Ferrous iron and 2-oxoglutarate-dependent oxygenases and related enzymes catalyse a range of oxidative reactions, possibly the widest of any enzyme family. Their catalytic flexibility is proposed to be related to their nonhaem iron-binding site, which utilizes two or three protein-based ligands. A possible penalty for this flexibility is that they may be more prone to oxidative damage than the P450 oxidases, where the iron is arguably located in a more controlled environment. We review the evidence for autocatalysed oxidative modifications to 2-oxoglutarate-dependent oxygenases, including the recently reported studies on human enzymes, as well as the oxidative fragmentations observed in the case of the plant ethylene-forming enzyme (1-aminocyclopropane-1-carboxylic acid oxidase).
The family of dioxygenases that employ 2-oxoglutarate (2OG) as a cosubstrate and ferrous iron as a cofactor is ubiquitous in aerobic life forms. Most members of the 2OG oxygenase family couple the two-electron oxidation of their substrates with the reaction of oxygen and 2OG to give carbon dioxide and succinate (Fig. 1A). In animals, the currently identified reactions catalysed by 2OG oxygenases appear to be limited to hydroxylation (and subsequent oxidation of alcohol products) and demethylations via hydroxylation. Although the most common reaction catalysed by family members is likely still to be hydroxylation, in microorganisms and plants, 2OG oxygenases catalyse a very wide range of oxidative reactions, including N-methyl and O-methyl demethylation via initial hydroxylation, desaturation, oxygen insertion, epoxidation, oxidative rearrangement/epimerization [1–3], oxidative halogenations  and, in a recently reported case involving oxidation of a clinically used inhibitor [3-(2,2,2trimethylhydrazine)propionate], a Stevens type reaction .
The diversity of biological roles of the 2OG oxygenases reflects their catalytic flexibility. In humans and likely most other higher animals, including the simplest animal Trichoplax adhaerens , they play diverse biological roles, including in collagen biosynthesis (prolyl- and lysyl-hydroxylation of procollagens), lysyl-hydroxylation of RNA splicing-related proteins, DNA repair, RNA modification, chromatin regulation/histone modification (Nε-methyl lysine demethylases), epidermal growth factor-like domain modification (aspartyl- and asparaginyl-hydroxylation), hypoxia sensing (prolyl- and asparaginyl-hydroxylation of a transcription factor) and fatty acid metabolism . They likely also have additional roles because not all of the 60–70 human 2OG oxygenases have been assigned functions. In microorganisms and plants, 2OG oxygenases also play roles in the biosynthesis of secondary metabolites and signalling molecules; some of their roles in animals also occur in plants.
The catalytic flexibility of 2OG oxygenases in terms of the reaction type catalysed is perhaps best illustrated by considering some of their roles in antibiotic biosynthesis. During the biosynthesis of the cephalosporin nucleus, they catalyse the formation of the bicyclic penicillin ring structure (isopenicillin N synthase; IPNS), the oxidative rearrangement of the penicillin nucleus to give a cephem ring, and modification of the latter by allylic hydroxylation (Fig. 2A), with the latter two steps catalysed by deacetoxy- and deacetyl-cephalosporin synthases) [8,9]. In the case of clavulanic acid biosynthesis, a single 2OG oxygenase, clavaminic acid synthase (CAS) catalyses three steps comprising hydroxylation, oxidative ring closure and desaturation reactions (Fig. 2B) . A similarly high degree of flexibility occurs in 2OG oxygenase-catalysed reactions in plants (e.g. in the biosynthesis of the giberellin signalling molecules and of the flavonoids) [11,12]. The 2OG oxygenases also accept a wide range of substrate types, including structurally varied small molecules, structured and unstructured proteins, nucleic acids and lipids. Several members of the 2OG oxygenase family that catalyse oxidative reactions do not use 2OG as a cosubstrate, although they have clearly related structures. These enzymes include IPNS  and 1-amino-1-cyclopropanecarboxylic acid oxidase (ACCO), which catalyses the final step in ethylene biosynthesis in plants [14,15]. This diversity in substrate type is proposed to render 2OG oxygenases amenable to selective inhibition by small molecules for therapeutic intervention . Indeed, some human 2OG oxygenases are being targeted for medicinal purposes, including the hypoxia-inducible factor hydroxylases, 2OG-dependent histone demethylases and enzymes involved in carnitine biosynthesis .
The available evidence is that all 2OG oxygenases have conserved structures based on a double-stranded β-helix fold (also known as jelly roll or cupin fold) [17–19] that supports the conserved iron- and 2OG-binding sites. It thus appears likely that most 2OG oxygenases arose by divergent evolution. However, there are substantial differences in the available structures, including in the substrate-binding region and in the presence/absence of adjunct noncatalytic domains. The iron cofactor is normally octahedrally coordinated by three protein ligands comprising the HXD/E…H/triad formed by the well conserved motif of two histidine imidazole groups and a carboxylate, either from aspartate or glutamate (Fig. 1B) [20,21]. However, in the case of the 2OG-dependent halogenases, only two protein ligands are employed . Recent studies also reveal that, at least in one case, only two ligands are necessary for hydroxylation in vitro . The apparent unusually high degree of catalytic flexibility exhibited by the 2OG oxygenases is proposed to reflect their active site chemistry/mechanism . In the currently proposed consensus mechanism, although there may be some exceptions (e.g. the deacetoxycephalosporin synthase reaction) (Fig. 2A) , ferrous iron binds first to the apo-enzyme followed by 2OG (Fig. 1C). There is evidence, at least in some cases, that binding of iron and 2OG results in parts of the overall structure becoming more ordered [25,26], perhaps to aid in the binding of the ‘prime’ substrate that occurs next . The 2OG binds to the iron in a bidentate manner via its 1-carboxylate and 2-oxo groups. The 2OG binds in a pocket of varying size (this variation can be exploited in the design of selective inhibitors for particular 2OG oxygenase) [16,27]. Extensive crystallographic studies have revealed that the 2OG C-5 carboxylate is, at least typically, positioned to form an electrostatic interaction with a lysine- or arginine-residue and a hydrogen-bonding interaction with one or more alcohol/phenol side-chains . The available evidence implies that there is more variation in the 2OG-binding pocket than in the iron-binding ligands, both in terms of identity and the positions of the residues involved in binding in the overall structure. Binding of the substrate is proposed to prime the metal for binding of oxygen by weakening the binding of a water molecule that completes iron coordination . There is some uncertainty as to whether the oxygen binds initially in the vacant coordination position closest to the substrate (i.e. that trans to the distal histidine of the HXD/E…H motif) or in the position trans to the proximal histidine of the HXD/E...H motif [18,29]. This arises because, although the coordination position of the 2-oxo oxygen of 2OG is (almost) invariant, being located trans to the acidic residue of HXD/E…H motif in reported 2OG oxygenase crystal structures, the observed position of the C-1 carboxylate varies between the two possible remaining octahedral coordination sites [17,18]. Indeed, in one case, the position of the 2OG C-1 carboxylate changes upon binding of nitric oxide . The structural observations of variations in 2OG binding further illustrate flexibility in the active sites of the 2OG oxygenases. Subsequent to oxygen binding, oxidative decarboxylation of 2OG occurs to give carbon dioxide, succinate and a ferryl-oxo intermediate for which there is spectroscopic evidence [1,30,31]. The ferryl-oxo species can then react with the substrate to effect hydroxylation or other oxidative chemistry, in a manner similar to that proposed for P450 oxidases [32,33]. Release of product followed by succinate then occurs. The timing of CO2 release is unclear in most cases. There is evidence that the relative rates of the individual steps may vary, with, in some cases, the reaction of the ferryl-oxo species being rate limiting and, in other cases, the release of product or succinate being rate limiting [1,34,35]. Not all members of the 2OG oxygenase family employ 2OG as a cosubstrate. IPNS, which is structurally related to 2OG oxygenses , catalyses the four-electron oxidation of its tripeptide substrate to give the bicyclic penicillin nucleus (Fig. 2A). Notably its mechanism involves direct coordination of the cysteinyl thiol of its peptide substrate to the ferrous iron cofactor . The mechanism of ACCO, which is probably not entirely resolved, is also proposed to involve bidentate chelation of the prime substrate (ACC) to the active site iron via its α-amino acid [36,37].
The proposal that the 2OG oxygenases can relatively easily evolve new product selectivities (e.g. hydroxylation versus desaturation) is consistent with their di- and trifunctional roles in some biosynthetic pathways (Fig. 2); it can be envisaged that gene duplication followed by mutation can lead to the evolution of new but related activities, acting on the product of the first catalysed reaction or a metabolite thereof. One example is the role of CAS in clavulanic acid biosynthesis. In the first CAS catalysed reaction, hydroxylation occurs, whereas, in the second and third, oxidation ring closure and desaturation reactions occur (Fig. 2). The different types of CAS catalysed oxidation reaction are likely governed by subtle differences in the substrate binding mode at its active site. Substrate analogue studies on IPNS also illustrate how the product selectivity of the family can be dramatically relaxed by use of appropriate substrate analogues [8,38]. The range of substrates and substrate analogues accepted by 2OG oxygenases further supports the proposal that evolving new selectivities might be unusually readily achievable for this enzyme family, consistent with their common use in the biosynthesis of secondary metabolites. The question arises as to whether or not the 2OG oxygenases are especially suited for carrying out oxidative reactions and, in particular, those that are viewed as being ‘chemically’ difficult (e.g. hydroxylation of an activated C-H bond).
Although there are similarities between the proposed mechanisms for the 2OG oxygenases and the P450 oxidases, particularly in terms of the generation of ferryl and related intermediates and their reaction, there are clear differences. Perhaps the most important aspect of these is that, in 2OG oxygenase catalysis, there is a potential for substrates and cosubstrates to chelate to the metal during catalysis. This cannot (at least easily) occur in P450 catalysis because the oxygen binds to one site and the haem ring both occupies four coordination sites and blocks the approach of the substrate to the distal coordination site [32,33]. It is proposed that the more flexible coordination chemistry of the 2OG oxygenases, which involves only two or three protein ligands, may be useful in terms of evolving new catalytic activities. So why is it that the 2OG oxygenases are not more prevalent? One possibility is that the ‘flexibility’ inherent in their coordination chemistry makes them more vulnerable to oxidative damage via the ‘inadvertent’ reaction of high-energy intermediates. Conversely, the haem ring of the P450 oxidases may render them relatively more robust by better isolating reactive intermediates as a result of more ‘rigid’ metal chelation. Thus, because they may generally be more robust, some P450 oxidases may be better suited than the 2OG oxygenases for catalysing the oxidation of multiple substrates, as occurs in the liver metabolism of small molecules/pharmaceuticals [39,40]. It should be noted that there is evidence that haem proteins, including P450 enzymes, are also prone to oxidative damage .
We are aware of the dangers and difficulties in linking the biochemical properties of isolated enzymes with their in vivo roles. Furthermore, there is no doubt that, within cells, 2OG oxygenases can be very efficient and robust catalysts, as demonstrated by the remarkably efficient fermentation titres that can be achieved for antibiotic biosynthesis. Nevertheless, there is some evidence that some 2OG oxygenases can be rather fragile in their isolated form (in some cases as a result of oxidative modifications) and, to date, this property may have limited their use as biocatalysts in purified forms.
Auto-oxidation of 2OG oxygenases
Loss of catalytic activity for some 2OG oxygenases in purified form is common. Indeed, this is likely the main reason why, to date, they have not been developed as useful preparative scale catalysts in isolated forms. Loss of activity can occur as a result of the removal of iron from the active site or oxidation of active site ferrous iron to the ferric (or higher) oxidation state. These problems can be partly overcome by the addition of ferrous iron or a reducing agent (commonly ascorbate), respectively. As with other enzymes, precipitation or aggregation can also be a problem, particularly at higher concentration and during catalysis. There is also accumulating evidence that residues of 2OG oxygenases can undergo auto-oxidation. In the case of 2,4-dichlorophenoxyacetate oxygenase (TfdA), which catalyses a step in the degradation of a herbicide 2,4-dichlorophenoxy acetate , and taurine/2-oxoglutarate dioxygenase (TauD), which catalyses the oxidation of taurine to give bisulfite , detailed spectroscopic studies have provided evidence for the hydroxylation of (an) active site aromatic residue(s).
In the case of TfdA, an enzyme Fe(II)-2OG complex was reported to react with oxygen (in the absence of its substrate 2,4-dichlorophenoxy acetate) to form a blue species . MS analyses led to the proposal that the blue species results from hydroxylation of a tryptophan residue (Trp112) (Fig. 3A) adjacent to one of the iron-binding ligands. Evidence was provided showing that the hydroxylated tryptophan can chelate to a ferric ion located in the active site . The blue species was inactive even when excess ascorbate and 2OG were added. Substantial activity (∼ 60%) was restored when the ‘blue-enzyme’ was treated with dithionite. However, an even higher level of activity (∼ 81%) was restored when the dithionite-treated blue species was dialyzed with EDTA, and then reconstituted by adding Fe(II). The restoration of activity was interpreted as resulting from the displacement of the hydroxylated tryptophan from the iron; notably the hydroxylated tryptophan does not appear to ligate to the iron in the active reconstituted enzyme, presumably as a result of kinetic–structural barriers . However, it is not certain that the blue colour arises from the same (form of) hydroxylated tryptophan observed in the MS analyses and the regiochemistry of tryptophan hydroxylation remains undetermined. 16O/18O isotope labelling experiments provide interesting mechanistic insights. Spectroscopic studies imply that the oxygen of the hydroxylated tryptophan (or at least the ‘blue species’) is derived from H2O, and not O2. Thus, although MS data on the incorporation of labelled oxygen into the hydroxylated tryptophan are not reported, spectroscopic studies imply that the transfer of oxygen to Trp112 occurs with complete exchange with H2O . A stepwise mechanism may occur, potentially involving one-electron oxidation of Trp112 by an iron peroxo/oxo species to form a tryptophan radical, which can react with an Fe(III)-OH species that equilibrates with water  (Fig. 3B).
In analogous studies on TauD, the enzyme-Fe(II)-2OG complex was observed to form a greenish–brown chromophore via a transient yellow species , which (on the basis of EPR studies) was considered to arise from a tyrosine-based radical. The greenish–brown chromophore was assigned as an Fe(III)-catecholate (i.e. dihydroxy phenylalanine, residue), arising from hydroxylation of a tyrosine residue, proposed on the basis of mutational studies to be Tyr73, which is located close to the iron-binding site (Fig. 4A). As with the studies on TfdA, combined spectroscopic and labelling studies imply that the oxygen of the introduced alcohol group is derived from H2O and not O2.
In the case of factor-inhibiting hypoxia-inducible factor (FIH), a transcription factor asparginyl-hydroxylase, evidence for autohydroxylation of a tryptophan residue  (Figs 3A and 4B) that is important in substrate binding has also been reported . In the absence of a FIH protein substrate, formation of a blue colour was observed in a similar manner to that observed for TfdA, which was assigned as a Fe(III)-O-Trp chromophore. After proteolytic MS assays, Trp296 was identified as being hydroxylated.
In the case of 2OG oxygenases involved in DNA repair (or regulation), there is evidence for oxidation of a leucine residue [data reported for human AlkB homologue (ABH)3, reported but not shown for AlkB, ABH2 and ABH6]  (Figs 3A and 4C). These results are interesting because they involve the oxidation of an unactivated C-H bond. Evidence for the modification of Leu177 in the ABH3 active site came initially from an analysis of electron density maps arising from crystallographic analyses and was supported by MS analyses; intact tryptic fragments displayed a +16 and +14 mass shift relative to the predicted mass of unmodified peptide corresponding to the occurrence of hydroxylation and possibly carbonyl formation . A separate study on AlkB reported observations analogous to those for TfdA and FIH in that, in the presence of Fe(II), 2OG and oxygen, a blue chromophore is formed, which is proposed to arise from the formation of a Fe(III)-hydroxy-tryptophan complex . MS analyses supported the oxidation of Trp178, which is positioned at the active site (Fig. 4D).
An interesting MS-based study on oxidative modifications of a viral collagen prolyl-hydroxylase has revealed the potential of ‘top-down’ MS-based approaches for studying low-level oxidative modifications to proteins . Evidence was presented for three (and possibly four) oxidations (16 Da mass shifts in the peptide fragment) in the region of the active site. Of particular note was the observation that Lys216, which is proposed to be involved in 2OG binding, is oxidized.
Crystallographic studies on an alkyl sulfatase (AtsK) have provided evidence that an aromatic active site residue can directly coordinate to the iron . In the crystal structure for the AtsK-Fe-succinate complex (but not in other complexes studied), a side-chain of Tyr168 was observed to coordinate directly to the Fe (Fig. 4E). Direct evidence for the self-hydroxylation of Tyr168 was not described for AtsK; however, as described above, there is such evidence for other 2OG oxygenases. The structure also highlights the potential for iron coordination other than by the stereotypical HXD/E...H motif. It is notable that, in each of the cases for TfdA, TauD, FIH, AlkB and AtsK, an aromatic residue (tyrosine or tryptophan) close to the active site is oxidized.
Recent studies from our group have shown that jumonji domain-containing protein 6 (JMJD6) is a lysyl hydroxylase acting on RNA splicing-related proteins , which also undergoes autohydroxylation . However, unlike the aforementioned cases (TfdA, TauD and ABH3), JMJD6 catalysed autohydroxylation occurs at the same residue type (i.e. lysine residues) that is oxidized during substrate hydroxylation (Fig. 3A). Importantly, amino acid analyses demonstrate that the stereospecificity of the autohydroxylation reaction is the same as that observed for substrate hydroxylation (i.e. the 5S-product is produced in both cases) . Two autohydroxylation sites have been identified in JMJD6 (Fig. 4F); however, combined amino acid and MS fragmentation analyses reveal that there are likely sites of autohydroxylation that were not identified in the ‘proteomic’ MS analyses. The results also imply that factors in addition to the local sequence are involved in regulating JMJD6 autohydroxylation. At this stage, it is unclear whether the JMJD6 autohydroxylation is inter- or intra-molecular. Notably, the identified JMJD6 autohydroxylation sites are further away from the active site iron (∼ 22 Å) than those identified in the aforementioned cases, as shown (∼ 10 Å) in Fig. 4F. JMJD6 from human cells (HeLa) also undergoes autohydroxylation. This contrasts with most of the other reported related studies where experiments were carried out with recombinant enzymes, although there is evidence that the observed TauD modification (at Tyr73) occurs in the wild-type enzyme isolated from aerobic cells . This observation, in addition to the fact that JMJD6 hydroxylation occurs in a stereospecific and 2OG-dependent manner suggests that JMJD6 autohydroxylation might have a biologically relevant role.
ACCO catalyses the final stage in the biosynthesis of the signalling molecule ethylene in plants (Fig. 2C). Although ACCO is not 2OG-dependent, it is closely related in structure and sequence to 2OG-dependent oxygenases and IPNS . Early studies indicated that ACCO undergoes inactivation both in the case of recombinant tomato ACCO  and in fruit tissues  (inactivation is apparently irreversible) . Inactivation was shown to be promoted by the addition of ferrous iron and ascorbate, and not a result of co-product inhibition by dehydroascorbate or cyanide. Ascorbate-dependent incubation was prevented, although not reversed, by the addition of catalase . Subsequent mutagenesis studies on recombinant tomato ACCO demonstrated that ACCO undergoes metal catalysed oxidative fragmentation dependent on the presence of an intact iron binding site [54,55]. The rate of iron/ascorbate-mediated inactivation was increased by the addition of ACC, unaffected by the addition of bicarbonate (which stimulates catalytic activity) and decreased by the addition of catalase (as in the case reported by Smith et al. ) or ACC and bicarbonate . The fragmentation pattern present was condition-dependent (e.g. altered by the addition of ACC). Edman sequencing resulted in the identification of cleavage sites (after Leu186 and Val214) , which, in subsequent crystallographic studies , were shown to be close to the active site (Fig. 5). Based on the overall results, it was proposed that ACCO undergoes hydrogen peroxide-mediated damage that is catalase protectable. It was also considered that ACCO can undergo active site modification (sometimes) leading to fragmentation (which is not catalase protectable) and inactivation by unfolding/aggregation. With respect to the latter, it is important to note that, although the present review focuses on oxidative damage leading to covalent modifications, a loss of activity as a result of conformational changes or aggregation may be an important factor with 2OG oxygenases (as is the case with many other enzymes).
Possible mechanisms of auto-oxidation by 2OG oxygenases
Metal catalysed oxidation of proteins is a well recognized biochemical ageing process [56–58] and has been used to ‘map’ the active sites of metallo-enzymes [59,60]. Oxidative modifications of proteins (and other biomolecules including nucleic acids) by iron and ascorbate reaction are also well known [57,61,62]. The selective oxidation of particular residues by reactive oxygen species is less well established, although there are examples including alcohol dehydrogenases , malic enzyme , glutamine synthase [64,65] and 1,2-propanediol oxidoreductase . In some cases, backbone fragmentation has been observed, as for ACCO.
The mechanisms by which the autohydroxylation of tryptophan (TfdA and FIH), tyrosine (TauD) and leucine (ABH3) residues are hydroxylated and fragmentation of ACCO occurs is not yet clear. It is possible that different mechanisms operate in different cases. The 2OG oxygenase substrate ‘uncoupled’ turnover of 2OG to succinate has long been known . The extent of uncoupled 2OG turnover varies between enzymes. One role for ascorbate in promoting 2OG catalysis in vitro (although there are likely others) [67,68] is proposed to be to reduce high (e.g. ferryl) oxidation state intermediates arising from uncoupled turnover, thus returning the enzyme to an active form . However, evidence for specific requirement of ascorbate in vivo is lacking. It is possible that an Fe=O species is generated by uncoupled 2OG turnover and can cause autohydroxylation. However, as described above, at least in some cases (TauD, TfdA), oxidation of aromatic residues occurs without incorporation of 18O from 18O2. This contrasts with the case of substrate hydroxylation where the alcohol oxygen comes (minimally predominantly) from O2. Thus, as described above for TfdA and TauD, it is possible that exchange of an O2-derived Fe-O species occurs with water before autohydroxylation (Fig. 3B). Support for this proposal comes from the observation that, for some 2OG oxygenases, labelling studies indicate substrate hydroxylation occurring with incomplete incorporation of oxygen from O2 . However, alternative mechanisms are possible. It is of interest to determine the origin of the alcohol product of autohydroxylation; in the case of ABH3 because hydroxylation occurs at an unactivated position. In the case of JMJD6, it is most probable that autohydroxylation of lysyl residues occurs via a mechanism closely related to that occurring for lysyl hydroxylation of substrates (i.e. via Fe=O mediated hydroxylation) .
In the case of fragmentation, it appears possible that an α-amidation pathway type mechanism occurs (Fig. 6). Pioneering studies on the mechanism of oxygen-dependent radiolysis led to the proposal of the α-amidation pathway  and an analogous mechanism may occur for oxygenase/oxidase fragmentation in the case of ACCO . The α-amidation pathway involves initial abstraction of the α-hydrogen followed by reaction with oxygen to give a peroxide radical, which then abstracts a hydrogen (Fig. 6A). Hydrogen peroxide can then be eliminated to give an imine, which undergoes hydrolysis to give an N-terminal fragment with a C-terminal amide modified by the addition of an α-keto amide. Variations on this classic α-amidation pathway are possible (Fig. 6B,C). In the case of the 2OG oxygenases, it has been argued that the hydrophobic nature of the active site coupled to conformational constraints helps to prevent oxidative damage [13,71]. However, this situation may be compromised in the case of partially misfolded enzymes or in the presence of active site binding redox agents (e.g. ascorbate). Thus, C-α hydrogen abstraction by a hydroxide radical, an iron bound peroxide/peroxide radical or ferryl-oxo species may be responsible for C-α hydroxylation.
In conclusion, there is accumulating evidence that 2OG oxygenases and related enzymes undergo autocatalysed reactions that can lead to oxidative modification. At present, it is unclear whether these observations have any functional biological relevance. In some cases, there is evidence that modification can occur in cells (ACCO and JMJD6), although its physiological role, if any, is unclear. The observation that JMJD6 undergoes self-hydroxylation  in a manner similar to substrate hydroxylation is biologically intriguing and it will be of interest to determine whether such self-hydroxylation occurs in the case of other 2OG-dependent hydroxylases. Regarding whether 2OG oxygenases are more prone to oxidative damage than their P450 counterparts , there is insufficient evidence to give a reasoned view. One line of evidence that may be interesting to pursue comprises proteomic studies, perhaps employing top-down technology , which, over the longer term, may yield quantitative information on oxidative modifications in vivo.