Cloning and Analysis of the Genes for a Novel Electron-transferring Flavoprotein from Megasphaera elsdenii

The genes that encode the two different subunits of the novel electron-transferring flavoprotein (ETF) fromMegasphaera elsdenii were identified by screening a partial genomic DNA library with a probe that was generated by amplification of genomic sequences using the polymerase chain reaction. The cloned genes are arranged in tandem with the coding sequence for the β-subunit in the position 5′ to the α-subunit coding sequence. Amino acid sequence analysis of the two subunits revealed that there are two possible dinucleotide-binding sites on the α-subunit and one on the β-subunit. Comparison of M. elsdenii ETF amino acid sequence to other ETFs and ETF-like proteins indicates that while homology occurs with the mitochondrial ETF and bacterial ETFs, the greatest similarity is with the putative ETFs from clostridia and withfixAB gene products from nitrogen-fixing bacteria. The recombinant ETF was isolated from extracts of Escherichia coli. It is a heterodimer with subunits identical in size to the native protein. The isolated enzyme contains approximately 1 mol of FAD, but like the native protein it binds additional flavin to give a total of about 2 mol of FAD/dimer. It serves as an electron donor to butyryl-CoA dehydrogenase, and it also has NADH dehydrogenase activity.

dimensional structure of the recombinant protein shows that the FAD interacts mainly with the ␣-subunit, while the AMP is buried in the ␤-subunit (6). A deficiency in the human enzyme causes the disease glutaric aciduria type II, an often fatal inborn error of fatty acid and amino acid metabolism, emphasizing the importance of this protein (7).
Proteins similar to mammalian ETF have been isolated from bacteria. ETFs from Paracoccus denitrificans and the methylotroph W3A1 are also heterodimers with one FAD and one AMP per molecule (3, 8 -10). The mammalian and P. denitrificans enzymes share very similar catalytic properties (11). Although the physicochemical properties of methylotroph W3A1 ETF are similar to those of the other ETFs, the methylotroph protein is not able to replace the other ETFs in catalytic assays (10,12). In both organisms, the genes are arranged in tandem with the gene for the ␤-subunit preceding the ␣-subunit gene (13,14). Sequence analysis identified similar genes in a variety of other organisms. Thus, the proteins encoded by the fixB and fixA genes of nitrogen-fixing bacteria have similar sequences to the ␣and ␤-subunits, respectively, of ETF, and their genetic organization is similar to that of bacterial ETF (5,(13)(14)(15). Although there is evidence that the proteins encoded by the fixAB genes are involved in nitrogen fixation, their precise function has not been established (16,17).
The ETF in the anaerobic bacterium Megasphaera elsdenii functions in a similar way to the ETFs described above, but it differs in several of its physicochemical and catalytic properties. The organism is found in the rumen of sheep and cattle, where it ferments lactate (18), disposing of excess reducing equivalents either by forming molecular hydrogen or through the production of short-chain fatty acids (19 -21). The ETF mediates electron transfer from the flavoprotein D-lactate dehydrogenase to a third flavoprotein, termed butyryl-CoA dehydrogenase (BCD), that functions physiologically by reducing enoyl-CoA to acyl-CoA (21)(22)(23)(24). M. elsdenii ETF has two subunits of different molecular mass, and it contains FAD, which acts as its sole redox group (24). However, in contrast to the ETF preparations from other sources described above, M. elsdenii ETF contains approximately 1.4 mol of FAD in the form in which it is isolated, and it binds additional FAD so that when saturated with flavin it has 2 mol of FAD/dimer (24). A variable fraction of the flavin in this ETF comprises two hydroxylated derivatives of FAD: 6-OH-FAD (6-hydroxy-7,8-dimethyl-10(5Ј-ADP-ribityl)-isoalloxazine) and 8-OH-FAD (7-methyl-8-hydroxy-10(5Ј-ADP-ribityl)-isoalloxazine) (24 -29). The hydroxyflavins usually occur in small amounts, but at the highest levels observed they make up about half the flavin in the enzyme. There is evidence that they are artifacts generated during isolation of this ETF (22). Their optical properties differ from those of FAD, and they also affect the catalytic properties of the enzyme. They have not been observed in preparations of ETF from other sources. In further contrast to other ETFs, the M. elsdenii enzyme does not contain AMP (26). In addition, it catalyzes the oxidation of NADH, thus allowing electron transfer from NADH to BCD and to other electron acceptors including 2,6-dichlorophenolindophenol (DCPIP) (22,24).
Incubation of the apoenzyme of M. elsdenii ETF with 8-halogenated FAD derivatives led to covalent attachment of flavin to the ␤-subunit, suggesting that the flavin sites are on this subunit (30,31). The analogue-substituted ETF is reducible by NADH, and it retains diaphorase activity, but it is inactive with BCD. Similar experiments using triazine dyes as affinity labels led to the conclusion that the NADH binding site is also on the ␤-subunit (32). M. elsdenii ETF stabilizes the red anionic form of the flavin semiquinone during reductive titration with dithionite ion, as also occurs with ETFs from other sources (1,24). However, semiquinone is not observed during electrochemical reduction of the enzyme in the presence of mediator dyes, leading to the conclusion that stabilization of the semiquinone is kinetic rather than thermodynamic (33). The midpoint potential for the overall two-electron reduction of the enzyme is Ϫ0.259 V.
The properties of M. elsdenii ETF show that although overall it resembles ETF preparations from other sources, it is unique in binding two molecules of FAD, in lacking AMP, and in catalyzing NADH oxidation. With the aim of using structural analysis to further elucidate the similarities with and differences from other ETFs, we have cloned the genes encoding the M. elsdenii protein. The base sequences of the genes and the predicted amino acid sequences of the subunits have been compared with those of other ETFs. Purified recombinant protein has been produced using an Escherichia coli expression system, and the properties of the recombinant protein have been compared with those of native ETF. These studies lay the foundation for more detailed structural analysis by x-ray crystallography (34) and site-directed mutagenesis. A preliminary report of some of the findings has been published (35).

EXPERIMENTAL PROCEDURES
Growth of Bacteria-Cells of M. elsdenii (strain LC1, NCIMB 8927) were prepared as described previously (24). Escherichia coli strains TG1 and DH5␣ were maintained and propagated aerobically in Luria-Bertani medium (36). Anaerobic cultures of E. coli were also grown in Luria-Bertani medium using the techniques described for M. elsdenii. A large scale culture of E. coli (40 liters) was grown under anaerobic conditions to obtain sufficient quantities of the recombinant protein for biochemical characterization. The growth medium (1% (w/v) tryptone, 0.5% (w/v) yeast extract, 0.5% (w/v) NaCl, 0.5% (w/v) glucose, 17 mM KH 2 PO 4 , 72 mM K 2 HPO 4 ) was supplemented with 200 g/ml ampicillin, and the culture was induced by adding IPTG to a final concentration of 0.5 mM in early log phase when A 590 ϭ 0.13.
Protein Purification-Native ETF was purified from M. elsdenii as described previously (24). Recombinant M. elsdenii ETF was purified from E. coli transformed with pBETFfl.1 as follows. All procedures were carried out at 4°C. The buffers contained 0.3 mM EDTA, and they were at pH 6 unless otherwise noted. A crude extract of E. coli was made by suspending cell paste in 20 mM potassium phosphate buffer, pH 7.0 (3 ml/g); lysing the cells by sonication (Branson sonicator, type 7532B operated at 80% full power); and centrifuging (23,500 ϫ g for 20 min). The extract (350 ml) was applied to a column of Q-Sepharose (42 ϫ 2.1-cm diameter) equilibrated with 50 mM potassium phosphate buffer. The column was washed with 0.1 M potassium phosphate buffer (460 ml), and the ETF eluted using a linear salt gradient made by continuously diluting 500 ml of 0.1 M potassium phosphate buffer with 500 ml of 0.33 M potassium phosphate buffer. Fractions that contained diaphorase activity were pooled, diluted to 50 mM potassium phosphate, and applied to a second Q-Sepharose column (37.5 ϫ 1.6-cm diameter). It was washed with 300 ml of 0.1 M potassium phosphate buffer and eluted with the gradient used earlier except for the total volume of buffer in the gradient, which was 600 ml. Fractions with an absorbance ratio (A 272 :A 450 ) less than 10 were pooled. Solid ammonium sulfate was added to a concentration of 1 M. The solution was then applied to a column of phenyl-Sepharose CL-4B (44 ϫ 2.1-cm diameter) equilibrated with 50 mM potassium phosphate and 1 M ammonium sulfate. A yellow band with bright yellow-green fluorescence was bound tightly to the top of the column. The column was washed with 600 ml of the same buffer, and the ETF was eluted with a linear gradient in which 500 ml of 1 M ammonium sulfate in 50 mM potassium phosphate buffer was continuously diluted with 500 ml of 50% (v/v) ethylene glycol in the same buffer. Fractions were analyzed by SDS-PAGE. Those that gave only two bands corresponding to the two subunits of ETF were pooled, concentrated by ultrafiltration using a PM30 (Amicon Corp.) membrane, dialyzed versus 0.1 M potassium phosphate, and stored at Ϫ20°C.
Enzyme Assays-ETF activity was measured by coupling the oxidation of NADH to the reduction of crotonyl-CoA in the presence of butyryl-CoA dehydrogenase (21,24). The assay contained in a final volume of 1 ml at 25°C 30 M crotonyl-CoA, 60 mM potassium phosphate, pH 6, 0.1 mM NADH, 2.3 M BCD, and ETF. One unit of ETF activity in this assay is defined as an absorbance change of 1 per min at 340 nm.
Diaphorase activity was measured by coupling the oxidation of NADH to the reduction of DCPIP. The reaction mixture contained 0.14 mM NADH, 32 M DCPIP, 0.1 M potassium phosphate buffer, pH 7.0, and ETF in a final volume of 1 ml at 25°C. The decrease in absorbance at 600 nm was measured. One unit of diaphorase activity is defined as an absorbance change of 1 per min.
Analytical Methods-Protein was determined (37) using bovine serum albumin as a standard and a Biuret coefficient of 0.833 (24).
The flavin chromophore was extracted from recombinant ETF by heat treatment in a sealed 1.5-ml centrifuge tube at 90°C (27). The precipitated protein was removed centrifugation (20,500 ϫ g, 10 min).
The enzyme was prepared for N-terminal sequencing by first using SDS-PAGE to separate the subunits. The areas of the gel that contained the two subunits were then excised separately, and the two proteins electroeluted (40). They were precipitated with 12.5% (w/v) trichloroacetic acid. N-terminal amino acid sequencing was carried out by automated Edman degradation using an Applied Biosystems 4778 protein sequencer.
PAGE and SDS-PAGE analyses were performed (39) using a vertical slab gel (ATTO AE-6450). The separated proteins were stained with Coomassie Blue or for NADH dehydrogenase activity as described later.
DNA Manipulations-Cloning and transformation techniques were carried out essentially as described (36). Plasmid DNA was isolated by the alkaline lysis method (36). M. elsdenii genomic DNA was isolated and purified using the procedure described (40). Approximately 2.5 mg of DNA was obtained from 0.12 g of cell paste when the culture was harvested after 10 -12 h of growth.
The polymerase chain reaction (PCR) was used to amplify genomic DNA sequences. Reactions were carried out in a reaction volume of 100 l that contained 0.1 g of genomic DNA; a 1 M concentration each of the primers A and B as given below; 2.0 mM MgCl 2 ; 50 mM KCl; 10 mM Tris-HCl buffer, pH 9.0, 25°C; 0.1% (w/v) Triton ® X-100; a 0.25 mM concentration each of the four dNTPs; 5 units of Taq polymerase (Promega Corp.). PCR amplifications were carried out in a Techne thermocycler. The conditions were as follows: 30 cycles of denaturation, 94°C for 1.5 min; annealing, 40°C for 1 min; extension, 72°C for 3 min. Primer A is nondegenerate: GATTTTAGGAGGCAAAACG. Primer B is degenerate with the sequence T(G/A/T/C)CC(C/T)TC(A/G)AA(C/T)TG-(C/T)TC(G/A/T/C)GC. The reaction products were analyzed by agarose gel electrophoresis. The amplified fragments were electroeluted from the gel and subcloned into the pCR ® 2.1 plasmid (Invitrogen) to generate plasmid pB900. The nucleotide sequences of the fragments were determined, and the correctly amplified sequence was identified by comparing the open reading frames with the N-terminal amino acid sequences of the two ETF subunits.
Nucleotide sequencing was carried out by the dideoxy chain-termination method (41) with [␣-35 S]dATP and Sequenase ® version 2.0 with double-stranded template DNA. Subclones were generated using restriction sites within the cloned segment of DNA. These plasmids were sequenced using M13 primers (Sigma). The DNA sequence was used to design synthetic primers, which in turn were used to sequence the complete fragment. dITP replaced dGTP as necessary when secondary structure in the DNA made it difficult to read the sequence.
Identification and Cloning of ETF Subunit Genes-The DNA fragment amplified by PCR was used as a probe in Southern blot analysis and colony hybridization. The probe was labeled with digoxygenin-11-dUTP (DIG) in a random-primed DNA-labeling reaction (Boehringer Mannheim kit, catalogue number 1175033). Purified genomic DNA was digested with restriction enzymes both singly and in pairs to generate DNA molecules that could be cloned into pBluescript SK ϩ (Stratagene). The restriction digests were separated by electrophoresis in 0.8% aga-rose gel and transferred to nylon membrane (Biodyne ® (42). DNA fragments that contained the genes that encode ETF were identified by direct hybridization of the blots with the DIG-labeled probe and subsequent detection in the specific antibody-mediated chemiluminescence reaction. Larger amounts of DNA were digested with the enzymes that produced fragments of DNA that hybridized to the probe and were likely to contain the full-length genes. Size-selected fragments were isolated from agarose gel and ligated with pBluescript SK ϩ that had been digested with the same enzymes. The ligation products were transformed into E. coli TG1, and the colonies that contained recombinant plasmids were screened by colony hybridization using the DIGlabeled probe. Plasmid DNA was isolated from transformants that contained DNA that hybridized to the probe. These plasmid preparations were than screened by restriction mapping analysis and DNA sequencing. One plasmid with a 2.1-kb insert was found to contain the complete genes for ETF except for the 5Ј-end that encodes the N terminus of the ␤-subunit. A plasmid with the full-length genes was constructed by ligating this insert with the 5Ј-end of pB900 and pBluescript SK ϩ . The plasmid was termed pBETFfl.1.
Expression of Recombinant Genes-E. coli was grown either aerobically or anaerobically at 37°C in the presence or absence of IPTG as indicated. Crude lysates of E. coli were made by resuspending the cell pellets in lysis buffer (50 mM Tris-HCl, 40 mM EDTA, pH 8.0), to which lysozyme was added (1 mg/ml). After incubation on ice (30 min) and centrifugation (18,500 ϫ g, 20 min, 4°C), the supernatants were analyzed by native PAGE using 10% gels and a discontinuous buffer system (39). The gels were stained for NADH dehydrogenase activity using 0.01 mM p-iodonitrotetrazolium violet and 0.1 mM NADH in 0.1 M potassium phosphate buffer, pH 7.
Computer-based Methods-DNA sequence analysis was carried out using the Macmolly package (Softgene, Berlin). The amino acid sequences were aligned using ClustalW (43). The secondary structure of polypeptide sequences was determined using Predict Protein (44 -46).

RESULTS
Cloning and Sequence of ETF Genes-The relative physical proximity and arrangement of the genes for the two ETF subunits within the M. elsdenii genome was not known at the beginning of the investigation. The N-terminal sequences of the ␣-subunit (30 residues) and the ␤-subunit (32 residues) were determined with the intention of using them to generate primers for PCR. However, a sequence similarity search of the protein/DNA data bases using the BLAST search algorithm (47) revealed that the N-terminal sequence of the ␤-subunit of M. elsdenii ETF is encoded by a region downstream from the gene encoding butyryl-CoA dehydrogenase (bcd) (48). The assumption was then made that the gene for the ␣-subunit (etfA) would occur directly downstream from the ␤-subunit gene. Consequently, two oligonucleotide primers were designed and synthesized for use in a PCR amplification of the ETF genes. Primer A was derived from the sequence of bcd (48), beginning 22 bases upstream from the initiation codon of etfB. Primer B was degenerate (256-fold), and it corresponded to the amino acid sequence of the ␣-subunit from residue 19 to 13. The amplified products were cloned into pCR ® 2.1 (Invitrogen). Nucleotide sequencing showed that a 900-base pair PCR product contained base sequence that corresponded to parts of the two ETF subunits; this was designated pB900.
The sizes of the subunits of ETF predicted that the complete ETF genes would occur on a 2.1-kb fragment of DNA. A Southern blot was carried out with M. elsdenii genomic DNA that had been digested with restriction enzymes that were selected from restriction maps of the base sequences for bcd (48) and pB900. Size-selected fragments (2.0 -2.2 kb) produced from the digestion of genomic DNA with PstI and HindIII were ligated into pBluescript SK ϩ to produce a subgenomic DNA library. The labeled 0.9-kb PCR product was used to screen the library. Of 2059 colonies screened, 38 were found to contain DNA that hybridized to the probe. One of these plasmids, with a 2.1-kb PstI-HindIII insert, was found to contain the complete genes for ETF except for the 5Ј-end that encodes the N terminus of the ␤-subunit. A plasmid that contained the full-length genes was constructed by ligation of the 2.1-kb fragment of DNA together with the 5Ј-end of pB900 and pBluescript SK ϩ . This plasmid was designated pBETFfl.1.
The partial restriction map and sequencing strategy used to establish the complete nucleotide sequence of the cloned fragment of DNA is shown in Fig. 1. Each base in the 2102-base pair fragment was sequenced an average of 2.96 times. The amino acid sequence of the ␤-subunit was derived from bases 23-832 (Fig. 2). A putative ribosome-binding site similar to those found in E. coli is positioned 10 bases upstream from the initiation codon. No other possible start sites for translation occur upstream from the ATG codon. The gene is terminated by a single TAA stop codon. It encodes a polypeptide with 270 amino acids and a molecular mass of 29,081 Da. The gene etfA that encodes the ␣-subunit of ETF is separated from etfB by 20 bases. The amino acid sequence ␣-ETF was derived from bases 856 -1869. This gene is also preceded by a putative ribosomebinding site (bases 838 -845), 10 bases upstream from the ATG initiation codon, and it is terminated by a single TAA stop codon. It encodes 338 amino acids, and the polypeptide has the molecular mass 36,101 Da. The two genes are translated in different reading frames, indicating that although they might be transcribed as a single polycistronic mRNA molecule, they are translated as individual polypeptides similar to other bacterial ETFs. A nucleotide sequence that resembles a -independent transcription termination site that occurs in E. coli is located downstream from etfA between bases 1887 and 1925 (Fig. 2). Although several stem-loop structures were identified, only one fulfills the requirements of a -independent terminator (49). The free energy of formation calculated for this RNA hairpin structure using the program Mfold (50) is Ϫ15.1 kcal/ mol. This value is similar to values calculated for similar structures in E. coli (51). The G/C content of etfA and etfB is 50.15 and 49.45%, respectively. There is a slight bias in favor of G (16.8%) and C (37.4%) at the third position of codons as reflected in the G/C content of M. elsdenii genomic DNA (53.6%).
Amino Acid Sequence Comparisons-The BLAST search algorithm (47) was used to search the GenBank TM , Swiss-Prot, and PIR data bases for amino acid sequences that show similarity to the amino acid sequences of the subunits of M. elsdenii ETF. The two subunits were treated as separate polypeptides and used independently to search the data bases. Earlier analyses identified similarity between ETF-like sequences from eukaryotes and prokaryotes (14,52,53). The present analysis shows that the sequences of the subunits of M. elsdenii ETF are similar not only to these sequences, but also to additional sequences that have become available more recently (Table I). They are most similar to putative ETFs from the anaerobic bacteria Clostridium acetobutylicum and Clostridium thermosaccharolyticum. The genetic organization of the etfA and etfB genes in all three organisms is also similar, suggesting that the proteins have similar physiological functions. All of these ETF genes resemble the fixAB genes of N 2 -fixing bacteria. The fixAB genes are proposed to function in nitrogen fixation, but their role has not been established, nor have they been investigated at the protein level. It is interesting that although M. elsdenii does not fix nitrogen, its ETF resembles the fixAB gene products more closely than the well characterized ETFs from other bacteria and mammals.
Confirming previous analyses (14,53), multiple alignment of the amino acid sequences of the larger of the two subunits of ETF and FixB polypeptides shows that there is much sequence similarity between the C-terminal halves of these proteins but little similarity toward the N terminus. In contrast, the corresponding alignment for the small subunits shows that similarities occur throughout the polypeptide.
The FAD-binding site of human ETF involves mainly the C-terminal region of the ␣-subunit (6). Eighteen residues occur within hydrogen bonding distance of the FAD molecule, and of these, 16 are contributed by the ␣-subunit and 2 by the ␤-subunit. Fourteen of the residues are conserved in M. elsdenii ETF (Fig. 3A), making it likely that one of the two FAD molecules is bound at this site. Two substitutions occur for residues that in the human enzyme interact with the isoalloxazine moiety; residue His ␣286 (alignment position 301) that bonds to the oxygen of the carbonyl at C(2) of the flavin is replaced by proline in M. elsdenii ETF (Pro ␣287 ), and residue Tyr ␤16 (not shown) whose side chain is 3.6 Å from the methyl group at flavin C(8) in the human protein is replaced by Thr ␤13 in M. elsdenii ETF. The residue His ␣286 is within 2.9 Å of the carbonyl oxygen at C(2) of the isoalloxazine moiety in the human protein and is postulated to function in the stabilization of the anionic semiquinone (6). only kinetic, while in other bacterial ETFs and mammalian ETF both kinetic and thermodynamic stabilization of the anionic semiquinone occurs.
The other two replacements that occur are for residues involved in binding the AMP part of FAD in the human enzyme, and both are conservative changes: Lys ␣301 (alignment position 239) to Asn ␣225 and Leu ␣319 (alignment position 343) to Val ␣320 . As has been identified in other ETFs (14), this part of the M. elsdenii ETF sequence is similar to a fingerprint sequence that is diagnostic of a ␤␣␤ fold that is known to be involved in the binding of the ADP moiety of FAD and nicotinamide dinucleotides ( Fig. 4A) (54). Its consensus sequence has three highly conserved glycine residues flanked by small hydrophobic residues, with an acidic residue occurring at the C terminus of the sequence (55) (6,14) and also in alcohol dehydrogenase from yeast (56). The substitution for the second glycine of the consensus sequence has not been noted before in an FAD-binding protein. An aliphatic residue in the N-terminal end of the consensus sequence is replaced by tyrosine in M. elsdenii ETF, as also occurs in other FTFs and in glutathione reductase of human erythrocytes (57). Furthermore, the secondary structure of the residues surrounding these amino acid residues is in the form of a ␤␣␤ fold, which is diagnostic of a flavin-binding site (Fig. 4A).
The sequence similarities with the human enzyme around the FAD-binding site and the presence of a dinucleotide-binding motif provide strong evidence that an FAD binds in this part of the M. elsdenii sequence. However, two additional dinucleotide-binding sites are likely in M. elsdenii ETF, one for the second molecule of FAD and one for NADH. A possible site was identified close to the putative FAD-binding site described above toward the N terminus of the ␣-subunit (Fig. 4B). In this second sequence, only the first Gly of the GXGXXG motif is conserved. However, the sequence is similar to that of the NAD(P)H binding site of mercuric reductases (55). The second Gly is replaced by a serine, and the third Gly is replaced by an alanine, as also occurs in the NAD(P)H binding site of glutathione reductases from E. coli and humans (55). A hydrophobic residue replaces the C-terminal acidic residue of the consensus sequence. Again, a similar substitution occurs in the NAD(P)H binding site of mercuric reductases. It is possible that this site in M. elsdenii ETF binds FAD or NADH. While at first sight it appears unlikely that two FAD sites could involve two sets of amino acids so close in the sequence, there is spectroscopic evidence that the two flavins are not too far apart. Thus, binding of the second FAD to ETF causes an unexpectedly large increase in absorbance around 405 nm (Ref. 24; see later), and measurements by time-resolved polarized fluorescence anisotropy decay suggest that the centers of the isoalloxazine moieties of the two flavins are 12-23 Å apart (58). The secondary structure around the residues at the second putative dinucleotide-binding site on the ␣-subunit is not predicted to be in the form of a ␤␣␤ fold. If FAD is bound at this site, the lack of a classic nucleotide-binding fold might result in weaker binding of the flavin and its consequent loss during enzyme isolation. It should be noted, however, that the human enzyme does not bind a second FAD but that it shares much sequence similarity in this region with M. elsdenii ETF.
The M. elsdenii subunit sequences were also analyzed for two different motifs known to be associated with the binding FAD and/or NAD in other proteins. One of these is a proline-rich sequence that is thought to be involved in binding the AMP moiety of FAD in human ETF-CoQ oxidoreductase and in succinate dehydrogenase from E. coli, yeast, and ox (57). The other is a sequence that is common in flavoprotein hydroxylases with a putative dual function in binding FAD/NAD(P)H molecules (59). Sequences that match these motifs do not occur in M. elsdenii ETF. The AMP molecule in the human enzyme is buried in the ␤-subunit. The eight residues in human ETF that are in hydrogen bonding distance of the AMP molecule were located using Insight II (60). Three of these residues are completely conserved in the M. elsdenii enzyme, and one other residue occurs as a conservative substitution (Fig. 3B). In addition, much similarity between the human and M. elsdenii proteins occurs in residues surrounding these positions. Although M. elsdenii ETF does not bind AMP (26), it is possible that it binds either the second FAD at this site or the substrate NADH. The residues concerned are also moderately conserved in other ␤-ETFs and in FixA sequences, suggesting that these proteins also may have a dinucleotide-binding site in this region.
It was of interest to determine whether the amino acid sequence similarities between ETF and the other proteins of Fig.  3 extend to structural homology. Predictions of secondary structure were made for the two subunits of ETF from M. elsdenii and were compared with those predicted by the same method for human ETF as well as those observed in the crystal structure. Predictions were also made for the FixAB polypeptides of Azorhizobium caulinodans, and these are included for comparison (Fig. 5). The secondary structures determined for the human enzyme by theoretical analysis of the amino acid sequence with Predict Protein (44 -46) are in good agreement with those determined from the three-dimensional structure (6,61), thus providing confidence in the theoretical analysis of sequences for which three-dimensional structures are not available. The computer program used to analyze the amino acid sequences recognizes ␣-helices or ␤-sheets and loop regions, but it does not recognize 3/10 helices or ␤-bridges. A bend or a H-bonded turn observed experimentally appears as a loop in the theoretical analysis.
The two subunits of human ETF comprise three separate domains: domain I, the N-terminal part of the ␣-subunit; domain II, the C-terminal part of the ␣-subunit and a small C-terminal part of the ␤-subunit; and domain III, composed of most of the ␤-subunit (6). As noted above, the sequences of the larger polypeptides (␣-ETF/FixB) are identical at many posi- Table I. Me., M. elsdenii. The sequences were aligned using ClustalW. A dash indicates a position at which a gap is inserted to optimize the alignment; an asterisk indicates a residue that is invariant; a colon indicates a residue that is highly conserved; a period indicates a residue that is moderately conserved in the sequences. ¥, q, and ‡ indicate residues that are involved in binding the isoalloxazine, ribityl chain, and AMP moieties of FAD, respectively. Ϯ indicates a residue that interacts with AMP in the ␤-subunit. The residue numbers in the alignment sequence are given, with the numbers for M. elsdenii ETF in parentheses (see Fig. 2). The open reading frames of the fixA genes from E. coli k12, 1 min and 32 min, are longer than those given above (63); we suggest that their initiation codons were incorrectly assigned and that methionine at alignment position 7 is the correct start codon.

FIG. 4. Amino acid sequences in the ␣-subunit of M. elsdenii ETF that are similar to the dinucleotide-binding consensus sequence for FAD/NAD(P)H-binding proteins.
A and B show the proposed dinucleotide-binding sites in the ␣-subunit. The amino acid positions according to the sequence of M. elsdenii ETF are given (see Fig. 2). The secondary structure was derived using Predict Protein (46 -48). H, E, and L represent ␣-helix, ␤-sheet, and loop, respectively. Only secondary structures that are predicted with a reliability Ͼ6 (scale 0 -9) are shown. tions in domain I, and this is reflected in the very similar secondary structure predicted for this region. Furthermore, the proteins from the three different sources are all predicted to have a Rossmann fold (62) at the conserved amino acids that in human ETF are involved in binding the ADP moiety of FAD ( Fig. 5A; alignment positions 294 -329). Very little sequence identity occurs in domain II, indicating that a more rapid sequence divergence has occurred in this domain. Nevertheless, the secondary structure predicted for this region is almost identical for the three proteins, suggesting that the function of domain II is common to all three proteins.
The secondary structures of the small subunits (domain III) are also strikingly similar throughout their sequences (Fig.  5B). The conserved sequences around the AMP-binding site of the human protein (alignment positions 121-155) have a ␤␣␤ structure that is similar to a Rossmann fold, a feature that is not revealed by analysis of the amino acid sequences. As described above, it is possible that in M. elsdenii ETF this is either the NADH binding site or the site at which the second FAD binds. Furthermore, a loop (alignment positions 227-235) has been extended in the M. elsdenii sequence that may allow greater access to such a dinucleotide-binding site. The loop is also extended in the A. caulinodans sequence, with the implication that this enzyme may also bind a second dinucleotide and/or catalyze the oxidation of NADH.
Expression of Recombinant ETF-An overall aim of the present investigation was the expression of the recombinant genes for M. elsdenii ETF in E. coli and purification of the recombinant protein in an amount sufficient for biochemical characterization. The genes that encode M. elsdenii ETF were inserted into pBluescript in the appropriate orientation for transcription to be controlled by the lac promoter (pBETFfl.1). The native ribosome-binding sites are located upstream from each subunit gene. However, a promoter region is not present. It was shown that the plasmid is lost from E. coli during culture at 37°C with the inference that the plasmid or its gene product is toxic to the cells. An enzyme stain that measures diaphorase activity showed that when extracts of E. coli are analyzed by PAGE, expression of M. elsdenii ETF is not detectable in cultures grown aerobically with or without IPTG (0.1-1.0 mM). However, when E. coli is grown anaerobically, the plasmid is retained by the cells, and recombinant protein is expressed at a low level. The level of expression is increased by the addition of IPTG to early log phase cultures, indicating that the genes are under the control of the vector promoter. Maximal expression of protein was observed with 0.5 mM IPTG (data not shown).
Purification and Properties of Recombinant ETF-Recombinant M. elsdenii ETF was purified from an extract of E. coli TG1 (pBETFfl.1) that had been grown anaerobically (Table II). The final preparation gave only two bands of protein after SDS-PAGE, indicating that the protein was at least 95% pure (Fig. 6). The absorbance ratio, A 272 :A 450 , was 6.1, similar to that reported for native ETF (24). The yield of pure protein was The secondary structure was derived using the Predict Protein program (46 -48). H, E, and L represent ␣-helix, ␤-sheet, and loop, respectively. Only secondary structures that are predicted with a reliability Ͼ6 (scale 0 -9) are shown. The secondary structure in the crystal structure of human ETF (HsA) was derived using the "sstruc" module of Procheck (61). B, ␤-bridge; E, ␤-sheet; G, 3/10 helix; H, ␣-helix; S, bend; T, hydrogen-bonded turn; e, extension of ␤-sheet; g, extension of 3/10 helix; h, extension of ␣-helix. An asterisk indicates a position at which the amino acid is identical. A dash indicates a position where a gap was inserted to optimize the alignment of the amino acid sequences. approximately 1 mg/g of E. coli cell paste, and it is therefore greater than the yield of native ETF from M. elsdenii (0.27 mg/g cell paste). The molecular masses of the subunits of the recombinant protein were estimated from SDS-PAGE analysis to be 35.5 and 29.5 kDa for the ␣and ␤-subunit, respectively (Fig. 6). These values are in close agreement with those calculated from the predicted amino acid sequence (36.1 and 29.1 kDa). The electrophoretic mobilities of native and recombinant ETF are identical (Fig. 6). Hence, the somewhat greater values reported earlier for the subunit molecular masses of the native form of this ETF are erroneous (41 and 33 kDa; Ref. 24).
The absorption spectrum of the recombinant ETF has maxima at 273, 373, and 451 nm and minima at 314 and 398 nm. These values agree closely with those determined for native ETF in the present investigation and with those reported earlier (24) ( max ϭ 272, 375, 450, and 660 nm; min ϭ 312 and 398 nm) (Fig. 7). However, some differences occur between the spectra of the different preparations. The absorbance at 660 nm of the native enzyme is much greater that than of the recombinant protein. The spectrum of the native enzyme shows pronounced shoulders at 430 and 510 nm as well as higher absorbance at 400 nm compared with that of the recombinant enzyme. These features indicate that the preparation of native enzyme contained hydroxylated flavins (24). The long wavelength absorbance at 660 nm is characteristic of 6-OH-FAD (green), and the shoulder at 510 nm is characteristic of ETF that contains 8-OH-FAD (orange) (27). It is clear that the recombinant protein contains little if any hydroxylated FAD.
The absorption spectrum of the flavin extracted by heat treatment of recombinant ETF was found to be identical to that of authentic FAD. The extinction coefficient of the bound flavin was calculated to be 11,300 M Ϫ1 cm Ϫ1 . The value is lower than reported for native ETF (24), but the difference may reflect the differences in the flavin composition of the two preparations. The FAD content of the recombinant protein was determined using the absorption coefficient, protein analysis using the Biuret method, a Biuret coefficient of 0.833 (24), and a value of 65,958 Da calculated for the molecular mass of ETF from the predicted amino acid sequences of the two subunits plus one molecule of FAD. The flavin content of the isolated recombinant ETF was calculated to be 1.06 mol of FAD/mol of protein. This is lower than the flavin content reported for native enzyme (approximately 1.4 mol/mol; Ref. 24), a difference that may reflect the different values for extinction coefficient and molecular mass determined in the present investigation. The value becomes 1.17 when the extinction coefficient (12,500 M Ϫ1 cm Ϫ1 ; Ref. 24) determined earlier is used.
Recombinant ETF binds additional flavin after purification, as also occurs with native ETF. The binding of additional FAD was monitored by the difference spectrum between bound and free FAD. This shows maxima at 405 and 500 nm and minima at 360 and 445 nm (Fig. 8). The flavin-saturated recombinant protein contains 1.6 -1.8 mol of flavin/mol of protein, a value that is taken to indicate that, as reported for the native enzyme (24), M. elsdenii ETF contains two FAD-binding sites. The absorbance at 450 nm of the extra flavin that is bound does not change relative to that of free FAD, but a large increase occurs around 400 nm with the result that the trough at 400 nm in the spectrum of isolated ETF is filled in (Fig. 7). Again, the changes seen with the recombinant protein are similar to those reported earlier for the interaction of FAD with native ETF (24). The dissociation constant calculated for the extra FAD bound (0.5-0.6 mol/mol) is approximately 1 M (Fig. 8). The visible absorption spectrum of the purified recombinant protein does not change when one molar equivalent of FMN or AMP is added, suggesting that these compounds are not able to replace the additional FAD.
The recombinant ETF catalyzes the transfer of electrons from NADH to BCD and to DCPIP. Specific activities were determined for ETF as it was isolated and for ETF to which FAD had been added. Two assays were carried out: a diaphorase assay (coupling the oxidation of NADH to the reduction of DCPIP) and an ETF assay (coupling NADH oxidation to the reduction of crotonyl-CoA via BCD) ( Table III). The diaphorase activity of the isolated recombinant ETF is similar to that determined in the present work for the native enzyme and also similar to the value reported by Whitfield and Mayhew (24). The effect of saturating the ETF preparation with FAD seems to vary somewhat with the preparation. Very little effect was observed in the present study and in an earlier report (32); in contrast, Whitfield and Mayhew (24) reported that the activity/ flavin ratio in this assay decreases when additional FAD binds.
The activity of recombinant ETF in the ETF assay with BCD is similar to that of the native enzyme as reported by Whitfield and Mayhew (24). The activity of the native ETF made in the present work is low by comparison. These differences probably reflect to some extent the difficulties inherent to this assay. The activity measured in the ETF assay depends on the activity of the BCD used, and it is therefore difficult to compare the activity measurements made in different laboratories. A 1.5-1.7-fold increase in the activity occurs when native ETF is saturated with FAD, whereas the change observed with the recombinant ETF is much smaller. It appears that the additional flavin that binds has very little effect on the activity of the enzyme in the ETF assay. The ETF activity of the native preparation used in the present investigation was low, but the preparation had been stored for several years at Ϫ20°C before assay, and it is possible that a partial loss in activity may have occurred during storage. Another factor that affects the activities of ETF preparations from M. elsdenii is their content of hydroxylated FAD. Preparations with 6-OH-FAD and 8-OH-FAD have diaphorase activity, but they fail to catalyze BCD reduction in the ETF assay (25). DISCUSSION The structural genes that encode the two different subunits of M. elsdenii ETF have been cloned, sequenced, and expressed in E. coli. Comparison of the deduced amino acid sequences of the subunits with those of the other well characterized ETFs dispels any doubt that this ETF is a member of the ETF family. However, it has also been shown that M. elsdenii ETF is more closely related to the putative ETFs of clostridia and to the fixAB gene products of nitrogen-fixing bacteria. Alignment of the M. elsdenii sequence with that of human ETF provides strong evidence for a nucleotide-binding site in each of the two subunits. Sequence homology to the consensus dinucleotidebinding motif and homology to similar sites in mercuric reductases provide circumstantial evidence that the third nucleotidebinding site is on the ␣-subunit. These tentative conclusions do not support earlier analyses with reactive flavins and dyes that proposed that all three binding sites would be found on the ␤-subunit (30 -32). As was pointed out by Roberts et al. (6) for the human enzyme, the orientation of the two subunits may mean that binding of modified flavins and dyes to the ␣-subunit allows their reaction with groups on the ␤-subunit.
The analysis of the secondary structure in the human, M. elsdenii, and A. caulinodans polypeptides highlights the struc- a One unit of activity is defined as an absorbance change of 1 per min at 600 nm and 25°C in a 3-ml standard assay. A 1-ml assay was used in the present study and was corrected accordingly. tural similarities between these proteins. Although the overall level of amino acid sequence identity between the proteins is low, the secondary structural similarities are high throughout the length of the polypeptides. This suggests that the conformations of the M. elsdenii and A. caulinodans proteins are very similar to that of the human protein and provides further evidence that ETF proteins have evolved from a common ancestor.
The genes that encode M. elsdenii ETF are only expressed in E. coli TG1 when the cells are grown under anaerobic conditions. The underlying reason for this unexpected finding has not been investigated. It is possible that M. elsdenii ETF short circuits electron-transfer pathways in E. coli by coupling the oxidation of a crucial intermediate such as NADH or a flavoprotein dehydrogenase to O 2 . Genes that are similar to the genes for M. elsdenii ETF have been identified in E. coli (63). It is proposed that they function in electron transfer to carnitine and that they are only expressed under anaerobic conditions, lending support to the view that a protein such as M. elsdenii ETF is toxic to aerobic E. coli.
The chemical and catalytic properties of the recombinant protein are very similar to those reported for native ETF. The differences in the spectroscopic and enzymic properties of the recombinant and native proteins are no greater than have been observed between different preparations of native ETF. They may be due in part to differences in the flavin composition of the different preparations. The total flavin in the enzyme isolated varies between 1.06 and 1.4 mol of FAD/mol of protein, but all preparations bind additional flavin to give approximately 2 mol/mol of protein. In addition, the content of 6-OH-FAD and 8-OH-FAD varies with each preparation for reasons that are not yet clear. The AMP that is bound to the ␤-subunit of human ETF is thought to have only a structural role in the protein (2). It is possible that one of the two FAD molecules that binds to M. elsdenii ETF plays a similar structural role.
The function of M. elsdenii ETF is to transfer electrons from D-lactate dehydrogenase and NADH to butyryl-CoA dehydrogenase. This ETF may therefore be useful as a model to investigate inter-and intramolecular electron transfer processes. The successful cloning of the genes that encode the ETF and their expression in E. coli will allow a more comprehensive investigation of the physicochemical and catalytic properties of M. elsdenii ETF than was possible previously. The native enzyme has been crystallized (34) with the expectation that it will be possible to determine the three-dimensional structure of the protein by x-ray crystallography. The crystal structure should show whether the possible FAD and NADH binding sites identified in the present study are correct. In addition, it will be possible to carry out complementary site-directed mutagenesis to explore the structure and function of the protein.