Isolation and Functional Characterization of CYP71AJ4 Encoding for the First P450 Monooxygenase of Angular Furanocoumarin Biosynthesis*

The biosynthesis of linear and angular furanocoumarins is still poorly understood at the molecular level, with only psoralen synthase (CYP71AJ1) identified from Ammi majus. Using cDNA probes inferred from CYP71AJ1, three orthologs were isolated from Apium graveolens (CYP71AJ2) and Pastinaca sativa (CYP71AJ3 and -4) and functionally expressed in yeast cells. CYP71AJ2 and CYP71AJ3 displayed psoralen synthase activity, whereas CYP71AJ4 only catalyzed the conversion of (+)-columbianetin to angelicin and negligible amounts of a hydroxylated columbianetin by-product. CYP71AJ4 thus constitutes the first fully characterized P450 monooxygenase specific for the angular furanocoumarin pathway. The angelicin synthase exhibited an apparent Km of 2.1 ± 0.4 μm for (+)-columbianetin and a kcat of 112 ± 14 min–1. Moreover, the use of 3′-deuterated (+)-columbianetin as substrate led to an almost complete “metabolic switch,” resulting in the synthesis of anti-3′-hydroxy-3′-deuterated(+)-columbianetin. This confirms that angelicin synthase attacks columbianetin by syn-elimination of hydrogen from C-3′. Sequence comparison between psoralen synthase (CYP71AJ3) and angelicin synthase (CYP71AJ4) showed 70% identity, whereas the identity dropped to 40% in those regions thought to provide the substrate recognition sites. Accordingly, CYP71AJ3 and CYP71AJ4 might be derived from a common ancestor of unknown functionality by gene duplication and subsequent molecular evolution.

Furanocoumarins are natural plant metabolites characterized by a furane moiety fused to benzopyran-2-one. The position of the furane substitution distinguishes two large groups of compounds, the linear (psoralens) and the angular furanocoumarins (angelicin and derivatives) ( Fig. 1) (1, 2). The psoralens in particular are known for their photosensitizing and phototoxic effects and have been used in photochemotherapy of skin disorders (psoriasis, vitiligo, and mycosis). In addition, numer-ous other bioactivities have been attributed to furanocoumarins, which are the focus of many pharmacological screenings, although their mutagenic potential was recognized many years ago (3,4). Furanocoumarins are formed by only four plant families (i.e. Apiaceae, Rutaceae, Fabaceae, and Moraceae) (5,6). They are allelochemicals that have been assigned ecological functions, such as the defense against phytopathogens, germination inhibition of competitor plants, or herbivore deterrence (7)(8)(9)(10). Their toxicity is based on their ability to create covalent adducts of double-stranded DNA, impeding DNA replication (9,11), as well as to inhibit cytochrome P450 enzymes (12)(13)(14). It is noteworthy that plants synthesize either linear furanocoumarins (psoralens) only or both linear and angular furanocoumarins, which are less widely distributed and mainly confined to some Fabaceae and Apiaceae. The association of both forms was demonstrated to be more toxic for insect larvae than the presence of linear furanocoumarins alone, which could reflect the evolutionary adaptation of plants to new herbivores that have acquired the capacity to detoxify psoralens (15)(16)(17).
Due to the complex bioactivity of furanocoumarins, its biosynthesis has received continuous attention. The basis of linear furanocoumarin formation was mostly established by the end of the 1980s by a combination of precursor feeding experiments and the biochemical characterization of major enzymes of the pathway. Nevertheless, until recently, the pathway was thought to depend on the ortho-hydroxylation of trans-or cis-4-coumaric acid and cyclization of the product to umbelliferone (see Ref. 6 for a review). However, a dioxygenase has now been cloned from Arabidopsis and expressed; this enzyme converts feruloyl-CoA to 2,4-dihydroxy-5-methoxycinnamoyl-CoA and scopoletin (18). To a lesser extent, it also accepts 4-coumaroyl-CoA, which yields umbelliferone (Fig. 1). The data prove for the first time the essential role of CoA-esters. These serve as an active substrate in coumarin biosynthesis, whereas unconjugated cinnamic acids are inactive. This pivotal role of CoAesters in coumarin synthesis is consistent with the report of a unique 4-coumaroyl-CoA ligase that was specifically assigned to coumarin synthesis in Ruta graveolens (19). Umbelliferone is subsequently prenylated to demethylsuberosin by a 6-prenyltransferase also assigned to plastids in R. graveolens (20). The conversion of demethylsuberosin to marmesin and further to psoralen is then catalyzed by two distinct P450 enzymes (i.e. marmesin synthase and psoralen synthase) in Ammi majus and Petroselinum crispum (21)(22)(23). Psoralen is next hydroxylated at C-5 by yet another P450 monooxygenase-generating bergaptol, which is subsequently O-methylated to bergapten (5-methoxypsoralen) (24,25). Alternatively, the formation of xanthotoxin (8-methoxypsoralen) from psoralen is supposed to proceed via analogous reactions (i.e. successive hydroxylation at C-8 and O-methylation), but this remains to be established. The pathway to angular furanocoumarins has not been as well documented, especially at the level of enzyme activity. This pathway branches off from that for psoralens by divergent prenylation of umbelliferone ( Fig. 1) to form osthenol. The two subsequent steps leading to angelicin were assumed to be very similar to the steps leading to the conversion of demethylsuberosin to (ϩ)-marmesin and psoralen. This assumption is particularly convincing when the catalytic mechanisms of psoralen synthase and angelicin synthase are compared. Previous studies using stereospecifically deuterated (ϩ)-marmesin confirmed that psoralen synthase activity catalyzes a single concerted reaction, initiated by the abstraction of the C-3Ј-hydrogen synconfigured to the isopropyloxy group and release of acetone as a by-product (Fig. 1) (26,27). An equivalent study demonstrated by feeding leaves of Heracleum mantegazzianum with stereospecifically 3Ј-deuterated (ϩ)-columbianetin that the conversion of (ϩ)-columbianetin to angelicin (i.e. angelicin synthase activity) also proceeded by a syn-elimination at C-3Ј (28), although this was inferred by mass spectral analysis of metabolites rather than by in vitro enzyme assays. The mechanistic and stereochemical similarities of the two pathways and the limited distribution of angular furanocoumarins to a few plant species that already synthesize linear structures support the assumption that the two pathways are driven by structurally related enzymes.
The molecular genetics of furanocoumarin formation has remained elusive, due to the high lability of the specific activities in crude plant microsomal preparations (21), until the very recent isolation of the first furanocoumarin-committed P450 gene, CYP71AJ1, which encodes psoralen synthase. This gene was cloned from A. majus (Apiaceae), which produces only linear furanocoumarins (23). Notably, the recombinant enzyme efficiently catalyzes the conversion of (ϩ)-marmesin to psoralen but is unable to metabolize the angular substrate (ϩ)-columbianetin, although the angular compound acts as a competitive inhibitor of psoralen synthase activity. Modeling studies and docking calculations suggested that (ϩ)-columbianetin fitted properly into the catalytic pocket, as required for competitive inhibition, but the distance of C-3Ј to the active site porphyrin-iron was too large to initiate the process of abstracting hydrogen (23). The in silico results raised the possibility that in plants producing both linear and angular furanocoumarins, a rather limited number of mutations in CYP71AJ1 paralogs could have generated angelicin synthase; this would have been as a pivotal step in the evolution of the angular furanocoumarin pathway, provided that umbelliferone 8-prenyltransferase for the synthesis of osthenol was also present.
In the present work, we describe the cloning and functional expression of two new orthologs of A. majus CYP71AJ1, encoding psoralen synthase, from Apium graveolens (celery CYP71AJ2) and Pastinaca sativa (parsnip CYP71AJ3). Additionally, the closely related CYP71AJ4 was isolated from parsnip and annotated as angelicin synthase, the first P450 gene committed to angular furanocoumarin biosynthesis. Alignments of the paralogs (CYP71AJ3 and CYP71AJ4) revealed that the sequence identity dropped to only 40% in substrate recognition sites (SRSs) 3 at an overall identity of 70%. Assays conducted with the recombinant enzyme and syn-[3Ј-D]columbianetin demonstrated a reaction mechanism analogous to that of psoralen synthase initiated by the abstraction of syn-C-3Јhydrogen. These results demonstrate that the CYP71AJ subfamily is involved in the synthesis of both linear and angular furanocoumarins and suggest that CYP71AJ3 and CYP71AJ4 derived from a common ancestor gene.

EXPERIMENTAL PROCEDURES
Plant Material-Parsnip (Pastinaca sativa) plants var. "Demi-Long de Guernesey" (Caillard, Avignon, France) were grown in the university greenhouse for 5 weeks. Celery (A. graveolens) plants were bought at a local market.
(ϩ)-Columbianetin was preparatively isolated from dried roots of Angelica pubescens. The roots (Huisong Pharmaceuticals, Hangzhou, China) obtained from Chongde Sun (Zheijang University, Hangzhou, China) were coarsely chopped, redried in an oven at 80°C for ϳ5 h, and then ground to a fine powder by mill and mortar. The powder (30-g aliquot) was stirred for 60 min in acetone (100 ml) at room temperature, the acetone extract was evaporated to dryness, and the oily residue was partitioned in a mixture of n-hexane/methanol/water 10:1:9 to remove lipids. The polar phase was reduced by evaporation to remove methanol, and the remaining aqueous phase was extracted with ethyl acetate, yielding the organic crude extract. This extract was applied to a silica gel 60 (Merck) column (3 ϫ 35 cm), and the elution was performed in two steps, with chloroform (250 ml) followed by chloroform/acetone 9:1 (500 ml). After the first 200 ml, fractions of 10 ml each were collected and examined by silica F 254 thin layer chromatography (Merck) using toluene/acetic acid 4:1 as the mobile phase. Fractions containing blue-fluorescing compounds under UV 254 irradiation and co-migrating on TLC with authentic columbianetin (Rf ϳ0.3-0.4) were pooled, evaporated to dryness, and subjected to high speed countercurrent chromatography on an HSCCC instrument (P.C. Inc.) equipped with a 400-ml Teflon coil and 10-ml sample loop switch. Solvents were delivered to the coil, the flow was maintained by an L6000 pump (Merck-Hitachi), and fractions were collected manually. The partition coefficients and purity of fractions were determined on an HPLC system (Merck-Hitachi) consisting of a D6000 interface, L6200A gradient pump, AS4000 autosampler, and L4500 diode array detector. Samples were analyzed on an EC 125/4 Nucleodur Sphinx RP column (Machery-Nagel, Düren, Germany) operated at a flow rate of 1 ml/min. The two-solvent system employed for separation was composed of 1.5% (v/v) aqueous phosphoric acid (solvent A) and 80% (v/v) acetonitrile (MeCN) in water (solvent B) and using linear gradients of 80 to 50% A in B within 20 min, followed by 50 to 20% A in B over 20 -25 min before a final re-equilibration to the starting conditions. The elution was monitored at 330 nm. Solvents chosen for HSCCC separation were mixed at appropriate ratios in a separation funnel, shaken vigorously, and allowed to stand overnight for complete phase separation. The HSCCC coil was then equilibrated with the stationary phase, and the sample was dissolved in 10 ml of the mobile phase and injected into the coil for separation at 890 -900 rpm at room temperature and a flow rate of 1 ml/min. Fractions of 2-10 ml were collected, fluorescent substances were detected in small aliquots spotted on filter paper under UV irradiation, and fractions of interest were examined further by TLC and HPLC analyses. The substance identity was finally verified by negative mode mass spectral fragmentation on a VG Micromass 7070 mass spectrometer at 70 eV ionization energy and comparison with authentic (ϩ)-[3Ј-D]columbianetin (G. Laufenberg and N. Zitzer, Institute of Pharmaceutical Chemistry, Philipps-University Marburg). Columbianetin-containing fractions were pooled and concentrated to remove methanol; the remaining aqueous phase was extracted with ethyl acetate and dried.
cDNA Cloning-Total RNA was extracted from leaves of 6-h mechanically wounded celery and parsnip with the RNeasy plant extraction kit (Qiagen). The cDNA was generated using SuperScript TM III RT (Invitrogen) and random primers. A fulllength cDNA sequence from celery and two full-length cDNA sequences from parsnip were PCR-amplified using CYP71AJ1 (accession number AY532373.1) specific primers (forward, 5Ј-CGGTACCATGAAGATGCTGGAACAGAAT; reverse, 5Ј-GGAATTCTCAAACATGTGGTCTGGCAACCACCAAG-AGAGG) and proofreading polymerase FideliTaq (GE Healthcare). The PCR amplification included a touch-down (A) and a classical (B) PCR as follows: 5 min at 94°C, followed by 10 cycles (30 s at 94°C, 45 s at 60°C Ϫ1°C/cycle, and 90 s at 72°C), 20 cycles (30 s at 94°C, 45 s at 50°C, and 90 s at 72°C), and a final 10-min extension step at 72°C. The CYP71AJ2 PCR product from celery (1498 bp) and those from parsnip (CYP71AJ3, 1485 bp; CYP71AJ4, 1503 bp) were purified and cloned in the pCR8-TOPO vector (Invitrogen) prior to sequencing. In order to identify the specific 5Ј-and 3Ј-ends of CYP71AJ3 and CYP71AJ4 orthologs, an inverse PCR strategy adapted from Stemmer and Morris (29) was used. The genomic DNA from parsnip leaves was isolated using the DNeasy plant extraction kit (Qiagen). One g of genomic DNA from each plant was digested during 2 h at 37°C by 6 units of the restriction enzyme DraI. After heat inactivation of DraI during 20 min at 65°C, 0.1 g of digested gDNA was self-ligated overnight at 16°C using 4 units of T4 DNA ligase (Invitrogen) in a final volume of 50 l. This ligated gDNA was used as a matrix for PCR amplification using specific primers. The 5Ј-and 3Ј-ends of the celery CYP71AJ2 were amplified using forward primer 5Ј-GCAATG-GCTGTTAATGACCTTGTAGTTGCAAATGTC and reverse primer 5Ј-ATCATGAGTTTTCAAAACCTCACGAGCCGC. The 5Ј-and 3Ј-borders of parsnip CYP71AJ3 and CYP71AJ4 were amplified with the forward primers 5Ј-GCTATGGC-TGTTAATGAGCTTGCTGTAGCAAATGTT and 5Ј-GCA-ATCTCTGTTAATGAGCTTGTAGTAGCAAAT and the reverse primers 5Ј-GACTACATTTCTGGTTAGTTGGATT-AATTGCTCGCC and 5Ј-AGAATTTTCAATATTATCG-AGTAACAGAGCAAC, respectively. In each case, the temperature cycles consisted of 5 min at 94°C, 30 cycles (30 s at 94°C, 45 s at 55°C, 150 s at 72°C). PCR products were purified and then cloned into pCR8 TOPO TA GW vector prior to sequencing.
Yeast Expression and Enzyme Assays-The celery and parsnip CYP71AJ1 orthologs corresponding ORF were ligated in the yeast expression vector pYeDP60 (30) between KpnI and EcoRI restriction sites. Yeast transformation was performed as described elsewhere (31). Propagation of yeast cells and preparation of microsomes were conducted as described previously (23). Protein amounts were determined according to Bradford (32), and the microsomal P450 contents were determined as described by Omura and Sato (33). Enzymatic screening was done with microsomes incubated in 0.1 M sodium phosphate buffer containing 100 M substrate and 1 mM NADPH. The incubation was performed for 1 h at 27°C and 800 rpm agitation (Eppendorf Thermomixer compact). The reaction was stopped by adding 75 l of MeCN/HCl (99:1). Product formation and substrate consumption were analyzed by reverse phase HPLC on a Purospher 5-m, 4 ϫ 250-mm end-capped column (Merck). The column was equilibrated in water/acetic acid (99.9/0.1) at a flow rate of 0.9 ml min Ϫ1 , and the elution was performed under diode array detection (220 -400 nm) using a linear gradient of MeCN/methanol (1:1) from 0 to 60% for 25.0 min, followed by 100% MeCN/methanol (1:1) for an additional 10 min.
Kinetic Analyses-Kinetic parameters for psoralen synthase activity were determined as previously described (23). The activity of angelicin synthase was measured for 10.0 min in 0.1 M sodium phosphate buffer, pH 7.0 (150.0 l total), containing 0.5 mM NADPH, 0.5 pmol of the recombinant CYP71AJ4 and varying the concentration of (ϩ)-columbianetin or syn-[3Ј-D]columbianetin. The apparent K m values were determined by Lineweaver-Burk extrapolations. Mechanism-based inactivation assays were performed as described previously (23).
Gas Chromatography-Mass Spectrometry Analyses-The product eluting from HPLC at a retention time corresponding to angelicin was collected from 10 incubations, the solvent was evaporated in an air stream, and the pellet was resuspended in chloroform/MeCN (1:1) for gas chromatography-mass spectrometry analysis. The sample (1.0 l) was injected into a Varian Star 3400 CX spectrometer fitted with a Varian Factor Four 5MS column (15.0 m ϫ 0.25-mm inner diameter, 0.1-m film) using helium at 5.5 p.s.i. as carrier gas. The column temperature was initially 60°C for 2 min, ramped to 90°C at 40°C min Ϫ1 and held for 4 min and then to 200°C at 10°C min Ϫ1 , and finally to 300°C at 10°C min Ϫ1 and held for 10.0 min. Mass spectra were recorded at 70 eV, scanning from 60 to 400 atomic mass units.

HPLC-Solid Phase Extraction (SPE)-1 H NMR Analyses-
The sample was subjected to coupled HPLC-SPE-1 H NMR analysis using the following equipment and experimental conditions. HPLC was performed on an Agilent 1100 chromatography system (quaternary pump G1311A; autosampler G1313A; J&M photodiode array detector operating at a wavelength range between 200 and 620 nm, monitoring wavelength 254 and 325 nm). A LiChrospher 100 RP-18 column (250 ϫ 4 mm, particle size 5 m) with MeCN/water (0.01% trifluoroacetic acid) as a mobile phase (flow rate 1.0 ml min Ϫ1 ) in linear gradient mode from 10 to 90% MeCN in 40 min was applied. After photodiode array detector detection, the eluting solvent was diluted with H 2 O by a makeup pump to decrease eluotropic strength. The peak eluting at R t 10.25 min was trapped on a poly(divinylbenzene) SPE cartridge (HySphere resin GP) using a Prospekt 2 SPE unit (Spark Holland) as an LC-NMR interface. The cartridge containing the trapped peak was dried with a stream of nitrogen gas for 30 min. Then the compound was eluted with MeCN-d 3 and transferred through the connecting capillary from the Prospekt 2 SPE unit to the NMR spectrometer (Bruker AV 500). The NMR spectrometer was equipped with a 5-mm TCI cryoprobe and Cryofit TM flow insert (30-l active volume). The 1 H NMR spectrum of the trace sample amount of the enzyme product was measured in MeCN-d 3 at 300 K, and 2024 scans were added to the FID. The central line of the solvent signal at ␦ 1.93 was used for referencing. Residual solvent signals were suppressed using presaturation. 1  Bioinformatic Analysis-Multiple sequence alignments were generated with ClustalX.

RESULTS
Isolation of (ϩ)-Columbianetin-The recent characterization of A. majus CYP71AJ1 (psoralen synthase) (23) opened the door to further investigations of furanocoumarin biosynthesis in other Apiaceae. The reaction in the angular pathway analogous to psoralen synthase (angelicin synthase), however, depends on (ϩ)-columbianetin as a substrate (Fig. 1), but this was not available commercially or from laboratory collections and therefore had to be isolated from natural sources. Based on the literature (34,35), several two-phase solvent mixtures were used to examine the separation of columbianetin extracted from A. pubescens roots and prepurified by silica column chromatography, and the partition coefficient k ϭ [columbianetin] stat /[columbianetin] mob (36) was analyzed by HPLC. Columbianetin was enriched in the aqueous phase of systems composed of n-hexane, ethyl acetate, methanol, water at volume ratios 5:5:5:5 (I) and particularly 5:5:7:4 (II) with k values of 0.40 and 0.11, respectively. The dry prepurified root extract (221 mg from 30 g of dry root) was therefore dissolved in solvent system II (10 ml) and loaded onto a HSCCC coil for separation at 890 rpm, and the  FEBRUARY 20, 2009 • VOLUME 284 • NUMBER 8 eluate (20-ml fractions) was analyzed by TLC ( Fig. 2A). Fractions 6 -9 (ϳ85 ml) were combined; columbianetin was extracted from the aqueous phase after the removal of methanol, evaporated to dryness, and subjected to another round of HSCCC separation in solvent system I at 900 rpm. Fractions containing columbianetin only (Ͼ95% purity) as judged by TLC were pooled and examined by HPLC (Fig. 2B). Unequivocal identification of columbianetin was achieved by mass spectral fragmentation and comparison with authentic [3Ј-D]columbianetin (Fig. 2, C and D), revealing the [M ϩ ] at m/z 246 with major fragments at m/z 187 (base peak) and 160. A yield of 9.48 mg (38.53 mol) columbianetin was recovered and assigned the (ϩ)-configuration, because only (ϩ)-columbianetin derivatives have been reported from A. pubescens (38). The preparative purification also revealed significant amounts of conjugated columbianetin (i.e. the angeloylester columbianadin identified by co-chromatography with an authentic sample); although unconjugated (ϩ)-columbianetin had never been described from A. pubescens roots, our yield corresponds to 0.03% of the dry herbal drug.

P. sativa Angelicin Synthase
Isolation and Sequence Analysis of CYP71AJ1 Orthologs-Psoralen synthase (CYP71AJ1) possesses a narrow substrate specificity and binds to (ϩ)-columbianetin as a competitive inhibitor of (ϩ)-marmesin only (23). Three-dimensional docking simulations indicated a nearly perfect fitting of the (ϩ)columbianetin core into the active site pocket (23) and suggested that a limited number of evolutionary mutations might have conferred some angelicin synthase activity on the CYP71AJ1 sequence. Such mutations must have developed to encompass the isopropyloxy side chain and C-2Ј/3Ј of (ϩ)-columbianetin, which are oriented far from the iron-porphyrin active center in the CYP71AJ1-columbianetin model. Under these conditions, angelicin synthase could be spotted beside further psoralen synthases by cloning CYP71AJ1 orthologs. A. graveolens (celery) and P. sativa (parsnip) belong to the Apiaceae, just like A. majus, but are capable of producing linear and angular furanocoumarins when their tissue is wounded (1,2). Furthermore, their common codon usage (available on the Kazusa World Wide Web server) is an essential prerequisite for a successful nondegenerated PCR approach. The amplification of CYP71AJ1 orthologs was therefore accomplished using RNA from wounded celery and parsnip leaves as a template for reverse transcription-PCR, followed by a two-step strategy combining the PCR with CYP71AJ1-specific primers and inverse PCR techniques. Three full-size cDNA sequences, one from celery and two from parsnip, were isolated; these ranged from 1482 to 1503 bp in length (Fig. 3). Comparison of the translated polypeptides revealed more than 65% identity with CYP71AJ1, which groups the enzymes into the same P450 subfamily. Classification by the international P450 nomenclature committee (available on the cytochrome P450 home page World Wide Web server) assigned the sequences as CYP71AJ2 (celery; GenBank TM accession number EF191022), CYP71AJ3, and CYP71AJ4 (parsnip; GenBank TM accession numbers EF191020 and EF191021). The CYP71AJ2 and CYP71AJ3 polypeptides are closely related to CYP71AJ1 (Fig. 3), with an identity of 89 and 82%, respectively, but contain a four-amino acid insertion at position 258 (CYP71AJ2) or one deletion at position 28 and a Pro 257 insertion (CYP71AJ3). In contrast, CYP71AJ4 showed only 70% identity to CYP71AJ1 with two single deletions at positions 28 and 370 and two four-amino acid insertions at positions 192 and 261 (Fig.  3). The divergence in polypeptide sequences becomes even more pronounced when the six substrate recognition sites (SRS1 to -6) (39) previously defined in CYP71AJ1 (23) are compared (Table 1). Most notably, the level of sequence identity with CYP71AJ1 in SRS1, -4, and -5 of CYP71AJ2 and CYP71AJ3 was higher than the level of sequence identity for the full sequence average, whereas the inverse tendency was observed for the sequence identity of CYP71AJ4 dropping to 37, 40, 42, and 43%, respectively, in SRS1, -2, -3, and -6.
Functional Expression of CYP71AJ2, CYP71AJ3, and CYP71AJ4-The genes CYP71AJ2 to -4 were individually expressed in yeast strain WAT 11, as described earlier for the functional characterization of CYP71AJ1 (23). A low level of expression was recorded for each of the transformants, as was true of CYP71AJ1 (23) and revealed by weak absorbance at 450 nm in the differential CO spectrum (data not shown). Nevertheless, the expression was sufficient for activity screenings of the crude yeast microsomal fractions. In the search for the substrate, enzyme assays were performed with either of two intermediates of the furanocoumarin pathway (demethylsuberosin, (ϩ)-marmesin, 5-hydroxymarmesin, psoralen, bergaptol, xanthotoxol, bergapten, xanthotoxin, isopimpinellin, (ϩ)-columbianetin, or angelicin) as well as with cinnamic acids (cinnamic acid, 4coumaric acid, 2-coumaric acid, or ferulic acid) and simple coumarins (coumarin, herniarin, scopoletin, or umbelliferone). CYP71AJ2 and CYP71AJ3 were shown to encode NADPH-dependent psoralen synthases with maximal activity in sodium phosphate buffer at pH 7.5 and substrate affinities to (ϩ)-marmesin of K m ϭ 0.54 Ϯ 0.21 M and 1.3 Ϯ 0.40 M, respectively (Table 2). Analogous to CYP71AJ1 (23), both recombinant enzymes were capable of converting 5-hydroxymarmesin to bergaptol with much lower affinities of K m ϭ 54 Ϯ 16 M and 19 Ϯ 7 M ( Table 2); none of the other compounds included in the assays were used. These data clearly proved that CYP71AJ2 and CYP71AJ3 are psoralen synthases. In contrast, recombinant CYP71AJ4 did not exhibit psoralen synthase activity but catalyzed the NADPH-dependent conversion of (ϩ)-columbianetin (Fig. 1) to a major product and a minor by-product, eluting at 27.5 and 21.3 min, respectively, from HPLC (Fig. 4, A and B). Neither was detected in control assays employing microsomes from yeast transformed with the empty plasmid vector. Maximal conversion occurred at pH 7.0 in phosphate buffer and a temperature range of 27-30°C. The major product was unequivocally identified as angelicin due to its ability to absorb max 300 nm and electron impact mass spectrum with an [M ϩ ] of m/z 186 and fragments at m/z 158, 130, 102, and 51, which fully matched the fragmentation of commercial angelicin (Fig. 4, C-F). Mass spectral analysis of the minor by-product revealed an [M ϩ ] of m/z 263 and fragments corresponding to a hydroxylated columbianetin derivative. The fragmentation pattern (m/z fragments at 249, 229, and 187) clearly indicated that hydroxylation of the isopropyloxy side chain had occurred (Fig. 4G). Further incubation of the isolated hydroxyl derivative of columbianetin with CYP71AJ4 failed to give angelicin (data not shown), ruling out the possibility that this derivative was an intermediate in angelicin formation. Thus, CYP71AJ4 is the first gene reported from the angular branch of furanocoumarin biosynthesis and encoding angelicin synthase.
Characterization of Recombinant CYP71AJ4-Microsomes harvested from yeast cells expressing CYP71AJ4 showed narrow substrate specificity for (ϩ)-columbianetin, since none of the furanocoumarins, coumarins, or cinnamic acids listed    (23) and other plant enzymes of the CYP71 family (40,41). The formation of the hydroxycolumbianetin by-product by angelicin synthase occurred 3-4 times less efficiently than the formation of angelicin, as indicated by the k cat /K m ratios (Table 3). Angelicin synthase assays were also conducted in the presence of increasing concentrations of (ϩ)marmesin (30 -200 M). Surprisingly, no inhibition was observed (data not shown). Angelicin synthase assays were also performed in the presence of psoralen, 5-or 8-methoxypsoralen (100 M), and angelicin (100 M). All of these compounds failed to inhibit angelicin synthase in vitro (data not shown), which is consistent with the proposal that P450s from furanocoumarin-producing plants are resistant to the mechanismbased inhibition by furanocoumarins reported for various other P450s (12,42,43). Mechanistic Considerations-Precursor feeding experiments using leaves of H. mantegazzianum had shown that the conversion of (ϩ)-columbianetin to angelicin proceeds by 2Ј,3Ј-synelimination of the isopropyloxy side chain. Although no direct proof was given, this reaction is probably initiated by the abstraction of the respective C-3Ј-hydrogen (28). The analogous reaction was confirmed in vitro for the conversion of (ϩ)-marmesin by psoralen synthase (27). The availability of recombinant angelicin synthase and specifically deuterated (ϩ)columbianetin offered the opportunity to extend the in vivo investigations of angelicin formation by in vitro studies. Incubation of syn-[3Ј-D]columbianetin with recombinant angelicin synthase revealed a new product. This compound is synthe-  sized at the expense of a relatively large decrease in the formation of angelicin and hydroxycolumbianetin by-product, as compared with control assays employing natural (ϩ)-columbianetin as substrate (Fig. 5A). HPLC and gas chromatographymass spectrometry (n) analyses unambiguously assigned the new product as another hydroxylated columbianetin derivative retaining the deuterium label (Fig. 5B). Roughly 2 g of the new product was collected from several incubations for NMR analysis. The LC-SPE- 1  (J H-3Ј-H-2Ј ϭ 9.8 Hz) to a singlet. This was apparently not the case. Instead, the signal of H-3Ј disappeared from the spectrum of the new product, indicating that the proton had been abstracted. Thus, hydroxylation at C-2Ј can be excluded, and the deuterium label at C-3Ј must be retained. Moreover, compared with the deuterated substrate (44), which was measured under the same conditions as the product (see "Experimental Procedures"), the signal of H-2Ј was shifted upfield to ␦ 4.36 and collapsed to a singlet, confirming loss of the adjacent proton at C-3Ј (Fig. 5C). The unresolved signals between ␦ 3.2 and 3.9 ppm can be ascribed to impurities of the minute product sample. Moreover, hydroxylation at C-2Ј is highly improbable in light of the literature (1,2). Taken together, these results are fully consistent with the new product as syn-[3Ј-D](anti-3Ј-hydroxy)-columbianetin (Fig. 5D).
Kinetic studies revealed that the affinities of recombinant angelicin synthase to natural columbianetin or syn-[3Ј-D]columbianetin were almost identical (Table 3). However, the conversion rates k cat to angelicin and its derivative hydroxylated in the isopropyloxy residue observed in assays with natural columbianetin were dramatically affected when the C-3Ј deuterium-labeled substrate was used, because the ratio k cat /K m dropped from 55.2 to 11.3 (Table 3). Furthermore, gas chromatography-mass spectrometry analysis of angelicin isolated from the CYP71AJ4 assays with syn-[3Ј-D]columbianetin revealed the loss of deuterium (data not shown), demonstrating that syn-3ЈD was the site of attack by the iron-oxo center of angelicin synthase. In contrast, formation of the new product hydroxylated in anti-orientation at C-3Ј in these assays was considerably favored, as indicated by a k cat of 234 Ϯ 66 min Ϫ1 (Table 3).

DISCUSSION
Despite the considerable interest in furanocoumarins as therapeutics or plants' herbivore defense factors, the biosynthesis and genetics of these compounds, particularly the angular furanocoumarins, are still poorly understood (45,46). Evolution of the angular pathway is thought to have developed later from genes of the pathway leading to psoralens and as a consequence of herbivore selection pressures (47); such a later devel-