Limonene dehydrogenase hydroxylates the allylic methyl group of cyclic monoterpenes in the anaerobic terpene degradation by Castellaniella defragrans

The enzymatic functionalization of hydrocarbons is a central step in the global carbon cycle initiating the mineralization of methane, isoprenes, and monoterpenes, the most abundant biologically produced hydrocarbons. Also, terpene-modifying enzymes have found many applications in the energy-economic biotechnological production of fine chemicals. Here, we describe a limonene dehydrogenase that was purified from the facultatively anaerobic betaproteobacterium Castellaniella defragrans 65Phen grown on monoterpenes under denitrifying conditions in the absence of molecular oxygen. The purified limonene:ferrocenium oxidoreductase activity hydroxylated the methyl group of limonene (1-methyl-4-(1-methylethenyl)-cyclohex-1-ene) yielding perillyl alcohol ([4-(prop-1-en-2-yl)cyclohex-1-en-1-yl]methanol). The enzyme had a DTT:perillyl alcohol oxidoreductase activity yielding limonene. Mass spectrometry and molecular size determinations revealed a heterodimeric enzyme comprising CtmA and CtmB. Recently, the two proteins had been identified by transposon mutagenesis and proteomics as part of the cyclic terpene metabolism (ctm) in C. defragrans and are annotated as FAD-dependent oxidoreductases of the protein domain family phytoene dehydrogenases and related proteins (COG1233). CtmAB is the first heterodimeric enzyme in this protein superfamily. Flavins in the purified CtmAB are oxidized by ferrocenium and are reduced by limonene. Heterologous expression of CtmA, CtmB, and CtmAB in Escherichia coli demonstrated that limonene dehydrogenase activity required both subunits, each carrying a flavin cofactor. Native CtmAB oxidized a wide range of monocyclic monoterpenes containing the allylic methyl group motif (1-methyl-cyclohex-1-ene). In conclusion, we have identified CtmAB as a hydroxylating limonene dehydrogenase and the first heteromer in a family of FAD-dependent dehydrogenases acting on allylic methylene or methyl CH-bonds. We suggest placing in Enzyme Nomenclature as new entry EC 1.17.99.8.

Monoterpenes constitute a large and diverse group of hydrocarbons ubiquitous in nature. Over the years, around 1000 individual monoterpene structures have been identified (1)(2)(3). These 10 carbon atom compounds are mainly produced by plants as major components of essential oils (4). Minor amounts are also synthesized by insects and fungi (5,6). Monoterpenes are secondary metabolites that act principally as allelochemicals. Limonene (4-isopropenyl-1-methyl-cyclohex-1-ene, see Fig. 1) is by far the most readily available monoterpene in nature. It is found as the main component of the essential oils of citrus plants. Between 30,000 and 50,000 tons of limonene are extracted from natural sources per year, basically as a by-product of citrus juice processing (3,7). Because of its olfactory and well-known antimicrobial properties, it is often added to food, cosmetics, and household products (8,9).
In the global carbon cycle, the large annual production of monoterpenes by plants is balanced by photooxidation and microbial mineralization. Aerobic bacteria use oxygenases to introduce a hydroxyl or epoxide group at different positions of limonene (10). In Mycobacterium sp. HXN-1500, Pseudomonas putida KT2440 and Geobacillus stearothermophilus (formerly Bacillus stearothermophilus), a cytochrome P450 monooxygenase, hydroxylates the methyl group of limonene yielding perillyl alcohol (11)(12)(13).
Perillyl alcohol was also formed in an anaerobic bacterium as product of a limonene biotransformation. Castellaniella defragrans 65Phen, a facultatively denitrifying betaproteobacterium, uses a wide range of monoterpenes as sole substrate (14). Proteomic data and transposon mutants suggested a degradation pathway from limonene via perillyl alcohol and aldehyde to perillic acid. A deletion mutant in the putative perillyl aldehyde dehydrogenase revealed the co-metabolic formation of perillyl alcohol from limonene during growth on acetate. C. defragrans' genome contains a gene cluster named cyclic terpene metabolism (ctm) within a genetic island coding for the majority of the monoterpene metabolism genes (14). Mutants with a transposon insertion in the genes ctmA, ctmB, or ctmE failed to grow on limonene, yet they grew on perillyl alcohol as efficiently as the WT. The gene cluster ctmABCDEFG codes for two presumably flavin-containing oxidoreductases, CtmA (CDM25290) and CtmB (CDM25289) and an electron transfer system consisting of a 2Fe-2S ferredoxin (CtmE) and a NADH:ferredoxin oxidoreductase (CtmF). These proteins, but not one of the other putative proteins of unknown function (CtmC, -D, or -G), This work was supported in part by the Max Planck Society. The authors declare that they have no conflicts of interest with the contents of this article. This article contains Figs. S1-S6. 1 Supported by a grant from the DAAD in Germany and from the Corporación para la Investigación de la Corrosión and COLCIENCIAS in Colombia. 2  cro ARTICLE were expressed in larger quantities in ␣-phellandrenegrown cells compared with acetate-grown cells (14). CtmA and CtmB affiliate with COG1233 (phytoene dehydrogenase and related proteins), a group of flavoenzymes involved mainly in carotenoid biosynthesis. Members of this group act with electron acceptors with a positive reduction potential on the dehydrogenation of methylene groups in a diallylic motif, a hexa-1,5-diene moiety yields as an oxidation product a hexa-1,3,5-triene structure, or the oxidation of an allylic methylene group (-CHϭCH-CH 2 -) yielding an alk-2,3-en-1-one motif. Structural information is available for phytoene desaturase and ␤-carotene ketolase (15,16). In this study, we characterized the limonene dehydrogenase enzyme activity present in C. defragrans 65Phen.

Enzyme activities in soluble extracts
The oxidation of limonene was tested with protein extracts from limonene-grown cells of C. defragrans 65Phen prepared in a molecular oxygen-free environment without the addition of reducing agents. The dialyzed soluble fraction catalyzed in vitro the formation of perillyl alcohol from limonene when the ferric iron-containing ferrocenium hexafluorophosphate (FHP) 3 (EЈ 0 ϭ ϩ0.38 V) was present as electron acceptor. The specific activity was 108 pkatal (mg protein) Ϫ1 . Chiral GC of the product revealed the formation of (S)-(Ϫ)-perillyl alcohol from (S)-(Ϫ)-limonene and of (R)-(ϩ)-perillyl alcohol from (R)-(ϩ)limonene with similar rates ( Fig. 1 and Fig. S1). The reverse reaction, the formation of limonene from perillyl alcohol, was detected in the presence of the reducing agent dithiothreitol (DTT) (EЈ 0 ϭ Ϫ0.33 V) with a lower specific activity of 0.63 pkatal (mg protein) Ϫ1 . A nonenzymatic reduction of perillyl alcohol with DTT was not observed.

Enzyme purification
The limonene dehydrogenase activity was purified as limonene-dependent ferrocenium reductase activity from soluble extracts of C. defragrans by protein chromatography in an oxygen-free chamber. During anion-exchange chromatography in phosphate buffer a total loss of activity was observed. FAD at concentrations of 20 M restored the enzyme activity com-pletely during a pre-assay incubation for 4 h at 4°C. No reactivation was observed upon incubation with FMN. Addition of FAD in the separation buffers failed to prevent enzyme inactivation during anion-exchange chromatography. This finding and the small increase in purity during the anion-exchange chromatography ( Fig. 2A, quantitative data not shown) suggested as the best purification method a combination of hydrophobic-interaction and size-exclusion chromatographies on phenyl-Sepharose and Superdex 200 columns, respectively. The purification yielded a nearly homogeneous protein (Fig.  2B) and a 7-9-fold increase in specific activities of the forward and reverse reaction (Tables 1 and 2). A separation of CtmA from CtmB did not occur during the purification.
The active fraction had an apparent molecular mass of 152 kDa in the size-exclusion chromatography. A similar result was observed via dynamic light scattering (166.2 kDa). Denaturing polyacrylamide gels revealed two dominant bands, one at 59 and another at 57 kDa (Fig. 2). For identification of the proteins, the Coomassie-stained protein bands were excised from acrylamide gels and digested with the protease trypsin, and the oligopeptide mixture was analyzed by MALDI-TOF mass analysis. The larger protein band of the SDS-PAGE contained protein CtmB (CDM25289, gene-based predicted mass 60.7 kDa), whereas the smaller protein band was identified as CtmA (CDM25290, 61.7 kDa). The N-terminal peptides of CtmA were not detected in the MALDI-TOF mass spectra. Hence, we characterized the N-terminal amino acids of CtmA by Edman degradation and both proteins by LC/MS-MS of tryptic peptides to obtain evidence for a post-translational modification by endoproteases. The sequence obtained by Edman degradation confirmed the N-terminal protein sequence predicted in the ORF of CtmA. LC/MS-MS analyses detected peptides close to the N and C termini of both proteins (Fig. S2). This strongly excludes a post-translational cleavage as the cause for CtmA's apparently smaller size in denaturing acrylamide gels.

Catalytic properties of the limonene dehydrogenase
The purified native CtmAB had an optimum temperature of 40°C for the photometric limonene dehydrogenase activity (Fig. S3), and at this temperature its optimal pH ranged between 7.5 and 8.0 in potassium phosphate buffer (Fig. S4). The kinetic parameters (K M and V max ) were determined at the temperatures for optimal bacterial growth (28°C) and for optimal enzymatic activity (40°C) ( Table 3). The limonene hydroxylation was faster than the perillyl alcohol reduction. The substrate affinity toward limonene was larger than with perillyl alcohol, which is physiologically meaningful for limonene utilization.
CtmAB did not use molecular oxygen as an electron acceptor. All experiments were performed in an anoxic chamber or in a closed glass container containing nitrogen as gas phase. The presence of molecular oxygen (21% volume in the gas phase) in the reaction cuvette caused a 40% reduction in specific activity (from 1099 Ϯ 22 to 589 Ϯ 50 pkatal (mg protein) Ϫ1 ). The reestablishment of the anoxic conditions restored the specific activity to 993 Ϯ 9 pkatal (mg protein) Ϫ1 .

Characterization of a limonene dehydrogenase Substrate spectrum
Several monoterpenes and analogous compounds were tested as substrates in the photometric limonene dehydrogenase assay (Fig. 3). Monocyclic monoterpenes were oxidized with highly-specific activities irrespective of the presence and position of a second double bond in the cyclohex-1-ene ring or in the isopropyl substituent. The loss of the allylic character of the methyl group correlated with inactivity of the enzyme as demonstrated with isolimonene, which does not promote growth of C. defragrans (17). The p-isopropyl group as a structural element likely increases the substrate binding as indicated by the low activity on 1-methyl-cyclohex-1-ene compared with that on monoterpenes. The substrate-binding site does not seem to be suitable for polar compounds (␣-terpineol and terpinen-4-ol) and nonplanar, spacious structures close to the allylic group (pinene and carene). The aromatic compound cymene was not oxidized.

Heterologous expression
To obtain individual information on CtmA and CtmB, Escherichia coli BL21 Star (DE3) was used to express CtmA, CtmB, or CtmA and CtmB together from the plasmids pET42a(ϩ)  Table 1 Purification of the limonene dehydrogenase activity (detected as ferrocenium reductase activity) from soluble protein extracts of C. defragrans 65Phen Activity was tested at 28°C, and purification was started from 1.2 g of wet biomass of C. defragrans 65Phen.   Table 3 Catalytic features of the limonene dehydrogenase purified from C. defragrans 65Phen at 28 and 40°C

Purification step
All reactions were carried out in phosphate buffer (10 mM), pH 8.0, and contained 100 g ml Ϫ1 of purified WT protein. Limonene dehydrogenase activity was measured using 10 -1000 M limonene and ferrocenium hexafluorophosphate (200 M) as electron acceptor. Perillyl alcohol reduction was observed using dithiothreitol (2 mM) as electron donor and 0.5-10 mM perillyl alcohol as electron acceptor. The reactions were started by the addition of the monoterpenes pre-dissolved in Tween 20-containing buffer.

Characterization of a limonene dehydrogenase
ctmA, pET42a(ϩ) ctmB, or pET42a(ϩ) ctmA ctmB, respectively. Protein expression upon IPTG addition was observed in all three genetic constructs, and CtmA, CtmB, and CtmAB were visualized as ϳ60-kDa protein bands in denaturing gels (Fig. S5A). The identity was verified by MALDI-TOF MS. The overexpressed protein(s) formed inclusion bodies even by induction at low temperatures, low cellular densities, and low IPTG concentration (Fig. S5B). Several detergents were tested to reactivate fractions with inclusion bodies, but these experiments failed to recover detectable catalytic activity. Small enzyme activities were recovered in the soluble fraction after cell lysis. The forward limonene dehydrogenase activity was exclusively detected in soluble extracts containing coexpressed CtmA and CtmB (31 pkatal (mg protein) Ϫ1 ). The reverse reaction, the perillyl alcohol reduction, was catalyzed by all genetic constructs at similar reaction rates (coexpressed CtmA and CtmB: 330 Ϯ 110 fkatal (mg protein) Ϫ1 ; CtmA, 303 Ϯ 42 fkatal (mg protein) Ϫ1 and CtmB 280 Ϯ 20 fkatal (mg protein) Ϫ1 ). A mixture of expressed and purified CtmA with expressed and purified CtmB did not recover the limonene dehydrogenase activity, only the perillyl alcohol reductase activity was observed.

Spectroscopic properties and FAD content
Purified native limonene dehydrogenase as purified under anoxic conditions had no absorption bands except the ones of aromatic amino acids (Fig. 4, spectrum 6). Ferrocenium has an absorption band at 620 nm (18). Upon addition of ferrocenium ions, the typical absorption spectrum of an oxidized flavin appeared with characteristic maxima at 365 and 465 nm. This suggested that the purified enzyme contained reduced flavins. A weak band appeared at 560 nm coinciding with the absorption bands of flavin semiquinone radicals in this spectral region (19,20). A nonenzymatic reduction of ferrocenium ions (100 M) with DTT or dithionite did not form a band at 560 nm. Limonene addition to the oxidized protein initiated a gradual decrease of the flavin absorbance at 365 and 465 nm as well as the bands at 560 and 620 nm. In this experiment, the amount of FAD formed accounted for on average 80% of one of the presumably two flavins in a CtmAB heterodimer using the extinction coefficient of free FAD (⑀ 450 , 11,300 M Ϫ1 cm Ϫ1 (21)).
A spectrophotometric analysis of flavin released from heattreated protein was used for quantifying the flavin content in CtmAB as described by Aliverti et al.

Characterization of a limonene dehydrogenase
supernatant. This finding is well known for a covalent bond between the flavin cofactor and the protein. The detection of flavin autofluorescence in denaturing polyacrylamide gels is widely accepted as evidence for a covalently bound flavin. Our enzyme preparation had, according to the autofluorescence in SDS-denatured acrylamide gels, covalently bound flavin in both CtmA and CtmB purified from C. defragrans (Fig. 5 and Fig.  S6). In heterologously expressed proteins, only CtmB showed flavin autofluorescence. This indicated that E. coli was unable to efficiently incorporate a flavin into CtmA. Typically, proteins with covalent bonds to flavins do not release flavins in heat treatments. To resolve this contradiction, we attempted to identify flavinylated oligopeptides in mass spectra of CtmA and CtmB. Several enzyme preparations of WT CtmAB did not contain flavinylated oligopeptides according to MALDI-TOF or LC-ESI/MS-MS analyses: MS-identified peptides covered 62 and 67% of the amino acid sequence of CtmA and of CtmB, respectively (Fig. S2). So far, these observations support a very tight flavin binding to the proteins CtmA and CtmB showing the characteristics of a covalently-bound flavin in acid-treated protein solutions and denaturing acrylamide gels.

Bioinformatics analysis
Genes ctmA and ctmB encode for proteins related to the family of the phytoene dehydrogenases (COG1233) (23). CtmA and CtmB share a 27% amino acid identity and have predicted molecular masses of 61.7 and 60.7 kDa, respectively. Neither membrane-spanning regions nor signal peptides for transport beyond the cytoplasmic membrane were predicted (TMHMM (24) and PSORTb (25)). The enzyme activity was completely in the soluble fraction (data not shown). Therefore, a cytoplasmic localization is expected for both proteins.
Sequence identities between the query proteins (CtmA and CtmB) and the aligned phytoene dehydrogenases ranged between 20 and 28%. Identical amino acids present in the six aligned sequences accounted for 3% of all amino acids (Fig. 6). Residues with similar properties accounted for another 11%. Despite the low overall sequence similarity, all six proteins showed a highly conserved N terminus. This region contains the GXGXXG motif that forms hydrogen bonds with the phosphate groups of FAD in COG1233 proteins (27,31). Three additional putative FAD-interacting regions were predicted in CtmA and CtmB. Altogether, these protein sections configure a hypothetical FAD-binding domain resembling the ones present in NdCrtD and CrtI (15,27). A pair of cysteines localized within the FAD-binding region near the proteins' C terminus are not conserved in the other enzymes of COG1233 suggesting a specific role of the amino acids in the oxidation of limonene. Apart from FAD-binding domains, two additional domains were predicted: a substrate-binding domain and a nonconserved helical domain.

Discussion
The purification of the limonene:ferrocenium oxidoreductase activity resulted in the isolation of CtmA and CtmB. This concurs with conclusions of previous proteomic analyses of C. defragrans 65Phen and physiological studies with different transposon and deletion mutants (14). CtmA and CtmB belong to the FAD-dependent oxidoreductases related to the phytoene dehydrogenases (COG1233) (23). The latter enzymes attack allylic methyl and methylene groups, CH 3 and CH 2 groups, adjacent to a carbon-carbon double bond. Among the products of the phytoene dehydrogenase superfamily are carboncarbon double bonds, aldehydes, and ketones (15,16,29). Similarly, the purified CtmAB showed dehydrogenase activity exclusively on monoterpenes carrying an allylic methyl group. In the case of (R)-(ϩ)-limonene and (S)-(Ϫ)-limonene, such dehydrogenation resulted in the formation of the respective enantiomeric form of perillyl alcohol.

Characterization of a limonene dehydrogenase
Three-dimensional modeling and a multisequence alignment indicated that CtmA and CtmB have domain architectures typical of COG1233 proteins. Like other members of the cluster, CtmA and CtmB each bear an FAD-binding domain, a nonconserved helical domain, and a putative substrate-binding domain. An atypical finding in the limonene dehydrogenase was the detection of tightly bound flavin in CtmA and CtmB from C. defragrans. So far, we have no molecular evidence for a covalent binding of FAD to the protein. Until now COG1233 proteins were reported to carry an FAD that dissociated either during purification, heat treatments, or acidic protein precipitation (15,23,32,33). In CtmAB, only heat treatment partially releases the flavins but not acid treatment or protein denaturation by SDS as shown by the detection of autofluorescence in denaturing protein gels. A lack of flavin incorporation during heterologous expression has also been observed for the phytoene desaturase of Myxococcus xanthus, where 0.5 mol of FAD was detected per mol of protein (23). Hence, the heterologous expression of some COG1233 proteins in E. coli seems to be accompanied by an incomplete incorporation of FAD molecules.
The catalytic efficiency (k cat /K M ) of the WT enzyme for limonene oxidation strongly suggests that the limonene dehydrogenase activity is the physiologically relevant reaction. The reverse reaction, the perillyl alcohol reductase activity, was observed with heterologously expressed CtmA, CtmB, and CtmAB. According to the flavin quantification and the lack of fluorescence in protein gels, the flavin content in heterologously expressed CtmA is very low. This observation can be seen as an indication that perillyl alcohol reduction is a FADindependent reaction. With DTT as electron donor, the cysteine residues in the C-terminal FAD-binding domain may  (27)), ␤-carotene monoketolase (CrtO) from Synechocystis sp. PCC 6803 (NCBI: YP_005652552 (16)), and 4,4Ј-diapolycopene oxidase (CrtNb) from Methylomonas sp. strain 16a (NCBI: AAX46185 (29)). The secondary structure elements predicted are indicated above the alignment. The colored bars below indicate the domain organization as follows: the FAD-binding domain (red), the substrate-binding domain (blue), and the nonconserved helical domain (orange). Identical and highly conserved residues are highlighted in blue and gray, respectively. Two pairs of conserved cysteines in CtmA (Cys-495 and Cys-508) and CtmB (Cys-488 and Cys-501) are indicated by an asterisk. The alignment was prepared with Clustal O (28). The structural features were predicted using Phyre2 and 3DLigandSite (26,30).

Characterization of a limonene dehydrogenase
transfer the reducing equivalents. Besides the flavin binding, the heteromeric character of the limonene dehydrogenase is unusual. It is the first reported heteromer within the COG1233 enzymes, likely a heterodimer. So far, only monomers or homodimers have been identified. In general, phytoene desaturases share a central nonconserved helical domain. In most cases, this region is appointed as responsible for a monotopic membrane association (15,27,34); however, it may also serve in CtmA and CtmB as a surface for dimer interaction.
Ferrocenium ions substitute in enzyme assays for the natural electron acceptor in the re-oxidation of the tightly bound FAD (Fig. 7). We suggest that the completely oxidized FAD accepts a hydride from limonene. This initial hydride transfer is also postulated for other COG1233 enzymes (16,27,35,36). The reaction results in the formation of an allyl cation in which the positive charge is stabilized by delocalization over the allylic moiety. The lack of activity on isolimonene demonstrates the necessity for an allylic alkene bond for this process. Then, the stabilized carbocation is expected to react with water accompanied by a proton transfer yielding the corresponding alcohol. Structurally related phytoene desaturases utilize in vitro electron acceptors that, like ferrocenium, have positive reduction potentials, i.e. molecular oxygen or benzoquinone (15,27). In C. defragrans, the ctm cluster within the genetic island for monoterpene utilization contains genes for CtmAB, an NADH:ferredoxin oxidoreductase (CtmF), and a ferredoxin (CtmE) as well as a heterodimeric electron transfer flavoprotein (ETF) (CDM25301 and CDM25302 (14). ETFs function as an electron shuttle between flavoprotein dehydrogenases involved in fatty acid and amino acid degradation and membrane-bound ETFquinone oxidoreductases (ETF-QO), but also in electron bifurcation complexes (37,38). Transposon mutagenesis identified an ETF-QO (CDM23589) as essential for the cyclic monoterpene degradation, likely as entrance point to the respiratory chain ending in denitrification (14). The ETF ␤-subunit CDM25301 was found to be highly expressed during growth on monoterpenes (14). This ETF seems to be the natural electron acceptor of CtmAB, because ferrocenium is used in vitro as an ETF substitute for flavin dehydrogenases (38,39). An electron transfer from NADH via ferredoxin and ETF to ETF-QO may provide energy for the limonene degradation via a conproportional bifurcation of electrons.
The hydroxylating limonene dehydrogenase affiliates with the enzyme class 1.17.99, collecting all oxidoreductases acting on CH or CH 2 groups with unknown physiological electron acceptors. This class includes other key enzymes of the anaerobic hydrocarbon metabolism that use water rather than molecular oxygen as cosubstrate for the CH activation. Examples include the ethylbenzene and cholesterol hydrolases, two molybdenum-dependent enzymes (40 -42). Another well characterized member of the group is the p-cresol methylhydroxylase, a flavocytochrome c enzyme containing one FAD and two hemes. It hydroxylates p-cresol to 4-hydroxybenzyl alcohol, which is then further oxidized to 4-hydroxybenzaldehyde with phenanzine as in vitro electron acceptor (43,44). We propose for the limonene dehydrogenase CtmAB a placement as EC 1.17.99.8.

Experimental procedures
Cultivation C. defragrans 65Phen was cultivated in a 10-liter fermenter with anaerobic artificial freshwater medium as described previously (45). Carbonate buffer was replaced by 10 mM K 2 HPO 4 , pH 7.2. Vitamin addition was omitted, and nitrogen was used as headspace gas. (R)-(ϩ)-Limonene (Ͼ97% purity, Sigma, Germany) was directly added to the medium to a final concentration of 30 mM in the aqueous phase, a virtual concentration describing the two-phase system. The fermenter was inoculated with 1 liter of a freshly grown culture, incubated at 28°C, and stirred at 150 rpm. After 6 -7 days (OD 600 nm Ϸ1.9), the cells were harvested by centrifugation at 16,000 ϫ g and 4°C for 40 min. The biomass was resuspended in a 3-fold volume of 25 mM potassium phosphate buffer, pH 8.0, prior to disruption by three passages through a French pressure cell press (SLM Aminco, Rochester, NY) at 8.6 megapascals. Ultracentrifugation at 230,000 ϫ g for 30 min at 4°C yielded a soluble protein fraction.

Heterologous expression
The putative limonene dehydrogenase genes ctmA and ctmB were PCR-amplified from genomic DNA of C. defragrans 65Phen using the primer pairs ctmA_NdeI_F (TATCATATG-GCAAATCCGAAAAGCGA) and ctmA_SalI_R (AAAGT-CGACTCAATCGACACTGGTCCGTCGT); ctmB_NdeI_F (TAACATATGTCTGAAGTCAAACAATG) and ctmB_SalI_ R (TAAGTCGACTCATAGGAGGAGCCCCTTTT). The amplicon ctmAB for the coexpression of both genes was obtained using the primer pair ctmA_NdeI_F and ctmB_SalI_R. All three amplicons were ligated into the vector pET-42a(ϩ) (Novagen, Merck KGaA, Darmstadt, Germany) and subsequently transformed into E. coli BL21 Star (DE3) (Invitrogen, Karlsruhe, Germany). The correctness of the genetic constructs was verified by sequencing using the BigDye Terminator ver-

Enzyme purification
Protein purification was performed in an ÄKTA LC system (GE Healthcare, Freiburg, Germany) installed in an anoxic chamber at 4°C. The first purification step was hydrophobicinteraction chromatography on a phenyl-Sepharose column (20-ml volume and 2.6-cm diameter). The soluble extract received concentrated salt solutions to final concentrations of 75 mM potassium phosphate, 60 mM ammonium sulfate, and 20 mM potassium chloride, pH 8.0. This was also the starting buffer. This optimized buffer composition guaranteed protein stability and column binding. The protein eluted using 10 mM potassium phosphate, pH 8.0. The active fractions were concentrated with a centrifugation filter (Amicon Ultra 15 ml, 10 kDa, Merck Millipore, Darmstadt, Germany) and further purified on a Superdex 200 column (120-ml volume and 1.6-cm diameter) equilibrated with 10 mM potassium phosphate, pH 8.0. The fractions with limonene dehydrogenase activity were pooled and used for enzymatic activity measurements.

Enzyme activity
Continuous and end-point analyses were used for measuring the reversible oxidation of limonene in protein extracts. All assays were prepared inside an anaerobic chamber at 4°C. To facilitate monoterpene availability in the aqueous phase, these were dissolved in 10 mM K 2 HPO 4 , pH 8.0, containing 2.5% v/v Tween 20. The final content of Tween 20 in the assays was 0.5% v/v.
Limonene oxidation (forward reaction) was assayed spectrophotometrically in 1-ml quartz cuvettes. A typical reaction contained in a 1-ml volume of 100 -200 g of protein, 200 M ferrocenium hexafluorophosphate and limonene (10 -1000 M) in 10 mM phosphate buffer, pH 8.0. The reaction was started with the addition of the monoterpene-containing buffer and proceeded at either 28 or 40°C. The absorption decrease of the ferrocenium ion was followed at 290 nm (⌬⑀ ϭ 7100 M Ϫ1 cm Ϫ1 ). The molar extinction coefficient of ferrocenium was calculated from a calibration curve prepared in phosphate buffer 10 mM, pH 8.0 (y ϭ 0.0071x; R 2 ϭ 0.9916). Limonene consumption and perillyl alcohol formation were monitored by GC. For this, 200 l of n-hexane were added to each sample and vortexed for 30 s. After 10 min of shaking at 60 rpm, the samples were centrifuged at 13,000 ϫ g for 10 min for complete phase separation. One l of the organic phase was analyzed in a GC coupled to a flame ionization detector (Auto System XL, PerkinElmer Life Sciences, Überlingen, Germany). Analyte separation was performed on an Optima-5 column (50 m ϫ 0.32 mm ϫ 0.25 m; Macherey-Nagel, Düren, Germany) with an injection port temperature of 250°C, a detection temperature of 350°C, and the following column program: an initial column temperature of 40°C for 2 min, increasing to 100°C with a rate at 4°C min Ϫ1 , staying constant for 0.1 min, further increasing to 320°C at 45°C min Ϫ1 , and finally a constant temperature for 3 min. The split ratio was 1:9. All concentrations refer to the aqueous phase. For analyte quantification, calibration curves with authentic standards were prepared. For stereospecific GC analysis, 1 l of sample was separated on a Hydrodex-␤-6TBDM column (25 m ϫ 0.25 mm; Macherey-Nagel, Düren, Germany) by the following program: injection temperature of 200°C and flame ionization detection temperature of 230°C; the column temperature was 80°C for 1 min, increasing to 130°C at a rate of 5°C min Ϫ1 , after 0.5 min further increasing to 230°C at 20°C min Ϫ1 , and stationary for 2 min.
Perillyl alcohol reductase activity, the physiologically reverse reaction, was tested in 1-ml reactions containing 100 -200 g of protein, 2 mM DTT, and perillyl alcohol (0.5-10 mM). Reactions were performed in 3-ml glass vials closed with a butyl rubber septum. Incubation took place at 28 and 40°C for 4 h. Limonene was extracted with hexane and quantified by GC as described above.
Optimal conditions such as protein concentration, temperature, incubation time, and pH were determined for forward and reverse reactions. The pH optimum was tested using several buffers with pH values adjusted near their specific pK a at 37°C in a range of 4 -10 (50 mM sodium citrate, pH 4 -6; 25 mM potassium phosphate, pH 6 -8; 25 mM Tris, pH 7.5-8.5; 50 mM glycine/HCl, pH 9 -10).

Analytical methods
Proteins were quantified according to Bradford (47) with BSA as standard and visualized in 10 -12% v/v acrylamide SDS-PAGE stained with Coomassie Blue (48). For protein analysis by MS, bands from SDS-PAGE were excised, in-gel-digested with trypsin, and analyzed on a 4800 MALDI-TOF/TOF Analyzer (Applied Biosystems, Darmstadt, Germany). Additionally, CtmAB purified from C. defragrans was subjected to in-solution tryptic digestion, separated in a reversed phase C18 column on a nano-ACQUITY-UPLC (Waters Corp., Milford, MA), and analyzed on an LTQ-Orbitrap Classic mass spectrometer equipped with an ESI ion source (Thermo Fisher Scientific). For protein identification, the mass spectra were analyzed against a protein database from C. defragrans using the Mascot search engine version 2.4.0 (Matrix Science Ltd., London, UK). For post-translational modifications, the spectra were searched against the parent amino acid sequences of proteins CtmA and CtmB allowing FMN-and FAD-flavinylation in all amino acid residues using Sequest version 27, revision 11 (Thermo Fisher Scientific). The N terminus of native CtmA was sequenced by Edman degradation (Proteome Factory AG, Berlin, Germany) after one-dimensional separation by SDS-PAGE and a semi-dry blotting onto a polyvinylidene difluoride membrane. In-gel fluorescence was recorded prior to Coomassie Blue staining. The denaturing polyacrylamide gels were immersed in 10% acetic acid for 10 min. Fluorescence was recorded on a Typhoon 9400 (GE Healthcare, Freiburg, Germany) with an excitation at 457 nm and emission at 526 nm. Protein absorption spectra were recorded using a DU 600 UVvisible spectrophotometer (Beckman Coulter, Krefeld, Germany). FAD was extracted and quantified as described by Aliverti et al. (22). The molecular mass of the enzyme was estimated on a Superdex 200 column (120 ml, 1.6-cm diameter) using thyroglobulin (669 kDa), ferritin (440 kDa), catalase (232 Characterization of a limonene dehydrogenase kDa), aldolase (158 kDa), albumin (66.3 kDa), ovalbumin (46 kDa), chymotrypsinogen A (25 kDa), and RNase A (13,7 kDa) as standards. An additional molecular mass determination was conducted by dynamic light scattering on a DynaPro Plate Reader II (Wyatt Technology Corp., Santa Barbara, CA) at 21°C as indicated by the manufacturer. All results in this contribution are reported as the average of triplicate measurements Ϯ S.D.

Bioinformatics analyses
Nucleotide and amino acid database searches were carried out with NCBI BLAST (49). Three-dimensional protein modeling was conducted with Phyre2 (26) and 3DLigandSite (30). The amino acid sequence alignment was prepared with Clustal Omega (28). Subcellular localization of the proteins was predicted using TMHMM Server version 2.0 and PSORTb version 3.0.2 (24,25).