Soret region (404, 418, and 423 nm, respectively) of optical absorption spectra in various concentrations (0–5 mM) of l-Trp.

The K_d values of l-Trp for the ferrous CO complex were deduced from the CO binding rates that were measured in laser flash photolysis experiments, which were performed in 0.1 M potassium phosphate buffer (pH 8.0), 0.1 M NaCl, and 2 mM l-Trp at 20°C by using an instrument constructed by UNISOKU (Osaka). The emission from a dye solution (rhodamine 6G in methanol at a concentration of 0.24 g/liter) was used. Changes in absorption after CO photolysis were monitored at 435.4 nm by a light from a xenon lamp.

Crystallization, Structure Determination, and Refinement.

IDO was concentrated to 40 mg/ml in a buffer containing 10 mM Mes (pH 6.5), 25 mM NaCl, and 1 mM PI. Crystals of IDO–PI complex were obtained by vapor diffusion from hanging drops (6 μl) containing a 1:1 (vol/vol) ratio of protein solution to reservoir solution (10% polyethylene glycol 8000, 200 mM ammonium acetate, 100 mM CHES, pH 9.0). Crystals formed in 2 weeks at 20°C in the space group P2₁2₁2₁ (cell dimensions of a = 86.1, b = 98.0, and c = 131.0 Å) with two IDO monomers in the asymmetric unit and a solvent content of 55%. The crystals were transferred to a cryo buffer consisting of the reservoir with an additional 30% xilitol. X-ray diffraction data of PI form were collected by using a Jupiter210 charge-coupled device detector (Rigaku, Tokyo) in BL26B1 at SPring-8. Data collection statistics are shown in Table 2. Multiwavelength anomalous dispersion data of the PI form were measured at the absorption edge of the Fe atom. The determination of the iron position and initial phasing was performed by using the autosharp program (25). The initial 2.8-Å phases were improved, and the poly(A) models were partially built with the arp/warp program (26). The protein and heme model was then built manually with the program o (27). This model was refined with the cns program (28) by using higher-resolution data (20–2.3 Å). The positions of the S atom in the CHES molecules in the active site were confirmed by an anomalous difference Fourier map. The N-terminal three residues (G-S-H) from a cloning artifact and residues 1–10 and 360–380 of IDO could not be defined in electron density. Noncrystal symmetry (NCS) restrains were not applied in the final stage of the refinement. The diffraction data for the CN⁻-bound form with the unit cell dimensions of a = 86.4, b = 97.1, and c = 129.4 Å were collected in BL44B2 at SPring-8. The form was prepared by soaking the PI-bound crystal in the cryo buffer containing 5 mM KCN for 5 min. Because the soaking experiments damaged the diffraction quality, we searched various conditions of soaking time and concentration of KCN. Five minutes are a minimum to exclude the electron density of PI. NCS restrains were applied in the whole refinement process for the CN⁻-bound form. The refinement statistics are shown in Table 2. Figures were created by o (27), molscript (29), raster3d (30), pymol (31), and grasp (32).