The cytochrome subunit is necessary for covalent FAD attachment to the flavoprotein subunit of p-cresol methylhydroxylase.

When p-cresol methylhydroxylase (PCMH) is expressed in its natural host Pseudomonas putida, or when the genes of the alpha and beta subunits of the enzyme are expressed together in the heterologous host Escherichia coli, flavin-adenine dinucleotide (FAD) is covalently attached to Tyr384 of the alpha subunit and the correct alpha 2 beta 2 form of the enzyme is assembled. The apoflavoprotein has been expressed in E. coli in the absence of the beta cytochrome c subunit and purified. While noncovalent FAD binding to apoflavoprotein in the absence of the cytochrome subunit could not be directly demonstrated, circumstantial evidence suggests that this indeed occurs. Covalent flavinylation requires one molecule each of FAD and cytochrome for each flavoprotein subunit. The flavinylation process leads to the 2-electron-reduced form of covalently bound FAD, and the resulting alpha 2 beta 2 enzyme is identical to wild-type PCMH. This work presents clear evidence that covalent flavinylation occurs by a self-catalytic mechanism; an external enzyme or chaperon is not required, nor is prior chemical activation of FAD or of the protein. This work is the first to define the basic chemistry of covalent flavinylation of an enzyme to produce the normal, active species, and confirms a long standing, postulated chemical mechanism of this process. It also demonstrates, for the first time, the absolute requirement for a partner subunit in the post-translational modification of a protein. It is proposed that the covalent FAD bond to Tyr384 and the phenolic portion of this Tyr are part of the essential electron transfer path from FAD to heme.

Since the discovery of enzymes containing covalently bound flavin-adenine dinucleotide (FAD) or flavin mononucleotide (FMN) over 40 years ago, five types of covalent bonds between flavin and proteins have been revealed (1, 2) (Fig. 1). In all this time, two major questions have remained unresolved: (a) what is the reason for covalent flavin binding in enzymes, and (b) what is the mechanism for covalent attachment of flavin?
No completely satisfying explanation has been provided for the first question. The attachment of an aminoacyl group to the 8␣or 6-position of riboflavin does not confer any unusual properties on the flavin, either free in solution or in an enzyme, with the exception of the ultraviolet-visible spectrum of 6-Scysteinylriboflavin, which is quite different from that of other forms of free and bound aminoacyl flavins (2,3). While the oxidation-reduction potentials (E m 7 ) are about 50 -60 mV more positive for aminoacyl flavins than for unmodified forms (2), this increase in potential can also be achieved by noncovalent interactions with protein. Additionally, enzymes with covalently bound flavins do not catalyze a unique or specific set of reactions.
As to the second question, it is known that 2-electron reduction of protein-free 8␣-O-tyrosylriboflavin or 8␣-S-cysteinylsulfonylriboflavin cause expulsion of the aminoacyl groups, thus producing unmodified, oxidized riboflavin (2,4). The principle of microscopic reversibility suggests that the reverse reaction could occur within an enzyme, with or without intervention of an external enzyme. A mechanism for covalent flavinylation, which has long been in the literature, is shown in Fig. 2 (5,6). An analogous mechanism was proposed for nonenzymic basecatalyzed nucleophilic attack at the 8␣-carbon of riboflavin derivatives in organic solvents (7).
It has been suggested that covalent tethering might require prior activation of the flavin or proteins, e.g. a high energy phosphate bond (8). Of several enzymes studied to date, no specific enzyme has been implicated in the covalent modification process, which contrasts with examples of nonflavin-cofactor covalent attachment to apoenzymes (9 -13).
The structural genes of several bacterial enzymes containing covalently bound flavin have been cloned into vectors for expression in new hosts: succinate dehydrogenase, fumarate reductase (14,15), sarcosine oxidase (16), 6-hydroxy-D-nicotine oxidase (6-HDNO) 1 from Arthrobacter aerogenes (15) (all containing 8␣-N 3 -histidyl-FAD), trimethylamine dehydrogenase (6-S-cysteinyl-FMN) (17), and p-cresol methylhydroxylase (PCMH) (8␣-O-tyrosyl-FAD) (18) (Fig. 1). In some cases, normal enzymes with covalently bound flavin are produced. For the latter three, enzymes without bound flavin could be recovered. Further, only apo-6-HDNO and apo-PCMH could be subsequently converted to the holo forms in vitro. Analytical amounts of apo-6-HDNO were obtained from a pure Escherichia coli-expressed ␤-galactosidase-6-HDNO fusion protein (19). The in vitro flavinylation of 6-HDNO required Mg 2ϩ and mercaptoethanol and was accelerated by millimolar amounts of three-carbon compounds such as glycerol 3-phosphate, glyceraldehyde 3-phosphate, phosphoenolpyruvate, or 45% glycerol (1,15,20). The phosphorylated compounds do not chemically activate FAD, but possibly act allosterically. It also has been reported that citric acid cycle intermediates accelerate covalent flavinylation of succinate dehydrogenase and fumarate reductase (14). These observations, together with results from GroEdependent 6-HDNO refolding and flavinylation studies (21), and the characterization of a series of Cys to Ser replacements (22) in 6-HDNO, suggest that in vivo covalent incorporation of FAD occurs during the folding process after the entire protein chain has been synthesized. Contrary to expectation (21), a specific GroE/6-HDNO complex is not formed. It is not clear why such a myriad of factors should affect covalent flavinylation of 6-HDNO, and the relationship of these agents to in vivo A. aerogenes factor(s) is unknown. No experiments reported heretofore directly addressed the chemistry of covalent flavinylation.
In experiments with fumarate reductase, His 44 , which covalently binds to FAD, was replaced with Cys, Tyr, Arg, or Ser, amino acids capable of covalently binding to FAD at the 8␣carbon. All site-specifically altered forms displayed lowered activity, relative to the normal enzyme, but in no case was FAD covalently bound (14). Obviously, precisely defined conditions are necessary for covalent flavinylation, in order to properly position the aminoacyl nucleophile, and the electrophilic center of the flavin.
The flavocytochrome PCMH oxidizes p-cresol (4-methylphenol) to 4-hydroxybenzyl alcohol and then oxidizes this alcohol to 4-hydroxybenzaldehyde in the periplasmic space of Pseudomonas putida and related bacteria (23). Electrons from these oxidations are funneled into the membrane electron-transport chain of the bacterium for ATP production. PCMH has an ␣ 2 ␤ 2 structure consisting of two ␣ flavoprotein subunits (PchF), each with covalently bound FAD, and two ␤ c-type cytochrome subunits (PchC). PCMH and its related 4-ethylphenol methylenehydroxylase are the only proteins known to contain FAD covalently bound via the 8␣-carbon of the flavin to the phenolic oxygen of a tyrosyl residue. The genes of both subunits (pchF and pchC) of form A of PCMH from P. putida NCIMB 9869 were cloned, sequenced, and expressed in E. coli (18). Whether PchC is expressed in the absence of or together with PchF, heme is covalently bound. When pchF and pchC are expressed together in E. coli, PchC is produced at about 20% the molar level of PchF. E. coli extracts were combined so that the final mixture contained equimolar amounts of PchC and PchF. PCMH purified from this mixture was identical to PCMH expressed by wild-type P. putida 9869. Gram quantities of pure heterologously expressed PCMH could be obtained in this way.
Fractions containing PchF or PchC from chromatography steps were identified using nitrocellulose dot-blot antibody detection. The dot-blots were developed as for Western blot analysis described below.
PchF and PchC (partially purified from extracts of E. coli/pKK-CF) were separated from each other by a modified isoelectric focusing procedure (27). One gram of washed Sephadex G-200 Superfine was suspended and hydrated overnight in a 1:16 (v/v) solution of Pharmacia Pharmalytes (equal amounts of pH 4 -6.5 and pH 6.5-9 Pharmalytes). The slurry was layered onto a glass plate (13 ϫ 7.5 ϫ 0.2 cm, length ϫ width ϫ depth) and dried to 75% of the original weight. The samples (1-50 mg of protein in 50 -500 l) were applied directly to the gel and focused at a constant power of 7 watts at 4°C. The separated subunits were eluted from the Sephadex matrix by repeated extractions with 16 mM potassium phosphate, pH 7.0, and then concentrated.
Immunodetection and Activity Staining-For native gels, cell-free extracts were prepared by a modified published method (28). E. coli cultures grown overnight at 37°C were harvested and resuspended with cell suspension buffer (10 mM Tris HCl, pH 7.4, 1 mM EDTA) containing 2 mg/ml lysozyme. After a 1-h incubation at room temperature, cell debris was centrifuged away and washed with cell suspension buffer, and the supernatants were combined. The centrifuged supernatants were diluted with 4 volumes of sample buffer (75 mM Tris-HCl, pH 6.8, 50% glycerol, 0.25% bromphenol blue) and loaded onto a 7.5% native gel.
Following PAGE of the cell lysates under nondenaturing conditions (29), gels were soaked in 50 mM Tris-HCl buffer, pH 7.5, for 10 min with occasional shaking, transferred to an activity-staining solution (50 mM Tris-HCl, pH 7.5, 5 mM p-cresol, 2 mg/ml nitro blue tetrazolium, and 0.2 mg/ml phenazine methosulfate), and incubated for 30 min in the dark, at room temperature. The gels were rinsed with distilled H 2 O and washed three times, for 5 min each time, with 10% (v/v) acetic acid to fix the color.
For the denaturing gels, cells were lysed in SDS sample buffer, centrifuged, and the soluble fraction separated by SDS-PAGE (30). Protein concentrations were determined by the bicinchoninic acid method (Pierce). For immunodetection, PCMH and its subunits were electrophoretically transferred to nitrocellulose after electrophoresis (31). Rabbit polyclonal antibodies against PchF or PchC were used in conjunction with the secondary antibody-alkaline phosphatase conjugate. Color development reagents were obtained from Promega Corp. (Madison, WI), and used according to the supplied protocol.
Analyses for Covalently Bound FAD-Aliquots containing 1 mg of protein were brought to 1 ml with H 2 O and then mixed with 0.1 volume of 55% trichloroacetic acid. Samples were put on ice for 10 min and then centrifuged. Supernatants were recovered and analyzed for free flavin. The pellets were washed twice with 1% trichloroacetic acid and resuspended with 0.1 ml of Tris (pH 8.2). Samples were digested for 4 h at 37°C by adding 10 l of a solution containing 10 mg/ml each trypsin and chymotrypsin. The solution was brought to 5% trichloroacetic acid and centrifuged. Recovered trichloroacetic acid solutions were heated at 100°C for 5 min to convert FAD to FMN/riboflavin. After cooling, the solutions were neutralized to pH 6 with 2 M KHCO 3 . Flavin contents were measured by fluorescence with a Hatachi F-4010 Fluorescence Spectrophotometer (Hatachi Instruments, Inc., San Jose, CA), using riboflavin as a standard: excitation ϭ 445 nm; emission ϭ 525 nm. After the initial measurement, a known volume of a standardized riboflavin solution was added to each sample. A fluorescence intensity increase less than that measured for the same amount of riboflavin in buffer alone would have indicated internal fluorescence quenching in a particular sample. No such quenching was observed.
Flavin Peptide Analysis-PCMH was generated by incubating 128 nmol of E. coli-expressed apo-PchF 2 , 188 nmol of E. coli-expressed PchC, and 325 nmol of FAD in 2.1 ml of 16 mM potassium phosphate buffer, pH 7.0, for 2 h at room temperature. The buffer and excess FAD were removed by three concentration/H 2 O dilution steps using a Centricon-10 (Amicon, Danvers, MA). PchF and PchC were dissociated from each other by isoelectric focusing at 4°C, in a 13 ϫ 7.5 ϫ 0.2-cm (length ϫ width ϫ depth) bed of hydrated Sephadex G-200 Superfine, using 1:16 diluted Pharmacia Pharmalytes (pH 4.5-6.0) (see above) (27). After focusing, the yellow, covalently flavinylated PchF band was removed from the bed and the gel eluted with 16 mM potassium phosphate buffer, pH 7.0. The yield of recovered PchF was 131 nmol in 2 ml. Trichloroacetic acid (200 l of 55% (w/v) solution) was added to the solution of PchF. The centrifugation pellet was washed with H 2 O and recentrifuged. The pellet was suspended and sonicated in 1 ml of H 2 O ϩ 20 l of 88% formic acid. Pepsin (50 l of a 10 mg/ml solution) was added to the suspension, which was incubated at 37°C. The reaction was followed by HPLC using a 15 ϫ 0.4-cm C-18 Spherex octadecylsilane silica gel column, 3-m particle size (Phenomenex, Torrance, CA), eluting with a gradient from 0 to 100% CH 3 CN in H 2 O over 30 min. The solvents contained 0.1% trifluoroacetic acid, and the flow rate was 1 ml/min. The reaction was quenched after 4.5 h by adding trifluoroacetic acid to 0.1% at 0°C. Two partially pure flavin-containing peptide fractions were obtained by this HPLC method with three injections of the peptic digest (retention times: 1, 13. Steady-state Kinetic Assays-For these assays, reduction of PES by p-cresol-reduced PCMH was monitored at 600 nm due to the subsequent reduction of 2,6-dichlorophenol indophenol. The assays were done in 50 mM Tris HCl, pH 7.6, I ϭ 0.05 at 25°C (32). In all assays, the sodium 2,6-dichlorophenol indophenol concentration was 92 M. The concentration of PES was varied from 0.6 to 3.6 mM, and the concentration of p-cresol was varied from 5.25 M to 0.6 mM. As in past studies with PCMH purified from P. putida (32), steady-state kinetic assays with the various forms of recombinant enzyme produced parallel 1/v versus 1/[PES] plots for various p-cresol concentrations, and parallel 1/v versus 1/[p-cresol] plots for different PES concentrations. These patterns are indicative of ping-pong type kinetic behavior. The steady-state parameters reported in Table I  X-ray Structural Analysis-The x-ray structure of PCMH, form A from P. putida 9869, was initially solved at 3.0 Å resolution by the multiple isomorphous replacement method using the known amino acid sequence for the cytochrome subunit and polyalanine for PchF (34). The deduced amino acid sequence for PchF (18) was subsequently fit to the electron density map and the structure of the native enzyme was partially refined to 2.5 Å resolution.
A crystal of PCMH was soaked in a solution of 25 mM p-cresol, from which data were recorded to 3.0 Å resolution, and the structure was refined directly using the coordinates from the native structure. The current R factors (⌺͉F o ͉ Ϫ ͉F c ͉/⌺͉F o ͉) for the native and p-cresol derivatives are 0.219 and 0.211, with root mean square deviations of 0.018 and 0.008 Å from the ideal bond lengths, and 3.98°and 1.66°from the ideal bond angles. Further refinements of both structures are in progress and a manuscript with complete details is under preparation. 2 The x-ray coordinates for the PCMH/p-cresol complex were used for the GREENPATH calculations (35). The structure of this complex is little different from that of oxidized PCMH. It was assumed that there is no structural reorganization for the electron transfer process.

RESULTS AND DISCUSSION
Apo-PchF Requires PchC for Its Covalent Flavinylation-In the course of E. coli DH5␣ expression experiments (18), antiflavin (25), anti-PchC, or anti-PchF antibodies were used for Western blot analyses of whole cell extracts electrophoresed on denaturing SDS gels. It was determined that PchF was heterologously expressed as a noncovalently associated dimer (␣ 2 or PchF 2 ), while PchC was expressed as and exists as a monomer (␤). For E. coli/pKK-CF, which produces both PchF 2 and PchC, high levels of flavin were found associated with the PchF band on a denaturing gel. This indicated that FAD was covalently bound to this subunit. In earlier work, it was demonstrated that heme was covalently attached to E. coli-expressed PchC and that the protein displayed the properties of a typical c-type cytochrome (18). In analyses of extracts of E. coli expressing PchF 2 but not PchC (E. coli/pKK-F), nearly undetectable amounts of flavin were found associated with PchF. Prolonged incubation of the extract with excess FAD did not alter the outcome. In contrast, after a 20-min incubation of an E. coli/ pKK-F extract with an extract of E. coli expressing PchC but not PchF 2 (E. coli/pKK-C), we detected a huge increase in the level of covalently bound flavin associated with the PchF band on an SDS gel. Addition of excess pure PchC to the E. coli/ pKK-F extract yielded the same results. In either case, addition of excess FAD had no effect on the outcome. This indicated that E. coli extracts contain a relatively large amount of available FAD.
As progressively more pure PchC or an extract of E. coli/ pKK-C was added to an extract of E. coli/pKK-F, progressively more FAD became covalently bound to PchF. Increasing the amount of E. coli/pKK-C extract or pure PchC added to the E. coli/pKK-CF extract also resulted in progressively more FAD becoming covalently bound. It was concluded that PchC was essential for covalent FAD attachment to PchF 2 . Subsequently, it was estimated that E. coli/pKK-CF manufactured PchF and PchC in a 5:1 ratio, which accounts for the incomplete flavinylation of PchF in E. coli/pKK-CF extracts.
A partially purified preparation of PchC/PchF from E. coli/ pKK-CF was subjected to preparative isoelectric focusing using a pH gradient from 4 to 9. A red heme-containing PchC band focused at pI ϭ 4.5, while two yellow bands focused at pI ϭ 5.0 (F A ) and 5.4 (F B ), respectively, and a broad colorless band focused at pI Ϸ 6.7. F A was shown to be ␣ 2 flavoprotein (PchF 2 ) with FAD covalently bound to both ␣ subunits, and F B was found to be ␣ 2 flavoprotein with one ␣ subunit containing covalently bound FAD and the other ␣ subunit devoid of flavin (␣ Ϫ ). The colorless protein was ␣ 2 flavoprotein devoid of FAD (apo-PchF 2 or ␣ 2 Ϫ ). Addition of FAD to pure apo-PchF 2 did not result in covalent flavin attachment.
Using PAGE, addition of increasing amounts of PchC to ␣ 2 Ϫ , in the presence of excess FAD, indicated that a 1:1 ratio of ␣ and ␤ was required for complete flavinylation (Fig. 3). Incremental addition of FAD to ␣ 2 Ϫ , in the presence of excess PchC, indicated that 1 mol of FAD bound covalently per 1 mol of flavoprotein subunit (data not shown). Covalent flavin binding did not occur when ␣ 2 Ϫ was incubated with FMN, FMN and ATP, FMN and ADP, FMN and AMP, riboflavin, riboflavin and ATP, riboflavin and ADP, or riboflavin and AMP, in the presence of a large excess of PchC, even after 24-h incubations. PchC does not bind to ␣ 2 Ϫ in the absence of FAD (see below). Horse heart cytochrome c, P. stutzerii cytochrome c 551 , P. aeruginosa cytochrome c 551 , or halophilic Paracoccus (ATCC 12084) cytochrome c 554 did not substitute for PchC in promoting covalent binding of FAD to apo-PchF 2 . Soluble flavin was not released from trichloroacetic acid-precipitated samples containing holo-PchF, indicating that all FAD was covalently bound. Furthermore, the fluorescence quantum yield of FMN/ riboflavin-containing, trypsin/chymotrypsin-digested PchF samples was much lower than expected for free FMN/riboflavin, but is in accord with the low quantum yield for 8␣-Otyrosylflavin (4).
Attempts to establish noncovalent binding of FAD to apo-PchF 2 were inconclusive. A K d value of 15 M at 25°C and pH 7.0 was determined by fluorescence quenching of bound FAD; however, the maximum fluorescence quenching was about 4 times greater than expected if specific binding of FAD to the flavin binding site apo-PchF 2 had caused complete quenching of the FAD fluorescence. It was concluded that there was significant nonspecific quenching by apo-PchF 2 , possibly due to nonspecific FAD binding. The changes that occurred in UVvisible spectra during a FAD titration of apo-PchF 2 were minor. Unless the K d for specific FAD binding was extremely small ([apo-PchF 2 ] ϭ 19.7 M), which apparently was not the case, the spectra of free FAD would always have swamped out the spectra of bound FAD; thus, these data were impossible to analyze. Additionally, there would be no way the segregate spectral perturbations resulting from specific and nonspecific binding. Other experiments, such as equilibrium dialysis, sim- ilarly would not discriminate specific and nonspecific binding. Band shifting was not observed on native gel electrophoresis gels of apo-PchF 2 , in the absence of FAD, with and without PchC present, nor was a change in the retention time seen for molecular sieving chromatography of apo-PchF 2 in the absence and presence of excess PchC. Finally, PchC had no specific effect on the tryptophan fluorescence of apo-PchF 2 . From these observations it was concluded that PchC does not bind to apo-PchF 2 when FAD is absent. This implies that FAD must bind to apo-PchF 2 before PchC can bind to catalyze the covalent flavinylation. Alternatively, there could be a fleeting complex between PchC and apo-PchF 2 to which FAD binds.
The ␣␤␣ Ϫ form of PCMH was made by adding excess PchC to F B , in the absence of FAD. The ␣ 2 ␤ 2 form PCMH was prepared either by mixing ␣ 2 Ϫ with a 10-fold excess of FAD and a 2-fold excess of PchC, or by mixing PchC with F A . Excess PchC and FAD were removed by gel filtration. Based on the heme or FAD content, within experimental error, the values for steady-state kinetic parameters were identical for these three forms of PCMH, and these parameters were the same as those for PCMH produced by wild-type P. putida 9869. It is interesting that the ␣␤ portions of the ␣␤␣ Ϫ variant and ␣ 2 ␤ 2 enzyme display identical kinetic properties (Table I).
A peptic flavin-containing peptide was isolated from PCMH, prepared by mixing FAD, heterologously expressed apo-PchF 2 and PchC. The sequence of the peptide was Xaa-Trp-Asn-Arg-Gly-Gly-Gly-Gly-Ser-Met. The peptide was treated with sodium dithionite to induce reductive elimination of the flavin, as expected for 8␣-O-tyrosylflavin. No other naturally occurring 8␣-modified flavin undergoes this reductive-cleavage reaction (4). The flavin-free peptide also was sequenced, and the Nterminal amino acid, Xaa, was identified as Tyr. This peptic flavin peptide is identical to the one previously isolated from wild-type PCMH purified from P. putida extracts (4,18). These results prove that FAD attaches correctly to Tyr 384 of PchF.
The Chemical Mechanism of Covalent FAD Attachment-In PCMH, 1-or 2-electron-reduced flavoprotein-bound FAD rapidly transfers an electron to the heme in PchC (k Ն 200 s Ϫ1 ) (32, 36). The long proposed mechanism for covalent FAD binding to the apo-PchF 2 subunit requires a 2-electron reduction of the isoalloxazine ring (Fig. 2) (5, 6). Thus, it would be expected that the reaction between FAD, apo-PchF 2 , and PchC would result in reduction of the PchC-bound heme. Preliminary kinetic studies have shown that PchC is indeed reduced, and heme reduction provided a sensitive and convenient measure of the flavinylation reaction (⌬⑀ 417(redox) ϭ 91.4 mM Ϫ1 cm Ϫ1 , and ⌬⑀ 552(redox) ϭ 19 mM Ϫ1 cm Ϫ1 for the cytochrome subunit). Rate constants for reactions run at 30°C, in 16 mM potassium phosphate buffer, pH 7.0, are presented in Table II.  (Table II). This suggests that chemical or conformational processes, not binding, have a disproportionate effect on the rate of covalent flavin binding at the conditions used for these studies. Bands c represent an activity-stained 7.5% native gel. In the figure, Fla ϭ PchF, Cyto ϭ PchC, and the lane labeled pKK-CF indicates that extracts of E. coli/pKK-CF were loaded on the various gels. Notice that protein in the lane containing only PchF (Fla) does not react with anti-flavin antibodies, nor does it produce an activity stain. Bands a, as the PchF 2 /PchC ratio changes from 1:0 to 1:2, there is an increase in the intensity of the anti-flavin antibody reaction. There are no further changes of significance at higher relative levels of PchC. Bands c, at a PchF 2 /PchC ratio of 1:0.5, a single weak activity-stained band is seen, while at a 1:1 ratio, two bands are apparent. When the ratio is 1:2 to 1:60, the slower band is not seen. At a PchF 2 /PchC molar ratio of 1:2, the PchF/PchC subunit ratio is 1:1. It is assumed that the slower band on this native gel is due to the ␣␤␣ Ϫ form of PCMH, and the faster band is due to the ␣ 2 ␤ 2 form. Notice that the E. coli/pKK-CF extract contains a significant amount of the ␣␤␣ Ϫ form, but a lesser amount of the ␣ 2 ␤ 2 . Catalytic groups on the enzyme participating in flavinylation are unknown. We speculate that Arg 477 stabilizes the negative charge that develops at N-1 nitrogen/C-2 oxygen of the flavin, when the iminoquinone methide form of FAD develops just prior to nucleophilic attack at the 8␣-carbon by Tyr 384 phenolate (Figs. 2 and 5). A guanidino nitrogen of Arg 477 is 2.5 and 3.3 Å from the C-2 oxygen and N-1 nitrogen of FAD, respectively (Fig. 5).
The inability of PchC to bind to apo-PchF suggests that noncovalently bound FAD alters the structure of PchF such that PchC can bind. This structural change could be global, or local, only affecting the environment of PchF around the FAD site and the surface region where PchC binds. On the other hand, PchC binding to the apo-PchF/FAD complex must confer further structural changes in order to properly position catalytic groups, the 8␣-carbon of FAD, and Tyr 384 phenolate oxygen for efficient covalent tethering of flavin.
The mechanism in Fig. 2 presumes that a base is required to remove an 8␣-hydrogen to produce the iminoquinone methide form of the dimethylbenzene portion of the isoalloxazine ring of FAD. The closest group that could possibly act as a base is Asp 440 , in its anionic form (Fig. 5). One of its carboxylate oxygens is 5.5 Å from one of the 8␣-hydrogens. Perhaps Asp 440 moves subsequent to the formation of the FAD-Tyr 384 bond, or there is an intervening H 2 O molecule that provides a bridge between Asp 440 and the 8␣-hydrogen. Alternatively, Tyr 384 phenolate could remove the 8␣-hydrogen and then lose the proton to another base such as Asp 440 to regenerate the phenolate form for attack of the 8␣-carbon.
Attack of Tyr 384 phenolate at the 8␣-carbon of FAD is a Michael-type addition that produces reduced covalently bound FAD (Fig. 2). The optimal stereochemistry for this addition would result in a dihedral angle of 90°between the plane of the flavin ring and the plane formed by the covalent bonds from the flavin to the phenolic oxygen (38). Inspection of the structure of covalently bound FAD on PCMH indicates a dihedral angle of 88°. This is also an optimal angle for elimination of Tyr 384 from 2-electron-reduced PCMH-bound FAD. It is known that 2-electron reduction of enzyme-free, 8␣-O-tyrosylflavin, results in facile elimination of tyrosine (4). However, 2-electron-reduced FAD, covalently bound in ␣ 2 ␤ 2 PCMH or ␣ 2 flavoprotein, is not subject to elimination (18). The reason for stability of covalent linkage in the protein is speculative. Perhaps a base that is required to protonate Tyr 384 phenolate as the 8␣-ether bond is broken has moved as a result of a conformational change occurring when the covalent bond forms. Possibly newly formed protein-FAD interactions decrease the propensity for shifting electron density into the benzenoid portion of the isoalloxazine ring system. Further refinement of the structure is under way. A more refined structure will provide valuable information regarding groups involved in the flavinylation reaction and may indicate a dihedral angle somewhat different from 90°.
Raison d'Être for Covalently Bound FAD in PCMH-As stated earlier, a major unanswered question concerns the purpose for covalent flavin binding in oxidoreductases. For PCMH, this linkage is likely involved in electron transfer from FAD to heme. On inspection of the 2.5-Å x-ray crystal structure of PCMH and of the 3.0-Å structure of the PCMH/p-cresol complex, it appeared that the most direct path for electron transfer from FAD in PchF to heme in PchC included the 8␣-carbon of FAD and the phenolic moiety of Tyr 384 . The computer program GREENPATH, version 0.97, was used to calculate the optimal pathway for electron transfer from the N-5 position of FAD to the iron at the core of the heme group (35). By far the best calculated pathway is the one displayed in Fig. 5. An electron tunnels from the flavin C-8 Tyr 384 phenolic ether bond through three other bonds of the phenol moiety, before "jumping" 2.96 Å to Ala 49 of PchC. From Ala 49 , the electron travels to Met 50 , an axial ligand of heme-bound iron. Without the FAD-Tyr 384 covalent bond, assuming minimal structural changes, and based on van der Waal's radii, an electron would need to "jump" at least 2.4 Å through space from FAD to Tyr, on its journey to the heme group. This would decrease the rate of electron transfer (39).
While the covalent flavin bond may accelerate electron transfer in PCMH, this is not the case for the two covalently bound flavins of other flavoproteins with known structures. Diheme flavocytochrome c sulfide dehydrogenase from a purple phototrophic bacterium contains FAD covalently linked at its 8␣carbon to sulfur of a Cys residue (40). Four potential electrontransfer pathways from FAD to the closest heme group were calculated. For two pathways, electrons jump to the polypeptide from the N-3 position of FAD; for another path, electrons jump to the peptide from the C-2 oxygen of FAD; for the fourth path, electrons exit via the N-5 position of FAD. The 8␣-FAD bond to Cys 42 is distal to the heme group and is not part of any "optimal" pathways (41).
For trimethylamine dehydrogenase from bacterium W3A1, the closest approach of covalently bound FMN and the Fe 4 S 4 cluster is at the 8␣-position of the flavin. The separation is about 6 Å (42). However, FMN is attached at its 6-position to  Table II). Trace C, the reaction between 0.70 M apo-PchF and 11.6 M PchC. FAD is not present. The slope of the linear change in trace C is the same as the slow linear portion of trace A (see Table II). The slow linear decrease in trace B is attributed to slow reoxidation of reduced heme in the aerobic reaction mixture. Flushing the assay solution with argon before the reaction is initiated eliminates the decrease (data not shown). the sulfur of Cys 30 (3,42). It is extremely unlikely that electrons pass through the 6-position of FMN on the way to the Fe 4 S 4 cluster.
We conclude that there is no dominant reason for covalent tethering of flavin in proteins. For PCMH, an obvious function revealed itself, but for other oxidoreductases, the reason remains obscure. At least in some cases, it is unlikely that covalent flavinylation is a fluke of nature. The linkage of FAD to the N-3 nitrogen of a specific, corresponding histidine is found in succinate dehydrogenase from all forms of life (14). If this linkage were not essential for stability, catalysis, electron transfer, etc., the linkage would not have survived eons of evolution.
The mechanism of post-translational covalent modification of PCMH is unique. We are not aware of any other example where a subunit seemingly stabilizes noncovalent binding of the modifying agent to its partner subunit, and then catalyzes formation of the covalent bond. In PCMH, this occurs without intermediating factors (e.g. external enzymes, chaperons, small molecule effectors), other than the subunits and FAD, nor is preactivation of apoprotein or flavin required.

FIG. 5. A cut-away stereo view of the FAD and heme binding sites in PCMH.
The backbone of PchC is pink, and the backbone of PchF is gray. FAD is orange, and Tyr 384 is yellow. Colored pink in FAD are phosphorus atoms. Heme is colored red, and the axial ligands of iron in the heme, Met 50 (in front of the heme) and His 19 (behind the heme), are colored light blue and dark blue, respectively. Arg 477 is at the bottom, in red-brown, and Asp 440 is to the left, in dark blue. The path for electron transfer from the N-5-position of FAD to heme iron is shown in green. The broken portion of the path indicates the 2.96-Å jump from an aromatic hydrogen of PchF-Tyr 384 to a lone pair of a peptidyl oxygen of PchC-Ala 49 . The path was calculated with the GREENPATH program. The image was generated using the RIBBONJR program from the MidasPlus group of molecular modeling programs (37).