O2‐independent demethylation of trimethylamine N‐oxide by Tdm of Methylocella silvestris

Bacterial trimethylamine N‐oxide (TMAO) demethylase, Tdm, carries out an unusual oxygen‐independent demethylation reaction, resulting in the formation of dimethylamine and formaldehyde. In this study, site‐directed mutagenesis, homology modelling and metal analyses by inorganic mass spectrometry have been applied to gain insight into metal stoichiometry and underlying catalytic mechanism of Tdm of Methylocella silvestris BL2. Herein, we demonstrate that active Tdm has 1 molar equivalent of Zn2+ and 1 molar equivalent of non‐haem Fe2+. We further investigated Zn2+‐ and Fe2+‐binding sites through homology modelling and site‐directed mutagenesis and found that Zn2+ is coordinated by a 3‐sulfur‐1‐O motif. An aspartate residue (D198) likely bridges Fe2+ and Zn2+ centres, either directly or indirectly via H‐bonding through a neighbouring H2O molecule. H276 contributes to Fe2+ binding, mutation of which results in an inactive enzyme, and the loss of iron, but not zinc. Site‐directed mutagenesis of Tdm also led to the identification of three hydrophobic aromatic residues likely involved in substrate coordination (F259, Y305, W321), potentially through a cation–π interaction. Furthermore, a crossover experiment using a substrate analogue gave direct evidence that a trimethylamine‐alike intermediate was produced during the Tdm catalytic cycle, suggesting TMAO has a dual role of being both a substrate and an oxygen donor for formaldehyde formation. Together, our results provide novel insight into the role of Zn2+ and Fe2+ in the catalysis of TMAO demethylation by this unique oxygen‐independent enzyme.


Introduction
Bacterial trimethylamine N-oxide (TMAO) demethylase (Tdm) is a key enzyme involved in bacterial degradation of trimethylamine (TMA) and TMAO [1][2][3]. The enzyme was first proposed in the 1970s and has been partially purified from Bacillus sp. PM6 [4] and Pseudomonas aminovorans (now Aminobacter amino-vorans [5]). Despite being purified from aerobic hosts, Tdm can convert TMAO anaerobically to equimolar amounts of dimethylamine (DMA) and formaldehyde (HCHO) (1 TMAO ? 1 DMA + 1 HCHO) [2,3,5]. The gene encoding Tdm has only been identified very recently and it is now known that tdm is widely distributed in nature, particularly in heterotrophic bacteria of the Roseobacter clade and the SAR11 clade in marine bacterioplankton [2].
Tdm is constituted of two domains, an uncharacterized DUF1989-containing domain at its N terminus and a tetrahydrofolate (THF)-binding domain (GCV_T) at its C-terminus. DUF1989 in Tdm shows modest sequence similarity (< 30%) to urea-carboxylase-associated proteins, whose functions in urea catabolism are as-yet unknown [6]. GCV_T domains, however, are found in several very well-characterized THF-dependent enzymes, such as glycine cleavage T protein [7] and dimethylsulfoniopropionate demethylase [8], with a function of binding THF to accept formaldehyde. Therefore, it has been proposed previously that the N-terminal DUF1989 domain of Tdm may play a role in substrate binding and subsequent catalysis, whereas its C-terminal GCV_T domain is responsible for HCHO conjugation with THF [3].
It has been suggested that metal ions may play a role in Tdm catalysis [4]. For example, the partially purified Tdm of Bacillus sp. PM6 was strongly activated by ferrous iron [4]. In agreement with the putative role of metals in catalysis, purified Tdm of Methylocella silvestris does not contain either flavin adenine dinucleotide (FAD) or nicotinamide adenine dinucleotide (NAD) [3]. Although a crystal structure for Tdm has yet to be solved, structures of three DUF1989-domain containing proteins (3ORU, 3SIY, 3DI4) available in the Protein Data Bank (PDB) database all contain zinc (Zn 2+ ). However, the types of metal(s) present in Tdm are yet to be established experimentally, and the metal stoichiometry is not known.
The aim of this study was therefore to determine the stoichiometry of the metal cofactors of Tdm in order to gain insight into the catalytic mechanism of oxygenindependent TMAO demethylation by this unusual bacterial Tdm.

Tdm is a novel zinc-iron-dependent protein
Early studies in the 1970s have suggested that bacterial Tdm is a metal-dependent enzyme. The Bacillus Tdm is strongly stimulated by ferrous iron and reducing agents such as ascorbate and glutathione [3]. In agreement with these previous studies, we observed~20% inhibition of Tdm activity when the purified enzyme was incubated with the metal chelator EDTA (Fig. 1A). To characterize the metal ion(s) in Tdm, we carried out ICP-MS metal scan analyses of purified recombinant Tdm from Methylocella silvestris, which detected the presence of Zn, Fe and Ni above background levels (data not shown).
To obtain a more accurate estimation of metal contents in Tdm, Zn, Fe, Ni as well as S (for accurate determination of protein concentrations) were quantified by ICP-OES. The results showed that 1 monomer of Tdm contained 0.97 AE 0.03 Zn 2+ and 0.35 AE 0.02 iron in the as-isolated Tdm (CK-Tdm, Table 1). Trace amounts of Ni 2+ (< 0.1 molar equivalents per Tdm monomer) were also found, potentially as a consequence of the Ni-IMAC purification step. To address whether Zn or Fe could be replaced by each other, Feand Zn-enriched Tdm (Fe-Tdm, Zn-Tdm) were purified from E. coli cultivated in media supplemented with either Fe(NH 4 ) 2 (SO 4 ) 2 or ZnCl 2 (0.5 mM final concentration). Tdm expressed in Fe-supplemented media showed slightly, yet significantly, higher Fe 2+ content (0.38 AE 0.02 mol per mol monomer) (P < 0.05), whereas Tdm expressed in Zn-supplemented media had reduced Fe 2+ content (0.12 AE 0.01 mol per mol monomer), in coincidence with an increase of Zn 2+ content (1.33 AE 0.03 mol per mol monomer) (P < 0.05), suggesting a replacement of Fe by Zn in purified Tdm. Additionally, Fe-enriched Tdm had a higher catalytic activity compared to that of CK-Tdm or Zn-enriched Tdm (Table 1). Purified Tdm had a K m value~4 mM, in line with that of the same enzyme purified from two other bacteria, Pseudomonas aminovorans (2 mM) and Bacillus sp. PM6 (2.85 mM) [4,5].
To further probe the iron species in as-isolated Tdm of M. silvestris, activity assays were performed with the addition of a reducing agent (ascorbic acid, Asc) or an oxidizing agent (hydrogen peroxide, H 2 O 2 ) at varying concentrations. The results demonstrated that Tdm was sensitive to H 2 O 2 and > 80% activity was lost upon incubation with 100 lM H 2 O 2 (Fig. 1B), while Asc did not show any inhibition until a final concentration of 8 mM (Fig. 1B). Furthermore, when isolated Tdm was incubated with various divalent metal ions, it was observed that Fe 2+ significantly enhanced Tdm activity, particularly in the presence of Asc (Fig. 1C). Interestingly, the activity of H 2 O 2pretreated Tdm can at least be partially restored by incubating with Asc and Fe 2+ (Fig. 1D). Taken together, our data suggested a role of ferrous iron in the native as-isolated Tdm during catalysis.
Due to the traditional protein overexpression and purification procedures, oxygen-sensitive ferrous iron is prone to loss [9,10]. Our ICP-OES analyses of asisolated Tdm may therefore have underestimated in-vivo iron contents in this protein (Table 1). To determine the optimal Fe 2+ stoichiometry, a Fe 2+ titration experiment was carried out. To eliminate the   nonspecific binding of metal ions to the 6*His-tag, 6*His-tag-free Tdm was used. The data presented in Fig. 1E gave a stoichiometry number of n = 0.91. Together, our data suggest that Tdm is a Zn 2+ -and Fe 2+ -dependent protein with Zn 2+ /Tdm and Fe 2+ / Tdm monomer ratios of 1/1.
Three cysteine residues (C263, C279, C343) contribute to Zn 2+ coordination in Tdm The C-terminal GCV_T sequence is best characterized in the T protein of the glycine cleavage complex, which is required for glycine catabolism [7]. GCV_T has a THF-binding site for the conjugation of HCHO and this domain is not known to contain any metal cofactor. In agreement with the role of GCV_T domain in HCHO conjugation, we observed a significant reduction (~10%) of HCHO formation when additional THF was added to the enzyme assay. Similar observation has also been made in other enzymes that contain a GCV_T domain, such as the N-methylglutamate dehydrogenase in Methyloversatilis universalis FAM5 [11]. We thus postulate that metal-binding sites for Zn 2+ and Fe 2+ are likely located in the N-terminal uncharacterized DUF1989 domain.
The PDB database contains three entries for the DUF1989 protein family (3SIY, 3ORU, 3DI4), which all contain Zn 2+ in their crystal structures. To predict the Zn 2+ -binding sites in Tdm, homology modelling was therefore applied. The SWISS-MODEL template library was searched by profile-hidden Markov models (HMMs) HMM-HMM-based lightning-fast iterative sequence search (HHblits). Both algorithms indicated that the N-terminal domain of Tdm (residues 128-352) gave the highest sequence identity (34%) to the sequence of a protein of unknown function 3ORU (1.1 A), which was therefore chosen as the reference structure for modelling.
The established model predicted a conserved Zn 2+ coordination motif in M. silvestris Tdm, despite its poor global and per-residue model quality, assessed using the QMEAN scoring function (Qualitative Model Energy Analysis) [12]. In agreement with the existing structures of DUF1989 family proteins in the PDB database, homology modelling predicted that Zn 2+ was coordinated by three cysteine residues in Tdm (Cys263, Cys279, Cys343) with the thiol S-Zn 2+ distance around 2.3 A ( Fig. 2A). Multiple sequence alignment of Tdm proteins from a range of microbes of terrestrial and oceanic origins covering a-, band CK-Tdm, Tdm purified from recombinant E. coli cultivated using the LB medium; Fe-Tdm, the culture medium was supplemented with Fe (NH 4 ) 2 (SO 4 ) 2 at a final concentration of 0.5 mM; Zn-Tdm, the culture medium was supplemented with ZnCl 2 at a final concentration of 0.5 mM. Values are means AE SD. Different superscript letters in the same row between samples denote significant differences between different metal enrichment (P < 0.05). c-proteobacteria revealed strict conservation of this 3-Cys Zn 2+ -binding motif in Tdm (Fig. 2E).
To probe the role of these residues in Tdm activity, site-directed protein variants were constructed. These variants were purified and their activities and metal contents were characterized. The results showed that all three single variants were inactive, along with significantly reduced Zn 2+ and Fe 2+ contents (P < 0.05) ( Table 2). Although CD spectroscopy revealed only very minor changes in overall secondary structure of these three variants (Fig. 2F), the native homohexamer, which dominates in wildtype Tdm, was virtually absent in the variants (Table 2). Together, the results suggested that C263, C279 and C343 are crucial for maintaining structural integrity, and as a consequence enzymatic activity, in native Tdm.

H276 is a potential Fe 2+ -binding ligand in Tdm
Due to the lack of a crystal structure of Tdm and the absence of Fe 2+ in existing crystal structures of DUF1989 family proteins, we performed multiple sequence alignment of Tdm sequences in order to gain insight into conserved residues, which may shed light on the residues involved in Fe 2+ coordination (Fig. 2E). A 2-His-1-carboxylate facial triad is a common motif in a number of non-haem Fe 2+ -containing enzymes, where two histidine residues and one carboxylate-containing side chain are arranged at one face of an octahedron, whereas the opposite face of the octahedron is available to coordinate a variety of exogenous ligands [13,14]. Our sequence alignment revealed the presence of two strictly conserved histidines (H256, H276) in all Tdm sequences analysed (Fig. 2E).
In order to test whether H256 and H276 were indeed Fe 2+ -binding sites, point mutation variants were constructed, and the variants were subsequently purified and characterized. Data presented in Table 2 revealed that the two variants (H256A, H276A) were completely inactive. Metal analysis showed that the H276A variant indeed had lost Fe 2+ , but its Zn 2+ content remained unchanged compared to that of the wild-type Tdm. Furthermore, the loss of Fe 2+ in the H276A variant cannot be attributed to a structural alteration caused by site-directed mutagenesis as its secondary and quaternary structure is comparable to that of the wild-type (Fig. 2F, Table 2).
The H256A variant was also inactive and had almost completely lost its Fe 2+ . It also had a significantly lower Zn 2+ content (Zn 2+ /Tdm monomer ratio was 0.72 AE 0.01, compared to 0.97 AE 0.03 for wildtype) ( Table 2). Although its secondary structure remained largely unchanged as revealed by CD spectroscopy, this variant did not form the native hexamer (Table 2). Therefore, H256 likely also plays a role in maintaining overall structure.

The 'bridging' nature of D198 in Tdm
The three existing crystal structures of DUF1989 family proteins (i.e. 3ORU) employ a 3-Cys-OH 2 Zn 2+binding motif with the fourth ligand being a water molecule in the crystal structure (Fig. 2B). Searching the immediately surrounding zone (< 2.5 A) [15,16] of the Zn 2+ ion did not reveal any other potential ligands. Values are means AE SD. Different superscript letters in the same column between samples denote significant differences between different metal reconstitution (P < 0.05).

3984
The Although cysteines are commonly found in structural Zn sites, the 3-Cys-OH 2 tetrahedral coordination motif is found in the catalytic sites of several enzymes, e.g. cytidine deaminase (CDA) from Mycobacterium tuberculosis (3IJF) [17], man (1MQ0) [18] and mouse (1ZAB) [19], and the CDA-related enzyme, Blasticidin S deaminase, from Aspergillus terreus (2Z3G, 1WN6) [20]. In these structures, the catalytic H 2 O forms a Hbond to a carboxylate oxygen of a glutamate and the main chain nitrogen of a Cys in the vicinity. In these enzymes, the zinc ion appears to act as electrophilic catalyst through this accompanying water molecule. Structure superimposition revealed that the 3-Cys-OH 2 motif of 3ORU is similar to CDA (Fig. 2B). D66 of 3ORU is in a similar position as the conserved Glu in CDA, i.e. E58 of 3IJF and E56 of 2Z3G (Fig. 2B), indicating that D66 may also H-bond to the Zn 2+bound H 2 O (Fig. 2C).
Multiple sequence alignment of various Tdm proteins and DUF1989 family proteins demonstrated strict conservation of this aspartate residue (corresponding to D198) in Tdm (Fig. 2E). To investigate whether D198 has any significance for TMAO demethylation by Tdm, three site-directed variants were made, D198A (no oxygen atom), D198N (one oxygen atom, neutral side chain) and D198E (two oxygen atoms, negatively charged). All three variants were inactive, and their Zn 2+ contents (Zn 2+ /Tdm ratios between 0.5 and 0.7) were significantly lower than that of the WT (0.97 AE 0.03) (P < 0.05) ( Table 2). Although their overall secondary structure remained largely unchanged, alteration of quaternary structure was observed and the variants were prone to aggregation (Fig. 2F, Table 2).
The Fe 2+ content of D198 variants varied considerably, following the trend of D198A (0.12 AE 0.01) < D198N (0.18 AE 0.02) < D198E (0.33 AE 0.01). Of the three D198 variants analysed, only the D198E variant retained its binding capacity for Fe 2+ , similar to that of the wild-type Tdm (0.33 AE 0.02). It also retained hexameric state as observed in the wild-type enzyme (Table 2). Therefore, our data suggest that D198 also likely provides a carboxyl group for Fe 2+ binding.
The substrate-binding pocket TMAO-binding pockets have been studied previously in two enzymes, namely the substrate-binding protein (TmoX) of the TMAO ABC transporter and TMAO reductase (TorT) [21,22]. In both proteins, a hydrophobic substrate-binding pocket, composed of three to four aromatic residues, was found to recognize and bind TMAO via cation-p interaction [23]. Sequence alignment of Tdm proteins indeed revealed the presence of several conserved phenylalanine (Phe), tyrosine (Tyr) and tryptophan (Trp) residues (Fig. 2E). In an attempt to identify TMAObinding sites in Tdm, we generated site-directed variants by individually replacing these Phe, Tyr and Trp residues with Ala. Three of these variants, F259A, Y305A and W321A, had completely lost activity (Table 2), while the overall secondary and quaternary structure was retained (Fig. 2F, Table 2), suggesting a role of these aromatic residues in substrate binding. Substitution of F259, Y305 and W321 with Ala may have changed hydrophobicity of adjacent Fe 2+ centre and lead to significantly less Fe 2+ content than WT (Fe 2+ /Tdm ratios between 0.05 and 0.15) (P < 0.05). The remainder of the variants (Y185A, Y237A, Y267A, Y273A, W298A, W327A, Y363A) were, however, still active, thus these residues are unlikely to contribute to substrate binding.
The catalytic mechanism of TMAO degradation by Tdm TMAO degradation catalysed by Tdm resembles Ndealkylation by other non-haem Fe 2+ -containing enzymes, such as DNA dealkylase (AlkB) [24], histone demethylase (JHDM1, JMJD6) [25,26] and Riesketype demethylase [27,28]. A high-valent Fe(IV)-oxo complex is a common active species for attacking the C-H bond of saturated carbon centres by non-haem Fe 2+ -containing proteins. O 2 is required as oxygen donor for all the aforementioned reactions to form the Fe(IV)-oxo complex. However, a previous study on bacterial Tdm suggested that TMAO demethylation is O 2 -independent [5]. Our enzyme assays performed using purified Tdm showed no difference in kinetics between aerobically and anaerobically conducted reactions, supporting that TMAO demethylation is indeed O 2 -independent (Fig. 3A). The O 2 -independency therefore suggests that TMAO may function as both the substrate and the oxygen donor for the subsequent demethylation. Although such a mechanism of surrogate oxygen donation has not been observed in non-haem iron enzymes, it has been found in haemcontaining P450 enzymes in the absence of NAD(P)H and oxygen (reviewed in ref. [29,30]). Based on the wellstudied P450 enzymes, we propose a putative Tdm catalytic cycle (Fig. 3B), where a high-valent oxidant (e.g. Fe(IV)-oxo) is generated through oxygen atom transfer from TMAO and a tertiary amine intermediate (i.e. trimethylamine, TMA) is formed, which then serves as a substrate for oxidative demethylation to DMA. To test this hypothesis, we performed a crossover experiment using a TMA analogue in order to trap the formation of putative secondary amine species that can be released during the Tdm catalytic cycle [31]. N, N-dimethylethylamine (DMEA), a structural analogue of the postulated TMA intermediate, was added to the enzyme assays with and without the substrate, TMAO. If the tertiary amine intermediate (TMA) is formed and acts as a substrate during the catalytic cycle, a postulated high-valent Fe species, e.g. Fe(IV)-oxo, may also demethylate its analogue DMEA, hence forming the corresponding secondary amine species, methylethylamine (MEA) (Fig. 4A). The secondary amine products (DMA, MEA) were derivatized by benzenesulfonyl chloride (BSC) and quantified by GC-MS. In the absence of TMAO, Tdm does not catalyse the demethylation of DMEA. However, in the presence of TMAO, both DMA (m/z: 77.1, 141.1, 185.1) and MEA (m/z: 77.1, 141.1, 184.1) were detected, and DMA formation was competitively reduced in the presence of DMEA (Fig. 4B-D). The results therefore supported the postulated mechanism, confirming that TMAO is required as the oxygen donor, and that the resulting tertiary amine is a substrate for oxidative demethylation during the catalytic cycle of Tdm.

Tdm is a novel Zn 2+ -and Fe 2+ -containing metalloprotein
In this report, we presented data suggesting that Tdm is a Zn 2+ -and Fe 2+ -dependent metalloenzyme and the Zn 2+ /Fe 2+ /Tdm monomer ratio is likely to be 1/1/1. It is noteworthy that both Zn 2+ and Fe 2+ are in the close vicinity of D198. Therefore, it is possible that Tdm has a cocatalytic binuclear Zn 2+ -Fe 2+ centre. Binuclear Zn/Fe sites have been found in many enzymes throughout nature, such as glycerophosphodiesterase [32], glyoxalase II [33], enamidase [34] and protein phosphatase 2B [35]. However, unlike Tdm, in these binuclear Zn/Fe hydrolases, Zn 2+ is predominantly coordinated by histidine side chains.
Our data suggest that in Tdm, Zn 2+ is coordinated by three cysteine thiolates (C263, C279 and C343) and one water molecule, which resembles the catalytic  Zn 2+ site of CDA [17][18][19]. However, caution should be applied when interpreting the Zn 2+ site in Tdm due to the low quality of the overall model obtained through homology modelling. We cannot rule out the possibility that the fourth Zn 2+ ligand may be provided by the D198 in Tdm (Fig. 2D). Indeed, we propose that D198 has a dual role of stabilizing both Zn 2+ and catalytic Fe 2+ (Fig. 2C,D). Mutagenesis study of the conserved D198 in Tdm supported that D198 is crucial for maintaining structural integrity through interaction with Zn 2+ , either directly or indirectly through H-bonding with Zn 2+ -bound water. Replacing D198 with Ala, Asn and Glu resulted in quaternary structural alteration as well as reduced Zn 2+ content in the Tdm variants ( Table 2).
D198 also likely contributes to Fe 2+ -binding, together with H276. Results presented in Table 2 demonstrate that a carboxyl group is important in maintaining Fe 2+ stoichiometry in Tdm and the Fe 2+ / Tdm ratio of the D198E variant, but not the D198A variant, was comparable to that of the WT. So far, it is unclear what other residue(s) could contribute to Fe 2+ coordination in Tdm. Tdm may employ a nonclassic 2-His-1-carboxylate triad. Variations of the classical 2-His-1-carboxylate triad motif for Fe 2+ coordination have recently been found in a number of enzymes. For example, in the halogenase SyrB2, the carboxylate is absent and instead, a halogen ion takes its place in the coordination sphere [36]. In diketonecleaving dioxygenase, Dke1 [37], and cysteine dioxygenase, CDO, a three-histidine triad is found [38,39], whereas in carotenoid oxygenase, a four-histidine motif is present [40,41]. Further structural and biochemical investigations are certainly warranted to conclusively map the ligands involved in Zn 2+ and Fe 2+ coordination in Tdm.

Substrate recognition and binding by a hydrophobic pocket
Our site-directed mutagenesis results (F259A, Y305A and W321A) suggest that these residues may contribute to form substrate pocket that recognizes and binds TMAO by cation-p interaction. This interaction has also been found in other TMAO-binding proteins, such as TmoX and TorT [21,22]. The cation-p interaction has long been recognized as an important noncovalent binding interaction relevant to structural biology [23]. The aromatic rings of Phe, Try and Trp provide negative electrostatic potential allowing interaction with cations [42]. Such cation-p interaction also seems common in proteins involved in quaternary amine transport and metabolism, e.g. choline-TMA lyase (CutC) from Desulfovibrio alaskensis G20 [43], acetylcholine esterase from Tetronarce californica [44,45], substrate-binding protein (ChoX) of choline/ acetylcholine from Sinorhizobium meliloti [46], and substrate-binding protein (ProX) of glycine betaine from Archaeoglobus fulgidus [47]. Interestingly, the Fe 2+ / Tdm ratios of these variants were also significantly reduced compared to that of the wild-type. It is therefore likely that these hydrophobic residues are located adjacent to the Fe 2+ centre and thus may directly or indirectly influence Fe 2+ coordination in Tdm.
Chemically, many single-oxygen atom donors are capable of generating Fe(IV)-oxo species (e.g. NaOX, X = Cl or Br, iodosylbenzene (PhIO)) [48,49]. However, a surrogate single-oxygen atom donor has also been found in biological systems, such as P450 enzymes. N,N-dimethylaniline N-oxide (DMAO) is an intermediate formed during P450-mediated demethylation of N,N-dimethylaniline. Recently, studies have demonstrated that DMAO can serve as a surrogate single-oxygen donor because DMAO gave identical isotope effects as the natural system using NADPH and O 2 [50,51]. We propose that, akin to DMAO for P450 enzymes, TMAO can act as a surrogate oxygen donor for Tdm. A potential mechanism for TMAO degradation catalysed by Tdm is shown in Fig. 3B.
We postulated that a high-valent Fe is responsible for TMAO demethylation (Fig. 3B) and the nature of the active Fe species remains unclear. The Fe(IV)-oxo species is common in haem and non-haem Fe 2+ -containing enzymes [52][53][54][55]. In addition, another high-valent ironoxo complex, Fe(V)-oxo, has been postulated as an active oxidant in Rieske dioxygenase enzymes [56,57] and authenticated in non-haem iron biomimetic systems [58]. Indeed, our crossover experiments did support the presence of a TMA-alike intermediate during the catalytic cycle as proposed in Fig. 3B. According to this hypothesis, the oxygen atom is transferred from the substrate TMAO to produce formaldehyde via the high-valent iron-oxo intermediate (e.g. Fe(IV)-oxo, Fe (V)-oxo). Clearly further investigations, such as using electron paramagnetic resonance spectroscopy [48,54], are required to identify the active Fe species involved in Tdm catalysis. The exact role of Zn 2+ in the progression of TMAO demethylation also remains unclear. Although our sitedirected mutagenesis data largely supported the role of Zn in maintaining Tdm structure, its involvement as a cocatalytic Zn is also possible. For example, one can envisage that Zn 2+ may facilitate oxygen atom transfer from TMAO to Fe through Lewis acid-base interaction with the O atom at step ② (Fig. 3B). Similarly, Zn 2+ may bind to the O atom of the complex formed at step ④ to mediate C-N bond cleavage (Fig. 3B). A similar role of Zn in the catalytic site of alcohol dehydrogenase (ADH) for C-H bond cleavage is well known [59]. Alternatively, Zn 2+ may also stabilize the reactive high-valent Fe-oxo species during catalysis through electrostatic interaction via a solvent oxygen, e.g. Fe(IV)-oxo (step ③, Fig. 3B). It has been shown that the conserved positively charged arginine residue can help to stabilize and polarize the negative charge on the reactive Fe(III)-OOH complex in peroxidase [60][61][62].
In summary, the combination of site-directed mutagenesis, homology modelling and analytical chemistry have provided insight into the structure-function relationship of a novel Zn 2+ -and Fe 2+ -dependent metalloprotein, Tdm. It carries out an unusual O 2independent oxidative demethylation utilizing the substrate as the oxygen donor. Determination of the three-dimensional structure is now required to validate the model we proposed.

Materials and methods
Cloning, expression and purification of Tdm and variants of M. silvestris Plasmids and strains used for cloning and overexpression of Tdm and its variants in E. coli are listed in Tables S1 and S2. Tdm expression and purification were carried out as described previously [3]. Briefly, E. coli cells were grown at 37°C to an OD 600 of 0.5, and isopropyl b-D-1-thiogalactopyranoside (IPTG) was then added to a final concentration of 0.2 mM. Protein expression was carried out at 18°C before harvesting. Tdm was purified by 6*His-tag affinity purification from recombinant E. coli according to the manufacturer's protocol (Merck, Nottingham, UK). Proteins eluted from the affinity column were further purified by desalting against 20 mM Tris/HCl (pH 7.9), 50 mM NaCl using a HiTrap desalting column (GE Healthcare, Uppsala, Sweden) followed by size-exclusion chromatography using a Superdex 200 10/300 GL column (GE Healthcare). Fractions were pooled and concentrated by ultrafiltration (Amicon Corporation, Danvers, MA, USA). Protein concentrations were determined using a protein assay kit (Bio-Rad, Watford, UK). Site-directed mutations in tdm were introduced by PCR and confirmed by DNA sequencing. The oligonucleotides used in this study are shown in the supplementary information.

Enzyme activity assay
Enzyme activity was measured by quantifying HCHO production from TMAO degradation. The reaction mixture for HCHO assays contained 2.5 lg of purified Tdm in 50 lL of 10 mM MES buffer (pH 6.0). The reactions were initiated by adding TMAO (0.6-8 mM final concentration) into the mixture and incubated for 10 min unless otherwise specified [3]. Michaelis-Menten constant (K m ), maximum velocity (V max ) and catalytic constant (k cat ) were calculated as described previously [3]. Tdm activity was also measured in the absence of oxygen in an anaerobic chamber (MACS-MG-500 anaerobic workstation; Don Whitley Scientific, Shipley, UK). All solutions were degassed prior to use. To test the impact of metal chelators on Tdm activity, a range of concentrations of the following compounds were used, including ethylenediaminetetraacetic acid (EDTA, 0-50 mM), ethylene glycol-bis(b-aminoethyl ether)-N,N,N 0 ,N 0 -tetraacetic acid (EGTA, 0-50 mM) and pyridine-2,6-dicarboxylic acid (PDC, 0-10 mM). Neither PDC nor EGTA affected Tdm activity. Ascorbic acid (0-10 mM) and H 2 O 2 (0-100 lM) were also tested in the enzyme assays with 5 mM TMAO to investigate their effects on Tdm activity.
For the metal replacement experiment, purified Tdm was diluted in a buffer containing 20 mM Tris/HCl (pH 7.8) and 50 mM NaCl to a final concentration of 1 mgÁmL À1 and incubated with various metal ions (MgCl 2 , MnCl 2 , Fe (NH 4 ) 2 (SO 4 ) 2 , CoCl 2 , NiSO 4 , CuSO 4 , ZnSO 4 ) at an M 2+ / Tdm ratio of 500 for 6 h at 4°C. Two millimolar ascorbic acid was used when Fe(NH 4 ) 2 (SO 4 ) 2 was applied to maintain Fe 2+ in its reduced state. About 2.5 lg of Tdm and 5 mM TMAO were used for HCHO assay as described above.
For the Tdm reactivation experiment, the 6*His-tag was removed using thrombin (GE Healthcare, Little Chalfont, UK) according to the manufacturer's instructions. Briefly, 1 mg purified recombinant Tdm was incubated with 10 units of thrombin at 4°C overnight (16 h). Thrombin and cleaved 6*His-tag were removed by size-exclusion chromatography. Tdm fractions were pooled, and diluted in buffer containing 20 mM Tris/HCl (pH 7.8) and 50 mM NaCl to a final concentration of 1 mgÁmL À1 , which was then incubated with 100 lM H 2 O 2 for 20 min on ice. Wherever necessary, ascorbic acid and Fe(NH 4 ) 2 (SO 4 ) 2 was added to a final concentration of 500 lM and 10-500 lM respectively. The samples were then incubated for 20 min on ice prior to the quantification of Tdm activity using the aforementioned HCHO assay.

Fe 2+ titration
The Fe 2+ titration (using Fe(NH 4 ) 2 (SO 4 ) 2 ) was performed with the 6*His-tag-cleaved Tdm. To maintain Fe 2+ in its reduced form, the titration assay was performed with 2 mM ascorbic acid with varying molar ratios of Fe 2+ /Tdm in the Tris/NaCl buffer (20 mM Tris, 50 mM NaCl, pH 7.8) for 4 h before HCHO essay was carried out. The titration data for metal stoichiometry (n) determination were analysed by nonlinear curve fitting using Eqn (1) [63,64].
where x is the concentration of total metal ion and y is the activity percentage to the maximal activity, y m is the maximal activity, y 0 is the activity percentage of as-isolated enzyme without added metal, K d is the dissociation constant and n is the number of binding sites (stoichiometry).
Inductively coupled plasma-mass spectrometry and optical emission spectrometry (ICP-MS/OES) About 3% (v/v) trace metal grade nitric acid purified in house by sub-boiling point distillation was used as the sample matrix. ICP-MS analyses were carried out on an Agilent Technologies 7500 ICP-MS instrument. ICP-OES was performed on a Perkin Elmer Optical Emission Spectrometer Optima 5300DV instrument. The standards for calibration were freshly prepared by diluting Zn, Fe, S, Ni stock solutions (at 1000 mgÁL À1 ; Sigma-Aldrich, Saint Louis, MO, USA) with 3% HNO 3 in doubly deionized water with concentrations from 0.2 to 1 mgÁL À1 for Zn, Fe and Ni, and from 4 to 20 mgÁL À1 for S. About 2.4 mg protein was diluted in 3% HNO 3 matrix for metal analysis. The content of S was quantified in order to determine the protein concentration. The contents of Zn, Fe, Ni and S were measured using the emission lines of 213.857 nm (Zn), 234.830 nm (Fe), 231.604 nm (Ni) and 180.669 nm (S) respectively.
Gas chromatography-mass spectrometry (GC-MS) determination of secondary amines were derivatized using benzenesulfonyl chloride (Sigma-Aldrich) and determined by GC-MS as described previously [65].

Secondary and quaternary structure determination
Secondary structure was determined by circular dichroism using a Jasco J-815 spectrometer (Jasco, Great Dunmow, UK) and the secondary structure components (a-helix, bsheet, turn and random coil) were estimated using the CDSSTR algorithm with reference set 7 from the DICHROWEB website as described previously [68,69]. Quaternary structure was determined by size-exclusion chromatography on a Superdex 200 10/300 GL gel filtration column (GE Healthcare, Uppsala, Sweden) equilibrated with buffer containing 20 mM Tris/HCl (pH 7.8) and 100 mM NaCl at 0.7 mLÁmin À1 using an AKTA FPLC system (GE Healthcare, Chalfont St. Giles, UK).

Multiple sequence alignments
Multiple sequence alignments were performed using the iterative alignment program MUSCLE [70].

Statistical analyses
Analysis of variance (ANOVA) and Tukey HSD post hoc tests were performed using the R software package version 3.2.1 [71]. Data are expressed as means AE SD.

Supporting information
Additional Supporting Information may be found online in the supporting information tab for this article: Table S1. Bacterial strains and plasmids used in this study. Table S2. Primers used for site-directed mutagenesis of Tdm.