The E-loop Is Involved in Hydrogen Peroxide Formation by the NADPH Oxidase Nox4*

In contrast to the NADPH oxidases Nox1 and Nox2, which generate superoxide (O2̇̄), Nox4 produces hydrogen peroxide (H2O2). We constructed chimeric proteins and mutants to address the protein region that specifies which reactive oxygen species is produced. Reactive oxygen species were measured with luminol/horseradish peroxidase and Amplex Red for H2O2 versus L-012 and cytochrome c for O2̇̄. The third extracytosolic loop (E-loop) of Nox4 is 28 amino acids longer than that of Nox1 or Nox2. Deletion of E-loop amino acids only present in Nox4 or exchange of the two cysteines in these stretches switched Nox4 from H2O2 to O2̇̄ generation while preserving expression and intracellular localization. In the presence of an NO donor, the O2̇̄-producing Nox4 mutants, but not wild-type Nox4, generated peroxynitrite, excluding artifacts of the detection system as the apparent origin of O2̇̄. In Cos7 cells, in which Nox4 partially localizes to the plasma membrane, an antibody directed against the E-loop decreased H2O2 but increased O2̇̄ formation by Nox4 without affecting Nox1-dependent O2̇̄ formation. The E-loop of Nox4 but not Nox1 and Nox2 contains a highly conserved histidine that could serve as a source for protons to accelerate spontaneous dismutation of superoxide to form H2O2. Mutation of this but not of four other conserved histidines also switched Nox4 from H2O2 to O2̇̄ formation. Thus, H2O2 formation is an intrinsic property of Nox4 that involves its E-loop. The structure of the E-loop may hinder O2̇̄ egress and/or provide a source for protons, allowing dismutation to form H2O2.

The class of Nox protein NADPH oxidases is a group of enzymes whose sole known function is the production of reactive oxygen species (ROS). 2 The enzyme family is named for the enzymatically active transmembrane protein Nox. All seven Nox proteins share highly conserved structural features; the C-terminal dehydrogenase domain contains binding sites for FAD and NADPH. The N-terminal transmembrane region consists of six ␣-helical transmembrane domains that contain four conserved histidine residues, located in the third and fifth transmembrane helices that coordinate two hemes. Electron transfer occurs from NADPH to oxygen via FAD and the two heme groups, with the second heme group reducing molecular oxygen (1,2). Because heme is an obligate 1-electron donor, it is generally accepted that the superoxide anion (O 2 . ) is the initial reduction product of oxygen, although the latter can also react with a second O 2 . to form hydrogen peroxide (H 2 O 2 ) plus oxygen. Despite these similarities, Nox proteins differ in their mode of activation, their interaction with the small transmembrane protein p22 phox , and the requirement for additional maturation and activation factors. The most extensively studied NADPH oxidase isoform is the phagocyte Nox2 (previously termed gp91 phox ), which depends on p22 phox and whose activation requires assembly with the cytosolic regulatory subunits p47 phox and p67 phox (3), along with GTP-loaded Rac1 or Rac2. Similar to Nox2, the homologue Nox1 requires regulatory subunits. Unlike all the other Nox proteins, Nox4 is constitutively active and is independent of cytosolic activator proteins or regulatory domains (4,5). Another interesting difference between Nox1/2 and Nox4 is that Nox1 and Nox2 produce primarily O 2 . , whereas most studies report that Nox4 generates H 2 O 2 (4,6,7). Given that the prosthetic groups are identical and the core protein structures are very similar among the Nox proteins, it has been suggested that the failure to detect O 2 . formation by Nox4 is a consequence of its intracellular location, resulting in problems detecting O 2 . within the cell and the inability of this ion to pass freely through the membrane. Indeed, although a significant fraction of Nox1 and Nox2 is located at the plasma membrane and thus would reduce extracellular oxygen, Nox4 is localized predominantly to intracellular membranes where any generated O 2 . might be cryptic. In fact, Nox4 protein has been reported in mitochondria (8), the nucleus (9), the cytoskeleton (10), and the endoplasmic reticulum (11 . production using ESR spin traps and the dihydroethidium method (13). In contrast, Nox4 was able to reduce nitro blue tetrazolium. However, the site of the electron efflux from Nox4 to nitro blue tetrazolium has not yet been determined but is potentially via the FAD-containing dehydrogenase domain, which is known to catalyze the direct reduction of various dyes (14). Based on the above, it has to be concluded that although mechanistically, heme reduction of oxygen must initially generate O 2 . , Nox4 releases H 2 O 2 without releasing free O 2 . . The molecular basis for this potentially physiologically important difference is unclear. Based on the presence of six transmembrane ␣-helical domains, the current model for NADPH oxidases predicts that the N-as well as C-terminal parts of the protein reside in the cytosol, giving rise to two intracellular loops (B-and D-loop) and three loops oriented away from the cytosol and toward the extracellular space or intracellular compartments (A-, C-, and E-loop). So far, little work has been devoted to the extracellular loops. Although asparagines within these regions are glycosylated in Nox2, to our knowledge, no mutations leading to chronic granulomatous disease have been reported for these loops. Also, the functional significance of glycosylation is somewhat uncertain as unlike human Nox2, the murine enzyme does not undergo this modification and glycosylation has not been reported for Nox1. Based upon their proximity to the site of oxygen reduction by the B heme, we hypothesized that differences in the extracellular loops are responsible for the unique ability of Nox4 to release H 2 O 2 rather than O 2 . .

EXPERIMENTAL PROCEDURES
Sequence Alignment-Nox sequences were aligned using the online program ClustalW2 from the European Molecular Biology Laboratory-European Bioinformatics Institute (EMBL-EBI) (15).
Generation and Transfection of Mutant Nox Constructs-Plasmids encoding human full-length Nox1, Nox4, and p22 phox were kindly provided by T. Leto (National Institutes of Health, Bethesda, MD). The plasmids coding for mouse Noxa1 and Noxo1 were generous gifts of B. Banfi (Iowa University, Iowa City, IA). The plasmids coding for the Nox4 deletion mutants were generated by overlap extension PCR. The plasmids coding for the cysteine and histidine mutants of Nox1 and Nox4 were generated by site-directed mutagenesis using the QuikChange mutagenesis kit (Stratagene, La Jolla, CA) according to the manufacturer's instructions. All cloned plasmids were confirmed with DNA sequencing. Determination of Reactive Oxygen Species by Cytochrome c Reduction-HEK293 cells were seeded on 12-well plates and transiently transfected. 24 h later, cells were washed once in HEPES-modified Tyrode's solution containing cytochrome c (from horse heart; Sigma; 1 mg/ml) and then incubated in this solution in the presence or absence of superoxide dismutase (Sigma, 100 units/ml) at 37°C. After 30 -60 min, supernatants were transferred to fresh tubes on ice, and the superoxide dismutase-inhibitable reduction of cytochrome c was quantified in a spectrophotometer (Uvikon, Kontron Instruments). Data were normalized to the isosbestic points at 542 and 558 nm, and O 2 . formation was calculated with the aid of the molar extinc- Western Blot Analysis-24 h after transfection, cells were incubated for 8 h with the proteasome inhibitor MG132 (10 M, Calbiochem) to stabilize and increase the Nox expression. Cell lysis, SDS-PAGE, and Western blot were carried out (16). The following antibodies were used: anti-Nox1 (Santa Cruz Biotechnology: Mox-1 H-15); anti-Nox4 (generated by us (17)); anti-␤-actin (Sigma); and anti-Erk1/2 and anti-phosphorylated Erk1/2 (Cell Signaling).
Confocal Fluorescence Microscopy-HEK293 cells stably expressing Nox4 ⌬218 -235 ϩ ⌬264 -273 were seeded on -dishes (ibidi, Martinsried, Germany). When the cells reached ϳ80% confluence, they were incubated for 8 h with the translation inhibitor anisomycin (Calbiochem; 20 M) to reduce potential localization of the protein to the endoplasmic reticulum due to de novo synthesis. Imaging was carried out as described (16) with the following antibodies: anti-Nox4 (generated by us) and anti-GRP78 (Santa Cruz Biotechnology) for endoplasmic reticulum staining.
Statistical Analysis-Unless otherwise stated, all data shown are mean Ϯ S.E. Statistical significance was determined by oneway analysis of variance followed by Newman-Keuls post hoc test or by paired or unpaired t test, if appropriate.

RESULTS
Role of the Length of the E-loop of Nox4-To identify a structural basis for the H 2 O 2 formation of Nox4, we compared this protein with Nox1 and Nox2. An alignment of the amino acid sequences of human and mouse Nox1, Nox2, and Nox4 revealed that the E-loop of Nox4 is 28 amino acids longer than the E-loop of Nox1 and Nox2 (Fig. 1A) as a consequence of two insertions of 18 and 10 amino acids, respectively. The sequence within the two insertions is similar among Nox4s from different species, suggesting a conserved Nox4-specific function. We deleted these Nox4-specific E-loop regions so as to approximate the length of the E-loop in Nox1 and Nox2 (Fig. 1B). These mutations, which did not affect the endoplasmic reticulum localization of Nox4 (Fig. 1C), had a strong effect on ROS production by Nox4; the mutants produced very little H 2 O 2 when compared with native Nox4 and instead released O 2 . (Fig. 1, D and E). The effect was more pronounced in the double deletion mutant, which lacked both sequence insertions, but deletion of amino acids 264 -273 was almost equally effective. Although the Nox4 deletion mutants produced O 2 . , their O 2 . formation rate was still lower than that of native Nox1 when equimolar amounts of the expression plasmid were transfected (data not shown). Given that Nox4 del218 -235ϩ264 -273 colocalized with the heat-shock protein GRP78 (Fig. 1C) and thus is expressed in the endoplasmic reticulum-like native Nox4, we conclude that the type of ROS formed by Nox proteins is an intrinsic feature of the protein and not a consequence of the intracellular localization.

Role of the Cysteines in the E-loop of Nox Proteins-
The E-loops of Nox4 as well as Nox1 and Nox2 both contain two conserved cysteines, although they are located in different positions in the two isoforms. In Nox4, both are lost upon the construction of the deletion mutants of Nox4. As the cysteines might form a disulfide bridge to maintain the integrity of the E-loop, we studied their role for the function of Nox4. Mutation of the cysteines to valines had a similar, although less pronounced, effect on ROS production by Nox4 as the deletion constructs (Fig. 2, A and B), as demonstrated by chemiluminescence as well as cytochrome c reduction and Amplex Red oxidation. A plasmid harboring the mutation of both Nox4 cysteines together showed a more pronounced effect on the switching of the ROS type generated than the single mutations (data not shown). Mutation of the two cysteines in Nox1 decreased the overall activity of the enzyme but did not change the type of ROS released (Fig. 2C) (Fig. 3).
Binding of an Antibody to the E-loop Affects H 2 O 2 Production of Nox4-A general problem with mutational analyses is that local changes in structure may affect the function of an enzyme . generation was determined by L-012 (D), and H 2 O 2 formation was determined by luminol ϩ HRP chemiluminescence (E). ROS production was normalized against Nox protein expression determined by Western blot analysis. To facilitate better comparison of the constructs, the ROS formation of the most active construct was set to 100% (corresponding to mean 271,217 cpm (D) and 426,172 cpm (E)) n Ն 3, mean Ϯ S.E., *, p Ͻ 0.05 versus wild-type construct. % of Max., percentage of maximum. at a remote site by some allosteric modulations. We therefore sought a second approach to interfere with the E-loop. Recently, a monoclonal antibody directed against this part of the protein was generated (mAb8E9) (18). When compared with an antibody that binds in the NADPH binding site (Fig.  3C), Western blot analysis confirmed that the antibody mAb8E9 binds the E-loop as labeling was lost in the Nox4 mutants (Fig. 4). As the antibody can only directly interact with extracellular portions of Nox4, this part of the study was carried out in Cos7 cells, which exhibit some plasma membrane expression of this NADPH oxidase (12).
The addition of the mAb8E9 decreased H 2 O 2 production by Nox4-transfected Cos7 cells by ϳ20% (Fig. 4), and in a pilot study, mAb8E9 also increased the O 2 . formation of Nox4 (ϩ28 Ϯ 11%, n ϭ 2). Importantly, a nonspecific mouse IgG control at the same concentration had no effect on ROS formation by Nox4, and also, mAb8E9 did not alter O 2 . production in  (Fig. 4).
Western blot analysis of the Nox4 deletion constructs (Fig.  3C) indicated that the protein Nox4 del218 -235 might be unstable as it was detected with a molecular mass of 6 kDa less than the calculated mass. As, however, ROS measurements of the other deletion constructs and the cysteine mutants in the same region yielded similar results, the reduced stability of this single construct might be of lesser importance for the interpretation of the data. and statistical analysis (B and C) of ONOO Ϫ production of HEK293 cells transiently transfected with the plasmids indicated as well as protein expression of the constructs. ROS formation was determined by luminol chemiluminescence. After reading the cells in the absence of NO, the NO donor DetaNONOate (100 mol/liter) was administered (first arrow). The subsequent signal was considered to arise from ONOO Ϫ formation. At the end of the experiment, horseradish peroxidase (ϩHRP) was added (second arrow) to determine H 2 O 2 formation. The Nox1 plasmid was cotransfected with Noxo1 and Noxa1. Normalization was as in Fig. 1. Exemplary Western blots of the construct are shown in B and C. n Ն 3, mean Ϯ S.E., *, p Ͻ 0.05 versus wild-type construct. 100% corresponds to mean 7625 and 7818 cpm (B and C, respectively). % of Max., percentage of maximum. Structure alignment revealed that in addition to those that are directly involved in heme ligation, Nox4 contains five additional highly conserved histidines in its membrane integrated part (Fig. 5A), and interestingly, three of these are located in the E-loop. We mutated these five histidines to glutamine and studied protein expression and ROS formation by the mutants. The mutations of His-47, His-246, and His-248 had no effect on ROS production, whereas H16Q reduced H 2 O 2 formation by Nox4 by ϳ50% without affecting O 2 . formation. In contrast, Nox4-H222Q, which carries the mutation at the beginning of the E-loop, produced almost no H 2 O 2 and instead released O 2 . (Fig. 5B). The expression of the histidine mutants was similar to that of native Nox4 as determined by Western blot analysis (Fig.  5C). These observations not only confirm that the E-loop is central for H 2 O 2 formation but also indicate that specific features of the E-loop are required for the process.

Mutation of the Cysteines in the E-loop Prevents
Nox4-induced Erk1/2 Phosphorylation-In the final step, we sought to demonstrate that the type of ROS released by Nox4 is physiologically relevant. Unlike Nox2 and Nox1, Nox4 activity has previously been linked to Erk1/2 phosphorylation (17). We therefore investigated the activation of this MAP kinase in response to the overexpression of Nox4 versus Nox4 cysteine mutants. Although overexpression of native Nox4 induced a robust Erk1/2 phosphorylation in HEK293 cells, this effect was not observed with the cysteine mutants, although they were expressed to a similar level and total Erk1/2 expression was not affected by this approach (Fig. 6). These data indicate that the direct formation of H 2 O 2 by Nox4 has important biological consequences for Nox-dependent signal transduction.

DISCUSSION
In this work, we studied the structural basis for H 2 O 2 formation by Nox4. We identified the extended E-loop of the protein as an essential structural feature for this process and showed that alterations of the E-loop switch Nox4 from an H 2 O 2 into an O 2 . -producing enzyme. The ROS product generated appears to be sensitive to minor structural perturbations in the E-loop because binding by a monoclonal antibody or cross-linking at . production in Cos7 cells transiently transfected with wild-type Nox4 or Nox1 (cotransfected with Noxo1 and Noxa1). The following substances indicated were added to the cells prior to measurements: monoclonal antibody directed against the E-loop (mAb8E9: 1 g/ml), normal control mouse IgG (1 g/ml), and the surface cross-linker bis-sulfosuccinimidyl suberate (BS3, 1 mmol/liter). Normalization was as in Fig. 1. n Ն 3 (19,20). The molecular mechanisms leading to Duox-dependent H 2 O 2 formation are not clear but appear to be distinct from that of Nox4. Based on homologue searches, it was suggested that Duox enzymes might bear an intrinsic superoxide dismutase activity localized to the peroxidase homology domain of these proteins (21), but recently, it was shown that although structurally similar to peroxidases, mammalian Duox enzymes are not able to directly dismutate O 2 . (22). Rather, the type of ROS generated appears to depend on the maturation state of the Duox enzymes. In the absence of the maturation factors Duoxa1 or Duoxa2, the Duox enzymes generate O 2 . , and only after interaction with the Duoxa proteins does H 2 O 2 formation becomes apparent (23). This process also involves relocalization of the Duox-Duoxa complex from the endoplasmic reticulum to the plasma membrane, and it is not clear whether localization impacts the observed reduced oxygen species (24). The enzymatic activity of the Duox enzymes is independent of p22 phox , whereas the maturation of Nox2 (25) as well as the activity of most of the Nox proteins (with the exception of Nox5) is dependent on p22 phox (7). Thus, it is not an understatement that, by analogy with the Duox-Duoxa proteins, p22 phox was referred to as a maturation factor for the Nox proteins (24). Divergent from this concept, it was observed that alteration of the maturation factors for the Duox proteins switch them from H 2 O 2 to O 2 . formation (24), whereas so far no mutation in p22 phox was reported to induce such a transition for Nox4 (7). Interestingly, based on truncation experiments, it can be concluded that at least some different regions of p22 phox interact with Nox1/2 when compared with Nox4 (7), but an in-depth molecular analysis has not yet been carried out. As for Nox4, a molecular explanation for H 2 O 2 formation by Duox is unclear. The lack of superoxide dismutase activity in the peroxidase domain may suggest that the enzymes can produce H 2 O 2 directly like dioxygenases or that whatever O 2 . is produced remains cryptic and inaccessible to assay reagents prior to its dismutation to form H 2 O 2 , for example, due to either spontaneous dismutation or transfer to a second, so far unidentified, interacting protein like a superoxide dismutase.
Obviously, these mechanisms could also account for the H 2 O 2 formation of Nox4. The localization of Nox4 differs among cell types, which is suggestive of the presence of interacting proteins directing the protein to different destinations, but so far, little has been published to support this view. The p22 phoxinteracting protein Poldip2 has been shown to affect Nox4 expression and activity, but it has not been reported that it alters the type of ROS formed by Nox4 (26). A direct interaction of Nox4 with one of the superoxide dismutase isoforms seems unlikely due to the orientation of the O 2 . exit site to the extracytosolic compartment. Indeed, by co-immunoprecipitation experiments, we found no evidence for such an interaction. 3 On the basis of the present data, we suggest that H 2 O 2 formation is an intrinsic function of Nox4. Thus, either the enzyme has endogenous superoxide dismutase activity or it acts as a dioxygenase and directly forms H 2 O 2 . For both processes, histidines and cysteines could be important. Superoxide dismutase activity requires the presence of metal cofactors such as zinc, manganese, or iron, which are coordinated by these amino acids. Indeed, zinc in CuZn-superoxide dismutase is coordinated by three histidines (27), and the zinc in zinc finger proteins is coordinated by two histidines and two cysteines. However, at least two coordinating histidines are likely to be required to form a metal binding site, and our mutational analysis of Nox4 provided evidence for only a single histidine (His-222) of importance. Such a scenario might, however, be viable if Nox4 forms a homodimer in the membrane, as has been suggested for Nox5 (35). According to this scenario, the metal binding site would be formed at the interface between the subunits. Interestingly, also p22 phox contains a single histidine, and this is not required for the function of Nox2 (28). This histidine is located in a region important for Nox4-p22 phox interaction (7) but appears to be located in the transmembrane region of the protein directed more toward the cytosolic side. Thus, it seems unlikely that this amino acid of p22 phox has a specific function for ROS formation by Nox4. Interestingly, His-222 of Nox4 is embedded in the sequence THPPGC, and deletion of this stretch or mutation of the Despite this speculation about metal coordination, histidine might also directly accelerate the formation of H 2 O 2 in the absence of a metal. This amino acid is frequently involved in enzyme catalysis, where it acts as a proton donor (29). The spontaneous rate of dismutation of superoxide is high, provided that the reaction is between O 2 . and its protonated form HO 2 ⅐ 3 I. Takac, unpublished observation. 4 R. Brandes, unpublished observation.  (pK a 4.9). The reported rate constant for the process is ϳ1 ϫ 10 8 M Ϫ1 s Ϫ1 , 7 orders of magnitude higher than the spontaneous dismutation rate for two superoxide anions (30). Thus, we suggest that His-222 could serve as a proton donor for a superoxide anion to form the perhydroxyl radical, which would accelerate spontaneous O 2 . dismutation by many orders of magnitude. Although this rate is still at least 200-fold less than that catalyzed by superoxide dismutases, it is likely to be more than sufficient to keep up with the turnover rate of Nox4, which has been reported to be around 200 min Ϫ1 (14). Another possible role for the E-loop (not mutually exclusive with the above) is to form a physical structure that slows the egress of O 2 . and/or HO 2 ⅐ from its site of formation at the heme,  (32), which was achieved by increasing the redox potential of FAD so that the rate constant of electron transfer from FADH* onto dioxygen was increased (32). Usually, dioxygen is reduced by FADH 2 in xanthine oxidase, and as this reaction in the native enzyme is much faster than the single electron transfer from FADH*, native xanthine oxidase produces much less O 2 . than H 2 O 2 (33). In this context, it is important to emphasize that H 2 O 2 formation indeed usually occurs at FAD sites and does not require the heme present in Nox proteins. Heme proteins such as cytochrome P450 monooxygenases, the NO synthase enzymes, and also the Nox enzymes in contrast produce O 2 . . Thus, one could speculate that H 2 O 2 formation by Nox4 occurs at the FAD site. Up to now, no evidence has been presented to suggest that the FAD in Nox enzymes is able to directly reduce oxygen. Expression of the isolated dehydrogenase domain of Nox4 demonstrated that this domain of Nox4 is able to reduce cytochrome c and several dye electron acceptors (14) but does not form H 2 O 2 directly. 5 From the physiological point of view, the functional role of Nox4 is still incompletely understood, and thus, it is also unclear whether the type of ROS generated by Nox4 impacts function. In overexpression experiments, Nox4 activated a different set of MAP kinases than Nox2, with a particularly strong activation of the Erk1/2 pathway (17)