Oxidation of the FAD cofactor to the 8-formyl-derivative in human electron-transferring flavoprotein

The heterodimeric human (h) electron-transferring flavoprotein (ETF) transfers electrons from at least 13 different flavin dehydrogenases to the mitochondrial respiratory chain through a non-covalently bound FAD cofactor. Here, we describe the discovery of an irreversible and pH-dependent oxidation of the 8α-methyl group to 8-formyl-FAD (8f-FAD), which represents a unique chemical modification of a flavin cofactor in the human flavoproteome. Furthermore, a set of hETF variants revealed that several conserved amino acid residues in the FAD-binding pocket of electron-transferring flavoproteins are required for the conversion to the formyl group. Two of the variants generated in our study, namely αR249C and αT266M, cause glutaric aciduria type II, a severe inherited disease. Both of the variants showed impaired formation of 8f-FAD shedding new light on the potential molecular cause of disease development. Interestingly, the conversion of FAD to 8f-FAD yields a very stable flavin semiquinone that exhibited slightly lower rates of electron transfer in an artificial assay system than hETF containing FAD. In contrast, the formation of 8f-FAD enhanced the affinity to human dimethylglycine dehydrogenase 5-fold, indicating that formation of 8f-FAD modulates the interaction of hETF with client enzymes in the mitochondrial matrix. Thus, we hypothesize that the FAD cofactor bound to hETF is subject to oxidation in the alkaline (pH 8) environment of the mitochondrial matrix, which may modulate electron transport between client dehydrogenases and the respiratory chain. This discovery challenges the current concepts of electron transfer processes in mitochondria.

In 1956, Crane et al. (1) identified the electron-transferring ability of an unknown flavoprotein from the pig liver, which they named electron-transferring flavoprotein (ETF). 3 Since then, numerous studies on ETF have been reported, and orthologs have been described in all kingdoms of life (2). The heterodimeric human (h) ETF serves as a central electron carrier in the mitochondrial matrix. hETF accepts electrons from 13 flavin dehydrogenases and transfers them to the human ETF-ubiquinone oxidoreductase (hETF-QO), an iron-sulfur cluster containing flavoprotein bound to the inner mitochondrial membrane that feeds these electrons into the mitochondrial respiratory chain (2,3). The flavin dehydrogenases are either part of ␤-oxidation, amino acid, or choline degradation, as shown in Scheme 1.
Interestingly, these dehydrogenases are structurally distinct with dehydrogenases operating either in the degradation of fatty or amino acids adopting the "acyl-CoA dehydrogenase"-fold, whereas both dimethylglycine dehydrogenase (hDMGDH) and sarcosine dehydrogenase are part of the amine oxidase protein family. Apparently, hETF has evolved a flexible mechanism to interact with various dehydrogenases as well as with hETF-QO (2). The protein exists in a closed non-productive and in an open productive conformation with a highly flexible upper protein domain (Scheme 1). The interaction with client dehydrogenases is initiated by a recognition peptide of the ␤-subunit leading to the exposure of the FAD and concomitant electron transfer (4).
Recently, Toogood et al. (4) have shown that a single amino acid replacement in the ␤-subunit (␤E165A) favors the open conformation by removing a side-chain interaction with ␣Asn-259 in the ␣-subunit. As a consequence, variant ␤E165A displayed higher affinity to the human medium-chain acyl-CoA dehydrogenase (hMCAD) (and to rat DMGDH) enabling cocrystallization (Scheme 1). In the course of our attempts to co-crystallize hETF with hDMGDH, we have reproduced the ␤E165A variant as well as generated the corresponding ␣N259A variant. However, characterization of the variants by UV-visible absorption spectroscopy indicated that the FAD cofactor undergoes a slow conversion affecting the isoalloxazine ring. A detailed analysis of the isolated flavin moiety showed that the observed conversion entails the oxidation of the 8␣-methyl to the formyl group, i.e. 8-formyl-FAD (8f-FAD) is the product of the spontaneous conversion observed in these two variants. This prompted us to extend the study to include wildtype hETF and several variants with amino acid replacements in the binding pocket of the isoalloxazine ring. In this study, we document that (i) conversion of FAD to 8f-FAD also occurs in wildtype hETF at alkaline pH, (ii) the 8f-FAD stabi-lizes an oxygen-insensitive semiquinone radical, and (iii) several highly conserved amino acid residues in the FAD-binding pocket are essential for the formation of 8f-FAD. In addition, we demonstrate that wildtype hETF bearing 8f-FAD as cofactor exhibits increased affinity to reduced hDMGDH under steadystate conditions suggesting a potential role in vivo. Taken together, our results report entirely novel properties of hETF that may have important implications for our understanding of electron handling in the mitochondrial matrix.

Time-dependent absorption changes of purified recombinant hETF
hETF produced in Escherichia coli host cells was purified by means of Ni-NTA affinity chromatography and yielded 2 mg of homogeneous protein per g of wet cell pellet (Fig. 1). To explore the effect of the pH on protein yield and stability, purifications were performed at pH 7, 7.8, and 8.5. This gave rise to different UV-visible absorption spectra as shown in Fig. 2A. At neutral pH, the absorption spectrum featured two maxima at 375 and 436 nm, which were very similar to spectral properties reported for hETF and pig liver ETF, respectively (5, 6). However, pronounced changes were observed when hETF was purified at pH 7.8 or 8.5. These spectra were characterized by a bathochromic shift of the maximum at 436 to 415 nm accompanied by an increase in absorption at a longer wavelength ( Fig. 2A). To check whether the observed spectral changes were caused by a chemical modification of the FAD cofactor, hETF purified at pH 7 and pH 8.5 was denatured, and absorption spectra were recorded. As shown in Fig. 2B, hETF purified at pH 7 yielded the typical absorption spectrum of free FAD with maxima at 370 and 450 nm, whereas the flavin moiety released from hETF purified at pH 8.5 featured spectral shifts to 354 and 456 nm, respectively.

Isolation and identification of the modified FAD generated at alkaline pH
To identify the nature of the chemical modification in the flavin chromophore, we extracted the cofactor(s) from hETF purified and kept at pH 8.5. After extraction and purification by HPLC (Fig. 3, A and B), the two main fractions were analyzed by mass spectrometry and NMR spectroscopy (Fig. 3, C and D, respectively). Mass spectrometric analysis of one of the major fractions (peak 2) clearly showed the typical fragmentation pattern (AMP, m/z ϭ 348; FMN, m/z ϭ 439 and 457) and mass of FAD (m/z ϭ 786; MassBank accession number KNA00248 (7)), whereas the other fraction (peak 1) exhibited a shift of 14 a.u. of all major peaks, except that for AMP. In agreement with the observed differences in the absorption properties, this result confirmed that the chemical modification has most likely occurred in the isoalloxazine ring of the FAD cofactor. Further analysis of the two flavin-containing fractions by NMR spectroscopy revealed the presence of a resonance at ϳ10.4 ppm in the unknown flavin species that is absent in FAD. Additional differences were observed in the resonances of the methyl groups at the 7␣and 8␣-positions (see arrows in Fig. 3D). Overall, the 1 H NMR spectrum possesses the same features as reported previously for 8-formyl-FAD isolated from formate oxidase (8). Taken together with the observed mass difference, we therefore conclude that the 8␣-methyl group of FAD was oxidized to a formyl group to yield 8f-FAD.

Time-dependent formation of 8f-FAD in hETF
To obtain further insights into the formation of 8f-FAD in hETF, a sample purified at pH 7 was diluted into buffer at pH 8.5, and the spectral changes were observed over time. As Scheme 1. Interaction of human flavin dehydrogenases with hETF. So far, 13 flavin dehydrogenases, involved in ␤-oxidation (short chain acyl-CoA dehydrogenase; medium chain acyl-CoA dehydrogenase; long chain acyl-CoA dehydrogenase; very long chain acyl-CoA dehydrogenase; and acyl-CoA dehydrogenase family member 9 -11), amino acid (short branched chain acyl-CoA, iso-valeryl-CoA, iso-butyryl-CoA, and glutaryl-CoA dehydrogenase), and choline degradation (dimethylglycine and sarcosine dehydrogenase) were identified to interact with hETF. hETF exhibits a flexible interaction mechanism and adopts a closed non-productive form (PDB code 1EFV) and an open productive conformation, here shown bound to human medium chain acyl-CoA dehydrogenase (PDB code 2A1T). The ␣and ␤-subunits of hETF are shown in raspberry and in marine cartoon view, respectively. hETF bound hMCAD is displayed in gray. hETF bound FAD is presented in yellow sticks, hMCAD bound FAD in pink stick representation.

8-Formyl-FAD formation in hETF
shown in Fig. 4A, the formation of 8f-FAD is a slow process with an approximate half-time of 20 h for hETF-WT and 4 h for hETF-␣N259A, respectively. The spectral changes are marked by a single set of isosbestic points at 340, 375, 420, and 469 nm and the appearance of a sharp absorption maximum at 415 nm as well as much less pronounced maxima at 650 and 710 nm (Fig. 4). The reverse reaction, the formation of FAD from 8f-FAD at neutral or acidic pH, was not observed. The formation of 8f-FAD was independent from the buffer used (HEPES, Tris, and phosphate buffer). Similarly, removal of the hexahistidine tag using tobacco etch virus protease had no effect on the oxidation of the 8␣-methyl group.
Because the formation of 8f-FAD does not occur free in solution, we assumed that the oxidation is promoted by amino acid residues in the isoalloxazine-binding pocket. Therefore, we generated four variants of hETF featuring single amino acid replacements: ␣R249C, ␣T266M, ␣H286A, and ␤Y16F (Fig. 5). Interestingly, the formation of 8f-FAD was substantially affected in all variants and decreased in the order ␤Y16F Ͼ ␣T266M Ͼ ␣R249C (Table 1). For the ␣H286A variant, 8f-FAD formation could not be monitored reliably since the FAD binding was apparently compromised by the amino acid replacement. Hence, we conclude that amino acid residues directly participate in the oxidation of the 8␣-methyl to the formyl group. In addition, we investigated two variants, ␣N259A and ␤E165A, that presumably increase the flexibility of the FADbinding domain to favor the formation of a productive open conformation for electron transfer (9). Both of the variants showed a much more rapid formation of 8f-FAD, which already occurred at neutral pH (Fig. 4B, data for ␣N259A as an example). This observation strongly suggests that the conformational dynamics of hETF affect the rate of conversion of FAD to the 8f-FAD.

Generation of 8f-FAD leads to the flavin semiquinone
Although our analysis has clearly shown the oxidation of the 8␣-methyl to the formyl group, the spectral characteristics observed for the generated 8f-FAD are vastly different from those seen after release of the flavin from hETF (Fig. 2B). As a matter of fact, the sharp peak at 415 nm as well as the long wavelength absorption is reminiscent of a flavin semiquinone species suggesting that the 8f-FAD is present as a semiquinone radical instead of the oxidized form. The presence of a flavin radical was confirmed by EPR spectroscopy of the hETF variant ␣N259A, which rapidly forms the 8f-FAD at pH 8.5. A g-factor of 2.0048 was calculated after calibration with DPPH (g-factor of 2.0036) and a peak-to-peak line width of 11-13 G was measured. This line width is rather narrow for a flavin radical and suggests the stabilization of the red anionic semiquinone (10). Expectedly, hETF variant ␤Y16F and wildtype hETF purified at pH 7.0 (no 8f-FAD formation) did not yield an EPR signal.

Redox behavior of wildtype hETF and the ␣N259A variant
The stabilization of the anionic (red) FAD semiquinone in hETF is a well-established phenomenon (5,6). As shown in Fig.  6A, photoreduction of wildtype hETF purified at neutral pH yielded the FAD semiquinone before full reduction to the hydroquinone was achieved with sodium dithionite (Fig. 6A, inset). In contrast to wildtype hETF, the ␣N259A variant purified at pH 8.5 was already present in the semiquinone form, and photoreduction yielded the fully reduced hydroquinone species featuring characteristic absorption maxima at 390 and 510 nm (11,12). The hydroquinone form of 8f-FAD is sensitive to oxygen and is reoxidized to the semiquinone but not to the oxidized form. Thus, hETF not only stabilizes the semiquinone form of 8f-FAD but also prevents its oxidation by molecular oxygen. This differential behavior of FAD versus 8f-FAD has important implications for the one-electron transfer processes between hETF, the serviced dehydrogenases, as well as the electron acceptor hETF-QO.

Physiological relevance of 8f-FAD, effects on stability, and protein-protein interaction
To evaluate the potential impact of FAD cofactor formylation on the electron transfer reaction, we investigated the steady-state kinetics of hETF with hDMGDH, which is one of

8-Formyl-FAD formation in hETF
its client dehydrogenases. For that purpose, wildtype and variant ␣N259A were both purified in 50 mM HEPES at pH 7.0 and pH 8.5, and afterward, the steady-state kinetic parameters were determined at pH 7.0. As shown in Fig. 7 and summarized in Table 2, we observed moderate changes in the velocity of electron transfer (k cat was reduced by ϳ50%) and a 5-fold decrease in the K m values in the proteins harboring 8f-FAD. It should be noted in this context that hETF-␣N259A purified at pH 7.0 already contains ϳ15-30% of 8f-FAD, which may contribute to the lower K m value compared with the wildtype hETF ( Table 2). In contrast, wildtype hETF purified at pH 8.5 contains only ϳ30 -40% 8f-FAD, and thus the obtained K m value reflects a mixed population of wildtype hETF, i.e. with the FAD cofactor in its methylated (native) and formylated structure, respectively. This may also contribute to the differences found between wildtype hETF and the hETF-␣N259A variant ( Table  2). The steady-state kinetics of protein variant ␤E165A purified at pH 7.0 resembled the values obtained for wildtype hETF purified at pH 7.0. In contrast to the corresponding variant ␣N259A, FAD formylation was absent in the freshly purified variant ␤E165A and was only observed during long-time storage at pH 7.

Discovery of an unusual FAD derivative in an old protein
In this investigation, we have unambiguously demonstrated the formation of 8f-FAD in hETF by means of various spectroscopies such as HPLC/MS, UV-visible absorption, and NMR and EPR spectroscopy. Furthermore, it was shown that the formation of 8f-FAD strongly depended on the pH such that wildtype hETF is resistant to the formation of 8f-FAD at acidic and neutral pH, whereas at alkaline pH the slow oxidation of the 8␣-methyl to the formyl group is observable (t1 ⁄ 2 ϭ 20 h, Figs. 2A . Analysis and comparison of the two main flavin-containing fractions isolated from hETF. A, HPLC reversed phase purification of the extracted cofactor(s) gave two major fractions, which were further analyzed by MS and NMR. B, spectra of the two peaks of HPLC purification featured the same shifts as seen in Fig. 2B. C, mass spectra of the two main fractions as separated by HPLC. The spectrum shown at the bottom exhibits the typical fragmentation and mass peaks of FAD. The mass spectrum at the top shows a mass shift of 14 a.u. D, in agreement with the mass analysis, the 1 H NMR spectrum at the bottom can be assigned to FAD, whereas the additional resonance at 10.4 ppm and the shifts observed for the methyl groups in position 7␣ and 8␣ indicate chemical changes in the dimethylbenzene ring moiety of the isoalloxazine ring. Both methods indicate that a small amount of a closed, hemiacetal form of 8f-FAD is present, as observed previously (12).

8-Formyl-FAD formation in hETF
and 4A). Interestingly, 8f-FAD was not present in the oxidized but in the semiquinone state, which was stable toward oxidation by dioxygen. The UV-visible absorption spectrum featured a sharp peak at 415 nm and two minor absorption maxima at longer wavelengths (650 and 720 nm, Fig. 2A). The latter spectral features are clearly indicative of the (blue) neutral semiquinone, although the peak-to-peak linewidth measured by EPR spectroscopy suggested that the major fraction is the (red) anionic semiquinone. Thus, we conclude that a mixture of the blue and red semiquinone is present in hETF suggesting that our measurements were conducted near the pertinent pK a of the flavin radical, as was also found earlier for a lysine to arginine replacement in the FMN-binding pocket of lactate oxidase (13). Overall, the spectral characteristics are very similar to those reported earlier by Yorita et al. (13) and by Maeda et al. (14) for formate oxidase from Aspergillus oryzae. In the latter case, it was shown that oxidation of the 8␣-methyl group significantly enhances enzyme activity, and thus it was argued that the 8f-FAD derivative is the cognate cofactor of the enzyme.
Because detailed studies on ETF date back to the 1950s, we were very surprised that our findings were apparently not observed previously. However, closer inspection of the literature revealed that isolation of ETF from pig liver yielded two forms (designated ETF R and ETF B ), which were reduced to the red (ETF R ) and blue semiquinone (ETF B ), respectively (11). Analysis of the isolated cofactors demonstrated the presence of a flavin species with absorption maxima at 463 and 352 nm, which are in fact very similar to the spectral features observed for the 8f-FAD spectrum after denaturation of wildtype hETF purified at pH 8.5 (Fig. 2B, red solid line). Furthermore, Lehman and Thorpe (11) reported the occurrence of a pink species upon reduction featuring an absorption maximum at 520 nm. Again, this is reminiscent of the absorption spectrum obtained when the ␣N259A hETF variant is photoreduced (Fig. 6B, dashed black line). Similarly, Yorita et al. (13) reported the same species upon reduction of 8-formyl FMN bound to lactate oxidase. Therefore, we assume that Lehman and Thorpe (11) have also

8-Formyl-FAD formation in hETF
isolated the 8f-FAD cofactor from pig liver ETF; however, neither the chemical nature of the flavin nor the mode of its generation was further investigated. Because the amino acid residues in the FAD-binding pocket of mammalian and bacterial ETFs are highly conserved (Fig. 8), it was tempting to assume that the generation of 8f-FAD is a common feature in this protein family. Thus, we have analyzed the crystal structures of previously published ETF structures, hETF (PDB code 1EFV), the ETF from Methylophilus methylotrophus (PDB code 1O96), Paracoccus denitrificans (PDB code 1EFP), and Acidaminococcus fermentans (PDB code 4KPU). Closer inspection of the electron densities obtained for the 8␣-position showed obvious deviations from the expected electron density of a methyl group for some reported crystal structures (Fig. 9). In fact, in those cases the additional electron density is indicative of a covalent modification of the 8␣-methyl group and is in line with the presence of (partially) formylated FAD (Fig. 9). With the exception of hETF, this was also found in cases where crystallization was carried out at or near pH 7, i.e. under conditions where oxidation of the 8␣-methyl to the formyl group is very slow. The structurally very similar bacterial ETFs from A. fermentans and M. methylotrophus show clear indications of FAD modification even for wildtype forms and a variant with a conserved positive charge on the re side of the flavin (Fig. 9, C, D, and F), whereas no indication of 8f-FAD formation is observed for the modification-incompetent ␣R236A variant of M. methylotrophus ETF (Fig. 9E). With regard to ETF from M. methylotrophus, it is also noteworthy that Byron et al. (15) clearly have observed the 8f-flavin radical species (cf. Fig. 4) but did not further elucidate the cause of the drastic change in the absorption spectrum. The occurrence of the 8f-FAD in M. methylotrophus ETF is potentially relevant as this could (partially) explain the unusually positive redox potential of the protein (ϩ196 mV for the ETF ox /ETF redox pair) (15), which is in accordance with previous reports that the redox potential of free 8f-FAD is ϳ130 -160 mV more positive than that of free FAD (8,12).

Proposed reaction mechanism for the formation of 8f-FAD
We propose that the oxidation of the 8␣-methyl group is initiated by proton abstraction, as proposed previously (16 -18), leading to the formation of a negative charge at the N(1)-C(2)ϭO locus of the isoalloxazine ring. After addition of water, the reduced intermediate is oxidized to yield the 8␣-hydroxymethylene intermediate, which is again subject to proton abstraction at the 8␣-group with the resulting enol tautomerizing to the reduced 8f-FAD. Taking the results of our mutagenesis study into account, it is conceivable that the phenolate side chain of ␤Tyr-16 may act as a base (instead of hydroxide). Such an involvement would be reminiscent of the formation of a covalent linkage established in 8␣-tyrosyl-modified flavoenzymes (19 -22). The side chains of ␣Arg-249, ␣Thr-266, and ␣His-286 support the reaction by hydrogen bonding to the N(1)-C(2)ϭO locus (␣His-286), N(5) (␣Thr-266), and by -cation interaction (␣Arg-249), respectively. Interestingly, variants favoring the open conformation of hETF, i.e. ␤E165A and ␣N259A (see under "Discussion"), also promote the oxidation of the 8␣-methyl group, probably due to greater solvent exposure of the relevant part of the isoalloxazine ring in line with the proposed role of hydroxide and water in the putative reaction mechanism. Because our experiments have shown that the generated 8f-FAD is preferentially present as the semiquinone, we propose that the 8f-FAD is rapidly reduced, e.g. in vitro by light and in vivo by one of the client dehydrogenases, to the very stable radical species.

Effect of cofactor oxidation to the 8f-FAD on electron transfer
The oxidation of FAD to 8f-FAD clearly affected the interaction with hDMGDH, which is one of the confirmed client dehydrogenases of hETF. The most striking effect was observed on the apparent K m value defined here as the concentration of hETF at which the rate of electron transfer proceeds at halfmaximal velocity. A 5-fold lower K m value was observed not only in the case of wildtype hETF but also for the ␤E165A and ␣N259A variants when 8f-FAD was present as the main cofac-

8-Formyl-FAD formation in hETF
tor moiety (Table 2). A smaller effect was also seen on the maximal rate of electron transfer from reduced hDMGDH to hETF as well as the ␤E165A and ␣N259A variants, which was ϳ50 -60% lower in the proteins containing 8f-FAD instead of FAD (Table 2). Thus, it is apparent that oxidation of the FAD cofactor in hETF significantly affects the interaction with hDMGDH. In fact, Toogood et al. (9) have reported that hETF variant ␤E165A has a higher affinity to hMCAD and rat DMGDH compared with the wildtype protein. Because the side chains of amino acid residues ␤Glu-165 and ␣Asn-259 stabilize a nonproductive, closed hETF conformation (Scheme 1), the replacement by alanine populates the open conformation, which is conducive to complex formation with an electron delivering dehydrogenase and enabled the crystallization of a protein complex comprising hETF and hMCAD (9). As a consequence of this perturbation of the conformational equilibrium of the closed and open conformation, the protein complex of hETF with its client enzymes exhibits higher affinity and thus rationalizes the observed kinetic effects on K m value and the rate of electron transfer under steady-state conditions ( Fig. 7 and Table 2).
Interestingly, formation of 8f-FAD appears to exhibit a similar effect on the interaction with client dehydrogenases as seen in the ␤E165A and ␣N259A variants. Therefore, we hypothesize that the presence of 8f-FAD in hETF leads to a significant shift of the conformational equilibrium toward the open (productive) conformation, which in turn increases the affinity to client enzymes. Currently, we are conducting hydrogen/deuterium exchange mass spectrometric experiments to test this hypothesis.

Potential physiological role of 8f-FAD in health and disease
Considering the apparent impact of the formylation of the FAD cofactor, the central question that emerges here concerns the physiological relevance of the cofactor modification. As the pH of the mitochondrial matrix was found to be close to 8.0 (23,24), formylation of wildtype hETF would clearly occur in this environment, albeit at a slow rate; thus, the overall lifetime of the protein will ultimately determine the fraction of hETFs containing 8f-FAD. Although, this lifetime is currently not known for hETF, mitochondrial matrix proteins were shown to exhibit half-lives in the range of 17 to more than 100 h (25). In the case that hETF belongs to the long-lived proteins, a significant fraction of the protein will in fact harbor the 8f-FAD cofactor. Moreover, it is conceivable that other proteins, such as the electron delivering dehydrogenases, promote the formation of 8f-FAD by stabilizing the open conformation of hETF as seen in the ␣N259A and ␤E165A variants.
In this context, it is very intriguing that several inborn mutations in the gene encoding human ETF are known to cause glutaric aciduria type II (GAII), also called multiple acyl-CoA dehydrogenase deficiency (OMIM entry no. 231680), a disease characterized by severe non-ketotic hypo-  To obtain protein with a high FAD content, purification was carried out at pH 7, whereas pH 8.5 was used to maximize the 8f-FAD content. All kinetic parameters were determined at pH 7.

8-Formyl-FAD formation in hETF
glycemia, metabolic acidosis, and excretion of large amounts of fatty acid and amino acid-derived metabolites (26,27). Strikingly, two of the hETF variants generated in our mutagenesis study, namely ␣R249C and ␣T266M, were reported as the cause of GAII, with the latter being the most abundant variant found in affected patients (28,29). Because both of these variants were severely impaired in the formation of the 8f-FAD derivative in their active sites, it is tempting to speculate that the inability to catalyze the oxidation of the 8␣-methyl to the formyl group contributes to the disease-causing effect of the underlying genetic mutation. In a previous study, Dwyer et al. (30) concluded that ␣Arg-249 plays a crucial role in stabilizing the flavin semiquinone state; however, at that time the occurrence of 8f-FAD and its mode of generation had not been recognized. The remarkable stability of the 8f-FAD radical may alter the operational basis of the electron transfer reactions of hETF with the serviced dehydrogenases and the terminal electron acceptor hETF-QO. In hETF with FAD as cofactor, it is assumed that the oxidized FAD receives one electron from a client dehydrogenase generating the FAD semiquinone, which is spontaneously disproportionate and subsequently reduces hETF-QO by two single electron transfer processes (29). Only the fully reduced hETF is then able to transfer two electrons to the hETF-QO. Thus, it will be interesting to see how formylation affects the electron transfer process between the hETF and the hETF-QO. In addition, 8f-FAD also possesses a much more positive redox potential, i.e. around Ϫ90 mV (15), and therefore formylation may also affect the rate of electron transfer between the client dehydrogenases, hETF and hETF-QO.

Conclusions
Our study has conclusively shown that hETF catalyzes the oxidation of FAD to 8f-FAD, an unusual cofactor modification that has not been reported for hETF or any other electrontransferring flavoprotein. Furthermore, we have demonstrated that the generation of 8f-FAD strongly depends on the pH, is catalyzed by amino acid residues in the FAD-binding pocket, and is favored in variants that preferentially adopt an open conformation. Depending on the lifetime of hETF in the mitochondrial matrix, formation of 8f-FAD will be physiologically relevant in particular because the 8f-FAD alters the interaction with electron-delivering dehydrogenases and the electron acceptor hETF-QO in the inner mitochondrial membrane. It also remains to be seen whether additional factors, such as the interaction with client dehydrogenases or the hETF-QO, affect the formation of 8f-FAD. In any case, it is apparent that our discovery of the spontaneous cofactor oxidation in a central protein of mitochondrial electron handling raises important biochemical and physiological questions with implications for human health and disease.

Enzymes and reagents
Restriction enzymes and Phusion DNA polymerase were from Thermo Fisher Scientific (Waltham, MA), and purification columns were from GE Healthcare (Chalfont St. Giles, UK). Salt-free purified oligonucleotides for site-directed mutagenesis were synthesized by VBC-Biotech (Vienna, Austria) or Sigma. All other chemicals and media were from Carl Roth GmbH (Karlsruhe, Germany) or Sigma and were of the highest grade available.

hETF-WT and hETF variants of gene expression design
The hETF sequence for expression of the mature hETF consisting of ␣and ␤-subunit was designed following a similar strategy as described by Bross et al. (31) and was optimized for expression in E. coli using GeneOptimizer (Thermo Fisher Scientific). The operon starts with a ribosomal binding site (AAGGAG), followed by a TATA box in front of the ATG start codon of the gene sequence of the ␤-subunit. After the stop codon of the ␤-subunit, a 69-bp spacer region between the ␤ and ␣ gene sequence was introduced, which again comprises the same ribosomal binding site and a TATA box in front of the ATG start codon of the ␣-subunit. The designed gene sequence was flanked by an XbaI and an XhoI restriction site and cloned for expression into a pET-28aϩ vector (Thermo Fisher Scientific). In agreement with Herrick et al. (5), the ␣-subunit starts with the first amino acid of the mature ␣ETF (␣Gln-20). For protein purification, a hexa-histidine tag was added to the N terminus of the ␤-subunit. The recombinant plasmid was transformed into E. coli BL21 (DE3) cells. Positive clones were selected by kanamycin resistance. Correct cloning and potent expression colonies were verified by automated sequencing. All investigated variants of hETF-WT (␣R249C, ␣N259A, ␣T266M, ␣H286A, ␤Y16F, and ␤E165A) were constructed by two-step site-directed mutagenesis with Phusion DNA polymerase and the mutation primers shown in Table 3 (the altered codons are highlighted in bold).
First, two separate PCRs using either forward or reverse primer with 10 ng of hETF-WT template DNA, 1ϫ Phusion HF buffer, 200 M dNTPs, 3% DMSO, 1 unit of Phusion DNA polymerase, and 0.5 M of each primer were run in 50-l volumes with (98°C (2 min) Ϫ (98°C (50 s) Ϫ 60°C (20 s) Ϫ 68°C (16.5 min)) ϫ 5 Ϫ 4°C ∞). Afterward, the separated PCRs were combined, and the same PCR program was further employed for another 20 cycles. The PCR was followed by a 2-h DpnI digestion step; the plasmid was then transformed into E. coli BL21 (DE3) cells, and the strain selection was done as above using the pET28aϩ kanamycin resistance.

hETF production
hETF expression was carried out in shake flasks in an HT Multitron Standard shaking system (Infors AG, Basel, Switzerland) at 150 rpm. Briefly, overnight LB medium cultures of E. coli BL21 (DE3) cells with the desired hETF variant were used to inoculate 1 liter of main culture in a baffled shake flask to an

8-Formyl-FAD formation in hETF
A 600 of 0.1. After reaching an A 600 of 0.6 -0.8, the protein production was started by induction with 0.1 mM isopropyl 1-thio-␤-D-galactopyranoside and expression took place overnight at 25°C. The cell pellet was collected by centrifugation (2704 ϫ g, 15 min, room temperature) and stored at Ϫ20°C.

hETF purification
In general, enzyme purification was carried out according to the following protocol. Cell lysates were prepared by four 3-min sonications (3-min cooling steps) using a Labsonic L sonication probe (B. Braun Biotech, Berlin, Germany) in a Sonopuls rosette cell RZ (Bandelin, Berlin, Germany). The wet cell pellet was suspended in lysis buffer (3 ml of 50 mM HEPES/NaOH, 15 mM imidazole, pH 7.0, 7.8, or 8.5 per g), and a spatula tip of FAD was added before sonication. The lysates were cleared by centrifugation (38,720 ϫ g, 45 min, 4°C) and filtration through a paper filter. Nickel ion affinity chromatography was performed by applying the cell lysates onto 5-ml HisTrap HP columns (GE Healthcare). Afterward, the columns were washed with at least 10 column volumes of lysis buffer, and the enzyme was stripped off with elution buffer (50 mM HEPES/NaOH, 200 mM imidazole, pH 7.0, 7.8, or 8.5). The purification was monitored by SDS-PAGE, and fractions containing hETF were concentrated using Amicon ultracentrifugal filter units (10-kDa cutoff, Merck-Millipore, Darmstadt, Germany) and rebuffered to storage buffer (50 mM HEPES/NaOH, pH 7.0, 7.8, or 8.5) using Sephadex G-25 PD10 desalting columns (GE Healthcare). After rebuffering, the enzyme solution was incubated at 37°C for 30 min and afterward cleared by centrifugation to remove aggregated protein and excessive hETF ␤-subunits. The obtained enzyme purity was sufficient for all kinetic and spectrophotometric studies.

hDMGDH production and purification
hDMGDH expression and purification for use in interaction studies for steady-state kinetic analyses was performed as reported previously by Augustin et al. (32).

SDS-PAGE
Enzyme samples were separated by SDS-PAGE with 12.5% separation and 5% stacking gels under reducing conditions (100 mM DTT in the sample buffer) as described by Laemmli (33). Gels were stained with Coomassie Brilliant Blue R-250 for purification control, and a PageRuler prestained protein ladder (Thermo Fisher Scientific) was employed as protein standard.

UV-visible absorption spectroscopy
UV-visible absorption spectra to assess protein concentration, activity, purity, and quality as well as for steady-state kinetic measurements and photoreduction were recorded with a Specord 210 spectrophotometer (Analytik Jena, Jena, Germany).

Protein quantification and calculation of the extinction coefficient
Protein concentrations of purified hETF wildtype and variants were determined using the characteristic absorption of protein-bound FAD at 469 nm (isosbestic point of FAD and 8f-FAD). A molar extinction coefficient (⑀) of 9900 Ϯ 700 M Ϫ1 cm Ϫ1 for hETF was determined using the method described by Macheroux (34) based on an ⑀ of free FAD at 469 nm of 9910 M Ϫ1 cm Ϫ1 .

Extraction and purification of 8f-FAD
Wildtype hETF or the variant ␣N259A was purified in 50 mM HEPES/NaOH buffer at pH 8.5 and concentrated to about 300 -400 M using Amico ultracentrifugal filter units (10-kDa cutoff). After complete denaturation, which was achieved by two times treatment at 70°C for 10 min and subsequent centrifugation at 18,500 ϫ g for 10 min, the supernatants were transferred to HPLC vials for purification. HPLC purification was done on a Dionex UltiMate 3000 HPLC (Thermo Fisher Scientific) equipped with an Atlantis dC18 column (5 m, 4.6 ϫ 250 mm, Waters, Milford, MA) equilibrated with H 2 O, 0.1% TFA, 7% acetonitrile, and a diode array detector for UVvisible monitoring ( ϭ 280, 370, 450, and 460 nm). An injection volume of 40 l, a temperature of 25°C, and a flow rate of 1 ml min Ϫ1 were used. 8f-FAD was separated from residual free FAD and other impurities for subsequent UV-visible absorption and NMR spectroscopy as well as HPLC/MS analyses using the following program: 0 -25 min, 7-12% acetonitrile; 25-30 min, 95% acetonitrile; 30 -35 min, 7% acetonitrile. 200-l fractions were collected between 12.0 and 18.0 min, and fractions with a pure and typical 8f-FAD UV-visible absorption spectrum were combined. The combined fractions were dried using an ISS110 Savant SpeedVac System at 45°C, Ͻ10 mbar vacuum, and an RH64-11 rotor (Thermo Fisher Scientific), and afterward stored at Ϫ20°C until further use. Purity of the cofactor was controlled by HPLC measurements using the same conditions as above.

H NMR spectroscopy
4 mM solutions of purified 8f-FAD or FAD dissolved in 20% D 2 O in water (v/v) were subjected to 1 H NMR analysis using a Varian INOVA 500 (499.82 MHz, Agilent) spectrometer. 1 H NMR spectra were recorded at 500 MHz at 30°C. The signal of water protons (␦ H 4.75 ppm) was used as the reference for the observed chemical shifts.

8-Formyl-FAD formation in hETF EPR spectroscopy
Electron paramagnetic resonance experiments were performed with 150 M hETF-␣N259A and hETF-␤Y16F purified in 50 mM HEPES/NaOH, pH 8.5, and with 150 M hETF-WT purified in 50 mM HEPES/NaOH, pH 7.0, with n X-band ECS 106 spectrometer (Bruker, Billerica, MA) with 9.45 GHz microwave frequency. A microwave power of 2 milliwatts was used, with a modulation amplitude of 2.0 G at a modulation frequency of 50 kHz. Samples were run at 295 K in 100-l capillaries and 10 scans with a conversion time of 5.12 ms, a time constant of 10.24 ms, and a sweep time of 20.97 s. The magnetic field was scanned for 100 G from 3340 to 3440 G. The magnetic phase and field modulation amplitude of the signal channel of the EPR machine was calibrated with solid DPPH according to the manual to a g factor of 2.0036.

HPLC analysis of 8f-FAD formation
Protein purified at pH 7.0 was diluted to a final concentration of 40 M to adjust the pH to 8.5 and then incubated at 25°C. Samples were taken after 0, (1), 2, 4, 6, 8, and 24 h, and the cofactor was isolated according to the protocol described above. After HPLC analysis, areas of the peaks corresponding to FAD and 8f-FAD were determined, and the ratio between the two chromophores was calculated.

Steady-state kinetics
Steady-state kinetic parameters were determined spectrophotometrically according to Okamura-Ikeda et al. (35) using DCPIP (2,6-dichlorophenolindophenol) as terminal electron acceptor. For the assays, 125 M DCPIP, 100 nM hDMGDH, and 0 -100 M hETF in 50 mM HEPES/NaOH, pH 7.0, were incubated at 25°C for 10 min before the reaction was started by addition of 25 mM dimethylglycine, and the change of absorption was monitored at 600 nm for 3 min. For each concentration, at least a triplicate measurement was performed. From these data the initial velocities were determined, and K m and k cat values were assessed using a non-linear hyperbolic fit in Origin 8.6 (OriginLab Corp., Northampton, MA).

Anaerobic photoreduction
Photoreduction of flavoproteins was done according to the method reported by Massey et al. (36). The experimental procedure of photoreduction and reoxidation was performed as described by Augustin et al. (32). Approximately 20 M purified enzyme in 50 mM HEPES/NaOH, pH 7.0, was reduced at 15°C. Further reduction of hETF-WT was achieved by adding a 10-fold excess of sodium dithionite to the solution.

Thermo FAD thermal stability
The temperature stability of the proteins was determined by monitoring the change in the intrinsic protein fluorescence of FAD in a Thermofluor assay (37). Thermo FAD measurements were carried out with an FX Connect real-time PCR system (Bio-Rad) in 25 l of 50 mM HEPES/NaOH, pH 7.0, and 30 M enzyme. The samples were pre-heated to 25°C, and then the temperature was increased in 0.5°C/min steps to 95°C. Fluorescence data were collected using the FRET channel.
Melting temperatures (T m ) were determined using the CFX Manager 3.0 software (Bio-Rad).