Insights into Flavin-based Electron Bifurcation via the NADH-dependent Reduced Ferredoxin:NADP Oxidoreductase Structure*

Background: Flavin-based electron bifurcation is a vital process in microbial energy metabolism. Results: The NfnAB complex structure determines the positions of the prosthetic groups and the substrates. Conclusion: The environment of the central FAD and its distance to the next redox centers of the two electron routes control electron bifurcation. Significance: The first complete structure of a flavin-based electron bifurcating enzyme provides insights into this ancient catalytic process. NADH-dependent reduced ferredoxin:NADP oxidoreductase (NfnAB) is found in the cytoplasm of various anaerobic bacteria and archaea. The enzyme reversibly catalyzes the endergonic reduction of ferredoxin with NADPH driven by the exergonic transhydrogenation from NADPH onto NAD+. Coupling is most probably accomplished via the mechanism of flavin-based electron bifurcation. To understand this process on a structural basis, we heterologously produced the NfnAB complex of Thermotoga maritima in Escherichia coli, provided kinetic evidence for its bifurcating behavior, and determined its x-ray structure in the absence and presence of NADH. The structure of NfnAB reveals an electron transfer route including the FAD (a-FAD), the [2Fe-2S] cluster of NfnA and the FAD (b-FAD), and the two [4Fe-4S] clusters of NfnB. Ferredoxin is presumably docked onto NfnB close to the [4Fe-4S] cluster distal to b-FAD. NAD(H) binds to a-FAD and NADP(H) consequently to b-FAD, which is positioned in the center of the NfnAB complex and the site of electron bifurcation. Arg187 is hydrogen-bonded to N5 and O4 of the bifurcating b-FAD and might play a key role in adjusting a low redox potential of the FADH•/FAD pair required for ferredoxin reduction. A mechanism of FAD-coupled electron bifurcation by NfnAB is proposed.

In 2008, a novel mechanism of energy coupling was discovered, namely flavin-based electron bifurcation that changed our views of the energy metabolism of many anaerobic microorganisms, both bacteria and archaea (1-3). The first example was the coupling of the endergonic reduction of ferredoxin (Fd ox ) with NADH to the exergonic reduction of crotonyl-CoA with NADH via the cytoplasmic butyryl-CoA dehydrogenase/electron transfer flavoprotein (Bcd/EtfAB) 6 complex from Clostridium kluyveri and other butyric acid forming Gram-positive bacteria (4). The Bcd/EtfAB complex contains per subunit one FAD and no other cofactors. The crystal structure of the EtfAB subcomplex was recently solved, and a mechanism of flavin-based electron bifurcation was proposed (5). The proposal is based on the property of free flavins (F) to have three redox potentials: E o Ј (F/FH Ϫ ) ϭ Ϫ219 mV; E o Ј (FH ⅐ /FH Ϫ ) ϭ Ϫ124 mV; and E o Ј (FH ⅐ /F Ϫ ) ϭ Ϫ314 mV (6). Upon downhill one-electron transfer from the fully reduced bifurcating FAD bound to EtfB to the oxidized FAD bound to EtfA, a flavin radical FADH ⅐ at EtfB is generated that has a redox potential sufficiently negative to drive the reduction of Fd ox (E o Ј ϭ Ϫ400 mV).
As mentioned above, the Bcd/EtfAB complex contains only FAD as prosthetic group, and therefore only FAD can be the site of electron bifurcation. All other electron-bifurcating complexes analyzed to date also contain, in addition to FAD or FMN, iron-sulfur clusters that, however, can accept and donate electrons only one at a time and are therefore not likely sites of electron bifurcation. In the case of electron-bifurcating hydrogenases and formate dehydrogenases, the site of H 2 activation ([FeFe] center) and that of formate activation (molybdopterin center) function as reversible 2e/1e switches and could therefore principally also be sites of electron bifurcation (21).
The electron-bifurcating enzyme complexes mentioned above are all composed of at least three different subunits with one exception: the electron-bifurcating transhydrogenase from C. kluyveri and M. thermoacetica is only a heterodimer of subunit NfnA (ϳ32 kDa) and subunit NfnB (ϳ50 kDa). The complex contains two FAD, one [2Fe-2S] cluster, and two [4Fe-4S] clusters. NfnA shows sequence similarities to plant ferredoxin: NADP ϩ reductase (Fnr) and bacterial dihydroorotate dehydrogenase (Dodh) from bacteria and NfnB to the ␤-subunit of bacterial NADPH-dependent glutamate synthase (Gls) (22) and eukaryotic dihydropyrimidine dehydrogenase (Dpdh) (23). The transhydrogenase complex catalyzes the reversible reduction of ferredoxin and NAD ϩ with 2 NADPH (Reaction 2). In vivo, its function is rather to catalyze the back reaction, the NADH-dependent ferredoxin:NADP reductase reaction (19,20). In both directions, the bifurcation process is based on the low NADH/ NAD ϩ ratio (1/30) and the high NADPH/NADP ϩ ratio (50/1) in the cell (24).

Reaction 2
NfnAB appears to be an ideal model system to explore the mechanism of flavin-based electron bifurcation because it has the simplest structure. Therefore, we overproduced the NfnAB complex from T. maritima and characterized it biochemically, kinetically, and structurally to acquire insights in its mode of action. NfnAB is the first flavin-based electron-bifurcating enzyme whose complete x-ray structure is characterized.
Heterologous Expression of nfnAB Genes and Purification of NfnAB-The nfnAB genes from T. maritima genomic DNA were amplified with the forward primer 5Ј-GACGACGACAA-GATGGGGGGGACGGCTTTG-3Ј and the reverse primer 5Ј-GAGGAGAAGCCCGGTCACTTTTGCCACGGATTCC-ATTC-3Ј (the inserted sequences specific for ligation-independent cloning are underlined). After purification with MinElute PCR purification kit, the blunt PCR product was treated with T4 DNA polymerase in the presence of dATP to generate specific vector-compatible overhangs. Then it was annealed into the linear pET-51b(ϩ)Ek/LIC vector (Merck) and thereby 5Ј-tagged with a Strep cassette. The vector was transformed into Escherichia coli C41 (DE3), which already harbored pCodonPlus and pRKISC vectors successfully used for the production of iron-sulfur cluster proteins (28,29).
Before inoculation, the medium was supplemented with carbenicillin (50 mg liter Ϫ1 ), chloramphenicol (25 mg liter Ϫ1 ), and tetracycline (10 mg liter Ϫ1 ) to maintain the plasmids and with cysteine (0.12 g liter Ϫ1 ), ferrous sulfate (0.1 g liter Ϫ1 ), ferric citrate (0.1 g liter Ϫ1 ), and ferric ammonium citrate (0.1 g liter Ϫ1 ) for the enhancement of iron-sulfur cluster synthesis. The recombinant E. coli C41 (DE3) cells were aerobically grown in 2 liters of Terrific broth (TB medium), and after nfnAB expression induced by isopropyl ␤-D-thiogalactopyranoside, the solution was continuously stirred at a speed of 250 rpm for 20 h at 37°C under aerobic conditions and further for 2 h under anaerobic conditions. After incubation for further 20 h at 4°C, the recombinant cells were harvested by centrifugation and washed with anaerobic 100 mM Tris-HCl, pH 7.5. The obtained cell pellet was stored at Ϫ80°C under N 2 .
All steps of purification were performed at room temperature under strictly anaerobic conditions in a Coy type B vinyl anaerobic chamber (Coy, Grass Lake, MI) that was filled with a mixture of N 2 and H 2 (95%:5%) and contained a palladium catalyst for continuous O 2 reduction with H 2 . All buffers were boiled to remove O 2 and then evacuated and filled with N 2 under slight overpressure. The E. coli cell pellet (10 g of wet cells) was resuspended in 100 mM Tris-HCl, pH 7.5, containing 2 mM DTT and 10 M FAD. The cells were disrupted by sonication, and the generated cell debris was removed by ultracentrifugation. The supernatant was then heated for 30 min at 80°C (growth temperature optimum of T. maritima). After removing the denatured protein by centrifugation at 10,000 ϫ g at 25°C for 10 min, the supernatant was applied to a 5-ml Strep-Tactin Superflow column (IBA, Goettingen, Germany), which was equilibrated with 20 ml of buffer A (100 mM Tris-HCl, 150 mM NaCl, 2 mM DTT, 10 M FAD, pH 7.5). Then the column was washed with 35 ml of buffer A. The recombinant protein was eluted with 15 ml of buffer A supplemented with 2.5 mM desthiobiotin. The fractions containing NfnAB activity were pooled and concentrated by ultrafiltration with 50-kDa cut-off Amicon filters. Subsequently, the enzyme was subjected to iron-sulfur cluster reconstitution in 100 mM Tris-HCl, pH 7.5, containing 8 mM DTT, 10 M FAD, 2 mM cysteine, and 1.5 mM FeSO 4 at room temperature for 1 h under strictly anaerobic conditions. After centrifugation at 52,000 ϫ g at 4°C for 30 min, the supernatant with the reconstituted enzyme was concentrated via 50-kDa cut-off Amicon filters to a volume of ϳ1 ml and then loaded on a HiPrep 16/60 Sephacryl S-200 high resolution column equilibrated with 100 mM Tris-HCl, pH 7.5, containing 500 mM NaCl, 2 mM DTT, and 10 M FAD. NfnAB activity was found in two peaks (Fig. 1A). Peak A eluted first was smaller and not always detectable. The fractions of peak B that eluted last were pooled, concentrated via 50-kDa cut-off Amicon filters, and washed with 10 mM MOPS-KOH, pH 7.0, 2 mM DTT, and 10 M FAD. The purified enzyme was then concentrated to ϳ25 mg/ml and stored at Ϫ80°C under N 2 .
Cofactor Content Determination-The flavin was identified by thin-layer chromatography on an RP-18 F254 aluminum sheet (Merck) using the supernatant of the denaturated protein.
Activity Assays-The assays were performed at 45°C in 1.5-ml anaerobic cuvettes closed with a rubber stopper and filled with 0.8-ml reaction mixtures and 0.7 ml of 100% N 2 or 100% H 2 at 1.2 ϫ 10 5 Pa as gas phase. The reaction mixtures contained 100 mM MOPS-KOH, pH 7.0, 10 mM 2-mercaptoethanol, and 12 M FAD as basal ingredients. The reactions were followed photometrically at the wavelength indicated. Measured activities (see Table 1) remained essentially constant for several days at 4°C under strictly anaerobic conditions and for more than 2 weeks when frozen at Ϫ20 or Ϫ80°C. (For estimated redox potentials under physiological conditions of the electron donors and acceptors, see "Results and Discussion").
Crystallization and Structure Determination-NfnAB was crystallized with the sitting drop method at room temperature inside an anaerobic chamber equipped with an OryxNano crystallization robot (Douglas Instruments Ltd.). Protein solution contained 17.5 mg/ml NfnAB, 10 mM MOPS-KOH, pH 7.0, 2 mM DTT, and 10 M FAD. For screening experiments, the JBScreen Pentaerythritol, JBScreen Classic 1-10 and SaltRx kits were applied. The best crystallization conditions were listed in Table 2. X-ray diffraction data were collected at Beamline PXII at the Swiss Light Source (Villigen, Switzerland) using XDS (33) for data processing. Multiple wavelength anomalous dispersion data sets were measured at the iron absorption edge at a wavelength of 1.73 Å. The iron positions were identified with SHELXD (34), and the phases were calculated with SHARP (35) and improved by solvent flattening with SOLOMON (36). Coot was used for electron density analysis and model building (37). Several secondary structure elements were clearly visible in the experimental electron density, and the related subunits of Dodh from Lactococcus lactis (38) and Gls from Azospirillum brasilense (22) were fitted into the electron density of NfnA and NfnB, respectively. Automatic model building in Phenix (39) correctly positioned ϳ50% of the residues of NfnA and NfnB into the electron density when incorporating the superimposed Dodh and Gls as reference models. The R free factor was 37%. Because refinement in space group P4 3 2 1 2 did not converge, the data were reprocessed in P4 3 , and partial twinning was introduced into refinement with Phenix. Partial twinning was also detectable with Phenix.xtriage (40). The temperature factors of several segments were higher than 100 Å 2 , and accurate fitting of the amino acids was not always possible, although the polypeptide chain could be traced except for segment C170-C182. Parameters describing the structure quality are listed in Table 2. The structure of the NfnAB-NADH complex was determined in the same crystal form after soaking the crystals with 5 mM NADH for 40 min. In addition to the highly occupied NADH in NfnA, one NfnB subunit of the asymmetric unit also contains a weakly occupied NADH. Its adenosine part is, however, highly disordered (B ϳ 100 Å 2 ). Figs. 2-5 were generated with PYMOL (Schrödinger, LLC).

Results and Discussion
Attempts to obtain a crystal structure of heterologously produced NfnAB from C. kluyveri and from M. thermoacetica, in which this enzyme was first discovered, have failed. Crystallization SEPTEMBER 4, 2015 • VOLUME 290 • NUMBER 36 of the two enzymes without a tag and tagged with His 6 or Strep cassettes were tried. We have therefore screened heterologously produced NfnAB complexes from various organisms, both bacteria (Thermoanaerobacter tengcongensis and T. maritima) and archaea (Pyrococcus furiosus) and were finally successful with NfnAB from the hyperthermophile T. maritima.

FAD-based Bifurcating NfnAB Complex
Molecular Mass and Cofactor Content of the NfnAB Complex from T. maritima-Expression of the genes from T. maritima in E. coli with a Strep tag at its N terminus led to a soluble and intact NfnAB complex after purification (Fig. 1). Purification of NfnAB under strictly anaerobic conditions from 10 g (wet mass) of recombinant E. coli cells resulted in ϳ8 mg of enzyme with a specific activity of ϳ4 units/mg at 45°C (Fig. 1A and Table 1). Via gel filtration chromatography on HiPrep Sephacryl S-200, NfnAB was separated in two peaks. The smaller peak 1 and the larger peak 2 elute at apparent molecular masses of ϳ170 and 80 -90 kDa, respectively, corresponding to a Nfn(AB) 2 heterotetramer and NfnAB heterodimer. The specific activity of the enzyme from peak B was 35% higher than that from peak A, contained much less NfnB proteolysis products, and was therefore used for all further studies.
The flavin content was calculated by UV-visible absorbance spectroscopy to be approximately two FAD per NfnAB heterodimer, which fits well to two flavin binding sites detectable in the protein sequence. From the 10 irons predicted in NfnAB, 6 -8 were detectable before and ϳ9 were detectable after in vitro iron-sulfur cluster reconstitution, which also led to an increase in absorbance of the enzyme between 350 and 500 nm (Fig. 1B) and a slight increase in the specific activity.
Catalytic Properties of T. maritima NfnAB-The enzymatic activity of NfnAB of T. maritima was routinely measured at 45°C in 100 mM MOPS-KOH at the pH optimum of 7.0 by following the NAD ϩ -dependent reduction of ferredoxin with NADPH (regenerating system). Under these conditions, the specific activity of the most active fractions was near 4 units/ mg, which is rather low ( Table 1). Measurements at 80°C, the optimal growth temperature of T. maritima, resulted in a 10 times higher specific activity of NAD ϩ -dependent ferredoxin reduction, which is in the order of the specific activities measured for NfnAB complexes from other organisms (19,20). NfnAB was found to catalyze the NADH-dependent reduction of NADP ϩ with reduced ferredoxin (Fd red ), the NAD ϩ -dependent reduction of Fd ox with NADPH, and the Fd ox -dependent reduction of NAD ϩ with NADPH (Table 1). These findings indicate a complete reversibility of the reaction. The apparent K m value of NADPH is lower than that of NADH, and that of NADP ϩ is lower than that of NAD ϩ . Notably, the NADP ϩ -dependent reduction of NAD ϩ with Fd red is not feasible.
Normally, Fd red was continuously regenerated via reduction with 100% H 2 catalyzed by ferredoxin-dependent [FeFe]-hydrogenase from C. pasteurianum. At pH 7, the redox potential of the C. pasteurianum ferredoxin is near Ϫ400 mV, and that of the H ϩ /H 2 couple is Ϫ414 mV. Therefore, in the experiments,  the ferredoxin was only reduced to ϳ50%, the rest remained oxidized. This is indicated in Table 1 by Fd red50% . In some experiments, reduced ferredoxin was regenerated in a system composed of ferredoxin, pyruvate, CoA, pyruvate:ferredoxin oxidoreductase from M. thermoacetica, phosphotransacetylase, and phosphate (100 mM). The redox potential of the acetyl-CoA ϩ CO 2 /pyruvate couple is near Ϫ500 mV. Therefore, ferredoxin was reduced to almost 100% indicated in Table 1 by Fd red100% . In addition, NAD ϩ or NADP ϩ cannot be reduced by Fd red without a high potential electron donor, although this is thermodynamically feasible and performed by plant Fnr.
The X-ray Structure-The crystal structure of the NfnAB complex was determined by the multiple anomalous dispersion method using the irons of the iron-sulfur clusters as anomalous scatterers and refined to R/R free factors of 18.7/21.9% at 2.3 Å resolution (Table 2). In the crystalline state, the enzyme is present as a NfnAB heterodimer (Fig. 2), which is most likely also the oligomeric state in solution despite the ambiguous gel filtration profile. In agreement with the biochemically determined FAD and iron contents of the T. maritima (see above) and also of the C. kluyveri and M. thermoacetica enzymes (19,20), NfnA contains one FAD (termed a-FAD) and one [2Fe-2S] cluster and NfnB one FAD (b-FAD) and two [4Fe-4S] clusters (Fig. 3).
NfnA architecturally belongs to the Fnr family (41), the most related structurally known member being the L. lactis Dodh with a root mean square deviation of 2.0 Å by a sequence identity of 24% (38). Accordingly, NfnA is composed of an FADbinding six-stranded antiparallel ␤-barrel domain (1-97), an NAD(P) binding ␣/␤ domain (98 -216), and a C-terminal extension (217-277) subdivided into a segment carrying the [2Fe-2S] cluster and a long C-terminal ␣-helix ( Fig. 2A). The planar isoalloxazine ring of a-FAD is sandwiched between the FAD and NAD(P) domains in the canonical manner; the siface is attached to strand 48:54 of the ␤-barrel, and the re-face points to the NAD(P) binding site (Fig. 3A). In high agreement with other family members, N 5 of FAD is hydrogen-bonded to Thr 53 -OG and -NH, O 2 is hydrogen-bonded to Lys 69 -NH, and N 1 has no contact to the polypeptide. Therefore, the redox potential of FAD from Fnr and Dodh of Ϫ342 mV (EЈ o (FAD/FADH ⅐ ) ϭ Ϫ350 mV; EЈ o (FADH ⅐ /FADH Ϫ ) ϭ Ϫ335 mV; pH 7, 10°C) (42) and Ϫ304 mV (EЈ o (FAD/FADH ⅐ ) ϭ Ϫ312 mV; EЈ o (FADH ⅐ /FADH Ϫ ) ϭ Ϫ297 mV) (43) might be approximate guidelines for NfnAB. For both enzymes, a neutral blue semiquinone was detectable. Fe1 and Fe2 of the [2Fe-2S] cluster are ligated to Cys 228 and Cys 240 , as well as to Cys 225 and Asp 220 , respectively (Fig. 3C). The aspartate might implicate a lower redox potential than E o Ј ϭ Ϫ212 mV measured for the [2Fe-2S] cluster of Dodh, which contains a cysteine at the equivalent position (43).
NfnB is a member of a disulfide oxidoreductase superfamily, and the structurally most related representatives are domains I-III of subunit A of pig Dpdh (23) and the ␤-subunit of A. brasilense Gls (22) (Fig. 3B). The NAD(P) binding site lies at the re-face. Interestingly, the guanidinium group of Arg 187 is positioned roughly parallel to the isoalloxazine ring and points toward the pyrazine and pyrimidine rings. Its guanidino group fixed by a bidentate salt bridge to Asp 293 interacts with O 4 and N 5 of FAD by its NH 2 and NH 1 groups, respectively. N 10 and O 2 are hydrogen-bonded to Val 441 -NH, the N-terminal residue of  (Fig. 3B). The redox potential of the equivalent FAD of Gls was determined to be Ϫ300 mV (44). A semiquinone was not detectable. The proximal [4Fe-4S] cluster is embedded into a rather hydrophilic pocket, and the irons are ligated to three cysteines (Cys 51 , Cys 90 , and Cys 96 ) and Glu 117 (Fig. 3D); the glutamate is conserved in Gls and replaced by a glutamine in Dpdh. A glutamate carboxylate as iron ligand is also reported for a few other [4Fe-4S] clusters (45) and should lower the redox potential compared with a cysteine thiolate ligand. The distal [4Fe-4S] cluster is flanked by predominantly hydrophobic residues and ligated to four cysteines (Fig. 3E). Although located close to the surface of the iron-sulfur cluster domain, it is completely shielded from bulk solvent. The redox potentials of both [4Fe-4S] clusters investigated in Gls and Dpdh are very low (ՅϪ400 mV) (44,46), and one of them could not be determined in Dpdh.
The Binding Site of the Substrates NADH, NADP ϩ , and Ferredoxin-Both NfnA and NfnB contain a nicotinamide coenzyme binding site, but the respective location of NAD(H) and NADP(H) has to be identified. An x-ray structure of an NfnAB-NADH complex at 2.4 Å resolution clearly determined NfnA as binding site for NADH ( Fig. 4A and Table 2). The nicotinamide ring of NADH is sandwiched between the isoalloxazine and Gly 114 , Gly 191 , and Pro 192 . Because the distances of 3.8 Å between their C 4 and N 5 atoms are relatively long, and the angle of ϳ45°between the nicotinamide and isoalloxazine ring significantly deviates from planar, the hydride transfer geometry appears not to be optimally adjusted. However, the corresponding rings in the Fnr-NADPH complex are also tilted 30°against each other.
The structural basis for the different NAD(H) and NADP(H) specificity of NfnA and Fnr (47), respectively, can be extracted from the NfnAB-NADH and Fnr-NADPH complex structure (Fig. 4A). The adenosine ribose group in NfnA is oriented in a manner that its hydroxy groups point toward strands 106:111 and 131:137, and their subsequent loops thereby interact with the main chain of Gly 111 , Gly 112 , Gly 138 , and Arg 139 . In comparison, the ribose-2Ј-phosphate of NADP in Fnr is rotated ϳ90°, and the phosphate group is hydrogen-bonded to Ser 228 , Arg 229 , Lys 238 , and Tyr 240 . In NfnA, These residues are replaced by Asp 162 , Gly 163 , Val 170 , and Val 171 , which are unsuitable for phosphate group binding. The different ribose or ribose-phosphate arrangement in NfnA and Fnr also implies a rotation of ϳ90°of their adenine rings relative to each other, indicating a redesign of the entire adenosine binding site in the absence of the extra phosphate group (Fig. 4A). Because of the described similarity between NfnAB and Dodh, the NADH binding mode of NfnAB also provides a reasonable model for Dodh that was not structurally characterized in complex with NADH yet. The unambiguous identification of the NADH binding site in NfnA defines NfnB as the binding site for NADP ϩ . This indirect identification is corroborated by the finding that the related Dpdh is also NADP dependent, and a superposition of NfnB and the Dpdh-NADP ϩ complex (48) indicates only a few collisions but several favorable and conserved side chain interactions between NADP ϩ and NfnB. Notably, the 2Ј phosphate group of ribose of Dpdh is hydrogen-bonded with Arg 364 and Lys 365 , which are equivalently exchanged by Arg 311 and Arg 312 in NfnB. A third salt bridge to the 2Ј-phosphate group is formed by Arg 371 in Dpdh and by Arg 371 in NfnB (Fig. 4B).
The binding site of the third substrate ferredoxin has not been experimentally explored. Structure comparison studies immediately exclude the known ferredoxin binding sites of Fnr because its position is occupied in NfnA (and Dodh) by the ironsulfur cluster carrying extension. A more conclusive model for ferredoxin binding to NfnAB could be derived from the Dpdh structure, which contains the mentioned NfnB-like domain in subunit A but also a ferredoxin-like domain in subunit B. A superposition of NfnB, and the structurally known ferredoxin of T. maritima carrying a single [4Fe-4S] cluster (49) onto Dpdh results in a model of the NfnAB-ferredoxin complex (Fig. 4C). The root mean square deviation between the C ␣ atoms of T. maritima and Dpdh ferredoxin is 1.6 Å using 92% of the residues for calculation. One main chain collision occurs between loop 11:14 in T. maritima ferredoxin and loop 40:46 in NfnB, which have to evade each other during complex formation. Moreover, this model for the NfnB-ferredoxin complex is in a high agreement with that based on protein-protein docking calculations (50). Accordingly, ferredoxin is attached to the outside margin of the iron-sulfur cluster domain of NfnB, and the edge to edge distance between its distal [4Fe-4S] cluster and the [4Fe-4S] cluster of ferredoxin is ϳ8 Å (Fig. 4C), which is sufficiently close to allow efficient electron transfer (51). The postulated ferredoxin binding site is also plausible because all redox centers are thus integrated into the electron transfer chain (see below).
Mechanism of FAD-based Electron Bifurcation-Kinetic characterization provided evidence that the NfnAB complex of T. maritima is an electron-bifurcating enzyme and reversibly catalyzes the reduction of NAD ϩ and Fd ox with 2 NADPH (Reaction 2 and Table 1). The obtained structural data revealed valuable information about how this enzymatic process works on an atomic basis (1). NAD(H) specifically binds to a-FAD and NADP(H) to b-FAD. Ferredoxin is most likely docked to the iron-sulfur cluster domain of NfnB adjacent to the distal [4Fe-4S] cluster (Fig. 4) (2). A one-electron transfer chain extends from a-FAD via the [2Fe-2S] cluster (8.5 Å) to b-FAD (15.0 Å) and from there via the proximal and distal [4Fe-4S] clusters (7.3 ϩ 9.4 Å) to the [4Fe-4S] cluster of ferredoxin (ϳ8 Å) (Fig.  2B). The edge to edge distances (in parentheses) are sufficiently short to ensure a fast electron transfer process (51) except for the borderline value between the [2Fe-2S] cluster and b-FAD (3). b-FAD is assigned as the bifurcating FAD because of its central position in the electron transfer chain (Fig. 2) and because of its binding site for the high potential 2e donor NADPH (Fig. 4B) whose redox potential (Ϫ370 mV) under physiological conditions is between that of NAD (Ϫ280 mV) and ferredoxin (ϽϪ400 mV). Kinetic measurement also dem-  SEPTEMBER 4, 2015 • VOLUME 290 • NUMBER 36 onstrated that NADH cannot replace NADPH as high potential 2e donor (Table 1) and NADP ϩ not NAD ϩ as high potential 2e acceptor, indicating specific binding sites for them in agreement with the structural data ( Fig. 4) (4). Because of related isoalloxazine binding modes, the determined redox potentials of Dodh and Gls were applied as approximate guidelines in the following mechanistic proposal.

FAD-based Bifurcating NfnAB Complex
In the bifurcating direction (Fig. 5), NADPH binds to NfnB and transfers a hydride to the bifurcating b-FAD serving as a 2 to 1e switch. According to energetic considerations, the first electron flows toward the site of the exergonic reaction and subsequently the second electron to the site of the endergonic reaction (21), implicating a high FADH Ϫ /FADH ⅐ and a low FADH ⅐ /FAD redox potential. The corresponding redox potentials of Gls (44) (52). Therefore, the redox potential of the b-FAD/b-FADH ⅐ pair have to be lower than Ϫ400 mV, which is in the range of the estimated value of Gls. In a second round, the described process is repeated (Fig. 5). Finally, NAD ϩ is reduced to NADH from a-FADH Ϫ , and a second Fd ox is reduced to Fd red .
Kinetic data clearly indicate that NADPH can only reduce NAD ϩ in the presence of Fd ox , despite the more unfavorable thermodynamics. Therefore, the downhill flow of the second electron from electron-rich b-FADH ⅐ toward a-FADH ⅐ has to be prevented in order to direct the electron to the uphill electron transfer branch for producing the energy-rich Fd red . How this process is governed is, so far, unresolved. According to the architecture of the NfnAB structure, our current model assumes a conformational rearrangement of NfnA relative to NfnB that prolongs or shortens the distance between the [2Fe-2S] cluster and b-FAD, thereby blocking or conducting electron flow (Figs. 5 and 6). The distance of 15.0 Å between them in the NfnAB structure is on the borderline for an efficient electron transfer. We speculate that NADPH binding shortens the distance and thus enables electron conduction to a-FAD, whereas the subsequent formation of b-FADH ⅐ associated with the loss of a negative charge and perhaps with the release of NADP ϩ implicates the corresponding return movement. An analysis of the NfnA-NfnB interface reveals rather few intersubunit contacts, which would allow a concerted rigid body movement perhaps induced by NADPH binding and NADP ϩ release (Fig. 6). In parallel, the thermodynamically unfavorable electron flow to  Bcd and EtfAB as separated proteins and the NfnAB complex are the only flavin-based electron bifurcating proteins structurally studied yet. In both systems, the reactions are embedded into a modularly composed protein apparatus with redox-center carrying domains/subunits that are adjustable relative to each other. Both enzymes contain bifurcating FADs that accept 2e from the high potential 2e donor NAD(P)H and are the start-ing point for two 1e transfer pathways into different directions to a high potential 2e acceptor and the low potential Fd ox . Additional 1e redox centers, i.e. iron-sulfur clusters or FAD, are spaced between the active sites of the three redox reactions if their distances are too long. It appears that the electron transfer routes between the bifurcating FAD and the [4Fe-4S] cluster of ferredoxin are optimized perhaps to enable a moderate uphill electron transfer and/or to compete successfully with the electron flow to the exergonic NAD ϩ reduction. The binding pocket for bifurcating isoalloxazine rings has to be designed to adjust a low redox potential of the FADH ⅐ /FAD pair. In EtfAB and the NfnAB complex, the oxidized state might be stabilized by an arginine placed in front of the N 5 and O 4 atoms that preferably interacts with a deprotonated oxidized FAD (Fig.  3B). For both Bcd/EtfAB and NfnAB, a subunit/domain rearrangement is postulated to switch on or off the electron flow between the bifurcating FAD and the first redox center of the exergonic electron transfer branch (Fig. 6). This scenario is related to that established for the quinone-based electron bifurcation process where the second electron transfer between the semiquinone and the high potential [2Fe-2S] center is interrupted by a conformational change of the Riske protein.  In the bifurcating direction, b-FAD is reduced with NADPH, and b-FADH Ϫ subsequently donates its first electron to a-FAD and its second electron to Fd ox . Therefore, a capability has to exist to block or conduct the downhill electron transfer (ET) from b-FADH ⅐ toward a-FADH ⅐ . The rearrangement of NfnA relative to NfnB would be a plausible model. The oxidation state of the iron-sulfur clusters in NfnAB is not indicated because of the instant change in electron delivery. In comparison with the Bcd/EtfAB complex, EtfB corresponds to NfnB carrying the bifurcating FAD, and Bcd corresponds to NfnA carrying the FAD for hydride transfer to the high potential 2e acceptor (crotonyl-CoA, NAD ϩ ). The varying distances between the bifurcating FAD and FAD of EtfA or the [2Fe-2S] cluster of NfnA, respectively, control the electron transfer route. FIGURE 6. Blocking (blue) and conducting (salmon) states of the exergonic electron transfer branch. Assuming the experimentally determined NfnAB structure in the blocking state, a conducting state might be adjusted by a conformational change of NfnA relative to NfnB (green), which reduces the distance between the bifurcating b-FAD of NfnB and the [2Fe-2S] cluster of NfnA. Modeling was performed in a manner to minimize the interference between the fixed NfnB and the moving NfnA.