Nitric-oxide Synthase Output State

Mammalian nitric-oxide synthases are large modular enzymes that evolved from independently expressed ancestors. Calmodulin-controlled isoforms are signal generators; calmodulin activates electron transfer from NADPH through three reductase domains to an oxygenase domain. Structures of the reductase unit and its homologs show FMN and FAD in contact but too isolated from the protein surface to permit exit of reducing equivalents. To study states in which FMN/heme electron transfer is feasible, we designed and produced constructs including only oxygenase and FMN binding domains, eliminating strong internal reductase complex interactions. Constructs for all mammalian isoforms were expressed and purified as dimers. All synthesize NO with peroxide as the electron donor at rates comparable with corresponding oxygenase constructs. All bind cofactors nearly stoichiometrically and have native catalytic sites by spectroscopic criteria. Modest differences in electrochemistry versus independently expressed heme and FMN binding domains suggest interdomain interactions. These interactions can be convincingly demonstrated via calmodulin-induced shifts in high spin ferriheme EPR spectra and through mutual broadening of heme and FMNH· radical signals in inducible nitricoxide synthase constructs. Blue neutral FMN semiquinone can be readily observed; potentials of one electron couple (in inducible nitric-oxide synthase oxygenase FMN, FMN oxidized/semiquione couple =+70 mV, FMN semiquinone/hydroquinone couple =–180 mV, and heme =–180 mV) indicate that FMN is capable of serving as a one electron heme reductant. The construct will serve as the basis for future studies of the output state for NADPH derived reducing equivalents.

and nNOS, are regulated by calcium/calmodulin (Ca ϩ2 /CaM) (4,5) and additional inputs including phosphorylation of specific residues (6); the NO produced by constitutive isoforms functions as a molecular signal. A cytokine inducible isoform (iNOS) is calcium insensitive, and produces much larger fluxes of NO as a cytotoxin in immune response (4).
Eukaryotic NOS isoforms are large modular enzymes. The monomer molecular mass is 120 -161 kDa; the dimer is the active form (7), and the dimerization interface includes the tetrahydrobiopterin (H 4 B) binding site in the oxygenase domain (8). The common elements are the heme and H 4 B containing the oxygenase domain and a complex reductase unit homologous to NADPH P450 reductase that consists of a NADPH binding domain, a FAD binding domain, and an FMN binding domain (9 -11). The reductase and oxygenase regions are linked by a polypeptide segment containing a CaM binding site (12). Evidence suggests that the oxygenase domain of one monomer is reduced by the reductase unit of the other (13) through an oxygenase domain surface that exposes the corner of the heme (8).
Ca ϩ2 /CaM controls constitutive isoforms through regulation of electron transfer between NADPH and heme (14). CaM has little or no effect on the thermodynamics of redox processes in NOS, suggesting that regulation is accomplished through modulation of electron transfer (15)(16)(17)(18). A 40 -50-residue autoinhibitory insertion in the FMN binding domain marks the major difference between cNOSs and iNOS (19), but electron transfer modulation appears also to involve a much smaller insertion present in the subdomain region as well as C-terminal tail differences (20,21).
Although no solved structure of the holoenzyme exists, both partial structures and structures of close homologs provide considerable information (22)(23)(24). In structures of P450 reductase and NOS reductase domains (22)(23)(24), the flavin isoalloxazine ring edges are within 5 Å of each other in the interior of the protein. The available structural and evolutionary information strongly suggests that this conformation can serve as the input state from electrons from NADPH but not as the output state to acceptors such as the oxygenase domain or cytochrome c. For example, FMN cannot be docked to within tunneling distance of heme. These enzymes evolved from ancestral systems in which the FMN domain was a separate protein that functioned as a free shuttle before gene fusion events tethered it to the FAD binding domain (25). Available crystal structures show only the input state of the shuttle. Mechanistic information suggests that the limiting step in catalysis is the formation of the output state; NAPDH reduction is 3 orders of magnitude faster than NO production, and cytochrome c reduction is intermediate between the two.
A construct containing just the FMN binding domain and the oxygenase domain would facilitate study of the association of these two domains free of the influence of the FAD and NADPH binding domains. An analogous construct has been expressed and characterized for cyto-chrome P450BM-3 from Bacillus megaterium, which has a P450 oxygenase domain unrelated to NOS linked to a reductase unit homologous to the corresponding NOS domains (26). P450BM-3 does not bind calmodulin. The B. megaterium construct was used to provide a simpler system for kinetic investigations. This communication describes the design and production of an oxyFMN NOS construct and provides preliminary characterization of its properties. We intend to use this construct to characterize the previously described (25) output state of the FMN binding domain of NOS.

Design of NOS OxyFMN; Strategy and Bacterial Expression Vectors-
The oxygenase and FMN binding domains of mammalian NOS have extended surface charge distributions that appear to define interaction surfaces. The surface of the oxygenase domain has a well defined ring of positive charge surrounding the most exposed edge of the heme. The FMN binding site is surrounded by a group of negatively charged residues, but these surfaces cannot interact when the FMN binding domain is associated with the reductase complex. When the constructs described here were originally designed, the reductase structure was not available, but a good approximation was made by modeling based on the structure of P450 reductase (22). Three logical termination points were selected; the shortest construct terminated directly at the end of the final ␣ helix of the FMN binding domain. The constructs described here were designed to include the initial two acidic residues in cNOS of a C-terminal negatively charged element, which closely wrapped the edge of the FMN binding domain (shown on the left of Fig. 1a terminating as E951) before forming the connector to the FAD binding domain. The extended connector was not included in any of the constructs. The solid green residues are acidic and define two negatively charged surfaces, including the one facing the viewer. Fig. 1b shows the face of the oxygenase domain in which a corner of the heme, shown in red, is exposed. The dark blue residues are positively charged, the green residues negatively charged, and cysteines are shown in yellow.
We decided to truncate the oxygenase domains N-terminally at residue 66 for eNOS, 65 for iNOS, and 290 for nNOS to minimize N-terminal proteolysis during purification, as previously described for oxygenase constructs (27,28). Fig. 2 shows a schematic sequence alignment for the three constructs used in this study. The C-terminal residues are FIGURE 1. a, nNOS FMN binding domain showing negatively charged residues in green solids on a red ribbon marking the main chain. FMN is shown in solid yellow. The figure was constructed using information in PDB entries 1AMO and 1TLL (22,24). b, surface of eNOS oxygenase domain showing conserved positive charge array (dark blue) surrounding heme (red). The figure was constructed using information in PDB entries 3NOS and 1D0O (39,40). Oxygenase domain at the left has binding determinants for Zn ϩ2 (mouse iNOS Cys-104 and -109) and heme (mouse iNOS axial Cys-194) indicated; additional sites are included but not marked. These include other residues involved in heme binding and binding sites for pterin and substrate arginine. FMN binding domain at the right includes all three regions involved in FMN binding (mouse iNOS residues 540, 587, and 625); cNOS constructs include autoinhibitory insertion within this domain, shown as a white diamond. All three constructs include a CaM binding site between the domains (filled black diamond). SwissProt accession numbers are P29474 (human eNOS), P29476 (rat nNOS), and P2477 (mouse iNOS). 947 for rat nNOS, 683 for mouse iNOS, and 715 for human eNOS. These constructs include the oxygenase domain with the N-terminal Zn ϩ2 binding site for all three cases, and the entire FMN binding domain with five additional associated C-terminal residues beyond the final ␣ helix. This short extension terminates in a pair of conserved acidic residues in cNOSs. All three constructs contain an unmodified and functional CaM binding site. eNOS and nNOS constructs contain the autoinhibitory control element but not the C-terminal tail or the subdomain small insertion hairpin. The schematic shows the location of some major structural features. The aligned sequences of the N-and C-terminal ends of the constructs are shown below to make the exact termination points clear; the entire sequence alignment was not shown to save space.
A Clontech high fidelity PCR kit was employed for cloning of the NOS oxyFMN construct. Using rat nNOS, murine iNOS, and human eNOS cDNA as a template, PCR was used to amplify cDNA encoding residues 291-947 (for rat nNOS), 66 -683 (for murine iNOS), and 67-715 (for human eNOS) and to add a sequence coding for a His 6 tag at the N-terminal end of the protein. PCR products were subcloned into the Escherichia coli expression vector pCWoriϩ (27,29). ⌬66 human eNOSox (residues 67-500, the oxygenase domain) was also made by PCR amplification followed by cloning in pCWoriϩ.
Recombinant Protein Production-The pCWoriϩ vector expressing human eNOS and rat nNOS oxyFMN was transformed into proteasedeficient E. coli strain BL21DE3 (Novagen) as described for the oxygenase domain (27,28). Cells were grown in Terrific Broth (Sigma) containing 125 g/ml ampicillin. Cultures were shaken at 250 rpm at 37°C until an A of 1.0 at 600 nm was reached. Protein expression was induced by adding 0.5 mM isopropyl-␤-D-thiogalactopyranoside plus 0.4 mM of the heme precursor ␦-aminolevulinic acid, and shaking was continued for 42 h at 22°C (27). Production of murine iNOS oxyFMN followed a similar procedure, except CaM coexpression was required, and therefore, 40 g/ml chloramphenicol was added to the culture medium in addition to ampicillin as described for holoiNOS expression (30).
Recombinant Protein Purification-Cells were harvested by centrifugation at 5000 ϫ g for 20 min at 4°C and kept at Ϫ80°C for a day or two. The frozen cell pellet was resuspended in a minimal volume of lysis buffer (40 mM EPPS for murine iNOS, pH 7.6, containing 2 mg/ml lysozyme, 10% glycerol, 150 mM NaCl, 0.5 mM L-Arg 10 M H 4 B, 2 M FMN, and 1 mM phenylmethylsulfonyl fluoride). For nNOS and eNOS Bis-tris propane was used instead of EPPS for lysis buffer with an otherwise identical composition. Cells were broken by sonication for three 20-s pulses with a 2-min rest on ice between pulses followed by 3 cycles of freezing and thawing as described for NOS holoenzyme (15,30,31) Cell debris was removed by centrifugation at 30,000 ϫ g for 30 min at 4°C. The supernatant was applied to a column containing Ni ϩ2 -nitrilotriacetic acid-Sepharose 4B metal chelate affinity resin (Novagen) preequilibrated with either 40 mM EPPS or Bis-tris propane, pH 7.6, containing 150 mM NaCl, 5% glycerol, and 0.5 mM L-Arg (buffer A). The column was washed with 5 bed volumes of buffer A and then with buffer A plus 25 mM imidazole. Bound protein was eluted by buffer A containing 150 -300 mM imidazole. Heme-containing fractions were pooled and concentrated by using centriprep-30 (Millipore).
The concentrated protein was dialyzed at 4°C against three changes of buffer A containing 4 M H 4 B and either 1 mM dithiothreitol or 2 mM 2-mercaptoethanol. eNOS and nNOS oxyFMN were purified further by using a column containing calmodulin-Sepharose (Amersham Biosciences) equilibrated with 40 mM Bis-tris buffer, pH 7.6, containing 1 mM L-Arg, 3 mM dithiothreitol, 4 M H 4 B, 5% glycerol, and 150 mM NaCl (buffer B), and 2 mM CaCl. The column was washed with buffer B plus 0.5 M NaCl to eliminate nonspecific binding, and nNOS and eNOS were eluted with buffer B containing 4 mM EDTA. The heme-containing fractions were pooled, concentrated, and dialyzed at 4°C against buffer B containing 0.5 mM L-Arg, 1 mM dithiothreitol, 4 M H 4 B, 2 M FMN, and 5% glycerol and stored in aliquots at Ϫ80°C until further analysis. Murine iNOS oxyFMN was further purified through a Superdex 200-HR (Amersham Biosciences) gel filtration column. For purification of NOS oxygenase domain, one-step Ni ϩ2 affinity purification was done as above except both cell lysis and purification were done in the absence of FMN (27).
NOS Activity-Both oxyFMN and oxygenase domain activity were measured by following H 2 O 2 -supported oxidation of N-hydroxy-L-Arg (L-NOHA) at 37°C for 10 min (32,33). The reaction was carried out in 100 l of total volume containing 40 mM EPPS, pH 7.6, 150 -500 nM oxyFMN or oxygenase domain protein, 1 mM L-NOHA, 0.5 mM dithiothreitol, 30 mM H 2 O 2 , 10 units/ml superoxide dismutase, 0.5 mg/ml bovine serum albumin, and variable concentrations of H 4 B (0 -1000 M). Reactions were initiated by 30 mM H 2 O 2 and stopped by the addition of 1500 units of catalase. Nitrite production was measured by the Griess assay as described previously (27,32,33).
Simulations of redox titrations were described in Gao et al. (15); here only terms for heme and FMN were retained since the constructs lack an FAD binding domain. Briefly, the Nernst equation describing the redox equilibrium of an electron carrier is The redox states for FMN are Titrations were simulated using summed contributions from two heme states and three flavin states. At any wavelength these contributions are of the form, MAY 19, 2006 • VOLUME 281 • NUMBER 20

Characterization of NOS OxyFMN Construct; Electron Output State
where R is the gas constant, F is the Faraday constant, T is the Kelvin temperature, and RT/F is ϳ59.6 mV at 25°C. Variables such as ⑀ FMN represent extinction coefficients for hemes and FMN redox states at specific wavelengths; A values are the contributions of individual components to the optical density. Variables of the form C heme represent the total concentration of all heme (or, equivalently, FMN) states, including redox, spin, and conformational states, in rapid equilibrium on the time scale of redox equilibration in the experiment. Eh is the standard potential relative to the hydrogen electrode, and midpoint potentials of one electron couples are specified as Emheme, E1FMN, and E2FMN. E1FMN and E2FMN are the midpoint potentials for the first and second one-electron reductions of FMN. In some cases multiple heme components differing in concentration, extinction coefficient, and midpoint potential have been introduced for simulation purposes.
Stoichiometric titrations were carried out in the same vessel as the potentiometric titrations. The apparatus was not intended to measure the reducing equivalents accurately, but by making small additions of buffered dithionite while monitoring the flavin absorbance at 455 nm, it was possible to observe intermediate reduction states without using a mediator system.
Spectroscopy-Binding of L-Arg, H 4 B, and imidazole was monitored by UV-visible spectral perturbation analysis. Light absorbance spectra of the purified NOS oxyFMN proteins were obtained with a Hitachi U2010 spectrometer equipped with computer-assisted data collection software (UV solutions, Wellesley Hills, MA). The NOS-ferrous-CO adduct absorbing at 444 nm was used to quantitate heme protein content by using an extinction coefficient of 74 mM Ϫ1 ⅐cm Ϫ1 (35,36).
Electron paramagnetic resonance spectra were carried out using a Bruker spectrometer with a flowing helium cryostat essentially as previously described (36). Samples were frozen quickly in cold isopentane or liquid nitrogen and kept on liquid nitrogen until the spectra were recorded.
Flavin Content-Flavin content was determined by measuring the absorbance of the released flavin at 450 nm using extinction coefficients of 17 mM Ϫ1 cm Ϫ1 for FMN. Twenty M enzyme in 200 l of 40 mM Tris-HCl buffer, pH 8.0, was placed in a boiling water bath for 10 min, chilled on ice, and centrifuged at 14,000 rpm for 20 min, and the supernatant was used for FMN quantitation using Sigma FMN as standard. The flavin content for all the NOS isoforms mentioned in the results section is an average of the values measured.

RESULTS
The optical spectra of three oxyFMN constructs are shown in Fig. 3, A-C. The heme is primarily in the high spin state after treatment with H 4 B and arginine during preparation as indicated by the Soret maximum near 400 nm. Features of the oxidized FMN optical spectrum are visible at 450 and 480 nm. The FMN appears to be more oxidized in these constructs than in freshly isolated holoenzyme, probably because the enzyme lacks an NADPH binding site. A band visible near 650 nm is due to the high spin ferriheme charge transfer band. The broad FMN blue neutral band, extending from 550 to 650 nm, that is characteristic of all three mammalian NOS holoenzymes, is very weak or absent. The spectra of corresponding constructs of the nNOS and iNOS are very similar, but in the eNOS oxyFMN construct a significant minority of the ferriheme remains in the low spin state in the presence of arginine. The top trace in Fig. 3D shows electron paramagnetic resonance spectra of nNOS oxyFMN construct replete with arginine and H 4 B. The high spin features at g ϭ 7.7, 4.1, and 1.76 are very similar to the features of the majority species of the arginine and H 4 B replete holoenzyme. There are very low levels of low spin heme (g ϭ 2.4 and 2.2), and almost all the heme is in a single high spin state. In particular, there is almost no g ϭ 6 axial ferriheme signal from heme sites that have lost their thiolate ligand. A small amount of adventitious rhombic ferric (g ϭ 4.3) iron is overlapped with the g ϭ 4.1 feature of the heme. There is no detectable FMNH ⅐ radical in the EPR spectra of the construct as isolated, consistent with the optical spectra. A small amount of copper contaminant is visible just above g ϭ 2.
The second trace in Fig. 3D shows a similar spectrum of iNOS oxy-FMN construct. Small differences in the g values of the heme spectra are not detectable in this wide scan (but see Fig. 9). The lower trace shows the corresponding spectrum of a nNOS oxyFMN preparation that has been passed through a cycle of reduction with a small amount of dithionite and air oxidized, producing an air stable FMN semiquinone, visible at g ϭ 2, and regenerating the high spin ferriheme. UV-visible spectra of a corresponding experiment are shown in Fig. 6.
All preparations of the eNOS, nNOS, and iNOS oxyFMN constructs contained between 0.9 and 1 heme/mol based on the CO difference spectra (not shown but indistinguishable from the CO difference spectra of holoenzyme and oxygenase domains); all of the heme content measured was in the native thiolate heme complex absorbing near 445 nm, with no detectable 420-nm band. The FMN content measured after extraction from the protein was in all cases at least 90% of the heme content, a difference that could result from loss of FMN during the process. The iNOS construct can be purified in higher yield, probably because it lacks an autoinhibitory element and is less likely to be destroyed by proteolysis. The high level of incorporation of the two major prosthetic groups suggests proper folding during expression.
Because there is no NADPH binding site, none of the oxyFMN constructs exhibit either NO production with NADPH as electron donor or any of the common NADPH reductase activities of NOS holoenzymes. However, oxyFMN constructs of all three isoforms are capable of NO synthesis from N-hydroxyarginine (L-NOHA) with peroxide as the electron donor. Fig. 4 shows the ability of all three NOS oxyFMN constructs and their oxygenase domain counterparts to catalyze the H 2 O 2 -supported nitrite formation from L-NOHA as a function of H 4 B concentration (0 -1000 M). The activity of both murine and human eNOS and iNOS oxyFMN is similar to their oxygenase domain activities, whereas nNOS oxyFMN showed increased activity compared with that of its oxygenase counterpart. The dependence of NO formation on H 4 B concentration is substantially the same for oxyFMN and oxygenase domain preparations. H 4 B is not a reactant, since it is substoichiometric with NO production, but the requirement for full activity is greater than expected from the K d , which is very low. H 2 O 2 -supported nitrite formation from L-NOHA by NOS oxyFMN also depends on H 4 B concentration at similar H 4 B levels, as reported earlier for the oxygenase domain and holoprotein (32,33). The requirement for H 4 B has not been elucidated but appears to be related to the use of peroxide as reductant. The activity measured by this method reflects the full functionality of the oxygenase domain catalytic site after expression.
Purified NOS oxyFMN derived from all three isoforms migrates as a dimer during gel filtration analysis (data not shown) without significant aggregation or monomer formation. The 280-nm/Soret absorbance ratio is ϳ2.2-2.5 in 80 -90% pure preparations and serves as a routine purification index. The iNOS construct migrates as a single majority band at 69 kDa on SDS-PAGE, with a minority band at 50 kDa due to a cleavage product and a 15.7-kDa band from co-expressed CaM. The eNOS and nNOS oxyFMN preparations have additional minor bands at 50 -55 kDa (Fig. 4, inset). The estimated molecular mass for nNOS oxyFMN is ϳ75 kDa. Fig. 5a shows a comparison between spectra of oxidized iNOS holoenzyme, oxygenase domains, and oxyFMN constructs in the presence of H 4 B and arginine. All three constructs are primarily high spin with a Soret maximum of ϳ395 nm and a well developed 650-nm band. The oxygenase domain spectrum lacks flavin features at 450 and 480 nm, whereas the holoenzyme spectrum has approximately twice the contribution from flavin relative to heme of the oxyFMN spectrum. Fig. 5b shows the effect of arginine addition on the spectrum of iNOS oxyFMN construct. Enzyme isolated in the presence of arginine diluted with arginine-free buffer is in a mixed spin state as indicated by the Soret shoulder at 420 nm; omission of arginine during purification produces a low quality preparation. Treatment with imidazole produces a characteristic low spin spectrum with a sharp Soret band at 425 nm. Dialysis with a change of buffers removes both imidazole and any remaining traces of arginine, which is displaced by imidazole binding, producing a mixed spin preparation with a Soret band at maximum near 400 nm and a pronounced shoulder at 420 nm. The addition of excess arginine produces a completely high spin preparation with a broad Soret maximum below 400 nm lacking the 420-nm shoulder. The inset shows the difference spectrum of arginine binding to a dialyzed preparation in the presence of H 4 B.
As judged by the effect of arginine on the heme Soret band, all three constructs bind H 4 B and arginine. Arginine-and H 4 B-replete iNOS and nNOS preparations are essentially 100% high spin by optical criteria, consistent with a high level of intactness of the cofactor and substrate binding sites and with accessibility of all sites to arginine and H 4 B; a low spin minority species persists in eNOS preparations. The dissociation constant of L-arginine (K d ) for all three NOS oxyFMN constructs, as In addition, all these spectra showed typical flavin (FMN) peaks at 450 and 480 nm and a characteristic high spin ferriheme charge transfer band at 650 nm. Panel D (top trace) shows the electron paramagnetic resonance spectrum of nNOS oxyFMN construct at a concentration of 45 M recorded at 10 K. Instrument settings were: microwave power, 2 milliwatts; modulation amplitude, 10 gauss; modulation frequency, 100 kHz; microwave frequency, 9.47 GHz; magnetic field scan, 0.4 millitorr; field center, 0.23 millitorr. The spectrum is dominated by a single high spin species with features at g ϭ 7.7, 4.0 and 1.76. Axial high spin heme at g ϭ 6 and low spin heme at g ϭ 2.4 are almost undetectable, and there is very little free radical at g ϭ 2. The conditions for the lower two traces are the same except that the temperature is 15 K, and the gain has been increased. The middle trace shows a spectrum of oxidized iNOS oxyFMN construct; the sample concentration was ϳ15 M. The lower trace shows a second spectrum of nNOS oxyFMN complex after reduction with dithionite and partial reoxidation by exposure to air; the heme is only about 50% in the ferric state. An FMN radical is visible at g ϭ 2. MAY 19, 2006 • VOLUME 281 • NUMBER 20 determined by spectral perturbation analysis in presence of imidazole (progressive titration spectra is not shown, but titration spectra reflect low to high spin transitions as in the difference spectrum shown in the inset of Fig. 5b), indicates a high affinity site (K d ϭ 0.5-1.0 M), confirming proper folding and dimerization of the oxygenase domain in these constructs. Fig. 6 shows spectra of the iNOS oxyFMN construct recorded during stoichiometric titration with small aliquots of sodium dithionite solution. After stepwise addition of enough reductant to bleach the FMN bands at 455 and 480 nm, UV-visible spectra were recorded, and additional dithionite was added until FMN and heme were fully reduced, as judged by the spectra. The first electron into the construct reduces FMN to the blue neutral semiquinone state with loss of the oxidized FMN bands at 455 and 480 nm; at the same time, the weaker semiquinone bands can be observed near 560 and 600 nm. An additional band at 480 nm is difficult to resolve because it is overlapped by the stronger oxidized FMN band at 480 nm, but the broader semiquinone absorbance contributes additionally near 520 nm.

Characterization of NOS OxyFMN Construct; Electron Output State
Further addition of reductant causes parallel bleaching of the semiquinone bands and conversion of the heme to the ferrous state. Heme reduction is visible as a red shift in the Soret and as loss of the high spin ferric charge transfer band at 650 nm; the latter is overlapped by the semiquinone bands. No red anionic semiquinone was detected. An additional cycle of oxidation and reduction produced spectra nearly identical to the original oxidized and reduced spectra, indicating that the construct was stable during the experiment. This method can be used to generate flavin semiqui-none, whereas the heme is completely oxidized, but heme reduction and reduction of FMN to FMNH 2 cannot be separated in the iNOS construct, which has calmodulin tightly bound. This is probably due to equilibration of heme and FMN faster than the time scale of the experiment (about 1 min). The stability of the construct to cycles of oxidation/reduction is critical to the design of kinetics experiments to demonstrate rapid heme/FMN electron transfer. Fig. 7 shows selected UV-visible spectra of iNOS FMN domain during a potentiometric titration. The midpoint potential for heme reduction is about Ϫ180 mV, slightly higher than observed in titrations of the holoenzyme or the independently expressed oxygenase domain. This can be readily observed in the Soret region; the peak position shifts from 390 nm to about 407 nm, and the band sharpens as the heme is reduced to the ferrous state. The charge transfer band at 650 nm disappears as the heme is reduced, and the ferrous ␣ band region increases in intensity. Very similar results were obtained for the eNOS and nNOS constructs.
The oxidized FMN bands at 450 and 480 nm decrease in intensity as FMN is reduced to the semiquinone state. The blue neutral semiquinone is again most easily observed as a group of broad bands centered near 560, 590 and 620 nm and, in addition, by increased absorbance at 520 nm. The inset shows an expanded view of the titration of the 600 -650-nm region, showing initially increasing intensity at 620 nm followed by loss in intensity at 620 and 650 as the potential is lowered further. The initial reduction of FMN occurs at approximately ϩ70 mV; semiquinone reduction is essentially complete by Ϫ260 mV. Fig. 8a shows a plot of the absorbance at 620 and 650 nm as a function  Fig. 8b. The major contributions to the changes in absorbance at 455 nm are the heme Fe ϩ2 /Fe ϩ3 couple and the FMN/ FMNH ⅐ couple. The data can be fit well with the potentials used to fit the data of Fig. 8a, but a slightly better fit was obtained with an E 1FMN of ϩ50 mV. This difference is not significant. Because FMNH ⅐ does not contribute significantly here, the FMNH ⅐ /FMNH 2 couple cannot be accurately determined at this wavelength, but it must be significantly lower than the FMN/ FMNH ⅐ couple.
The major contributions to the absorbance changes at 520 nm are the transient formation of semiquinone and the slightly greater contribution of ferric heme. The data have been fit with the same potentials as in Fig. 8a, except that the FMN/FMNH ⅐ couple is again better fit at ϩ50 mV. Similar results were obtained in titrations of nNOS (not shown).
Spectra taken during titration of eNOS and nNOS constructs are included in supplementary information. For eNOS and nNOS, calmodulin effects were examined by adding calmodulin and taking spectra at a few selected potentials. The results for both isoforms were consistent with a shift of about 20 mV to higher potential for both the heme and FMNH ⅐ /FMNH 2 couples, an effect at the edge of resolution in our experiments. An effect large enough to be responsible for the activation  FAD and FMN). The spectrum of holoprotein has been displaced upwards to facilitate comparison of the relative contribution of flavin and heme bands. b, UV-visible spectra of iNOS oxyFMN construct showing ligand-induced spin state changes. As indicated in legend, spectra shown are of iNOS oxyFMN exposed to arginine during purification but after chromatography and dilution (admixed spin with 420 nm shoulder), after treatment with imidazole to replace arginine with a relatively weak binding ligand (sharp low spin peak at 425 nm), after removal of imidazole and trace arginine by dialysis with two changes of buffer (admixed spin with shoulder at 420 nm), and after the addition of saturating (1 mM) arginine (broad high spin peak at 390 nm).
of electron transfer is ruled out, consistent with previous results with holoenzyme (15).
The results for titrations of iNOS, eNOS, and nNOS are summarized in Table 1. The potentials are generally similar, although the details are slightly different. In general, the first one-electron reduction of the FMN occurs at high potential as in holoenzyme and in reductase constructs. The second single electron reduction of FMN and heme reduction are nearly isopotential. The eNOS construct contains a significant minority species that is low spin in the presence of arginine. Fig. 9 shows effects of interdomain interactions on EPR signals of the high spin ferriheme. The three top traces are the g max features of nNOS constructs; the asymmetry is due to the details of strain broadening. The central peak is the g max signal from a nNOS oxygenase preparation replete with arginine and H 4 B; g max ϳ 7.72. The left peak is the corresponding signal from a nNOS oxyFMN preparation. g max for this peak is ϳ7.72. Although the peak position of the strain broadened line is only slightly shifted to high field relative to the oxygenase construct, the line is significantly sharpened, and the low field edge is noticeably shifted to the left (see "Discussion"). The right hand peak is the g max signal from the same nNOS oxyFMN preparation in the presence of Ca ϩ2 /CaM. The peak has shifted downfield to g ϳ 7.69, and the low field side of the line is broader. We attribute this shift in peak position and line shape to the effect of calmodulin-induced association of the FMN binding domain on the ligand geometry of the heme. This is the first report of a calmodulin effect on the spectrum of heme in any NOS construct.
The two lower spectra show the g max features of iNOS oxyFMN FMN has been reduced to the semiquinone state, and the blue neutral species is visible at 600 and 520. Note that there is significant absorbance at 550 -560 from blue neutral semiquinone. C, fully reduced with additional dithionite; note the loss of flavin bands and reduced Soret position. D, reoxidized with potassium ferricyanide; spectrum is nearly identical to A. E, re-reduced with excess dithionite; spectrum is almost identical to C.

of electron carriers in oxyFMN constructs
Potentials are accurate to within 20 mV for eNOS and nNOS and 10 mV for iNOS; heme potential is from fits to a single component. sq/hq, semiquinone/hydroquinone couple; ox/sq, oxidized/semiquinone couple.

FMN ox/sq FMN sq/hq Heme
construct. Because the iNOS construct must be coexpressed with a calmodulin, we couldn't perform the analogous experiment to demonstrate the change in the heme environment triggered by CaM binding. The solid trace at g max ϳ 7.72 is very close in peak position to the nNOS oxyFMN construct position after the addition of CaM. The dashed line shows the g max feature of the same preparation after the addition of 2 l of saturated dithionite solution to the 250-l sample. This generated a FMNH ⅐ radical in about 1 ⁄3 of the enzyme molecules. The high field side of the g max line is broadened in this sample. To produce an effect of this magnitude in a sample with only partial conversion to the radical state, a magnetic interaction of 5-10 gauss would be required. The spectrum of the flavin radical at g ϭ 2 is also broadened; we hope to present an analysis of the magnetic interactions in a subsequent communication.

DISCUSSION
The results presented here demonstrate that two-domain oxyFMN constructs of all three mammalian NOS isoforms can be expressed in E. coli and are well folded and bind functional prosthetic groups. The oxyFMN construct is designed to be the "output state" counterpart of the three-domain reductase construct and as such should be of value in probing the mechanism of the enzyme. The purpose of this study was to establish that such constructs could be expressed and purified in qualities and quantities suitable for detailed structural and functional studies. In addition, characterization of the constructs was undertaken to determine some critical parameters for internal electron transfer reactions.
It is clear from the spectroscopic and enzymology results that the oxygenase domain is stable and functional. By spectroscopic criteria, preparations of all three constructs are superior to most holoenzyme preparations in that they are homogeneous, bind substrate and cofac-tors stoichiometrically, and do not contain significant amounts of denatured material.
The titration results are of special interest. The FMN semiquinones are stable and are the blue neutral forms as in the holoenzymes. The midpoint potentials of the heme and the FMNH ⅐ /FMNH 2 couple appear, however, to be slightly above Ϫ200 mV in all three cases. This is 60 mV or more greater than the potentials obtained using individual domains and holoenzyme preparations, a modest but significant difference corresponding to a 1 order of magnitude change in equilibrium constants for electron transfer reactions. For example, in holoenzymes, typical values for arginine-replete heme and FMN are Ϫ240 mV for FeII/FeIII, 0 mV for FMN/FMNH ⅐ , and Ϫ250 mV for FMNH ⅐ /FMNH 2 (15). The FMN couples were reported by Noble et al. (17) to be about Ϫ50 and Ϫ275 mV in isolated reductase constructs.
In summary, as in holoenzyme the FMNH ⅐ /FMNH 2 and heme couples are an approximately isopotential group, whereas the FMN/ FMNH ⅐ couple is too high in potential to be an effective heme reductant in steady state experiments. This need not rule out the possibility of generating a transient one-electron reduced FMN species that would effectively reduce the heme by rapid electrochemical or photochemical methods.
Observation of higher potentials in a construct in which the electron carriers may be more exposed to solvent should not be surprising. The heme in these constructs, however, has a potential higher than that reported for individually expressed oxygenase domains (Ϫ220 mV in the presence of arginine) (38). This appears to be a special property of the construct and may reflect interactions between the domains. In nNOS and iNOS constructs, arginine generates nearly complete high spin ferriheme; small amounts of low spin sometimes persist. As Fig. 3 indicates, the eNOS construct is less homogeneous spectrally and contains a significant low spin component corresponding to about 30% of the total heme in the presence of arginine.
Some differences in opinion exist concerning the effects of arginine. Presta et al. (38) report little or no effects of arginine on apparent midpoint potentials of iNOS oxygenase domain constructs. We found that arginine shifted the midpoint potential of heme in iNOS and nNOS holoenzyme by at least 60 mV; preparations free of H 4 B did not titrate as a single component, suggesting a range of conformations and/or dimerization states with different properties (15,41).
It is particularly interesting to note that the potentials of both the heme and the FMNH ⅐ /FMNH 2 couple in the iNOS and nNOS constructs, which are the most homogeneous and native of the three expressed so far and are both slightly higher than the potentials of these carriers in holoenzymes but remain approximately isopotential. This suggests the possibility that the ϳ60-mV shift we observe reflects the effects of FMN/oxygenase domain interactions and that these interactions developed to retain the reversibility of the FMN/heme electron transfer in the absence of a reaction (e.g. NO production) pulling the equilibrium toward the oxygenase site. Potential differences are not strong evidence of FMN-heme interaction, but interactions between the two domains are a possible explanation for small differences in the observed potentials.
In contrast, the spectral evidence for interactions between the domains is very strong. The calmodulin-driven shifts in the EPR peak position and line shape of the nNOS oxyFMN complex shown in Fig. 9 indicate changes in the ratio of the rhombic and axial zero field splitting parameters E and D. As E/D increases, gx and gy split about 6; gz and (gx ϩ gy)/2 decrease as a second-order effect. High spin ferrihemes are well described by the S ϭ 5/2 spin Hamiltonian D{Sz 2 Ϫ 1 ⁄ 3[S(S ϩ 1)]} ϩ E(Sx 2 Ϫ Sy 2 ), where E and D reflect differential admix- ture of low-lying states into the ground state 6 A 1 sextet by spin-orbit coupling because of tetragonal and rhombic splitting of the t 2g d orbitals (see for example Ref. 37).
The strain-broadened line shape is produced by distributions of E/D centered around 0.083 that inherently produce a sharper high field side to the line because the dependence of the field position on E/D becomes weaker as E/D increases. All NOS EPR spectra observed so far result from E/D ratios in the range 0.058 -0.088. In this case E/D is probably close to the upper limit which can be obtained with thiolate ligation and the underlying molecular geometry.
It is not surprising that CaM-driven association of the FMN and oxygenase domains affects the distribution of E/D at the heme iron. Interaction of the surface charge distributions adjacent to the prosthetic groups is the likely source of the observed small distortion of the heme environment. CaM does not produce a comparable effect in holoenzyme, probably because the fraction of molecules in which the FMN domain associates with the heme site is too small.
The broadening of the ferriheme g max line produced by formation of FMN radical in iNOS oxyFMN is less obvious than the broadening of the radical because the heme line width at g max is about twice that of the radical line width and because only about 1 ⁄ 3 of the heme interacts with radical, whereas all or nearly all of the radical interacts with heme. Interactions of 5-10 gauss are likely to be due primarily to dipole-dipole coupling (e.g. Ref. 42). The interactions acquire anisotropy from both the heme g tensor and the (1-3cos 2 ⍜) dependence of dipolar coupling, where ⍜ is the angle between the magnetic field and the interspin vector; such interactions have an r Ϫ3 distance dependence. We hope eventually to obtain detailed distance and orientation information from simulation of spectra, but it is clear that the heme-FMN distance is in the 10 -15-Å range, favorable for electron transfer. This close association would be impossible for a holoenzyme conformation analogous to the reported crystal structure of nNOS reductase complex.
In isolated domains the potentials of the flavin couples have been reported to be somewhat lower than our results in holoenzymes. Reductase domain and holoenzyme FMN/FMNH ⅐ couples are high potential and similar to these couples in oxyFMN constructs (17), but in isolated FMN domains the FMN/FMNH ⅐ couples are significantly lower (Ϫ180 to Ϫ190 mV) in potential (18). The FMNH ⅐ /FMNH 2 couples are also lower in isolated FMN binding domains (Ϫ300 to Ϫ320 mV), but this effect is much more modest. This couple in holoenzymes is about midway between the values obtained in isolated domains and those obtained here in oxyFMN complexes. These effects may be coincidental, but the moderate shifts in potential are those expected if some redox energy is exchanged to drive the association and dissociation of domains. For example, if reduced FMN destabilized the input state, the FMNH ⅐ /FMNH 2 couple would be lower in potential in the reductase complex. If reduced FMN promotes formation of the output state and oxidized FMN promotes its dissociation, the FMNH ⅐ /FMNH 2 couple would be higher in potential in the oxyFMN complex.
An analogous FMN domain/oxygenase construct has been studied in P450-BM3, but the properties of that system are very different (43). Most obviously, the oxygenase domains of P450 and NOS are not even distantly homologous, and BM3 lacks a calmodulin binding site. The interdomain linker in the BM3 construct is sensitive to proteolysis and has been degraded in the crystal structures presented (43). In addition, the mechanism of the BM3 system is very different than that of NOS or P450 reductase in that the FMN binding domain of the BM3 system has a low potential FMNH ⅐ /FMN couple, and the semiquinone is the physiological reductant. The FMN binding domain in the BMN3 system is a flavodoxin homolog, but it has a very different interdomain interaction surface (43), lacking the flavodoxin-derived conserved clusters of negative charge found in P450 reductase (22) and NOS (24,44). It will be of interest to compare the structures and mechanisms of the output states of the NOS and BM3 systems.
In conclusion, the oxyFMN constructs of all three mammalian NOS isoforms have been expressed as stable proteins with activity and interesting properties. We have observed the first direct effects of CaM on NOS heme spectra and the first indications of FMN-heme magnetic coupling. They appear to be good candidates for further kinetic and structural studies of the NOS output state.