The flavin mononucleotide cofactor in α-hydroxyacid oxidases exerts its electrophilic/nucleophilic duality in control of the substrate-oxidation level

Structural and enzymological explorations of p-hydroxy-mandelate oxidase and its mutants uncover an unprecedented electrophilic/nucleophilic duality for the flavin mononucleotide cofactor as well as an intramolecular disproportionation mechanism for an oxidative decarboxylation reaction.


Introduction
p-Hydroxymandelate oxidase (Hmo) is a flavin mononucleotide (FMN)-dependent enzyme that oxidizes mandelate to benzoylformate. Its Y128F single mutant unexpectedly shows a new reactivity and is able to oxidize mandelate to benzoate via benzoylformate, a four-electron oxidation reaction that is typically catalysed by a monooxygenase. However, when using benzoylformate in place of mandelate the reaction becomes stuck in the absence or the presence of hydrogen peroxide (H 2 O 2 ; Yeh et al., 2019;Fig. 1). To the best of our knowledge, this is the second example after lactate monooxygenase (LMO) of an enzyme that performs a ThDP/PLP/NADPHindependent oxidative decarboxylation reaction at the expense of one molecule of O 2 with the concomitant production of CO 2 and H 2 O (Ghisla & Massey, 1989). It has been hypothesized that the H 2 O 2 generated at the active site of LMO acts on pyruvate to form acetate by H 2 O 2 -mediated oxidative decarboxylation because the dissociation of pyruvate is a slow step (Giegel et al., 1990;Lopalco et al., 2016). Aside from this non-ping-pong kinetic description, how H 2 O 2 ISSN 2059-7983 mediates the oxidative decarboxylation of an -ketoacid at the molecular level remains elusive (Choong & Massey, 1980;Ghisla & Massey, 1977;Lockridge et al., 1972;Walsh et al., 1973). It has been noted that the electron reactivity of FMN in flavin-dependent enzymes is the main factor governing the implementation of a given type of reaction by the dioxygen, substrates and enzyme-cofactor system. C4 and N5 of the isoalloxazine ring of FMN are the most reactive centers, electrophilically or nucleophilically interacting with a substrate through a covalent linkage or spatiotemporally conveying electrons during redox reactions (Walsh & Wencewicz, 2013). In a previous study (Yeh et al., 2019), benzoylformate was found to be able to adopt a pro-R orientation with reference to the si face of the isoalloxazine ring. This structural reorientation aligns the nucleophilic N5 or C4-OOH of reduced FMN (FMN red ) with an appropriate trajectory to the electrophilic -keto carbon of the -ketoacid, a reaction reminiscent of the suicide inhibition of monoamine oxidase by mofegiline by forming an inactive N5 adduct or the formation of indole-3-acetate in auxin biosynthesis by the NADPH flavin-dependent monooxygenase YUC via a C4-OOH decarboxylation-assisted mechanism (Wu et al., 2005;Stepanova et al., 2011;Milczek et al., 2008;Dai et al., 2013). In this report, both biochemical and structural measures were exploited in an attempt to address how a single O atom dictates the oxidation level in the reactions catalyzed by wildtype Hmo and its Y128F mutant.

Cloning and protein purification
The hmo gene (orf22) was amplified from Amycolatopsis orientalis genomic DNA by polymerase chain reaction and was subcloned into the expression vector pET-28a(+) to generate the protein with an N-terminal His 6 tag. The expression plasmid was transformed into Escherichia coli BL21(DE3) cells by electroporation and the cells were grown on LB agar plates containing 50 mg ml À1 kanamycin for 16 h at 37 C. A single colony was grown overnight in 5 ml LB medium containing 50 mg ml À1 kanamycin at 37 C. The cell culture was used to inoculate 1 l LB medium containing 50 mg l À1 kanamycin. For protein expression and purification, the transformed E. coli cells were cultured, induced with 200 ml 1.0 M IPTG at an OD 600 of 0.6 and grown for a further 24 h at 16 C. The cells were harvested by centrifugation, resuspended in 20 ml binding buffer (20 mM HEPES pH 7.5, 500 mM NaCl, 10 mM imidazole, 10% glycerol) and lysed using a microfluidizer or by sonication. The supernatant of the lysate after centrifugation at 16 000 rev min À1 for 30 min was applied onto an Ni 2+ -NTA resin column. The bound protein was sequentially washed with 10 ml binding buffer (50 mM HEPES pH 8, 200 mM NaCl, 20 mM imidazole) and 10 ml wash buffer (50 mM HEPES pH 8, 200 mM NaCl, 80 mM imidazole) before elution with 10 ml elution buffer (20 mM HEPES pH 8, 500 mM NaCl, 250 mM imidazole). Gel filtration was performed using an Ä KTA FPLC system equipped with a HiLoad 16/60 Superdex 200 column (Amersham Bioscience) under isocratic conditions (20 mM HEPES pH 8, 100 mM NaCl).

Crystallization and data collection
The purified proteins were crystallized using the hangingdrop vapor-diffusion method. Hmo and its Y128F, Y128C and R163L mutants were concentrated to 7 mg ml À1 in 50 mM HEPES pH 8.0 buffer solution and crystallized using a The oxidation reactions catalyzed by Hmo and its Y128F mutant. (a) Hmo catalyzes a two-electron oxidation reaction to form benzoylformate from (S)mandelate. (b) The Y128F mutant catalyzes a four-electron oxidative decarboxylation reaction from (S)-mandelate to benzoylformate and benzoic acid without freeing H 2 O 2 during the reaction. (c) When benzoylformate is used as the substrate, the decarboxylated product benzoic acid cannot be formed by Hmo or its Y128F mutant in the presence or absence of H 2 O 2 . solution consisting of 35% Tascimate, 0.1 M bis-Tris propane pH 7.0 in a 50:50 volume ratio. Crystals appeared within five days in VDX48 plates (Hampton Research) with sealant at 20 C. For Hmo and its mutants in complex with (S)-mandelate (SMA), benzoylformate (BF), benzoic acid (BA), phenylpyruvate (PPY), oxaloacetate (OAA) or mandelamide (MAAD), each crystal was soaked with 10-30 mM of the designated ligand dissolved in the mother solution for between 10 min and 24 h before data collection. All crystals were transferred to a solution containing cryoprotectants and flash-cooled in liquid nitrogen prior to data collection. The cryoprotectant-containing solution for Hmo and the Y128F mutant consisted of 20%(w/v) glycerol and 35% Tascimate. X-ray diffraction data were recorded at an operating temperature of 100 K using an ADSC Quantum 315 or an MX300HE CCD detector on beamlines 13B1, 13C1, 15A1 or 05A at the National Synchrotron Radiation Research Center, Taiwan or beamline 44XU at SPring-8, Japan. All of the crystals belonged to the same space group: I422.

Structure determination and refinement
Data were indexed and scaled using the HKL-2000 package (Otwinowski & Minor, 1997). The crystal structures were determined by the molecular-replacement (MR) method using Phaser-MR from the CCP4 suite (McCoy et al., 2007;. The crystal structure of hydroxyacid oxidase (PDB entry 3sgz; Chen et al., 2012) was used as the search model for solving the initial phase. The polypeptide structures were built and refined using REFMAC . Subsequent iterative cycles of model building and refinement were performed using Coot (Emsley et al., 2010) and PHENIX (Afonine et al., 2012). All non-H atoms were refined with anisotropic displacement parameters. Both protein structures and electron-density maps were generated using PyMOL (DeLano, 2002). Detailed refinement statistics are presented in Table 1.

Site-directed mutagenesis
The site-directed Hmo point mutants Y128C, Y128F and R163L were generated using QuikChange (Stratagene). The primers for mutant preparation are listed in Supplementary  Table S1. The mutations were confirmed by DNA sequencing. Each mutated protein was purified using the same protocol as used for wild-type Hmo.

Chemoenzymatic synthesis of 5-deazaflavin mononucleotide
To obtain 5-deazaflavin mononucleotide, the commercially available compounds 3,4-dimethylaniline, sodium cyanoborohydride and d-ribose were used as starting materials, following the synthetic procedure described in Supplementary Scheme S1. Chemically synthesized deazariboflavin was purified using column chromatography, characterized by mass spectrometry or NMR and converted to catalytically active 5-deaza-FMN by riboflavin kinase. The gene coding for riboflavin kinase was amplified from the genomic DNA of E. coli K12 and was subcloned into a pET-28a vector to carry an N-terminal His 6 tag. The constructed plasmid was then transformed into E. coli BL21(DE3) cells for overexpression. The overexpressed protein was purified using Ni 2+ -NTA affinity chromatography and concentrated to 15 mg ml À1 . The reaction was initiated by the addition of 2 mg ml À1 riboflavin kinase to a reaction solution consisting of HEPES pH 7.5 with 5 mM ATP, 5 mM MgSO 4 and 10 mM 5-deazariboflavin overnight. The sample solution was purified by HPLC and confirmed by mass spectrometry.

Preparation of 5-deaza-FMN-containing Hmo and its Y128F mutant
The deflavination and reconstitution of Hmo and its Y128F mutant were performed on an Ni 2+ -NTA resin column by circulating a solution of HEPES buffer containing 10 mM deazariboflavin (Hefti et al., 2003). The Ni 2+ -NTA resin column was loaded and saturated with a solution of the target protein at a flow rate of 0.2 ml min À1 using a peristaltic pump, which was installed between the sample reservoir and the column, to ensure maximal binding. Ten volumes of denaturing buffer (50 mM HEPES, 2 M potassium bromide, 2 M urea pH 8.0) were subsequently used to unfold Hmo or its Y128F mutant, allowing removal of the FMN prosthetic cofactor from Hmo or its Y128F mutant. On-column replacement with 5-deaza-FMN was completed by circulating a buffer solution consisting of 50 mM HEPES pH 8.0, 10 mM 5-deaza-FMN at 0.2 ml min À1 overnight. Hmo or its Y128F mutant containing 5-deaza-FMN was eluted using 10 ml elution buffer (20 mM HEPES pH 8, 200 mM NaCl, 250 mM imidazole). An analytical gel-filtration experiment using a Superdex 200 16/20 column confirmed that 5-deaza-FMNcontaining Hmo (5-deaza-Hmo) and its Y128F mutant have an unchanged tetrameric state after the refolding process. The purified enzymes were concentrated to 10 mg ml À1 for enzymatic assays and protein crystallization.

Enzymatic assays using 5-deaza-Hmo and its Y128F mutant
A typical enzymatic reaction (10 mg 5-deaza-Hmo or its Y128F mutant) was carried out in 200 ml buffer solution (20 mM HEPES, 100 mM NaCl pH 7.5, 5 mM S-mandelate, benzoylformate or benzoic acid) at 25 C for 2 h. Reactions were quenched using 6 N HCl and subjected to HPLC-MS analysis (using an Agilent 1260 Infinity Quaternary LC module connected to a Thermo-Finnigan LTQ-XL). Analytes were separated using a reverse-phase C 18 column (4.6 Â 250 mm, 5 mm, C 18 Prodigy, Phenomenex) at a flow rate of 1 ml min À1 in a mobile system programmed as a linear gradient from 2% to 40% solvent B against solvent A over 25 min followed by 98% solvent B for a further 8 min (solvent A, water with 1% formic acid; solvent B, acetonitrile with 1% formic acid). The data were recorded using a UV detector with the wavelength set to 254/280 nm and the mass spectrometer in positive mode. research papers Table 1 Data-collection and refinement statistics for Hmo and its Y128F, Y128C and R163L mutants.

Hmo
Hmo  (hkl)i is the average intensity value of the equivalent reflections. ‡ R work = P hkl jF obs j À jF calc j = P hkl jF obs j. § R free was calculated from 5% of data that were randomly excluded from refinement.

UV-Vis absorption measurements of reactions of Hmo and its Y128F mutant
Ultraviolet and visible absorption spectra were recorded using a Beckman spectrophotometer (DU-800). In aqueous solution, freshly prepared Hmo or its Y128F mutant (0.125 mM in 0.05 M HEPES, 0.1 M NaCl pH 7.5) were mixed with substrates (2.5-5 mM) and incubated in a quartz cuvette at ambient temperature for 2 h. The absorption spectra of the reactions were recorded from 300 to 600 nm. For the redissolved crystals/crystalloids, more than 100 crystals of Hmo or its Y128F mutant (after soaking with substrates at 5 mM for 2 h) were picked from hanging-drop crystallization plates and redissolved in the mother liquor in a UV quartz cuvette before spectral scanning.

Inhibition of a-ketoacids
To investigate the mechanism of the four-electron oxidative decarboxylation reaction catalyzed by the Hmo single mutant Y128F, we first solved crystal structures of the Y128F mutant in complex with different ligands such as (S)-mandelate, (S)-2-phenylpropionate, benzoylformate, benzaldehyde and benzoate. The ternary complexes of Hmo and its Y128F mutant were then compared, whereupon it was observed that the superpositioned complexes showed no apparent structural variations between Hmo and the Y128F mutant (r.m.s.d. of <0.1 Å ; Supplementary Fig. S2) in terms of chemical conformations and spatial positions of both the proteins and ligands. When the crystals were soaked with benzoylformate, the formation of an N5-benzoyl-FMN adduct with an N5-C 0 linkage was observed [ Fig. 2(a)]. MS analysis confirmed that the benzoyl moiety was covalently linked to FMN [ Fig. 2(b)]. When the crystals were soaked with benzaldehyde, the formation of a covalent adduct was not observed. This outcome suggests that the -ketoacid moiety is a prerequisite for the formation of the covalent adduct, in which the decarboxylation of the terminal carboxyl group of the -ketoacid is likely to take place after the formation of the N5-C 0 linkage. Moreover, when the crystals of the Y128F mutant were soaked with (S)-3-phenyllactate, phenylpyruvate or phenylacetate [ Fig. 2(a)], we observed that (S)-3-phenyllactate was oxidized to phenylacetate (a four-electron oxidative decarboxylation), phenylpyruvate ended up as an N5-phenylacetyl-FMN adduct and phenylacetate stayed as it was. When the crystals were soaked with oxaloacetate, an N5-malonyl-FMN adduct was found [ Fig. 2(a)], leading overall to the conclusion that the formation of the covalent N5 adducts is -ketoacid-dependent.
In most flavin-dependent oxidoreductases the FMN ox cofactor acts as an electron sink (electrophile) that accepts electrons or hydrides conveyed from a substrate or NADH/ NADPH in the reductive half-reaction. The nitroalkane oxidase from the fungus Fusarium oxysporum and the alkyldihydroxyacetone phosphate synthase in human fibroblasts are two rare cases in which a carbanion is generated at the active site prior to addition to N5 of the flavin cofactor as a covalent adduct (Razeto et al., 2007;Heroux et al., 2009). However, a question arises in the context of how two electrophiles (FMN ox and -ketoacid) are covalently associated in the Y128F mutant. One likelihood is that the -ketoacid undergoes decarboxylation in the first instance to form a localized C 0 carbanion that then acts on FMN ox , but anketoacid that spontaneously undergoes decarboxylation is chemically untenable. The second possibility is that the sp 2 N5 atom of FMN ox acts as a nucleophile, but this scenario likewise contradicts the current understanding: FMN ox is a strong electrophile that accepts electrons. Nevertheless, an extra chunk of electron density at the top of N5 of FMN ox was observed in unbiased difference electron-density maps, and this electron density was denser in the Y128F mutant than in the wild type [Figs. 2(c) and 2(d)]. This additional electron density suggests that the C4 N5 double bond in FMN ox is polarized to a C4 + -N5 À ylide (a tertiary carbocation and a tetrahedral sp 3 amine anion), as manifested by uneven wedgeshaped electron density for the -bond [Figs. 2(c), 2(d), 3(a) and 3(b)]. The extent of polarization appears to be a function of an active-site perturbation ensemble (e.g. the point mutation), reflecting cooperative interplay of the hydrogen-bond network between water, FMN and active-site residues as well as ligands [Figs. 3(c) and 3(d)]. The formation of the adduct is thereby proposed to take place as follows: the sp 3 N5 atom of the polarized FMN ox attacks the carbonyl C atom of the -ketoacid to form a covalent C 0 -N5 adduct, whereupon decarboxylation takes place, resulting in a localized C 0 carbanion. The lone pair of the C 0 carbanion subsequently hybridizes with the orbital of N5 to reinstate the neutrality of C4. Upon collapse of the C 0 oxyanion, N5-acyl-FMN red results via a series of bond rearrangements. This species has a hydroquinone-like structure, with the acyl moiety protruding out of the plane defined by the isoalloxazine ring of FMN ox [ Fig. 2(a)].
The UV-Vis spectrum of the Y128F mutant protein solution exhibits a typical FMN ox absorbance profile [two absorbances at 370 and 450 nm; Fig. 4(a), i]. FMN ox turned colorless when phenylpyruvate was added to the solution, suggesting the formation of an acyl-FMN species [ Fig. 4(a), ii]. The spectrum is dissimilar to that observed when phenyllactate was added [in which the two typical absorbances at 370 and 450 nm disappeared, suggesting the reduction of FMN ox to FMN red ; Fig. 4(a), iv] by a small hump at 340 nm [ Fig. 4(a), ii]. The redissolved solution of Y128F mutant crystals/crystalloids that had been pre-soaked with phenylpyruvate also turned colorless, with a profile similar to that in solution [with a smaller hump at 340 nm; Fig. 4(a), iii] (Sucharitakul et al., 2007;Thotsaporn et al., 2011). While O 2 is a small hydrophobic molecule that freely diffuses through spaces and tunnels in protein matrices Baron, Riley et al., 2009), Hmo may have evolved a discrete channel or pockets that temporarily limit the access of O 2 to C4 of N5acylated isoalloxazine. The metastable N5-alkyl-FMN red in aqueous solution is thus attributable to dysfunction of the charge-transfer cage in the absence of Hmo or its Y128F mutant. The inhibited Hmo and mutants identified here differ research papers Figure 2 Crystal structures of acyl-FMN red adducts and inhibition mechanism by -ketoacids. (a) Structures of acyl-FMN red adducts in crystals of the Y128F mutant soaked with benzoylformate (left), phenylpyruvate (center) or oxaloacetate (right). The flavin adducts all are at the si-face of the isoalloxazine ring. (b) LC traces and mass spectra of FMN (i) and phenylacetyl-FMN red (ii). (c, d) Weighted 2F o À F c electron-density maps (gray) and unbiased F o À F c difference OMIT electron-density maps (blue) for FMN in Hmo (c) and the Y128F mutant (d) without (left) or with a ligand (S-mandelate, center; benzoylformate, right), where the extent of polarization is justified by OMIT electron density (the wild type or Y128F mutant and the absence or presence of a ligand seem to be key factors). The 2F o À F c electron-density map is contoured at 2; the unbiased F o À F c OMIT difference electrondensity map is contoured at 4. Free ligands, FMN, FMN adducts and active-site residues are colored cyan, yellow, orange and green, respectively. See Supplementary Figs. S3(a), S3(b) and S2(c) for stereoviews and F o À F c difference electron-density maps. from the conventional inactivation of FMN ox , which requires chemically activated agents (Walsh, 1980(Walsh, , 1984.

5-Deaza-FMN in oxidation
A photoreduction mechanism has been proposed for the LMO-mediated oxidative decarboxylation reaction (Ghisla & Massey, 1977, 1989Ghisla et al., 1979), in which the reactants need to be activated by photosensitization. In the present case (Hmo and its Y128F mutant), the C4 + ÀN5 À ylide of FMN ox appears to be the key factor. To validate this proposition, we chemoenzymatically synthesized 1 g of the FMN analog 3,10dimethyl-5-deaza-isoalloxazine ribitol phosphate (5-deaza-FMN ox ) following previously reported methods [ Fig. 4(b); the modified method is described in the supporting information] (Carlson & Kiessling, 2004;Kittleman et al., 2007;Mansurova et al., 2008;Osborne et al., 2000). Biochemically, 5-deaza-FMN-containing Hmo or its Y128F mutant is able to oxidize (S)-mandelate to benzoylformate but not to benzoate, confirming that the enzyme was refolded successfully as the wild type and supporting N5 as the pivotal factor in the oxidative decarboxylation reaction [ Fig. 4(c)]. Similarly, benzoylformate but not benzoate was found in 5-deaza-FMN ox -containing crystals of Hmo or its Y128F mutant soaked with (S)-mandelate [Figs. 4(d)  The effect of 4-OH of Tyr128 on polarization, hydrogen-bond networking and reactivity. (a, b) A close-up view of the wedge-shaped electron density on top of C4 N5 of FMN ox shown by unbiased difference electron-density maps (blue, positive electron density) contoured at 4 in the wild type (a) and the Y128F mutant (b), suggesting that the electrons in the -orbital of C4 N5 are polarized from C4 to N5 (to form a C4 + -N5 À ylide), with this being more significant in the Y128F mutant than in the wild type. intervals. Furthermore, the lack of visible electron density at the top of C5 or between C5 and C 0 of benzoylformate (4.0 Å ) suggests that the C4 C5 double bond in 5-deaza-FMN is less polarizable. Our structural interrogation supports the decarboxylation of the -ketoacid taking place after or in concert with the formation of the C 0 -N5 bond. The R163L mutant (a low-activity mutant) was further examined using the nondecarboxylable substrates -(S)-mandelamide [2-(S)hydroxy-2-phenylethylamide] or benzoylamide (2-keto-2phenylethylamide), in which the former is oxidized to form the latter. When the R163L mutant crystals were soaked with benzoylamide, an -hydroxyamide-FMN adduct was formed [ Fig. 4( f)], leading to the unequivocal conclusion that N5 of FMN ox has a nucleophilic propensity and that the C 0 -N5 bond is formed prior to -ketoacid decarboxylation.

Four-electron oxidation to benzoate
We propose that C4-OOH-N5-acyl-FMN is the key intermediate in the four-electron oxidation of an -hydroxyacid mediated by Hmo and its Y128F mutant on the basis of the following facts: (i) Hmo and its Y128F mutant catalyze the four-electron oxidation of an -hydroxyacid via an -ketoacid to an acid with one O atom from O 2 incorporated into the terminal carboxylic group, (ii) H 2 O 2 is not able to oxidize the -ketoacid in the absence of Hmo or its Y128F mutant, (iii) the pro-R -ketoacid is covalently linked to FMN ox in the Y128F mutant, forming an N5-acyl-FMN red adduct, (iv) the oxidation cascade stalls at the -ketoacid using Hmo or its Y128F mutant with 5-deaza-FMN ox in lieu of FMN ox and (v) the sp 3 N5 in FMN red is highly reactive, as exemplified in UbiX, a flavin prenyltransferase involved in bacterial ubiquinone biosynthesis (White et al., 2015). The formation of the intermediate is somewhat similar to the mechanism proposed for EncM, which catalyses an oxidative Favorskii-type rearrangement reaction (Teufel et al., 2015). The major discrepancy, however, is that O 2 in Hmo and its Y128F mutant mediates the transient formation of C4-OOH-N5-acyl-FMN prior to its release as H 2 O 2 . (e) Hydrogen peroxide was modeled at C4, where the distance between Tyr128 and the terminal O atom of C-OOH is 2.7 Å . (f, g) The phenyl ring of benzoylformate, which is bulkier and takes up space, limits the access of dioxygen to the reaction center, while the methyl group of pyruvate, which is smaller and takes up less space, allows the access of dioxygen to the reaction center [the phenyl ring that bulges out at the substrate entrance in ( f ) prohibits exposure of FMN red to the bulk solvent, as opposed to the methyl group in (g) which allows exposure of FMN red to the bulk solvent]. Therefore, the size of the substrates is another factor in leverage of the oxidation cascade. (h) Aside from the active-site perturbation effect, the absence of the p-OH group also introduces some space allowing access of O 2 to the C4 redox-active center. (i) The sulfhydryl group (SH) of the Y128C mutant has been oxidized to a sulfenyl group (S-OH), as it is vulnerable to ROS generated in the active site. Free ligands, FMN and active-site residues are colored cyan, yellow and green, respectively. Superposition of the benzoylformate-liganded ternary complex of the Y128F mutant with that of the wild type shows no apparent discrepancies (r.m.s.d. of 0.06 Å ) except for the p-OH group of Tyr128 ( Supplementary Fig. S4). Given that the oxidative decarboxylation of an -hydroxyacid is catalytically executed by the Y128F mutant, the p-OH group ought to play a crucial role in leverage of the oxidation cascade. This effect is commensurate with a recent report that a single mutation, C65D, of phenylacetone monooxygenase converts a monooxygenase to an oxidase (in contrast to this report) by facilitating the discharge of H 2 O 2 (Brondani et al., 2014). C4-OOH-FMN red , which is a reactive intermediate in a typical monooxygenase/oxidase-catalyzed reaction, was modeled and optimized in the structure of Hmo, in which the p-OH group is within hydrogen-bonding distance (2.7 Å ) of C4-OOH [ Fig. 3(e)]. On the basis of this model, the p-OH group is in a position to protonate C4-OO À -FMN red to form C4-OOH-FMN red , thereby neutralizing or facilitating the discharge of H 2 O 2 from C4-OOH. In contrast, C4-OO À -FMN red may diverge in the absence of the p-OH group. One  implication is that the Y128F mutant works like a monooxygenase, whereby Baeyer-Villiger-type reactions result. That is, the C4-OO À anion attacks the -carbon (C 0 ) of pro-R benzoylformate to form a tetrahedral oxyanion species. Upon the collapse of the -oxyanion the terminal carboxyl group migrates to the distal O atom of C4-OO À to form a mixed-anhydride species; subsequent hydrolysis would give rise to benzoate and formate (Torres Pazmiñ o et al., 2010). This type of reaction, however, was ruled out because no benzoate was detected in the reactions catalyzed by the 5deaza-FMN-containing Y128F mutant. This fact, however, underscores the importance of the sp 3 N5 of C4-OO À -FMN red in the four-electron oxidation reaction, where the reduced or polarized sp 3 N5 actually has a better Bü rgi-Dunitz angle and is at a short distance from C 0 of pro-R benzoylformate, favoring the formation of C4-OO À -N5alkyl-FMN red before the release of H 2 O 2 .

Proposed catalytic mechanism of oxidative decarboxylation
In a general four-electron oxidation reaction, one molecule of (S)-mandelate should theoretically yield two equivalents of H 2 O 2 . The molar ratio of H 2 O 2 versus benzoate, however, did not follow this stoichiometry (it was much less than unity; Yeh et al., 2019). This fact, in contrast, is consistent with a disproportionation reaction of FMN peroxide, in which one O atom goes to benzoate and the other ends up as water. To search for clues, we re-examined the active-site geometry of Hmo and its Y128F mutant liganded with substrates (d, e) Crystal structures of the Y128F mutant harboring 5-deaza-FMN soaked with S-mandelate (d) or phenyllactate (e), which have been transformed into benzoylformate or phenylpyruvate, respectively. Unlike FMN in the wild type or the Y128F mutant, no electron density emerges at the top of C5 or between C 0 and C5. ( f ) The structure of an -mandelamide-N5-FMN red adduct in the crystal of the R163L mutant soaked with nondecarboxylablemandelamide (the chemical structure is shown). The flavin adduct is on the si-face of the isoalloxazine ring. The 2F o À F c electron-density map is contoured at 2. The unbiased F o À F c difference electron-density map is contoured at 4 in positive electron density. Free ligands, FMN, FMN adducts and active-site residues are colored cyan, yellow, orange and green, respectively. See Supplementary Figs. S2(j)-S2(m) for stereoviews and 2F o À F c difference electron-density maps.

Figure 5
The proposed mechanisms of oxidative decarboxylation catalyzed by the Y128F mutant.
(mandelate or lactate) or products (benzoylformate or pyruvate). The redox-active center C4 N5 of isoalloxazine is surrounded by a constellation of active-site residues [Val78 and Ala79 at the bottom and Tyr(Phe)128 and His252 at the top], where it is accessible only from the upper front side. The reaction center is sealed to form a narrow and low-dielectric milieu suitable for hydride transfer/electron tunneling when a substrate and redox-active center C4 N5 approach each other. Interestingly, the substrate pair mandelate/benzoylformate fits better than the alternative pair lactate/pyruvate because of the bulky phenyl group in the former [Figs. 3( f ) and 3(g)]. The redox chamber in the Y128F mutant, on the other hand, is not as tight as that in the wild type owing to the lack of the p-OH group [ Fig. 3(h)]. This flaw is exacerbated when lactate (with a smaller methyl group) is used [Figs. 3( f ) and 3(g)]. In this context, O 2 is relatively accessible to FMN red via a temporal space/tunnel to form C4-OO À -FMN before the release of the -ketoacid (the non-ping-pong mechanism). Meanwhile, the pro-R -ketoacid is accessible by sp 3 N5 to form C4-COOH-N5-oxyalkylate-FMN red . Upon decarboxylation, a C4-COOH-N5-aloxyl-FMN red C 0 carbanion results. The C 0 carbanion that intramolecularly attacks the distal O atom of C4-OOH then leads to heterolytic cleavage of the peroxide scissile bond. Upon return of the C 0 oxyanion, benzoate and H 2 O are formed in concert with the regeneration of FMN ox (Fig. 5).
The peroxide anion radical:FMN semiquinone caged pair is likely to proceed through a single-electron transfer from FMN red to O 2 in a given flavoenzyme, where the reactivity depends on the active-site polarity ensemble of factors including bound water, charge distribution, hydrogen bonds, van der Waal forces etc. (Fagan & Palfey, 2010). The Y128C mutant (in which the bulky phenyl group is replaced by a sulfhydryl group) that can transform (S)-mandelate to benzoate was used to assess the extent of active-site perturbation. The structure of the Y128C mutant crystallized and soaked with (S)-mandelate revealed that the sulfhydryl (SH) group of the Y128C mutant has been oxidized to a sulfenyl group (S-OH), in contrast to the other sulfhydryl groups, which are not changed [ Fig. 3(i)]. This result indicates that the sulfhydryl group of the Y128C mutant is relatively accessible and sensitive to the local unregulated reactive oxygen species (ROS; Chaiyen et al., 2012).

Conclusions
The present studies allow us to gain mechanistic insights into the reactions catalyzed by both Hmo and its Y128F mutant, in which substrate reorientation, active-site perturbation and spatiotemporal crowdedness are pivotal factors that influence the dioxygen accessibility and reaction order of the FMN red/ox : -ketoacid pair in the reactions mediated by Hmo and its Y128F mutant. Given the Y128F mutation, the original reactivity of Hmo is perturbed. One stark contrast is that the electrophilic FMN ox is polarizable to an ylide-like species. This species is capable of attacking an -ketoacid to form an N5acyl-FMN red dead-end adduct, providing evidence for the first time that FMN ox possesses a nucleophilic/electrophilic duality. Having confirmed the formation of the N5-acyl-FMN red adduct, both the nucleophilic propensity and positional preponderance of N5 of FMN red prompt us to propose that the N5-alkanol-FMN red C 0 carbanion is the key intermediate in the oxidative decarboxylation reaction. This intermediate reacts with dioxygen in place to form a C4-COOH-N5aloxyl-FMN red C 0 carbanion species that subsequently undergoes an intramolecular disproportionation reaction to yield benzoate and FMN ox , thus accounting for the ThDP/ PLP/NADPH-independent oxidative decarboxylation reaction. To this end, the p-OH group of Tyr128 that leverages the spatial and temporal leeway over the oxidation cascade was unexpected. The -substituent on the -hydroxy acid that influences the accessibility of dioxygen to the reaction center is another unexpected factor. A synthetic 5-deaza-FMN ox cofactor in combination with an -hydroxyamide or -ketoamide positively supports the proposed mechanism, in which the loose ends that benzoate is a minor product of Hmo and the major product of the Y128F mutant are tied up. An unequivocal consolidation of the proposed mechanism would be provided by the physical capture or visualization of the C4-COOH-N5-aloxyl-FMN red C 0 carbanion or other relevant intermediates, which however will require future studies using advanced spectroscopic and microscopic analysis on the submicrosecond time scale using, for example, the X-ray freeelectron laser technique. The present structural and biochemical elucidation nonetheless strengthens the idea that the FMN cofactor is versatile and cooperates with the activesite residues and substrates in dictating the oxidation cascade.