Evolution of Flavone Synthase I from Parsley Flavanone 3 b -Hydroxylase by Site-Directed Mutagenesis 1[W][OA]

Flavanone 3 b -hydroxylase (FHT) and ﬂavone synthase I (FNS I) are 2-oxoglutarate-dependent dioxygenases with 80% sequence identity, which catalyze distinct reactions in ﬂavonoid biosynthesis. However, FNS I has been reported exclusively from a few Apiaceae species, whereas FHTs are more abundant. Domain-swapping experiments joining the N terminus of parsley ( Petroselinum crispum ) FHT with the C terminus of parsley FNS I and vice versa revealed that the C-terminal portion is not essential for FNS I activity. Sequence alignments identiﬁed 26 amino acid substitutions conserved in FHT versus FNS I genes. Homology modeling, based on the related anthocyanidin synthase structure, assigned seven of these amino acids (FHT/ FNS I, M106T, I115T, V116I, I131F, D195E, V200I, L215V, and K216R) to the active site. Accordingly, FHT was modiﬁed by site-directed mutagenesis, creating mutants encoding from one to seven substitutions, which were expressed in yeast ( Saccharo- myces cerevisiae ) for FNS I and FHT assays. The exchange I131F in combination with either M106T and D195E or L215V and K216R replacements was sufﬁcient to confer some FNS I side activity. Introduction of all seven FNS I substitutions into the FHT sequence, however, caused a nearly complete change in enzyme activity from FHT to FNS I. Both FHT and FNS I were proposed to initially withdraw the b -face-conﬁgured hydrogen from carbon-3 of the naringenin substrate. Our results suggest that the 7-fold substitution affects the orientation of the substrate in the active-site pocket such that this is followed by syn - elimination of hydrogen from carbon-2 (FNS I reaction) rather than the rebound hydroxylation of carbon-3 (FHT reaction).

Flavones and flavonols are the predominant flavonoids found in tissues of Apiaceae species (Harborne, 1971;Harborne and Williams, 1972;Harborne and Baxter, 1999). Significant functions were ascribed to these metabolites for growth and propagation of plants, as well as for adaptation to ecological niches. Flavonoids have been shown to protect from UV radiation, provide pigmentation, mediate the plant's interaction with insects or microbes, and act as feeding deterrents and phytoalexins (Harborne and Williams, 2000;Martens and Mithö fer, 2005). Flavones (i.e. apigenin) are formed by direct 2,3-desaturation of natural flavanones such as (2S)-naringenin ( Fig. 1). In Apiaceae, this reaction is catalyzed by a soluble Fe 21 / 2-oxoglutarate-dependent dioxygenase (2-ODD), flavone synthase I (FNS I), whereas FNS II, a cytochrome P450-dependent monooxygenase, was found in all other flavone-producing plants investigated so far. Flavonols are formed from flavanones by sequential hydroxylation of carbon-3 and 2,3-dehydration involving flavanone 3b-hydroxylase (FHT) and flavonol synthase (FLS; Fig. 1), although in vitro FLS has the capability of catalyzing both steps . FHT and FNS I or II thus compete for flavanones as common substrates and the product of FHT may be delivered to the anthocyanidin branch pathway instead of desaturation by FLS (Fig. 1). FHT also belongs to the superfamily of 2-ODDs and is closely related to FNS I. Both enzymes had been proposed to attack the flavanone substrate in identical fashion and withdraw initially the b-configured hydrogen (trans-to B-ring substitution) from carbon-3 ( Fig. 1).
Considerable research has been dedicated to the mechanism of 2-ODD catalysis because of the relevance of these enzymes in the metabolism of microorganisms (antibiotics), plants (hormones, pigments), or mammals (connective tissue diseases, hypoxia-inducible factor). Two electrons are gained from the decarboxylation of 2-oxoglutarate and transferred to Fe(II) in the enzyme-active center, forming a highly reactive ferryl intermediate, which mobilizes molecular oxygen for hydroxylation, desaturation, epoxidation, ring closure, or expansion reactions (Prescott and Lloyd, 2000). Members of the 2-ODD superfamily do not always show close sequence identity, but rather appear to cluster in one of three groups of related enzymes or fall into a fourth group of unrelated sequences (Hogan et al., 2000;Prescott and Lloyd, 2000). FHT and FNS I were assigned to group I of this superfamily together with, for example, microbial isopenicillin N synthase (IPNS) and deacetoxycephalosporin C synthase (DAOCS), having in common a HXDX ;55 HX 10 RXS motif (Borovok et al., 1996;Prescott and Lloyd, 2000). These residues are particularly important for enzyme activity as revealed by site-directed mutagenesis (Lukacin and Britsch, 1997;Myllyharju and Kivirikko, 1997). Furthermore, IPNS (Roach et al., 1995 and DAOCS (Roach et al., 1995Valegard et al., 1998;Lloyd et al., 1999;Wilmouth et al., 2002), as well as anthocyanidin synthase (ANS) from Arabidopsis (Arabidopsis thaliana; Wilmouth et al., 2002) as another group I dioxygenase, were crystallized and revealed that these residues comprise the metallocenter and 2-oxoglutarate binding site. For detailed review of 2-ODDs, see Prescott and Lloyd (2000) and Clifton et al. (2006).
The evolution of FNS I in species of Apiaceae ascribes an essential role to flavones for the plant's existence and propagation. Moreover, the confinement of FNS I to one evolutionary advanced plant family, as compared to the more abundant expression of FHTs, suggested that FNS I developed much later than FHT. The close relationship of the FNS I polypeptide with those of FHTs and alignments with other 2-ODD sequences cloned from Apiaceae thus led to the hypothesis of gene duplication and subsequent change of function (Gebhardt et al., 2005). Very few conserved differences became apparent on alignment of FNS I and FHT sequences from parsley (Petroselinum crispum), which were likely to determine divergent catalytic activity. Apiaceae presumably benefit from stable maintenance of the new FNS I gene, which leads to accumulation of flavones, and the selective advantage has precluded any further diversification of 2-ODD functionality toward the destructive turnover of flavones. However, experimental evidence for this hypothesis has been lacking. To identify the amino acid residues essential for FNS I activity, we constructed chimera from fully functional parsley FHT and parsley FNS I  and generated step-by-step mutants of the FHT. The chimera functionally expressed in yeast (Saccharomyces cerevisiae) revealed the primary importance of the N-terminal enzyme portion for FNS I catalysis, and site-directed mutagenesis identified the minimal requirement of three amino acid substitutions to shift the FHT toward FNS I activity.

Cloning of 2-ODDs from Apiaceae and Sequence Alignment
In addition to the FHT and FNS I sequences already identified from various species of Apiaceae (Gebhardt et al. 2005), full-length FNS I from Aethusa cynapium (DQ683350), Angelica archangelica (DQ683352), and Cuminum cyminum (DQ683349), as well as FHT from A. cynapium (DQ683351) were cloned by PCR amplification and verified by functional expression. Alignments of all these translated polypeptides corroborated the previous finding of 27 amino acids differently conserved in FNS I versus FHT, albeit A. archangelica FNS I is exceptional with Val-312 instead of the otherwise common Ile (Fig. 2). Nevertheless, this conserved exchange unlikely bears functional consequences. Alignments of Apiaceae polypeptides with FHTs from other plant families recognized some of the exchanges conserved in FNS I also in these FHTs, which, however, cannot be disregarded for further study because these residues might negligibly affect FHT activity, but essentially support FNS I activity. The most striking difference among Apiaceae enzymes is a C-terminal triplet of FHT (Gln-348/Glu-349 or Asp-349/Trp-350 or Ala-350 or Val-350) deleted in FNS I (Fig. 2). However, this triplet was not conserved in non-Apiaceae FHTs, presumably due to weak conservation of the entire C terminus, and insertion of the triplet in Daucus carota FNS I did not change the activity (data not shown). Provided that FNS I has evolved from FHT, the triplet that does not affect FHT or FNS I activity was likely deleted shortly after gene duplication. The remaining conserved differences between FNS I and FHT noted on the alignments are scattered over the entire sequence; chimeric proteins were therefore constructed by swapping about 40% of the C-terminal portion between parsley FHT and FNS I.

Significance of the C Terminus
Previous studies concerning the relevance of the C-terminal portion of 2-ODDs on enzyme activity did not provide a coherent picture. Deletion of six amino acids from the C terminus of Aspergillus nidulans IPNS (Sami et al., 1997) or Streptomyces clavuligerus DAOCS (Valegard et al., 1998) significantly diminished the activity and it was suggested that the enzymes, which accommodate the active site in a b-sheet barrel, use the C terminus as a protective lid to maintain a hydrophobic environment and to enable proper cofactor binding (Lloyd et al., 1999). In the case of chimeric GA 20-oxidases, a pronounced influence of the C terminus on product selectivity was noticed (Lange et al., 1997). However, penicillin ring expansion by Acremonium chrysogenum DAOCS/deacetylcephalosporin C synthetase was not affected by the C-terminal deletion of 20 amino acids (Chin et al., 2003). The specificity of Petunia hybrida FHT was also retained on C-terminal truncation of 5, 11, or 24 amino acid residues, albeit the specific activity dropped by 56% to 72.6% (Wellmann et al., 2004). Petunia FHT activity was nearly lost, however, on deletion of 29 C-terminal amino acids (0.4% activity of wild type) or on swapping the C-terminal portion of 52 amino acids by the corresponding sequence from Citrus unshiu FLS (0.3% activity of wild type) without a change in specificity (Wellmann et al., 2004). Thus, the contribution of the C terminus to enzyme activity is variable for different 2-ODDs and this aspect was examined for parsley FHT and FNS I.
Two FHT/FNS I chimera (Pet_criChim I and II) were constructed from pYES2.1 clones harboring fully functional parsley FNS I or FHT sequences (Martens et al., 2001Supplemental Fig. S1). Pet_criChim I was composed of the N-terminal 219 amino acids of FNS I ligated to the C-terminal 149 residues of FHT, whereas in Pet_criChim II the C-terminal 146 amino acid residues of FNS I were joined to the N-terminal portion of 219 amino acids from FHT. Notably, however, amino acids 217 to 296 are highly conserved in FNS I and FHT, except for position 231 in FHTs outside Apiaceae; therefore, the last 72 amino acids of the chimera only differed from the wild-type enzymes. The constructs were overexpressed in yeast and the FHT or FNS I activity of crude extracts was determined in standard assays employing 14 C-labeled naringenin as substrate followed by thin-layer chromatography (TLC) separation and autoradiography . The effects of these swapping experiments on FHT versus FNS I activity (Fig. 3) differed considerably because recombinant Pet_criChim I mostly retained FNS I activity converting naringenin to apigenin with little FHT side activity forming dihydrokaempferol, whereas Pet_criChim II showed weak FHT activity compared to the wild-type enzyme without a trace of FNS I activity. The data rule out an essential contribution of the C-terminal enzyme portion on FNS I activity, which is supported also by fully functional truncated FNS I reported recently from D. carota, Apium graveolens, and A. cynapium (Gebhardt et al., 2005), although an effect of the C terminus on activity cannot be neglected without kinetic evidence. It is obvious, furthermore, that the C-terminal portion of the enzyme is not strictly Figure 2. Schematic overview of conserved substitutions in FHT and FNS I sequences. Amino acids that were assigned by homology modeling to the active site are underlined. I312V is shown in gray because Ang_arcFNS I preserved Valin in this position. The BamHI restriction site used for construction of Pet_criChimI and II is marked by an arrow. Black bars indicate sequence regions that are either identical or not conserved and white bar regions mark the position of amino acids responsible for cofactor binding conserved in all 2-ODDs (Lukacin and Britsch, 1997). required for FHT activity, but the side activity of Pet_criChim I and the suppression of Pet_criChim II suggest a significant beneficial contribution on FHT activity. Taken together, a potential gain of FNS I functionality likely requires amino acid substitutions in the N-terminal portion (positions 1-216) of the FHT sequence.

Homology Modeling and Choice of Mutations
The structures of some crystallized 2-ODDs have been solved and reviewed by Clifton et al. (2006). These include mostly microbial enzymes (e.g. IPNS [Roach et al., 1995], DAOCS [Valegard et al., 1998], clavaminic acid synthase [Zhang et al., 2000], carbapenem synthase [Clifton et al., 2003] or Pro 3-hydroxylase [Clifton et al., 2001], and taurine/a-ketoglutarate dioxygenase [Elkins et al., 2002]). Factor-inhibiting hypoxia-inducible factor (Elkins et al., 2003), phytanoyl-CoA 2-hydroxylase (McDonough et al., 2005), and ANS (Wilmouth et al., 2002) are examples from mammalian and plant sources. Each of these enzymes keeps its active-site iron center in a hydrophobic environment enclosed by a double-stranded b-helix or jelly roll topology. However, the extent and the periphery of the jelly rolls may vary and the enzymes show only little sequence similarity. Regardless of these limitations, a-helices and b-sheets appear to be analogously assembled, leading to almost identical circular dichroism spectroscopic profiles for P. hybrida FHT (Lukacin et al., 2000), C. unshiu FLS (Wellmann et al., 2002), or IPNS (Borovok et al., 1996;Durairaj et al., 1996). Alignment of the 2-ODDs that have been examined by x-ray scattering with parsley FHT and FNS I polypeptides revealed the highest sequence similarity of approximately 30% with ANS.
In an attempt to denote more closely those amino acids responsible for substrate binding, model calculations were done based on the ANS structure, although sequence similarity exceeding 30% had been postulated for reliable homology modeling (Sanchez and Sali, 1997). Nevertheless, in the case of UDPglucosyltransferases from Sorghum bicolor, for example, 15% similarity was sufficient (Thorsøe et al., 2005). ANS had been cocrystallized with quercetin or naringenin (Wilmouth et al., 2002;Welford et al., 2005) because the natural leucoanthocyanidin substrates are unstable. The structure of the ANS-naringenin complex (2brt) was preferred for modeling because FHT and FNS I use naringenin as substrate. Due to little sequence similarity of FNS I and FHT with ANS in the N-and C-terminal region, model calculations are based on residues 30 to 305, excluding four short N-terminal a-helices (a-helices 1-4) and two C-terminal a-helices (a-helices 16 and 17) of ANS. ANS is characterized by a jelly roll topology (b5-b12) with a long a-helical backbone (a-helix 12) as observed earlier for IPNS or DAOCS. Furthermore, the jelly roll motif of ANS is extended by two b-sheets (b3 and b4). Homology model regions corresponding to ANS b3 to b6 are represented as almost straight loops (Fig. 4) because Protein Data Bank and SWISS-MODEL software use slightly different protocols for the assignment of secondary structure, but those regions adopt a similar orientation as the corresponding b-sheets of ANS. The positions of residues for iron binding are strictly conserved in the homology models generated and revealed His-218, His-276, and Asp-220 in parsley FHT or FNS I, as compared to His-232, His-288, and Asp-234 in ANS, for almost octahedral coordination in conjunction with the C1 and C2 carboxyls of 2-oxoglutarate. Conversion of 14 C-labeled (2S)-naringenin with crude protein extracts from yeast transformed with Pet_criFNS I, Pet_criFHT, Pet_ criChimI, or Pet_CriChimII. Enzyme assays were carried out as described previously . Substrate and product positions vary in the presentation because the assays were separated on different thin-layer plates. However, products were unequivocally identified by cochromatography with authentic standards: 1, apigenin; 2, naringenin; 3, dihydrokaempferol .
Substrate binding in ANS is facilitated through p-stacking of the naringenin A-ring with Phe-304, corresponding to Phe-292 in FHT or FNS I (Fig. 4). Furthermore, the 7-hydroxyl of naringenin can form a hydrogen bond with the side chain of Glu-306, whereas the equivalent position in FHT or FNS I is held by Asn-294, which does not engage in hydrogen bonding. ANS fixes the B-ring of the substrate through hydrophobic interaction with Phe-144 and hydrogen bonding of the 4#-hydroxyl to Tyr-142. These residues are lacking from FHT or FNS I, which encode Ala-133 and Ile-131 (FHT) or Thr-133 and Phe-131 (FNS I) instead. Moreover, Lys-213 was proposed to participate in protonation and deprotonation in ANS catalysis (Wilmouth et al., 2002), whereas this residue is lacking in FHT and FNS I due to a gap of three amino acids between residues 200 and 201 (Fig. 5). These data suggest substrate binding in the active-site cavity of FHT and FNS I different from that in ANS and corroborate the previous proposal of a-face specificity of ANS and FLS versus b-face specificity of FHT and FNS I Turnbull et al., 2004;Welford et al., 2005). High-affinity binding of substrate is essential in both FHT and FNS I, as well as in ANS catalysis because of the radical mechanisms and is supported by the narrow substrate specificity of FHT and FNS I and the absence of side reactions. Conceivably, additional side chains enforce substrate affinity, but low sequence similarity of FHT and FNS I with ANS and putative inaccuracy of modeled side chain orientations generally associated with homology modeling ruled out more informative docking calculations. Thus, the projection of naringenin in FHT and FNS I is shown as determined for ANS (Fig. 4).
Both FHT and FNS I were proposed to initiate the loss of the b-configured hydrogen from carbon-3 of naringenin and the parameters of substrate binding unlikely explain the difference in product formation. However, subtle sequence differences must determine the fate of the remaining radical. Most of the differences conserved in parsley FHT or FNS I polypeptide were recognized in the periphery, except for seven residues at or close to the active-site cavity, assigning the substitutions M106T, I115T, V116I, I131F, D195E, V200I, L215V, and K216R as potential cause of FHT-to-FNS I conversion (Fig. 5). The M106T exchange concerns a flexible loop corresponding to ANS a-helix 9 near the enzyme surface, and the substitutions I115T, Figure 4. A, Structure of ANS complexed with naringenin (2brt; Welford et al., 2005). The jelly roll motif (b-sheet 5 to b-sheet 12) and its extending b-sheets are represented in yellow. B, Homology model of Pet_criFHT based on the ANS-naringenin structure (2brt). C, Homology model of Pet_criFNS I based on the ANS-naringenin structure (2brt). a-Helices and b-sheets in the homology models were numbered according to the model template. Due to the limited sequence similarities and mechanistic differences between template and FNS I or FHT, the substrate naringenin likely adopts spatially different positions in the active-site pockets. The substrate was not fitted in B and C, but the models clearly resolve the conserved differences between FNS I and FHT concerning the active-site residues. Residues conserved differently in Pet_criFHT and Pet_criFNS I and chosen for mutagenic studies are shown as sticks.
V116I, and I131F (corresponding to Tyr-142 in ANS) are part of b-sheets 3 and 4, whereas D195E in a-14 and V200I in b-sheet 5 are exposed to the catalytic pocket. L215V and K216R are located in b-sheet 6 close to the Fe 21 /2-oxoglutarate center. Conservative exchanges (i.e. V200I) are commonly irrelevant for enzyme function, but the exchanges V116I and L215V were examined further because of their proximity to the sites of I115T and K216R substitutions. Homology models based on the structure of the ANS-quercetin complex suggested some additional impact of the F320Y and R326K exchanges on enzyme activity (data not shown), but the low reliability of C-terminal modeling and the results obtained with Pet_criChim I and II excluded these residues from further investigation.

Site-Directed Mutagenesis
Each of the amino acids Met-106, Ile-115, Val-116, Ile-131, Asp-195, Leu-215, and Lys-216 conserved in FHTs was independently replaced by the corresponding amino acid found in the parsley FNS I sequence. Single and double mutants (M2-8; Table I) were constructed from Pet_criFHT-pYES2.1 and used as templates to generate multiple mutants (M9-15; Table I). For all single and double mutants (M2-8; Table I), the expression in yeast cells resulted in extracts with FHT activity and without any significant FNS I activity. Consequently, the substitution of one or two amino acids is insufficient to shift the naringenin 3b-hydroxylase activity toward flavone (apigenin) formation. Two of the triple mutants harboring M106T-I131F-D195E or I131F-L215V-K216R substitutions (M9 and M10) showed reduced FHT activity in comparison to the wild-type parsley FHT concomitant with the formation of a second product that was distinguished by TLC (Fig.  6A). This product was identified as apigenin by cochromatography with a reference sample in three solvent systems (Martens et al., 2001). However, the triple mutant D195E-L215V-K216R (M11; Fig. 6A) did not Figure 5. Alignment of Ara_thaANS and partial FNS I and FHT sequences. Similar amino acids are indicated by dots, identical amino acids by stars. Colored stars mark conserved amino acids necessary for cofactor or substrate binding. Conserved substitutions of FNS I and FHT are shown in pink (positioned in the active site) and green (peripheral position). The BamHI restriction site used for the construction of Pet_criChimI and II is indicated by an arrow. Secondary structure elements of ANS are underlined; a-helices are labeled in black, 3 10 -helices in bold gray, b-sheets of the jelly roll motif in bold black (Wilmouth et al., 2002).
gain FNS I activity, which emphasizes the essential role of Phe-131 for FNS I activity, although this substitution on its own (M3; Table I) was inefficient. Four or five substitutions, including I131F, were introduced in mutants 12 to 14 (Fig. 6A), which predominantly exhibited FHT activity with FNS I side activity. Finally, the full set of seven mutations inferred from homology modeling was introduced in FHT (M15; Fig. 6A). This recombinant mutant enzyme showed primarily FNS I activity, albeit at a reduced level as compared to wildtype FNS I, with very little residual FHT activity detected after two-dimensional TLC in two solvent systems (Fig. 6B).
It is thus obvious that amino acid residues in the N terminus or beyond residue 305 also contribute to FNS I activity. Swapping of the C-terminal domain of parsley FNS I by that of FHT had introduced some FHT activity, suggesting FHT-relevant residues in that region. Alignments of published FNS I and FHT sequences identified Asp-331 as strictly conserved in FHTs from Apiaceae and other plants, which is replaced by His in FNS I. The D331H substitution in Pet_criFHT (M16; Table I) and Pet_criChimI (M17) confirmed the importance of this residue for FHT activity because recombinant M15 displayed no enzymatic activity (data not shown) and M17 retained FNS I activity, but lost residual FHT activity (Fig. 6B).

DISCUSSION
Flavonoids are abundant plant secondary metabolites that have been reported even from primitive taxa, such as liverworts (Conocephalum conicum; Feld et al.,

2003) and horsetails (Equisetum arvense; Oh et al., 2004).
Their classification is based on the flavane skeleton and comprises a spectrum of compounds with flavones, flavonols, and anthocyanins as major components. The principles of flavonoid biosynthesis have been thoroughly studied regarding biochemistry, genetics, and molecular biology, but there is little information concerning the evolution of committed enzymes (Harborne and Williams, 2000). Sessile plants have to cope with multiple environmental changes that act as a driving force for the adaptation and evolution of enzymes. This process is believed to follow basically one of two routes. An existing structural gene may change and gain the capability of encoding an enzyme with broader substrate/product specificity or multifunctionality. Alternatively, gene duplication can lead to cumulative mutations in one of the copies due to relaxed functional constraints and often those copies are eliminated later, after pseudogenization. In some instances, however, mutated copies might be retained provided that the expression is of particular benefit, such as dosage effects, subfunctionalization, or the creation of a completely new function (Hughes, 1994;Lynch and Force, 2000;Ober, 2005).
Both concepts received support from investigations on the enzymology of secondary metabolites. For example, multifunctional 2-ODDs or terpene synthases are able to catalyze more than one step of a given biosynthetic pathway (Steele et al., 1998;Prescott, 2000). FLS from C. unshiu or ANS from Arabidopsis and G. hybrida exhibit several activities in vitro, although the significance in vivo remained uncertain Martens et al., 2003;Turnbull et al., 2004). On the other hand, duplications have been Table I. Pet_criFHTs and preferential product formation Standard 2-ODD activity assays were carried out in duplicate with 5,000 dpm 14 C-labeled naringenin (approximately 45 pmol) and 500 or 1,000 mg total protein as described previously . The substrate specificity of wild-type and mutant FHTs is compared by the ratio of dihydrokaempferol (FHT activity) to apigenin (FNS I activity) formation, which together represent the total product in each assay. Mutant enzymes showing both enzymatic activities were verified by expression of another clone carrying the same mutation and repeated activity assays. Asterisk (*) indicates wild-type FNS produced exclusively apigenin from (2S)-naringenin. documented for 2-ODDs of glucosinolate biosynthesis in Arabidopsis and various enzymes of flavonoid biosynthesis, such as chalcone synthase from G. hybrida and Ipomoea, chalcone isomerase and dihydroflavonol 4-reductase from Lotus, and dihydroflavonol 4-reductase from Ipomoea (Helariutta et al., 1996;Hoshino et al., 2001;Kliebenstein et al., 2001;Shimada et al., 2003Shimada et al., , 2005. The phenomenon of gene duplication is not restricted to secondary metabolism because genes of primary metabolism have also been recruited (i.e. deoxyhypusin synthase for the evolution of homospermidine synthase catalyzing the first committed step in pyrrolizidine alkaloid biosynthesis; Ober and Hartmann, 1999). Also, the evolution of FNS I from FHT by gene duplication was suggested, but the prime ancestor gene remains to be identified Gebhardt et al., 2005). Due to sequence similarity and substrate specificity, flavonoid dioxygenases were grouped to into 2-ODDs with low substrate specificity, which attack the a-face of the substrate, such as ANS and FLS, and 2-ODDs with high substrate specificity like FHT and FNS I, which attack the b-face Martens et al., 2003;Turnbull et al., 2004). Both FHT and FNS I withdraw the b-configured hydrogen from carbon-3 of naringenin, but then proceed on different routes despite their high sequence similarity (Fig. 7). FHT catalyzes 3b-hydroxylation through a rebound process, whereas FNS I affords the synelimination of hydrogen from carbon-2 in a cage-like setting without intermediate hydroxylation (Fig. 7B). The proposed FNS I reaction clearly differs from the mechanisms assumed for FLS or ANS, which likely Figure 6. Enzyme assays with crude protein extracts from yeast transformed with Pet_criFNS I, Pet_criFHT, or Pet_criFHT mutants were carried out as described previously . Substrate and product positions vary in the presentation because the assays were separated on different thin-layer plates. However, Flavonoids were unequivocally identified through cochromatography with authentic standards. 1, Apigenin; 2, naringenin; 3, dihydrokaempferol. A, Radio scan of one-dimensional TLC separation in 30% acetic acid. B, Radio scan of two-dimensional TLC separation in chloroform:acetic acid:water (50:45:5; CAW) and 30% acetic acid. hydroxylate carbon-3 or -2 of the substrate followed by antiperiplanar water elimination as indicated by small amounts of dihydrokaempferol and kaempferol byproducts (Welford et al., 2001), but is compatible with the previous finding that FNS I neither converts 2-hydroxynaringenin nor dihydroflavonols to flavones (Britsch, 1990;Martens et al., 2003). Indirect experimental support for the syn-elimination mechanism was provided recently by incubation of ANS with naringenin diastereomers (Welford et al., 2005) because the selectivity of ANS for substrates with a particular carbon-2 stereochemistry is greatly diminished in the absence of a carbon-3 hydroxy group. Mostly, dihydrokaempferol and kaempferol, besides traces of apigenin, were formed from (2S)-naringenin, whereas almost equivalent amounts of dihydrokaempferol and apigenin with little kaempferol resulted from unnatural (2R)-naringenin, which exposes the 2a-and 3a-configured hydrogens to the catalytic ferryl species in ANS (Welford et al., 2005; Fig. 8A). The crystal complex revealed that the 3a-hydrogen is closer to the ferryl species and presumably attacked first to release apigenin by syn-elimination (Welford et al., 2005). Overall, the precision of naringenin fixation with respect to the ferryl species in the active-site pocket of ANS determines whether syn-elimination is preferred over hydroxylation. These findings can be extrapolated to the FNS I and FHT reactions. Both enzymes exclusively accept flavanone substrates exposing a b-face hydrogen Martens et al., 2003;Turnbull et al., 2004). Following the argument for apigenin formation from (2R)-naringenin by ANS-catalyzed syn-elimination (Welford et al., 2005), the stereoconfiguration at carbon-2 of (2R)-naringenin interferes with FNS I catalysis and, in fact, FNS I does not convert (2R)-naringenin (Britsch, 1990). Thus, FNS I and FHT conceivably approach the common substrate (2S)-naringenin from the opposite site of the ring plane (Fig. 8B) as compared to ANS (Fig. 8A), which requires a mirror-image orientation of substrate and active-site residues. Two combinations of I131F with M106T/D195E (M8) or L215V/K216R (M9) have been shown to confer FNS I side activity; hence, these substitutions likely influence product specificity by adjusting the substrate position rather than actively participating in the reaction mechanism (e.g. through Figure 7. Reaction mechanism of FHT (A) and FNSI (B). Depending on the substrate orientation, synelimination and formation of flavones may proceed via a radical mechanism with initial attack at either carbon-3 or carbon-2 or both simultaneously in a concerted mechanism.
acidic or basic amino acids). Necessarily, carbon-3 should be closer to the ferryl species than carbon-2 for hydroxylation by FHT, whereas the protons at carbon-2 and carbon-3 may be equally distant from the ferryl species in FNS I. Overall, the conserved differences in FNS I appear to fit the substrate into the active-site pocket with maximal proximity of H-2 and bH-3 to the catalytic ferryl species. The essential voluminous Phe-131 as compared to Ile-131 in FHT supports the assumption.
FHT and FNS I are phylogenetically closely related and adopt a tertiary structure similar to ANS (Martens et al., 2001Gebhardt et al., 2005). Although the detailed effects of selective amino acid substitutions on the overall parsley FHT structure are unknown, this article assigns those residues proximal to the active site that control FHT and FNS I activity. Obviously, minor mutations of parsley FHT are sufficient to shift the activity and significant FNS I activity was conferred already by a triple mutation (M9 and M10), which still retained the capacity for dihydrokaempferol formation (FHT activity). Concomitantly, however, severe loss in specific enzyme activity was observed. Replacement of seven amino acids caused a nearly complete change toward FNS I activity. The process accomplished here by site-directed mutagenesis defines the minimal conditions for directed evolution in vivo to broaden the flavonoid spectrum. Plants following this route, like Apiaceae, might have gained the capacity of flavone accumulation without losing their flavonols, provided that gene duplication had occurred. It is likely that the efficiency and selectivity of the newly formed FNS I have improved with time through additional mutations and concomitant with the complete loss of FHT activity in the gene copy. The ease of change of function by only three mutations of FHT suggests that flavone biosynthesis may have evolved independently on this route more than once; however, other enzymes exhibiting FNS I activity have not been observed outside Apiaceae. The accumulation of flavones conceivably provides an advantage to the plant because expression of FNS I has been maintained in Apiaceae and other plants developed FNS II for the same purpose.
The capacity to form flavonoids is supposed to have developed gradually because the first flavonoid enzymes were probably not as effective or selective as today and the initially low flavonoid concentrations likely served in plant signaling rather than UV protection or defense (Stafford, 1991). In any case, the conservation of flavone biosynthesis indicates an advantage that eventually led to flavone concentrations sufficient also for UV protection of the plant, and the environmental impact likely furthered the evolution of FNS I in Apiaceae (Logemann et al., 2000;Solovchenko and Schmitz-Eiberger, 2003). Early ontogenetic expression of FNS I in parsley and the gradual replacement of flavonols through flavones in more advanced members of Apiaceae (Harborne, 1971;Gebhardt et al., 2005) support this assumption. Advantage of functional FNSs for plant families and the identification of FNS II in non-Apiaceae is a clear indication of convergent evolution. It remains to be established whether FNS I has also evolved in non-Apiaceae, taking into account that few mutations are sufficient to confer this activity on a FHT. The search for non-Apiaceae FNS I is an interesting challenge.
Whereas flavonols and flavones have been isolated from spermatophytic and primitive plants, anthocyanidins are confined to the more advanced gymnosperms and angiosperms. This seems to suggest that FHT and FNS I developed early followed much later by FLS and Figure 8. A, Positioning of naringenin in the active site of ANS (Welford et al., 2005). B, Proposed positioning of naringenin in the active site of FNS I. ANS (Prescott and John, 1996). However, Prescott and John (1996) also excluded FHT as a direct progenitor of ANS because of low sequence similarity and divergent gene structure. There is no experimental evidence so far for an early evolution of FNS I because confinement to Apiaceae and high sequence similarity with FHT suggest a fairly recent duplication event. It appears more likely that flavonoid 2-ODDs developed from a common multifunctional ancestor gene because FLS and ANS show low substrate specificity, which could be attributed to either incomplete or spreading evolution. The low stringency might be explained by channeling substrates in multienzyme complexes (Winkel-Shirley, 2001), releasing the evolutionary pressure for enzymes of narrow substrate specificity. Under these premises, the physiological function of ANS must be reevaluated because in vitro ANS predominantly converted leucoanthocyanidin to dihydroquercetin, (2S)naringenin to dihydrokaempferol, and dihydroquercetin to quercetin (Welford et al., 2001). The lack of anthocyanidins in less advanced plants could thus be a consequence of poor complex formation rather than lack of ANS. The capability of catalyzing several steps in the flavonoid pathway might furthermore qualify ANS also as a progenitor candidate of other flavonoid 2-ODDs and it is essential to determine the evolutionary distance of the various flavonoid 2-ODDs. The functional flexibility of these 2-ODDs by very few mutations was demonstrated in this study and highlights the evolutionary importance of 2-ODDs for the introduction of new enzymatic functions.

Yeast Strains and Growth Conditions
Yeast (Saccharomyces cerevisiae) INV Sc1 (Invitrogen) was used for standard cloning and overexpression. Medium and growth conditions are described elsewhere . All plasmids for chimeric FHT/FNS I and mutagenesis were constructed from pYES2.1 that contains the parsley (Petroselinum crispum) FHT (Pet_criFHT) or FNS I (Pet_criFNSI; Martens et al., 2001Martens et al., , 2003.

Molecular Techniques and Cloning of 2-ODDs
Plasmid DNA was isolated as described in Engebrecht et al. (2001). For sequencing, DNA was further purified by being passed through a NucleoSpin column according to the manufacturer's instructions (Machery-Nagel). Restriction digestions were performed as described by the enzyme suppliers (MBI Fermentas). Yeast transformations were performed according to Easy-Comp protocol (Invitrogen). Ligation reactions and agarose gel electrophoresis were performed by standard procedures (Sambrook and Russel, 2001).
Further putative FHT and FNS I cDNA clones were isolated and functionally verified as described in Gebhardt et al. (2005).

Chimeric Gene Construction
Highly conserved regions of the FHT and FNS I gene were identified by multiple sequence alignment of a number of 2-ODDs (Gebhardt et al., 2005). Chimeric constructs (Pet_criChim I and II) are based on functionally verified Pet_criFNSI and FHT pYES2.1 clones. BamHI is conserved between Thr-219 and Asp-220 in both sequences and was used in combination with XbaI (located 3# of the insert in the multiple cloning site of pYES2.1) to digest the cDNA clones, resulting in a long (pYES2.1 and N-terminal part of the insert) and a short (C-terminal part of the insert) fragment. All fragments were gel purified after restriction digestion, isolated from the gel via the NucleoSpin extract kit, and added to the ligation reaction. The long fragment of the FNS I digest was combined with the short fragment of the FHT digest resulting in Pet_criChim I, and vice versa resulting in Pet_criChim II as illustrated in Supplemental Figure S1. Full-length sequencing of the inserts confirmed successful ligation of the fragments and intact open reading frames for both chimerics.

Site-Directed Mutagenesis of Pet_criFHT
Single and multiple amino acids substitutions were generated by sitedirected mutation of Pet_criFHT or its previously mutated variants in pYES2.1 by Stratagene QuickChange and QuickChange Multi system according to the manufacturer's instructions (Stratagene). All primers and templates for mutagenesis are listed in Supplemental Table S2.
In detail, the plasmid Pet_criFHT/pYES2.1  was used as a primary template together with a respective mutation primer (Supplemental Table S2). For additional specific mutations, the confirmed mutants of previous rounds were used as indicated in Table I. Some mutagenic oligonucleotides were designed with a silent mutation to generate a new or destroy a restriction site (Supplemental Table S2) to facilitate the identification of clones carrying the desired mutation. All mutant FHTs were sequenced in full length to ensure the correct residue was changed and to confirm that no other unintended mutation was introduced (MWG-Biotech).

Expression of Cloned Genes and Analysis of Catalytic Properties
Enzymatic activities of the wild-type, chimeric, and mutated proteins were determined as previously described by heterologous expression in yeast (Gebhardt et al., 2005) with 500 and 1,000 mg total protein, respectively, as double tests. Protein concentrations were determined according to Bradford (1976) with bovine serum albumin as a standard.

Sequence Comparison
Related sequences were initially detected by BLAST and PSI-BLAST (Altschul et al., 1997) analysis. Multiple sequence alignments were generated with the ClustalW algorithm (Thompson et al., 1994).

Homology Modeling
To identify putative amino acids responsible for the different catalytic behavior of FNS I and FHT, homology modeling was performed with respect to parsley FNS I and FHT. Homology models were generated using the Webbased SWISS-MODEL server (Schwede et al., 2003). The crystal structure of ANS in complex with Fe 21 , 2-oxoglutarate, and naringenin (Protein Data Bank entry 2brt) was kindly provided in advance by the authors (Welford et al., 2005) and served as template structure. Even though the homology models obtained were truncated with respect to the N as well as the C terminus, these models clearly represented the substrate-binding pocket and, accordingly, allowed for the structural location of sequence mismatches within the binding cavities. Because SWISS-MODEL recognizes only protein atoms, naringenin, 2-oxoglutarate, and the iron ion were added to the models at similar positions as observed in the template structure. However, this step has to be regarded with some care because the binding mode of naringenin to ANS is certainly different from FNS I and FHT. Thus, insertion of the substrate was performed to evaluate putative interaction sites between naringenin and its binding pockets. Amino acid residues of the two models involved in iron and cosubstrate binding were manually adjusted using the program O (Jones et al., 1991). Figures were created by means of Pymol (DeLano, 2002).

Supplemental Data
The following materials are available in the online version of this article.