Structural Basis for a Distinct Catalytic Mechanism in Trypanosoma brucei Tryparedoxin Peroxidase*

Trypanosoma brucei, the causative agent of African sleeping sickness, encodes three cysteine homologues (Px I-III) of classical selenocysteine-containing glutathione peroxidases. The enzymes obtain their reducing equivalents from the unique trypanothione (bis(glutathionyl)spermidine)/tryparedoxin system. During catalysis, these tryparedoxin peroxidases cycle between an oxidized form with an intramolecular disulfide bond between Cys47 and Cys95 and the reduced peroxidase with both residues in the thiol state. Here we report on the three-dimensional structures of oxidized T. brucei Px III at 1.4Å resolution obtained by x-ray crystallography and of both the oxidized and the reduced protein determined by NMR spectroscopy. Px III is a monomeric protein unlike the homologous poplar thioredoxin peroxidase (TxP). The structures of oxidized and reduced Px III are essentially identical in contrast to what was recently found for TxP. In Px III, Cys47, Gln82, and Trp137 do not form the catalytic triad observed in the selenoenzymes, and related proteins and the latter two residues are unaffected by the redox state of the protein. The mutational analysis of three conserved lysine residues in the vicinity of the catalytic cysteines revealed that exchange of Lys107 against glutamate abrogates the reduction of hydrogen peroxide, whereas Lys97 and Lys99 play a crucial role in the interaction with tryparedoxin.

mitochondrion of the parasites (5). In the mitochondrial Px III sequence, Cys 47 replaces the selenocysteine residue. Substitution of Cys 47 or Gln 82 by a glycine residue resulted in inactive protein species, and the glycine mutant of Trp 137 had only 12% wild type activity (18), which seemed to confirm the presence of the catalytic triad. Interestingly, in a related enzyme from Chinese cabbage, the respective glutamine to glycine mutation had no effect on peroxidase activity (19). Nearly all of the cysteine homologues, but not the selenoenzymes, possess a second cysteine (Cys 95 in T. brucei Px III) that is essential for peroxidase activity (18,19). In the first half reaction of catalysis, the thiolate form of Cys 47 (reduced enzyme) reacts with the hydroperoxide substrate, resulting in the respective alcohol and probably a sulfenate at Cys 47 . Attack of Cys 95 generates an intramolecular disulfide bond (oxidized enzyme) that is in the second half reaction reduced by interaction with Tpx (Scheme 2).
Here we present the crystal structure of oxidized T. brucei Px III and the NMR structures of the protein in oxidized and reduced form. Contrary to previous observations on poplar TxP (16), the closest homologue for which a three-dimensional structure has been published, T. brucei Px III does not undergo any significant structural rearrangement upon reduction of the intramolecular disulfide bridge. Furthermore, residues Gln 82 and Trp 137 , the residues equivalent to the catalytic triad in classical GPXs, are located far away from the active site Cys 47 in our structures, ruling out the presence of the catalytic triad. Thus, the catalytic mechanism of Px III appears to differ fundamentally from any of the currently characterized enzymes. Analysis of the active site identified three lysines in the vicinity of the redox active cysteines that could play a role during catalysis. Kinetic analysis of three lysine mutants revealed that two of them, Lys 97 and Lys 99 , play a crucial role in the interaction with tryparedoxin, whereas Lys 107 is involved in the first part of the reaction, the reduction of hydrogen peroxide.

EXPERIMENTAL PROCEDURES
Materials-Recombinant T. brucei Tpx (20) and Trypanosoma cruzi trypanothione reductase (21) were prepared as described. Trypanothione disulfide was purchased from Bachem, and H 2 O 2 was from Merck. 15 NH 4 Cl, [ 13 C 6 ]glucose and selectively labeled amino acids were obtained from Spectra Stable Isotopes.
Cloning, Expression, and Purification of Px III-For crystallization, the Ser 76 mutant of Px III was expressed from the pQE30/c76s-px III plasmid which resulted in a protein species starting with Lys 7 . The protein was prepared as described (17). For NMR studies, the Px III gene was cloned into a modified pET 9d vector with an N-terminal thioredoxin-His 6 tag cleavable by tobacco etch virus protease. The mitochondrial signal sequence of Px III was removed, and Cys 76 was replaced by a serine residue which resulted in a fully active peroxidase (18) but excluded putative unspecific oxidation or dimerization. Details of the purification have been published elsewhere (22). The Ser 47 , Ser 95 , Glu 97 , Gln 97 , Glu 99 , Gln 99 , FIGURE 1. Sequence alignment of T. brucei Px III with related peroxidases for which the three-dimensional structures are known. The residues SeCys (U) or Cys, Gln, and Trp that form a catalytic triad in glutathione peroxidases and several cysteine-containing relatives are marked by triangles. Cysteine 95, which is essential for catalytic activity of Px III, is marked by a filled circle. The three Lys residues, Lys 97 , Lys 99 , and Lys 107 , that were mutated in this work are marked by a square. ␣ and 3 10 helices (rods) as well as ␤-sheet regions (arrows) above the alignment refer to the secondary structures of Px III in comparison to reduced and oxidized poplar TxP (pTxP red and pTxP ox ) as well as human GPX1 (hGPX1). Residues that are identical in all sequences are marked by an asterisk, and highly or partially conserved residues by are marked by colons and periods, respectively. Cysteine residues are given in bold letters. The alignment was generated with ClustalX (57). tPxIII, T. brucei Px III (17); pTxP, Populus trichocarpa glutathione peroxidase 5 (PDB codes 2P5R and 2P5Q (16)); hGPX4, human GPX4 (2GS3 and 2OBI (42)); hGPX7, human GPX7 (2P31); bGPX1, bovine GPX1 (1GP1 (12)); hGPX1, human GPX1 (2F8A); hGPX2, human GPX2 (2HE3); hGPX3, human GPX3 (11); hGPX5, human GPX-5 (2IY3). SCHEME 1. In African trypanosomes, hydroperoxides (ROOH) are detoxified by a cascade composed of trypanothione reductase (TR), trypanothione (T(SH) 2 ), Tpx, and the Px III or a 2-Cys-peroxiredoxin.
and Glu 107 mutants were generated using the QuikChange mutagenesis kit (Stratagene) and the pET9d/c76s-px III or pET9d/px III plasmid as template. Escherichia coli XL1-Blue cells were transformed and grown at 37°C in Luria-Bertani (LB) medium with 50 g/ml kanamycin. The plasmid DNA was purified using the Nucleobond PC100 kit (Macherey-Nagel). The recombinant Px III mutants were purified as described (22).
Structure Determination of the Oxidized State by X-ray Crystallography-Purified Px III was dialyzed against 100 mM Tris, 1 mM EDTA, pH 7.6, and concentrated to 13 mg/ml. The protein was crystallized by the hanging drop vapor diffusion method at 4°C using a droplet size of 4 l of Px III solution and 4 l of the reservoir solution of 2 M ammonium sulfate, 0.1 M sodium acetate, pH 4.6. A crystal with an approximate size of 50 ϫ 50 ϫ 400 m 3 was shock-frozen under liquid nitrogen using mother liquor containing 20% glycerol as cryo protectant. Data were collected on beamline ID14-EH4 at the European Synchrotron Radiation Facility, Grenoble, France, at a wavelength of ϭ 0.9393 Å (Table 1). A total of 500 frames with 0.25°o scillation were collected (i.e. 125°total). Toward higher resolution, diffraction spots were split, pointing to a potential twinning problem not readily detectable from the optical appearance of the crystals. Autoindexing, as implemented in denzo (23), suggested a centered orthorhombic cell with one molecule per asymmetric unit. Possible lower symmetries were centered monoclinic and primitive monoclinic with two or primitive triclinic with four molecules per asymmetric unit. Merging R-factors from scalepack (23) were 7.7% (C222), 6.1% (C2), 5.8% (P2), and 3.4% (P1). Data manipulation was carried out with the CCP4 package (24).
We used the ARP/wARP procedure of free atom density modification with subsequent auto-building (25) in the P2 lattice. Due to twinning, the model had to be completed in three steps by iterated model building in Coot (26) and ARP/wARP autotracing and refmac5 refinement (27). Twin refinement using the twin operator (-l, -k, -h) in shelxl (28) immediately reduced the R free by more than 5% (Table 1); the twin fraction refined to 0.473. Model correction and introduction of anisotropic refinement reduced the R free by a further 3%. Model building and refinement were carried out similarly to the serine protease example given in the shelx manual. The initial 491 water molecules, assigned by the ARP/wARP procedure, were reduced in the course of refinement to 365 water molecules to avoid fitting of noise; in refinement cycles 3 and 5, the automated water-divining procedure of shelx was used with default/ recommended settings.
NMR Experiments-The chemical shift assignment has been published elsewhere (22). For structure determination, two-dimensional NOESY in 100% D 2 O, 15 N HSQC-NOESY, and 13 C heteronuclear multiple quantum coherence-NOESY spectra were recorded at high field (800/900 MHz) with mixing times of 80 ms (oxidized form) and 100 ms (reduced form) at 22°C and a protein concentration of 1 mM. Data were processed with NMRPIPE (29) and analyzed using NMRVIEW (30). Structures were calculated with ARIA1.2/CNS (31) on the basis of the experimentally derived NOE restraints (Table 2). Additional dihedral restraints derived from TALOS analysis (32) of the chemical shifts and hydrogen bonds identified on the basis of characteristic NOE patterns ( Table 2) were added. The waterrefined structures were examined with PROCHECK (33). Figures were prepared with MOLMOL (34) and PyMOL (35).
To analyze the two conformations of the reduced enzyme, 300 M Px III in 50 mM sodium phosphate, 100 mM KCl, pH 6.8, was treated with 10 mM dithiothreitol, and the NMR-tube was purged with argon to prevent reoxidation. Constant reducing conditions were verified by subjecting an aliquot of the solution to thiol determination with 5,5-dithiobis(2-nitrobenzoate). The difference in chemical shift ⌬␦ was calculated with the formula ⌬␦ ϭ ͌(␦H 2 ϩ (␦N/10) 2 ). One-dimensional 1 H NMR spectra were recorded at 600 MHZ between 29 and 4°C in D 2 O and H 2 O. To evaluate if the appearance of two conformations was pH-dependent, the Px III solution was titrated to pH 5.0, 6.8, and 9.5 with 1 M HCl or NaOH. The role of Cys 47 and Cys 95 was studied by alkylating the protein at pH 8.0 with 15 mM iodoacetamide in the presence of 2 mM dithiothreitol.
Kinetic Analysis of the Px III Lysine Mutants-Peroxidase activity was measured at 25°C in a total volume of 150 l of 0.1 M Tris, 5 mM EDTA, pH 7.6, in the presence of 240 M NADPH, 100 M trypanothione disulfide, 150 milliunits of trypanothione reductase, 10 M Tpx, and 0.1-0.9 M concentrations of the different Px III species. The reaction was started by adding 100 M H 2 O 2 , and NADPH consumption was followed at 340 nm (5,17). The kinetic parameters were determined by Dalziel kinetics as described for the wild type enzyme (17). The Tpx concentration was varied between 2 and 10 M in the presence of 0.2 and 0.6 M Px III. The reactions were started by the addition of 60 M H 2 O 2 .
Modeling of the Complex between Px III and Tpx-The 10 lowest energy NMR structures of oxidized Px III and the x-ray structure of oxidized Trypanosoma brucei Tpx (PDB code SCHEME 2. Proposed reaction mechanism of Px III. In the first half reaction, reduced Px III (Px red ) reacts with the hydroperoxide substrate (ROOH), and Cys 47 becomes oxidized to a sulfenic acid (Px ox (sulfenic acid)). Attack of Cys 95 leads to formation of an intramolecular disulfide between Cys 47 and Cys 95 (Px ox (S-S)). In the second half-reaction, Cys 40 of reduced Tpx (Tpx red ) attacks Cys 95 of the disulfide bond, resulting in an intermolecular disulfide between both proteins (Px-Tpx mixed disulfide). Cys 43 of Tpx finally reacts with its vicinal Cys 40 to release Px red and oxidized Tpx (Tpx ox ). The three states of the catalytic cycle studied in this work, Px red , Px ox (S-S), and the complex between Tpx and PxIII are depicted in red, blue, and green, respectively. OCTOBER 31, 2008 • VOLUME 283 • NUMBER 44 1O73 (36)) were used to model the complex structure. The docking was performed using HADDOCK 2.0 (37,38). The solvated docking was carried out with the recommended parameters of HADDOCK. Cys 40 of Tpx and Cys 95 of Px III were defined as active residues. During the rotational and translational rigid body minimization docking, 1000 structures were calculated with solvation. The best 200 solutions according to intermolecular energy were used for the semi-flexible annealing in torsion angle space. This was followed by a final refinement with explicit modeling of hydrating water molecules. The resulting 200 structures were clustered on the basis of the intermolecular van der Waals, electrostatic, and restraint energy terms, buried surface area, desolvation energy, and binding energy as combined in the HADDOCK score versus the backbone r.m.s.d. from the lowest HADDOCK score structure. The HADDOCK score is a weighted sum of various energy terms obtained in different phases of the docking. The energy terms include van der Waals energy, electrostatic energy, ambiguous distance restraints energy, buried surface area, binding energy and desolvation energy. The structure with the smallest weighted sum, that is, with the smallest HADDOCK score, is ranked first. Solvent surface accessibility was determined by NACCES 2.1.1 (39). Intermolecular interactions were analyzed with LIGPLOT (40).

RESULTS
Overall Structure of T. brucei Px III-We determined the structure of oxidized T. brucei Px III both by x-ray crystallography at a resolution of 1.4 Å and by NMR spectroscopy (Fig. 2). For the structural analyses, Cys 76 of Px III was replaced by a serine residue which resulted in a fully active enzyme species (18). The structure of the reduced form was obtained only by NMR spectroscopy. The statistics of the structure determinations by x-ray and NMR are presented in Tables 1 and 2, respectively.
The x-ray and NMR structures of oxidized Px III ( Fig. 2A) are highly similar, with a r.m.s.d. over the backbone atoms of 0.63 Å. The crystal structure revealed a dimer in the asymmetric unit. The buried surface has been determined with CNS (41) to be 1160 Å 2 . The interface comprises the loops from Gly 81 to Glu 88 and from Gly 128 to Trp 137 as well as Asn 111 . Trp 137 is involved in hydrophobic interactions in the dimer; in classical GPXs, this residue is in hydrogen-bonding distance to the catalytic selenocysteine instead. The localization of Trp 137 at the dimer interface leads to two different conformations for the indole ring, neither of which corresponds to the conformation observed by NMR. To assay whether the dimer occurs at higher concentrations or is driven by crystal packing, 1 H T2 spin-echo relaxation measurements and pulsed-field-gradient NMR diffusion measurements were performed. They yielded a relaxation time and diffusion coefficient typical for a monomeric protein at 1 mM concentration (14.1 Ϯ 1.5 ms at 21°C, D s ϭ 0.953 Ϯ 0.008 ϫ 10 Ϫ6 cm 2 s Ϫ1 , supplemental Figs. 1 and 2, respectively) corroborating previous gel chromatography results in which Px III eluted as a monomer (17).
Minor differences between the x-ray and the NMR structure were observed for loop 1 (Ala 44 -Lys 51 ) and loop 3 (Lys 126 -FIGURE 2. Structures of oxidized and reduced Px III. A, stereo image of the superposition of the structures of oxidized Px III obtained by x-ray and NMR analysis (gray and black, respectively) and the NMR-structure of reduced Px III (light blue). ␣-Helices and 3 10 -helices are colored in red, and ␤-sheets are in blue. The two catalytically active cysteine residues Cys 47 and Cys 95 are depicted in orange. The ␣-helix in loop 2 only present in the crystal structure is shown in gray. B and C, ensemble of the 10 lowest energy NMR structures of 200 calculated structures after water refinement for the oxidized and reduced protein, respectively. The three most extended loops are given in cyan (L1, loop 1; Ala 44 -Lys 51 ), green (L2, loop 2; Ser 76 -His 116 ) and gray (L3, loop 3; Lys 126 -Thr 140 ). N and C mark the N and the C terminus of the protein.
Thr 140 , Fig. 2A). Well defined NOE contacts from Ser 45 to Pro 75 and Ser 76 lead to an altered orientation of loop 1 in the NMR structure. Because loop 1 and loop 2 (Ser 76 -His 116 ) are connected via the intramolecular disulfide between Cys 47 and Cys 95 , loop 2 and the entire catalytic region are shifted as a consequence ( Fig. 2A). Loop 3 is involved in the hydrogen bonding contacts between monomers in the crystal, which leads to a different conformation.
In the x-ray structure, Glu 88 -Glu 92 form an ␣-helical turn ( Fig. 2A), which coincides with ␣-helix 2 in GPXs ( Fig. 1) and reduced poplar TxP (16). Because NMR data reflect the average of all possible conformations, it is conceivable that this part of loop 2 has some potential of ␣-helix formation.
The monomeric state of Px III is in agreement with the fact that it lacks a long stretch (inserted between Ile 129 and Leu 130 ) that forms the subunit interface in the tetrameric GPXs (Refs. 11 and 12; Fig. 1). This loop is not present in the monomeric human GPX4 (PDB code 2GS3 and 2OBI (42)) as well as in most of the cysteine-containing glutathione peroxidase-type proteins including poplar TxP (16). Nevertheless, the latter forms a homodimer (42,43). This protein dimerizes through the C-terminal region (16). Most of the residues localized in the interface are conserved in related plant proteins but not in Px III.
Px III displays a high loop content of about 60% (Figs. 1 and 2). The overall fold of Px III is 3 10 -␤-␤-3 10 fold of other GPX-type proteins, because it lacks an ␣-helix and a short ␤-sheet found in distinct GPX structures between ␤-sheet 2 and ␣-helix 2. This region with the redox active Cys 95 forms a loop of more than 30 residues (loop 2; Cys 76 -Leu 118 ).
Active Site Structure of Px III-Cys 47 and Cys 95 form the redox active dithiol/disulfide couple, which in the oxidized enzyme forms an intramolecular disulfide (Ref. 18;Figs. 2A and 3A). In both oxidized and reduced Px III, the peroxidatic Cys 47 is located on the first loop (L1) between ␤-strand 1 and ␣-helix 1, whereas the resolving Cys 95 is part of the long loop 2 (L2) between ␤-sheet 2 and ␣-helix 2 that is relatively disordered in the bundle of NMR structures (Figs. 2, B and C). Both residues and F calc (h) are the observed and calculated structure factors, respectively. 5% of the data were excluded to calculate R free .

Data collection statistics
Space group   (31). The CNS E repel function was used to simulate van der Waals interactions with an energy constant of 25 kcal mol Ϫ1 Å Ϫ4 using PROLSQ van der Waals radii; r.m.s.d. for bond length, bond angles, and improper dihedral angels are 0.0034 (Ϯ0.0001) Å, 0.581 (Ϯ0.015)°, and 0.578 (Ϯ0.021)°before and 0.0052 (Ϯ0.0002) Å, 0.697 (Ϯ0.020)°, and 2.24 (Ϯ0.11)°after water refinement (oxidized form) and 0.0047 (Ϯ0.0001) Å, 0.657 (Ϯ0.011)°, and 0.634 (Ϯ0.017)°before and 0.0063 (Ϯ0.0002) Å, 0.761 (Ϯ0.016)°, and 2.27 (Ϯ0.10)°after water refinement (reduced form). c Distance restraints were employed with a soft square-well potential using an energy constant of 50 kcal mol Ϫ1 Å Ϫ2 . No distance restraint in the ͗SA ox ͘ and ͗SA red ͘ was violated by more than 0.2 Å. d Dihedral angle restraints derived from TALOS (32) were applied to , backbone angles using energy constants of 200 kcal mol Ϫ1 rad Ϫ2 . e Coordinate precision is given as the Cartesian coordinate r.m.s.d. of the 20 lowest energy structures in the NMR ensemble with respect to their mean structure for residues 15-179 of Px III. f Structural quality was analyzed using PROCHEK. OCTOBER 31, 2008 • VOLUME 283 • NUMBER 44 are surface-exposed and are, thus, freely accessible to large hydroperoxide substrates as well as to tryparedoxin. Px III contains three solvent-exposed aromatic residues (Tyr 49 , Phe 79 , and Phe 93 ). Two of them (Tyr 49 and Phe 93 ) are located close to the redox active cysteines. Because aromatic residues usually pack into the core of a protein, this may indicate that these residues are involved in the binding of protein partners. An aromatic residue at position 93 is found in most of the monomeric members of the protein family (Ref. 10, Fig. 1), whereas Tyr 49 is not conserved. Interestingly, Cys 47 and Cys 95 of Px III are located in the center of a negatively and positively charged surface patch formed by Glu 87 , Glu 88 , Glu 89 , Glu 92 and Lys 46 , Lys 97 , Lys 99 , Lys 107 , respectively. The residues of these charged patches are partially conserved (Fig. 1).

Structure of T. brucei Tryparedoxin Peroxidase
In the selenocysteine-containing peroxidases, the active site selenoate is in hydrogen bonding distance to the indole N ⑀ H of the tryptophan and the N ⑀ H amide group of the glutamine of the catalytic triad. The corresponding residues are conserved throughout the entire protein family (Fig. 1) (11-13, 42). However, in all of our three structures of Px III, Trp 137 and Gln 82 are not found in the immediate vicinity of the cysteines (the distance between the C ␣ atom of Cys 47 and the side chain nitrogen atoms of Trp 137 or Gln 82 are 16 and 16.7 Å; Fig. 3, A-C). According to the NMR data, Trp 137 is surface-exposed, with only a few NOEs to Leu 130 , Lys 136 , and Ser 155 . In the x-ray structure, the side chain amide of Gln 82 is in hydrogen-bonding distance to the carboxylate group of Glu 83 , which forms a hydrogen bond with the side chain amide of Asn 109 (Fig. 3A), concomitantly stabilizing loop 2 that harbors Cys 95 . In the NMR structures of both the oxidized and reduced protein, the side chain orientation of Gln 82 could not be determined to high precision due to spectral overlap of its resonances, but the rough orientation of Gln 82 defined by a few isolated NOEs in both structures superimposed well with the crystallographic data (Fig. 3, A-C). Thus, despite the conservation of Gln 82 and Trp 137 and the fact that the Gly 82 and Gly 137 mutants of T. brucei Px III had 2 and 12% wild type activity, respectively (18), in Px III the two residues are too far away from the redox active cysteine residues to contribute to a stabilization of the thiolate anion of Cys 47 .
Structure of Reduced Px III-Superposition of the 1 H, 15 N HSQC fingerprint spectra of the reduced and oxidized protein revealed that numerous residues underwent significant chemical shift changes, and several resonances disappeared due to exchange broadening (Fig. 4A). Most of these are located close to the redox active cysteines (Fig. 4B). To identify which peaks disappeared due to exchange broadening and which ones shifted into highly overlapping regions, selectively labeled samples of [ 15 Fig. S3).
Overall, the structures of the oxidized and reduced protein superimposed well ( Fig. 2A). The r.m.s.d. over the backbone atoms between reduced and oxidized structures was 0.59 Å. Thus, most of the secondary chemical shifts do not originate from structural changes, e.g. the signals of Ile 40 and Tyr 41 (Fig.  4A) shift although they are located in the core of the protein.
Large secondary chemical shifts can be attributed to the flanking regions of the two cysteines (Fig. 4B). Inspection of the structure revealed that they are essentially caused by the formation of the disulfide bridge and probably a slight rotation of the neighboring aromatic side chains of Phe 93 and Phe 98 . Substantial structural rearrangements were not observed for the catalytic region in accordance with comparable NOE patterns in the oxidized and reduced protein for residues of loop 1 and 2. The residues between Gln 82 and Lys 97 do not have long range NOEs either in oxidized or in reduced Px III, and their 13 C C␣-and C␤ chemical shifts were comparable in both redox states. This rules out that loop 2, rather separated from the core of the protein in our structures, rearranges into an ␣-helix in reduced Px III, as found for poplar TxP (16).
Trp 137 and Gln 82 are likewise unaffected by the redox state of the protein as both the NOE pattern and especially the more sensitive chemical shifts did not differ between oxidized and Gln 82 (magenta) on loop 2 is 16.7 Å apart from the disulfide bridge and is, thus, unable to interact with the cysteines. Instead, its side chain amide is in hydrogen bond distance to the carboxylate group of Glu 83 (red), which in turn is 2.9 Å distance from the side chain amide of Asn 109 (cyan). Thus, Gln 82 is part of a network of hydrogen bonds and thereby may stabilize the orientation of Cys 95 . B and C, details of the redox active part in a representative NMR structure of the oxidized (B) and reduced (C) Px III, respectively. reduced protein. In the reduced protein signals obtained for Trp 137 were exchange broadened (but still detectable), indicating higher flexibility. The tryptophan fluorescence emission spectra of reduced and oxidized Px III were identical (supplemental Fig. S4), corroborating that the protein environment around the single tryptophan residue does not change significantly. The spectra of the Ser 47 and Ser 95 mutants showed the same emission maximum, further highlighting that the environment of Trp 137 is unaffected by formation of the intramolecular disulfide bridge. The fluorescence emission maximum at 344 nm is in accordance with a rather solvent-exposed position of Trp 137 . In the presence of 6 M guanidinium hydrochloride, the emission maximum of the reduced protein was 360 nm (not shown). Thus, denaturing only slightly further increases the hydrophilicity around Trp 137 .
The Thus, the reason that the loops 2 and 3 are not well defined in both NMR structures is not only due to the fact that the respective residues have a low number of long range NOEs but reflects true mobility. It is not surprising that opening of the disulfide bridge leads to an increase in the mobility of the loops. This could also account for the overall low stability of the reduced protein. However, a significant motion is already present in the oxidized form and appears to be an intrinsic feature of the enzyme.
Evidence for Two Conformations in the Reduced Protein-When analyzing the two-dimensional NOESY spectra of reduced Px III acquired at high field (900 MHz, 22°C), we   OCTOBER 31, 2008 • VOLUME 283 • NUMBER 44 observed a duplication of at least 20 resonances that showed an exactly identical twin NOE pattern originating from two conformations. The intensities of the two conformations were comparable. Residues affected by the duplication extended well into the core of the protein. One-dimensional spectra acquired at different pH values or after carboxyamidomethylation of the cysteines revealed that the duplication arose from the redox active cysteine (supplemental Fig. S5). The analysis of the Ser 47 and Ser 95 mutants confirmed this. The ratio of the two conformations varied with temperature (studied range 4 -29°C), pH, and H 2 O versus D 2 O, e.g. at 4°C, one conformation was dominant, whereas at 29°C both populations were of similar intensity. Thus, instead of two irreversibly modified distinct species, we observed two interchanging conformations. Although the phenomenon awaits further accurate exploration, our data suggest that the two conformations observed in the reduced protein are caused by protonation/deprotonation of the cysteines, which might be relevant for the catalytic mechanism of the enzyme.

Structure of T. brucei Tryparedoxin Peroxidase
Conserved Lysine Residues Specifically Affect Catalysis-In the three structures of Px III, Lys 97 , Lys 99 (both conserved in trypanosomatid and related plant proteins), and Lys 107 (conserved throughout the entire protein family) are located in the vicinity of the redox active dithiol/disulfide on the surface of the protein pointing outward with their N z H 3 groups (Figs. 3, A-C). To elucidate if these residues play a role in catalysis, Lys 97 and Lys 99 were replaced by glutamine and all three lysine residues by glutamate. One-dimensional spectra confirmed that the mutations did not affect the overall protein structure.
Gln 99 , Glu 99 , and Gln 97 Px III showed 40 -60% and Glu 97 Px III 12% remaining activity, whereas Glu 107 Px III had only 2% wild type activity left ( Table 3). The expression system introduced in this work yielded Px III species that were significantly more stable than the previously prepared recombinant protein (5,17,18). In addition, the wild type enzyme obtained here exerted a three times higher specific activity and displayed saturation kinetics. To investigate which of the two half reactions, namely the reduction of H 2 O 2 by reduced Px III or the regeneration of reduced Px III by Tpx (Scheme 2), is affected by the replacement of the lysines, wild type, Glu 97 , Glu 99 , and Glu 107 Px III were subjected to a detailed kinetic analysis (as shown for Glu 97 and wild type Px III in supplemental Fig. S6). The K m values and the apparent second order rate constant k 1 Ј of wild type, Glu 97 , and Glu 99 Px III were in the same order of magnitude, whereas the k 2 Ј value of Glu 97 and Glu 99 Px III was 10 and 5 times lower than that of the wild type enzyme, respectively (Table 3). This suggests that Lys 97 and Lys 99 play a role in the interaction between oxidized Px III and reduced Tpx.
Strikingly, replacement of Lys 107 did not influence the interaction with Tpx (k 2 Ј) but caused a drop of the k 1 Ј-value by more than 3 orders of magnitude and, thus, severely affected the reduction of H 2 O 2 . Because Glu 107 Px III slowly reduced H 2 O 2 , the disulfide bond between Cys 47 and Cys 95 can still be formed. Otherwise Cys 47 should have been irreversibly over-oxidized. As shown previously, reaction of Ser 95 Px III with a stoichiometric concentration of H 2 O 2 generates a sulfinic acid at Cys 47 (18). Probably the nucleophilic attack of Cys 47 on the peroxide substrate is impaired.
Because our structures do not reveal direct contact between Cys 47 and Lys 107 , structural changes in loop 1 or loop 2 may account for the observed changes. In 14 of the 20 lowest energy NMR structures of reduced Px III, Lys 107 interacts with the backbone or side chain hydroxyl group of Ser 45 and/or Thr 96 and, thus, stabilizes the conformation of these loops. Ser 45 in its turn could activate Cys 47 by another hydrogen bond. However, since the side chain resonances of Lys 107 are found in highly overlapping areas of the NOESY spectra, this hypothesis needs a final confirmation by mutational analysis.
Modeling of the Interactions between Px III and Tpx-The structure of the complex between Px III and Tpx was modeled using HADDOCK (37,38). The starting structures were the lowest energy NMR structure of the oxidized Px III and the x-ray structure of oxidized T. brucei Tpx (PDB code 1O73 (36)). In the absence of a structure of reduced Tpx, this approach seems justified because reduction of Tpx from C. fasciculata has only minor effects (36,44). The expected structural change between oxidized and reduced Tpx was taken into account by defining the active site WCPPC motif fully flexible during the second stage of the docking.
Mass spectrometry data revealed that the intermolecular disulfide bridge is formed between Cys 95 of Px III and Cys 40 of Tpx (18). The two cysteines were, hence, treated as the center of the interaction. To create a tightly connected complex structure, the thiol hydrogen atoms of the reactive cysteines were removed, leaving more space for the sulfur atoms to approach each other. Lys 97 and Lys 99 of Px III were also incorporated in the interaction surface because of the mutagenesis results ( Table 3).
The complex structure (Fig. 6A) contains 4 salt bridges, 1 hydrogen bond, and 21 hydrophobic contacts as identified by LIGPLOT (40). The buried surface area at the contact between the two protein molecules is 1350.46 Å 2 . The HADDOCK score

Kinetic constants for the reduction of hydrogen peroxide by wild type and mutant Px III species
The Dalziel coefficient ⌽ 0 (the ordinate intercept in the secondary plot, see supplemental Fig. S6C) equals E o /V max that is 1/k cat . ⌽ 1 (the slope of the primary plot, supplemental Fig. S6, A and B) corresponds to 1/k 1 Ј, and ⌽ 2 (the slope of the secondary plot, supplemental Fig. S6C) corresponds to 1/k 2 Ј. The k 1 Ј and k 2 Ј values refer to the overall rate constants for the oxidation and reduction of the peroxidase, respectively. K m 1 (⌽ 1 /⌽ 0 ) and K m 2 (⌽ 2 /⌽ 0 ) represent the K m values for hydrogen peroxide and tryparedoxin, respectively. The values are the mean of three independent series of experiments. ND, not determined.

Px III
Activity

DISCUSSION
Eight structures of glutathione peroxidase-type enzymes are currently available. The majority of them (bovine GPX1 and human GPX1-4) represent the classical selenoenzymes, either the authentic proteins or mutants where the selenocysteine is replaced by different residues. The other two structures show cysteine-containing homologues (human GPX5 and GPX7), but these proteins lack a second redox active cysteine and, thus, are unable to form an intramolecular disulfide bridge. Thioredoxin-dependent poplar TxP is the only peroxidase that, as does Px III, cycles between a reduced dithiol form and an oxidized form containing an intramolecular disulfide bridge (16).
Poplar TxP undergoes a remarkable structural rearrangement upon oxidation previously observed for both typical and atypical 2-Cys peroxiredoxins (Refs. 45 and 46, and for a comprehensive review, see Ref. 47). The reduced form shows a catalytic triad with Glu 82 and Trp 133 stabilizing the peroxidatic cysteine analogous to GPXs (16). Upon oxidation, ␣-helix 3 unwinds, which leads to a long loop that extends from the core of the protein and allows the peroxidatic cysteine to approach the resolving one to form a disulfide bridge, disrupting the arrangement of the catalytic triad. Although the monomeric T. brucei Px III and the dimeric poplar TxP show 48% overall sequence identity, our structural data revealed that Px III does not undergo such a dramatic conformational change. Reduced and oxidized Px III have essentially the same structure, and the conserved Gln 82 and Trp 133 do not form a catalytic triad with Cys 47 . This was demonstrated by the unchanged chemical shifts for both residues and identical NOE patterns, especially for Trp 137 , loop 1, and loop 2.
Treatment of reduced Px III with 0.5 mM glutathione disulfide leads to glutathionylation of either Cys 47 or Cys 95 but not of both residues in the same protein molecule (48). Modification of one cysteine probably interferes sterically with the modification of the second one in accordance with the structural data that also in the reduced form of Px III both cysteines lie in close proximity. This would allow the rapid condensation between the sulfenic acid formed at Cys 47 upon hydroperoxide reduction and the thiol of Cys 95 generating the intramolecular disulfide bond of the oxidized enzyme.
The different catalytic mechanism of T. brucei Px III and the poplar TxP is further corroborated by the tryptophan fluorescence of the proteins. Whereas the spectra of poplar TxP reflect a switch from a hydrophobic environment in the reduced state to an exposed conformation in the oxidized state (43), the tryptophan fluorescence emission spectra of oxidized and reduced Px III are identical and in accordance  (35) showing the electrostatic interactions between charged residues at the interface. B, the side chains of Tpx that interact with Px III are displayed together with a surface representation of Px III. C, same for side chains of Px III that interact with Tpx. with an exposed position of Trp 137 in both forms (supplemental Fig. S4).
Is it possible that despite a sequence identity of nearly 50%, the structures differ so considerably? The paradigm of homology modeling efforts and structural genomic initiatives is that related sequences lead to homologous structures (49) hoping that prediction can mostly replace tedious structure determination. In general, a sequence identity of 50% is regarded as a good basis for a relatively safe prediction. However, a recent publication indicates that this goal is not always achievable. Bryan and co-workers inter-converted all residues of a G A and a G B domain of protein G, a cell wall protein from Streptococcus. 5 of 45 residues were shown to be sufficient to change the structure of an 3-␣-helix-fold into an ␣/␤-fold (50). Likewise, the insertion of three further residues into the sequence converts Gal3p, a protein of the GAL pathway in Saccharomyces cerevisiae, from a transcriptional inhibitor to an activator of the pathway, changing it into a galactokinase (51). Thus, small changes in the protein sequence can lead to huge and crucial differences in the three-dimensional structure or function that render predictions not entirely reliable.
The NMR structure of oxidized and reduced Px III revealed that the three loops containing Cys 47 , Cys 95 , and Trp 137 are highly flexible, especially in the reduced protein. In particular, the residues between Gln 82 and Phe 98 do not show long range NOEs, in accordance with loop 2 not being intimately connected to the core of the protein. Previous studies of Akke and Kern and co-workers (52) have revealed that enzymes may not have to rearrange their enzymatic pockets in an induced fit but that the substrate appears to select the correct one among many allowed conformations. A similar mechanism could be the case for Px III. Binding of a substrate could trap one conformation leading to a stabilization of the loops. Unfortunately, hydroperoxides are rather unstable, so that this could not be investigated by NMR or x-ray analyses. The fact that several residues show line-broadening in the 1 H 15 N HSQC spectra corroborates the idea that these loops and adjacent regions undergo micro-or millisecond motions. Upon formation of the disulfide bridge, the entire structure is stabilized, which is reflected in the higher thermal stability and the overall higher quality of the NMR spectra. When the disulfide bridge opens, the loops may undergo a swaying motion.
In the structures of Px III, Cys 47 , Gln 82 , and Trp 137 are located far apart from each other and not in direct contact, excluding the formation of the canonical catalytic triad. The specific role of these residues is also not easy to analyze in other members of the family because they are not necessarily conserved, at least in the cysteine-containing glutathione peroxidase-type enzymes. In TxP, the glutamine residue of the catalytic triad is replaced by a glutamate (Glu 79 ), but the structural arrangement is comparable with that in the selenocysteine enzymes. A respective mutation in GPX4 severely affected the catalytic efficiency of the protein (13,42), but in the TxP of Chinese cabbage (19), a glutamine to glycine mutation resulted in a fully active enzyme species.
One interesting aspect is how the peroxidatic Cys 47 is activated. In many cases the low pK value of a specific cysteine residue is difficult to explain. Despite a number of known struc-tures and detailed biochemical studies, it is still not clear why the N-terminal Cys of the CXXC motif in glutaredoxins has a pK value significantly lower than that of a free cysteine, whereas the pK value of the C-terminal active site Cys has a pK value higher than that of free cysteine (53). A common hypothesis is that conserved basic residues (Arg 26 and Lys 19 in pig glutaredoxin) (54) or an arginine in peroxiredoxins (55) stabilize the thiolate of the more N-terminal cysteine (54). However, a recent molecular dynamics simulation suggests that cationic side chains contribute little to the activation. Instead, the thiolate appears to be stabilized by the backbone NH groups of the two following residues (53). The structures of Px III did not reveal any positively charged residue in the vicinity of the active site thiolate, and one may speculate that the thiolate of Cys 47 in reduced Px III is also stabilized by interaction with the backbone. However, one should bear in mind that due to a lack of unambiguous NOEs of the residues close to Cys 47 , the exact orientation of several residues in the active site could not be determined by NMR.
The active site of Px III shows a very open conformation. The immediate environment of the redox active dithiol/disulfide is formed exclusively by residues of the loops 1 and 2 that contain Cys 47 and Cys 95 , respectively. In comparison, the active sites of poplar TxP as well as of GPX1, GPX3, and GPX4 also comprise residues of the loops between ␣-helix 2 and ␤-strand 3 as well as between ␤-strand 4 and ␣-helix 3 in Px III (11,16). Furthermore, the active site of these enzymes is shielded by an adjacent subunit. The exposed redox active cysteines of Px III would render the protein perfectly suited for interactions with bulky substrates and may also be a prerequisite for an efficient reduction by Tpx.
Glutathione peroxidase-type peroxidase-3 in yeast does not act only as a hydroperoxide scavenger but also as a hydroperoxide sensor (56). The sulfenic acid residue generated in the peroxidase reacts with a cysteine of the transcription factor Yap1. This intermolecular disulfide is then transposed to an intramolecular disulfide in Yap1, causing its activation. Future work will reveal if Px III is part of a redox signaling cascade in trypanosomes.
Note Added in Proof-While this manuscript was under revision, Fairlamb and colleagues published the crystal structure of reduced T. brucei Px II (2VUP; Alphey, M. S., König, J., and Fairlamb, A. H. (2008) Biochem. J. 414, 375-381). This structure is similar to the crystal structure of the poplar TxP but differs from our solution structure in the details of the catalytic site. Future work is needed to resolve the discrepancies between the structures.