ARTICLE Received 7 Jan 2016 | Accepted 11 Aug 2016 | Published 5 Oct 2016 DOI: 10.1038/ncomms12942 OPEN Elucidation of the biosynthesis of carnosic acid and its reconstitution in yeast Ulschan Scheler1, Wolfgang Brandt2, Andrea Porzel2, Kathleen Rothe1, David Manzano3,4, Dragana Božić5,w, Dimitra Papaefthimiou5, Gerd Ulrich Balcke1, Anja Henning1, Swanhild Lohse1, Sylvestre Marillonnet1, Angelos K. Kanellis5, Albert Ferrer3,4 & Alain Tissier1 Rosemary extracts containing the phenolic diterpenes carnosic acid and its derivative carnosol are approved food additives used in an increasingly wide range of products to enhance shelf-life, thanks to their high anti-oxidant activity. We describe here the elucidation of the complete biosynthetic pathway of carnosic acid and its reconstitution in yeast cells. Cytochrome P450 oxygenases (CYP76AH22-24) from Rosmarinus officinalis and Salvia fru- ticosa already characterized as ferruginol synthases are also able to produce 11-hydro- xyferruginol. Modelling-based mutagenesis of three amino acids in the related ferruginol synthase (CYP76AH1) from S. miltiorrhiza is sufficient to convert it to a 11-hydroxyferruginol synthase (HFS). The three sequential C20 oxidations for the conversion of 11-hydro- xyferruginol to carnosic acid are catalysed by the related CYP76AK6-8. The availability of the genes for the biosynthesis of carnosic acid opens opportunities for the metabolic engineering of phenolic diterpenes, a class of compounds with potent anti-oxidant, anti-inflammatory and anti-tumour activities. 1 Department of Cell and Metabolic Biology, Leibniz Institute of Plant Biochemistry, Weinberg 3, Halle 06120, Germany. 2 Department of Bioorganic Chemistry, Leibniz Institute of Plant Biochemistry, Weinberg 3, Halle 06120, Germany. 3 Program of Plant Metabolism and Metabolic Engineering, Centre for Research in Agricultural Genomics, Campus UAB, 08193 Bellaterra, Spain. 4 Faculty of Pharmacy, Department of Biochemistry and Molecular Biology, University of Barcelona, 08028 Barcelona, Spain. 5 Group of Biotechnology of Pharmaceutical Plants, Laboratory of Pharmacognosy, Department of Pharmaceutical Sciences, Aristotle University of Thessaloniki, 54124 Thessaloniki, Greece. w Present address: Institute for Biological Research ‘Siniša Stanković’, University of Belgrade, Bul. despota Stefana 142, 11060 Belgrade, Serbia. Correspondence and requests for materials should be addressed to A.T. (email: alain.tissier@ipb-halle.de). NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 T erpenoid-based drugs such as taxol and artemisinin have Results become indispensable in the treatment of cancer and Identification of HFSs. Using a Golden Gate modular cloning infectious diseases1,2. In addition, belonging to the large vector system developed for yeast (see Methods), we first recon- family of terpenoids are the phenolic labdane-type diterpenes stituted ferruginol biosynthesis in yeast using CYPs previously carnosic acid (CA), carnosol (CO) and pisiferic acid (PA; Fig. 1) characterized as FSs, namely CYP76AH1, CYP76AH4, from rosemary (Rosmarinus officinalis) and various sage species CYP76AH22, CYP76AH23 or CYP76AH24 (Supplementary (Salvia sp.). These compounds show diverse biological and Fig. 1 and Supplementary Table 1)14–16. Each of these P450s chemical activities, including anti-oxidative, anti-cancer, anti- was coupled with the upstream biosynthetic enzymes (CPS and inflammatory, as well as anti-microbial properties, and have MiS), the CYP reductase ATR1 (ref. 17) and a GGPP synthase been proposed as preventive agents and treatments for (GGPPS), the latter to allow for sufficient supply of GGPP. We neurodegenerative disorders3–9. However, the major use of CA call this group of enzymes the core module (CM) for the and CO is based on their strong antioxidative properties, which production of the diterpene precursors. To ensure efficient co- make them highly relevant and approved food additives in expression of all genes, we developed a library of synthetic Europe, China and Japan10. Furthermore, CA has been proposed galactose inducible promoters, which are not repressed by glucose as the precursor of a whole range of diterpenoids including the and whose members conferring the strongest expression were tanshinones from Salvia miltiorrhiza that are being actively selected (see Methods and Supplementary Table 2). As expected, investigated for their anti-cancer activities11. The first committed extracts of all obtained yeast strains (Supplementary Table 3) steps in the biosynthesis of this group of diterpenes are catalysed revealed the presence of miltiradiene, abietatriene and ferruginol by the terpene synthases copalyl diphosphate synthase (CPS) and in gas chromatography–mass spectrometry (GC–MS) a kaurene synthase-like enzyme called miltiradiene synthase measurements (Fig. 2a and Supplementary Figs 3 and 4). (MiS). CPS and MiS cyclize geranylgeranyl diphosphate (GGPP) However, expression of either one of CYP76AH4, CYP76AH22, to copalyl diphosphate and copalyl diphosphate to miltiradiene, CYP76AH23 or CYP76AH24 resulted in the production of a new respectively12 (Fig. 1). The oxidation of miltiradiene to compound, which could only be found in traces when expressing abietatriene occurs spontaneously, but can be accelerated by CYP76AH1 (Supplementary Figs 3–5). This product was also exposure to ultraviolet irradiation13,14. One key intermediate in present in small quantities in leaf surface extracts of R. officinalis the downstream pathway is ferruginol. Several cytochrome P450 and S. fruticosa. The product was purified from yeast after monooxygenases (CYP) of the CYP76 clan from S. miltiorrhiza, shake-flask expression of CYP76AH22 with the CM (Table 1) and S. fruticosa and R. officinalis, which are able to convert nuclear magnetic resonance (NMR) analysis established abietatriene to ferruginol, have been characterized14–16. These its identity as 11-hydroxyferruginol (Supplementary Table 4)18. enzymes are collectively called ferruginol synthases (FSs). In vitro Thus, CYP76AH4, CYP76AH22, CYP76AH23 and CYP76AH24, characterization of one of these enzymes (CYP76AH4 from which have been reported to be FSs, are in fact able to carry out a R. officinalis) indicated that abietatriene rather than miltiradiene second hydroxylation and are therefore HFSs. The activity of FS is the substrate14. We show here that the previously identified and HFSs was confirmed by in vitro assays with microsomal FS from S. fruticosa and R. officinalis carry out not a single fractions from yeast strains expressing CYP76AH1 and but two successive oxidations leading to the next intermediate CYP76AH22, respectively (Supplementary Figs 6 and 7). 11-hydroxyferruginol and were renamed 11-hydroxyferruginol During NMR analysis, we noticed increasing signals that did synthases (HFSs). Taking advantage of the similarity between not belong to 11-hydroxyferruginol. To check whether these these HFSs and the FS, we show that the exchange of three amino signals are due to instability of 11-hydroxyferruginol, the sample acid residues in the FS from S. miltiorrhiza by the corresponding was stored at  20 °C for 24 h and analysed again. Based on these residues in HFS is sufficient to convert the FS to a HFS. We also specific signals, this additional compound was identified as identify a group of related CYPs from R. officinalis and hydroxyferruginol quinone (Supplementary Table 4 and Supple- S. fruticosa with C20-oxidase (C20Ox) activity whose mentary Fig. 8). Consequently, 11-hydroxyferruginol sponta- co-expression in yeast with the diterpene synthases and neously auto-oxidized to the corresponding quinone18. This HFS leads to the production of CA, itself spontaneously was further supported by the simultaneous detection of oxidizing to CO. 11-hydroxyferruginol and hydroxyferruginol quinone in all OH OH HO Ferruginol CPS Spontaneous synthase GGPP MiS Miltiradiene Abietatriene Ferruginol 11-Hydroxyferruginol OH OH OH HO HO O HOOC HOOC O Pisiferic acid Carnosol Carnosic acid Figure 1 | Biosynthesis of PDs in Rosmarinus officinalis and sage species. State of the knowledge on the biosynthetic pathway from GGPP to CA, CO and PA before this study. The terpene synthases CPS and MiS cyclize GGPP to miltiradiene, which spontaneously oxidizes to abietatriene. Ferruginol synthases oxidize abietatriene to ferruginol. A further intermediate is presumed to be 11-hydroxyferruginol. Oxidations at the C20 position should lead to pisiferic acid (PA) from ferruginol and to carnosic acid (CA) from 11-hydroxyferruginol. Further oxidation of CA at the C7 position results in carnosol. 2 NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 ARTICLE yeast strains expressing the various HFSs (Supplementary Fig. 3). Triple-mutants of FS recapitulate HFS activity. To analyse the Such redox conversions of the catechol group to a quinone have functional differences between FS and HFS, we applied homology been shown for CA and its derivatives and are thought to be the modelling to CYP76AH1, CYP76AH4, CYP76AH22, main reason for their high antioxidant activity4,19–23. CYP76AH23 and CYP76AH24. The high degree of sequence a b 4 Relative abundance 189 21 4 6 5 175 69 5 191 271 205 Rosemary Yeast 286 149 201 115 217 302 Relative abundance 229 69 Relative abundance 287 HFS 115 165 100 150 200 250 300 350 50 231 m/z 6 Relative abundance 204 FS 115 55 128 229 91 141 69 189 300 285 Rosemary Control 100 150 200 250 300 350 50 24 25 26 27 28 29 30 m/z 100 150 200 250 300 350 50 Retention time (min) m/z c d 300 1. RoCYP76AH22 N303 2. SmCYP76AH1 320 330 1. RoCYP76AH22 E301 2. SmCYP76AH1 350 360 S303 D301 1. RoCYP76AH22 2. SmCYP76AH1 380 390 1. RoCYP76AH22 V479 2. SmCYP76AH1 F478 410 420 1. RoCYP76AH22 2. SmCYP76AH1 440 450 1. RoCYP76AH22 2. SmCYP76AH1 470 480 1. RoCYP76AH22 2. SmCYP76AH1 e f 4 5 5 4 CYP76AH22 CYP76AH22E301D,S303N,F478V Relative abundance CYP76AH1 CYP76AH1D301E,N303S,V479F Control 27 28 29 30 17 18 Retention time (min) Retention time (min) NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 identity between these proteins (on average 80%) resulted in in vivo co-expression assays with the CM (Supplementary Figs 13 predicted tertiary structures that were almost identical (see and 14). Mutations at position 117 and 118 in HFS prevented the Supplementary Fig. 9 as an example). These models allowed us to oxidation of ferruginol, but the reciprocal mutations in FS did not identify the amino acid residues that are likely to be involved in support the oxidation of ferruginol, indicating that these residues the binding of the substrates abietatriene or ferruginol. In all are not directly required for the oxidation of ferruginol. In con- enzymes the docking arrangement of abietatriene supports the trast, the double mutant FSD301E,N303S could produce hydro- oxidation reaction leading to ferruginol due to the short distance xyferruginol, but in smaller proportions compared with HFS. between the reactive oxygen atom bound to the haem and the Furthermore, a triple mutant FSD301E,N303S,V479F had a profile proton at C12 required for activation (Supplementary Fig. 10). indistinguishable from HFS (Fig. 2e,f), which was also confirmed Comparative analysis of the three-dimensional models led to the by in vitro enzyme assays (Supplementary Figs 6 and 7). The identification of three amino acids in the active site of HFS (E301, relevance of these three amino acids in the oxidation of ferruginol S303 and F478), which are different in FS (D301, N303 and is confirmed by the analysis of three-dimensional models with V479), and might be associated with the oxidation of ferruginol ferruginol docked into the active site (Fig. 2d and see (Fig. 2c,d and Supplementary Fig. 10). In addition, two amino Supplementary Fig. 10 for details). In the HFSs (CYP76AH4, acids of FS (A117 and S118) are located in a a-helix, whereas in CYP76AH22, CYP76AH23 and CYP76AH24), F478 interacts HFS the corresponding Gly–Gly sequence adopts a coil con- with ferruginol, allowing its correct positioning for further formation. Although both residues are rather far away from the oxidation. In contrast, in CYP76AH1, the hydrophobic interac- active site, the small conformational changes may also potentially tion between the substrate and the smaller side chain of V479 at influence the activity of the enzyme (Supplementary Figs 11 and the equivalent position combined with the repulsive interaction of 12). All reciprocal single mutations as well as several double, the carboxylic acid moiety of D301 with the isopropyl group of triple and quadruple mutant combinations were tested in yeast ferruginol, causes the adoption of a slightly different docking Table 1 | Quantification by NMR of products formed in engineered yeast strains. Expressed enzymes Product Concentration Concentration Ratio product/ Ratio product/ (lmol l  1) (mg l  1) miltiradiene ferruginol Empty vector control ND   GGPPS:CPS:MiS Miltiradiene 5.86±1.52 1.59±0.41 1 GGPPS:CPS:MiS:ATR1: Miltiradiene 0.93±0.18 0.25±0.05 1 0.15 CYP76AH1 Ferruginol 6.37±0.55 1.82±0.16 6.85 1 GGPPS:CPS:MiS:ATR1: Miltiradiene 0.86±0.22 0.23±0.06 1 3.74 CYP76AH22 Ferruginol 0.23±0.05 0.07±0.01 0.27 1 11-hydroxyferruginol 4.18±0.48 1.26±0.14 4.86 18.17 GGPPS:CPS:MiS:ATR1: Miltiradiene 1.15±0.11 0.31±0.03 1 2.80 CYP76AH1D301E,N303S,V479F Ferruginol 0.41±0.09 0.12±0.03 0.35 1 11-hydroxyferruginol 7.43±1.20 2.25±0.36 6.46 18.12 GGPPS:CPS:MiS:ATR1: Miltiradiene 3.72±0.25 1.01±0.07 1 0.558 CYP76AH22E301D,S303N,F478V Ferruginol 6.78±0.20 1.94±0.06 1.82 1 GGPPS:CPS:MiS:ATR1: Miltiradiene 0.73±0.14 0.20±0.04 1 1.87 CYP76AH22:CYP76AK8 Ferruginol 0.39±0.09 0.11±0.03 0.53 1 Carnosic acid 8.26±1.64 2.74±0.55 11.31 21.18 GGPPS:CPS:MiS:ATR1: Miltiradiene 2.41±0.25 0.66±0.07 1 3.60 CYP76AH1:CYP76AK8 Ferruginol 0.67±0.10 0.19±0.03 0.28 1 Pisiferal 3.03±0.84 0.91±0.25 1.25 4.52 Pisiferic acid 1.82±0.75 0.57±0.24 0.76 2.72 CPS, copalyl diphosphate synthase; GGPPS, geranylgeranyl diphosphate synthase; MS, miltiradiene synthase; ND, not detected. The concentrations of products, which could be detected with a signal sufficient for quantification, are given with their s.d. (n ¼ 3). The concentrations are normalized to the cell density as measured by OD600. The ratios product/miltiradiene or product/ferruginol were calculated with the molar concentrations. Figure 2 | Homology modelling-based mutagenesis of FS and HFS. (a) Part of the GC–MS profile of hexane extracts from rosemary leaf surfaces and from yeast strains co-expressing GGPPS, CPS, MS, ATR1 and indicated CYPs (selected m/z signals: 270, 272, 286, 300 and 302). Miltiradiene (1), abietatriene (2), ferruginol (4), 11-hydroxyferruginol (5) and hydroxyferruginol quinone (6). (b) Electron impact mass spectra of ferruginol (4), 11-hydroxyferruginol (5), which was extracted from rosemary or yeast cultures, and of hydroxyferruginol quinone (6). (c) Excerpts of the aligned amino acid sequences of CYP76AH1 and CYP76AH22 with the residues that were mutagenized indicated by red rectangles. (d) Models of the active site of CYP76AH22 (green) and CYP76AH1 (magenta) with bound ferruginol and haem (orange carbon atoms). The models were generated using 2HI4 (structure of human CYP1A2) as a template. Only in the case of CYP76AH22 can the hydrogen atom at C11 be abstracted by the reactive oxygen atom, thanks to the short distance of 3.1 Å (green dashed line). In contrast, the corresponding distance in CYP76AH1 is 4.6 Å (magenta dashed line), which is too large to support oxidation at this position. (e) GC–MS profile (selected m/z signals: 286, 300 and 302) and (f) LC–MS profile (selected m/z signals: 285.221 and 301.217) of yeast strains expressing GGPPS, CPS, MS, ATR1 and indicated CYP enzymes (wild type or mutagenized). The region of the chromatograms with signals for ferruginol (4) and 11-hydroxyferruginol (5) is shown. 4 NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 ARTICLE arrangement. In this configuration, the hydrogen atom at C11 of Supplementary Fig. 16). This compound was purified from ferruginol is located 4.6 Å from the reactive oxygen, a distance too rosemary plants, analysed by NMR and identified as carnosalde- large to support oxidation (Supplementary Fig. 10c). Strikingly, hyde (Supplementary Table 6). As CYP enzymes are known to the carboxylic moiety of E301 in HFS is involved in a salt bridge carry out multiple oxidations on terpenes, for example, in the with R481 and is thus preventing the repulsive interaction biosynthesis of abietic acid (CYP720B1) or artemisinic acid with the isopropyl group of ferruginol. On the contrary, this (CYP71AV1)26,27, the possibility that the C20Ox also convert this positioning of E301 favours hydrophobic interactions via the aldehyde to a carboxylic acid was considered. However, ethylene moiety of its side chain with the isopropyl group carboxylic acids, such as CA are not detectable by GC–MS. of ferruginol. Both exchanges combined, V479/F478 and Liquid chromatography–MS (LC–MS) analysis demonstrated that D301E, cause a slightly different docking orientation of CA was indeed formed in our strains (Fig. 3d,e and ferruginol in HFS with a short distance (3.1 Å) of the substrate Supplementary Figs 4 and 17) and was even converted to CO, C11 to the reactive oxygen atom bound to the heme, thereby which is in agreement with the literature19,21,23,28. Although CA allowing the catalytic oxidation. These two exchanges (D301E has been reported to be spontaneously degraded to CA quinone and V479/F478) are the most important ones, whereas from the or to epirosmanol, rosmanol and 7-methyl-epirosmanol upon model only very minor structural alterations of the N303S stress conditions29,30, none of these compounds were detected in substitution to the docking poses and therefore to the related our experiments. In coupled in vitro assays with microsome catalysis can be observed. fractions from strains expressing CYP76AH22 and CYP76AK8, Altogether, our results point to the essential contribution of complete conversion of 11-hydroxyferruginol to CA could be amino acid residues at positions 301, 303 and 478 for the observed (Supplementary Figs 6 and 7). In addition, as in yeast 11-hydroxylation of ferruginol in HFS and, remarkably, the engineering experiments, CO could also be detected, but not the reciprocal exchange of the corresponding residues in FS other oxidation products normally associated with non-enzymatic recapitulated the HFS activity. Interestingly, CYP76AH3, an oxidation of CA. Control incubations of CA with single CYP or enzyme from S. miltiorrhiza that was recently shown to with an empty vector control did not result in increased hydroxylate ferruginol at C11, has the same residues at these formation of CO, further supporting that this conversion is three positions as the HFS identified here24. non-enzymatic (Supplementary Fig. 7). To investigate whether C20Ox can, in addition to 11-hydroxyferruginol and miltiradiene, also oxidize ferruginol, C20 oxidation by CYP76AK6-8 leads to CA. To identify the the newly identified enzymes were co-expressed with CPS, MS missing steps in the pathway to CA, we searched for genes with a and FS (CYP76AH1). GC–MS analysis and comparative analysis similar expression pattern to those of the upstream steps, namely with published data31 revealed the presence of pisiferal in low CPS and MiS, which are preferentially expressed in trichomes on amounts, indicating that indeed ferruginol is also a substrate of young leaves12,16. We focused on CYP-encoding genes, as those the C20Ox (Fig. 3b,c and Supplementary Figs 4 and 18). As for are frequently involved in the oxidation of terpenes25. Real-time CA, PA could be found by LC–MS measurements (Fig. 3d,e). As PCR analysis of candidate genes identified from RNA-sequencing CYP76AH1 is able to produce traces of 11-hydroxyferruginol, data of R. officinalis and S. fruticosa was performed with RNA CA also co-accumulated with PA, albeit at much lower levels including isolated trichomes and leaves free from trichomes, as than with HFS (Supplementary Fig. 19), whereas CO and well as young and old leaves from R. officinalis and S. fruticosa. carnosaldehyde were not detected. Interestingly, co-expression CYP genes from R. officinalis (RoCYP76AK7 and RoCYP76AK8) of HFS and C20Ox did not result in the formation of pisiferal and and S. fruticosa (SfCYP76AK6) encoding highly similar proteins PA, although ferruginol and 11-hydroxyferruginol are still (Supplementary Fig. 2) showed increased transcript levels in detectable. This indicates that C20Ox preferentially uses young leaves compared with old leaves and were expressed 11-hydroxyferruginol as substrate. specifically in trichomes (Fig. 3a). Interestingly, those CYPs showed high sequence similarity to the FS and HFSs. As the precise order of the oxidations in the CA pathway was In vivo quantification of pathway products by NMR spectroscopy. not known, we first tested these CYPs by transient co-expression To evaluate the respective activities of the enzyme in vivo, we with CPS and MiS in Nicotiana benthamiana. Surface extracts of performed absolute quantification by NMR of the intermediates leaves expressing CPS, MiS, ATR1 and CYP76AK7 analysed by and products of the pathway from yeast strains expressing GC–MS contained a new product with a molecular mass of 286 different enzyme combinations (Table 1 and Supplementary and a base peak of 257, suggesting the presence of a diterpene Figs 20 and 21). NMR was chosen as a quantification method with an aldehyde group (Fig. 3b,c and Supplementary Fig. 15a). because of the lack of appropriate standards for 11-hydro- Co-expression of GGPPS, CPS, MiS, ATR1 and CYP76AK6, xyferruginol, the different responses of GC–MS and LC–MS to CYP76AK7 or CYP76AK8 in yeast led to the formation of the ferruginol and 11-hydroxyferruginol, and the possibility to same compound (Supplementary Fig. 15b). After purification and quantify compounds even without a standard (Methods). A strain NMR analysis, this product was identified as miltiradien-20-al expressing only the diterpene synthases produces miltiradiene at (Supplementary Table 5), thus providing first evidence that 5.86±1.52 mmol l  1. When the CYP enzymes are co-expressed CYP76AK6, CYP76AK7 and CYP76AK8 can oxidize miltiradiene with the diterpene synthases, comparable or even slightly larger at the C20 position and are thus named C20Ox (Fig. 4). Although amounts of the end products are produced (6.37±0.55 mmol l  1 miltiradien-20-ol is the expected intermediate, it could not be of ferruginol for FS, 4.18±0.48 mmol l  1 of 11-hydroxyferruginol detected, indicating its complete conversion to miltiradien-20-al. for HFS and 8.26±1.64 mmol l  1 of CA for HFS þ C20Ox). The relatively low conversion of miltiradiene to miltiradien- Whatever the CYP enzymes that are co-expressed, miltiradiene 20-al by the C20Ox enzymes suggested that ferruginol or can still be detected, albeit in significantly smaller amounts 11-hydroxyferruginol rather than miltiradiene are the preferred compared with the CPS/MS strain. The residual presence of substrates. We therefore co-expressed C20Ox together with HFS miltiradiene could be explained by the extreme hydrophobicity and the CM in yeast. GC–MS analysis of hexane extracts revealed of this compound, suggesting it could be sequestered in the a novel compound, which was also detected in leaf surface membrane fraction and thereby preventing its complete turnover extracts of R. officinalis and S. fruticosa (Fig. 3b,c and by CYP enzymes. In a strain expressing HFS (CYP76AH22) NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 residual amounts of ferruginol are present but the concentration 11-hydroxyferruginol to ferruginol of 18.12, confirming the ratio of 11-hydroxyferruginol to ferruginol is 18.17, indicating analysis by GC–MS and LC–MS. In the strain co-expressing an almost complete conversion of ferruginol to 11-hydro- the HFS and the C20Ox, neither 11-hydroxyferruginol nor any of xyferruginol. Interestingly, the strain expressing the triple mutant the intermediates of the C20 oxidation could be detected by FSD301E,N303S,V479F has very similar yields and ratios, notably NMR, indicating a quantitative conversion to CA by C20Ox. a 80 3 6 CYP76AK6 CYP76AK7 5 CYP76AK8 Normalized expression 60 2 4 40 3 1 2 20 1 0 0 0 O d le s O d le s O d le s tri ves es tri ves es tri ves es ch es tri ves ch es e O me e om om om om om av av a a a a o le le le ch ch ch ch g g g tri tri un un un l l l O O g g g ld ld ld Yo Yo Yo un un un Yo Yo Yo b c Relative abundance 3 257 21 4 6 5 78 133 161 117 Rosemary 187 243 286 100 150 200 250 300 350 50 Relative abundance HFS/C20Ox m/z 7 191 8 69 Yeast 205 Yeast 69 FS/C20Ox 217 Relative abundance Relative abundance 287 271 175189 115 55 3 231 316 55 115 201 229 300 C20Ox 27.0 27.5 Control Rosemary Rosemary 100 150 200 250 300 350 100 150 200 250 300 350 50 50 24 25 26 27 28 29 30 Retention time (min) m/z m/z d e 7 315 9 287 Yeast Yeast 287 331 Relative abundance Relative abundance 10 9 7 271 244 271 Rosemary CA standard Rosemary Standard Relative abundance 100 150 200 250 300 350 100 150 200 250 300 350 CO standard m/z m/z 285 269 HFS/C20Ox 10 11 Yeast Yeast 11 315 Relative abundance Relative abundance 329 285 PA standard 165 242 201 215 FS/C20Ox Control Standard Standard 100 150 200 250 300 350 100 150 200 250 300 350 12 13 14 15 16 17 18 19 20 Retention time (min) m/z m/z 6 NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 ARTICLE Furthermore, the concentration yield was the highest of all also lead to production of CA28. By identifying three amino acid products with 8.26±1.64 mmol l  1 of CA and a ratio of CA to residues in the FS that recapitulate HFS activity and with in vivo ferruginol of 21.18. The higher yields of CA could be explained quantification data, our results provide additional insights into both by the high activity of the C20Ox enzymes, but also by the the structure activity relationship of this family of CYP enzymes higher hydrophilicity of CA, which is more water soluble involved in the oxidation of diterpenes. than miltiradiene or ferruginol, and is therefore less likely to be In conclusion, the discovery of the missing genes for the trapped in membranes. In the combination of FS with C20Ox, biosynthesis of CA makes it possible to engineer yeast, but also which is not a natural combination, as the C20Ox comes from other organisms including plants, for the production at high rosemary and the ferruginol synthase from S. miltiorrhiza, the levels of this industrially important compound. It demonstrates situation is different as only 1.82±0.75 mmol l  1 of PA can be the power of enzymatic synthesis of high value products produced, with a ratio of 0.76 to miltiradiene. Interestingly, even compared with sophisticated chemical synthesis34. The higher concentrations of pisiferal, the aldehyde intermediate by combinatorial approach used for CA biosynthesis in yeast C20 oxidation of ferruginol, are produced (3.03±0.84 mmol l  1). constitutes a versatile and efficient platform for the elucidation This supports a higher specificity of the C20Ox for 11-hydro- and production of other related terpenoids, including the xyferruginol compared with ferruginol. The absence of the promising anti-cancer tanshinones. hydroxyl group at the C11 position further prevents the efficient conversion of the aldehyde (pisiferal) to the end product, PA. Methods A graphical overview of the results of phenolic diterpene pathway Plant material. R. officinalis was obtained from the Conservatoire National des engineering is provided in Fig. 4. Plantes à Parfum, Médicinales, Aromatiques et Industrielles (http://www.cnpmai.- net/). S. fruticosa plants were kindly provided by the group of Professor Angelos Kanellis16. N. benthamiana, R. officinalis and S. fruticosa plants were grown in the Phylogeny of CYP76AH and CYP76AK proteins. Using the greenhouse under long day conditions (16 h light/8 h darkness) with temperatures full-length amino acid sequences, a phylogenetic analysis was of 25 °C during day and 20 °C at night, with 53% humidity. Leaf material for gene applied to the investigated CYP76AH and CYP76AK enzymes isolation and chromatographic analysis was collected in the vegetative phase. (Supplementary Table 7 and Supplementary Fig. 22). The resulting maximum-likelihood tree confirmed that they belong to Chemicals. Authentic standards of CA, CO and PA were purchased from Santa the CYP76 clan whose members are often involved in the oxi- Cruz Biotechnology Inc., Sigma Aldrich and TCI Europe, respectively. dation of terpenes32. As expected, they share the highly conserved CYP motifs, that is, the I-helix (AGxDT), K-helix (KETLR), PERF Tissue collection and RNA preparation for quantitative RT–PCR. Leaves and haem-binding domain (PFGxGRRxCPG) (Supplementary from R. officinalis and S. fruticosa from Kavoussi, Crete, were collected in two developmental stages, young (1–2 cm) and aged (3–4 cm) leaves. Trichomes Figs 1 and 2)25. HFS and C20Ox cluster into subgroups, which from leaves were isolated using dry ice35. An additional brush-abrasion step form a subfamily with SmCYP76AH1, MgCYP76AH2, was used to remove any leftover trichomes from leaf tissue. The plant material MgCYP76AH9 and several CYP76 members from Coleus was homogenized and RNA isolation was performed using Spectrum Plant forskohlii, the latter involved in the biosynthesis of forskolin Total RNA Kit (Sigma-Aldrich) with additional ethanol washing steps. The complementary DNA was generated from 1 mg total RNA using Superscript and related compounds33. The HFSs (CYP76AH4, 22–24) III kit (Invitrogen) according to manufacturers’ specifications and with 1 mg constitute a clade distinct of the FS (CYP76AH1) reflecting random primers (Invitrogen). their functional differences. Interestingly, CYP76AH1 only shares 79% sequence identity with CYP76AH24, in contrast to 90–94% Quantitative real-time PCR analysis. The PCR mix was prepared using KAPA between CYP76AH22–24, although both originate from sage SYBR FAST qPCR kit and quantitative real-time PCR analysis was performed in species. Therefore, the presence of rosemary and sage sequences triplicates using the Applied Biosystems 7500 Real-Time PCR System (Invitrogen) in the HFS clade indicates an origin predating the divergence of and the following PCR conditions: 95 °C for 2 min; 6 cycles of 95 °C for 35 s, 64 °C these species. Whether HFS evolved from FS or the opposite for 30 s, 72 °C for 30 s; 35 cycles of 95 °C for 20 s, 62 °C for 20 s, 72 °C for 15 s; 72 °C for 10 min; plate read at 76 °C. To receive a melting curve from 70 °C to cannot be concluded with the current data set. The sequence 95 °C, data points were collected every 0.1 °C with a 10 s hold between them. The similarity within the C20Ox cluster ranges from 70 to 77%, data were evaluated using the delta Ct method36 with the elongation factor 4a as whereby CYP76AK8 from R. officinalis is more similar to reference gene. CYP76AK6 from S. fruticosa than to its rosemary homologue CYP76AK7. This analysis underscores the importance of the Construction of golden gate compatible yeast expression MoClo vectors. CYP76 clan in the oxidation of diterpenes, particularly in the Construction of Golden Gate compatible yeast expression vectors of level 1 and M Lamiaceae family, which is particularly rich in diverse labdanoid was carried out as for the MoClo system37,38. Their backbones were equipped with origins of replication for Escherichia coli (ColE1) and Saccharomyces cerevisiae diterpene structures. (2 mm ori) from pYES2 (Invitrogen), and in addition contained E. coli antibiotic We note that almost identical genes were identified and selection markers for replication (carbenicillin in level 1 vector and spectinomycin characterized from S. pomifera, which when expressed in yeast in level M vector). To select for insertion into the vectors, a LacZ gene flanked by Figure 3 | Functional analysis of C20Ox. (a) Expression profile of CYP76AK6-8 in young and old trichomes and leaves without trichomes. CYP76AK6 is from S. fruticosa, CYP76AK7-8 are from R. officinalis. Quantitative reverse transcriptase–PCR data were obtained from three technical replicates and normalized to the eukaryotic elongation factor 4a. (b) GC–MS analysis (selected m/z signals: 270, 272, 286, 300, 302 and 316) of extracts from rosemary leaf surfaces or from yeast strains co-expressing GGPPS, CPS, MS, ATR1, CYP76AH22 (HFS) and CYP76AK6 (C20Ox). The labelled peaks correspond to miltiradiene (1), abietatriene (2), ferruginol (4), 11-hydroxyferruginol (5), hydroxyferruginol quinone (6), carnosaldehyde (7) and pisiferal (8). The framed chromatogram is part of the GC–MS analysis (selected m/z signal: 286) of yeast expressing GGPPS, CPS, MS, ATR1 and C20Ox. The peak labelled (3) corresponds to miltiradien-20-al. (c) Electron impact mass spectra of PDs from yeast or rosemary. The spectra were isolated from yeast strains expressing GGPPS, CPS, MS, ATR1 and CYP76AK6 for miltiradien-20-al (3); GGPPS, CPS, MS, ATR1, CYP76AH22 and CYP76AK6 for carnosaldehyde (7); and GGPPS; CPS, MS, ATR1, CYP76AH1 and CYP76AK6 for pisiferal (8). (d) LC–MS analysis (selected m/z signals: 301.217, 315.196, 331.191 and 329.175) of rosemary extract, authentic standards and yeast extracts from strains expressing GGPPS, CPS, MS, ATR1 CYP76AH22 (HFS) and CYP76AK6 (C20Ox). Carnosaldehyde (7), CA (9), CO (10) and PA (11). (e) ESI mass spectra of PDs from yeast (expressing GGPPS, CPS, MS, ATR1, CYP76AH22 and CYP76AK6 for carnosaldehyde (7); GGPPS, CPS, MS, ATR1, CYP76AH22 and CYP76AK6 for CA (9) and CO (10); and GGPPS, CPS, MS, ATR1, CYP76AH1 and CYP76AK6 for PA (11) or rosemary and authentic standards. NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 OH FS CPS Spontaneous GGPP MiS HFS Miltiradiene (1) Abietatriene (2) Ferruginol (4) C20Ox HFS C20Ox OH HO OH HO O HO O Spontaneous Miltiradien-20-ol Pisiferol Hydroxyferruginol 11-Hydroxyferruginol (5) C20Ox quinone (6) C20Ox C20Ox O OH OH HO O HO Miltiradien-20-al (3) 11,20-Dihydroxy- ferruginol Pisiferal (8) C20Ox C20Ox OH OH OH OH HO HO HO O O HOOC HOOC C20Ox Spontaneous O Carnosol (10) Carnosic acid (9) Carnosaldehyde (7) Pisiferic acid (11) Figure 4 | Overview of the engineered CA pathway. FS (CYP76AH1) oxidizes abietatriene to ferruginol, whereas HFS (CYP76AH22-24) are able to further oxidize ferruginol to 11-hydroxyferruginol. 11-hydroxyferruginol spontaneously oxidizes to hydroxyferruginol quinone. The C20Ox (CYP76AK6-8) ensure conversion of 11-hydroxyferruginol to CA, which itself spontaneously oxidizes to carnosol. C20Ox can also oxidize miltiradiene and ferruginol, the latter leading to PA, albeit with much lower efficiency, indicating 11-hydroxyferruginol is their preferred substrate. The size of the arrows illustrates non- quantitatively the efficiency of the corresponding reactions. BsaI (in level 1 vector) or BpiI restriction sites (in level M vector) were introduced Enzyme mix (Invitrogen). Constructs were used for further cloning or for transient into the backbones. To allow assembling of up to six level 1 transcription units in a expression in N. benthamiana. defined order in level M vectors, level 1 vectors contained BpiI restriction sites, To produce plasmids for yeast engineering, Golden Gate compatible yeast which produced specific 4 bp overhangs dependent on the position on restriction37. expression vectors were used. CYP76AH22, CYP76AH23, CYP76AH24, The last positioned insert was connected to the vector M backbone by sequence CYP76AH4, CYP76AK6, CYP76AK7 and CYP76AK8 were PCR amplified in specific linkers. An URA3 gene completed the backbone for selection of yeast fragments, to remove internal BpiI and BsaI restriction sites by introducing transformants. single point silent mutations. The coding sequences were amplified from either plasmids or cDNA with KOD Hot Start DNA Polymerase (Merck) and gene-specific primers (Supplementary Table 1). PCR products were then purified Designing synthetic galactose-inducible promoters. Degenerated sequences of using QIAquick PCR Purification Kit (Qiagen) and subsequently sequenced. As synthetic galactose-inducible promoters (Supplementary Table 2) were created by the amino terminus of CYP76AH4 could not be amplified from cDNA, a mixed primer extension PCR (primers are given in Supplementary Table 1). They con- sequence out of a CYP76AH22-specific N terminus (253 amino acids) and a tained a degenerated consensus sequence of 147 bp of pGAL1 and pGAL2 as core CYP76AH4-specific carboxy terminus (243 amino acids) was used for further promoter and four GAL4-binding domains (three separated by 2 bp, the fourth analysis (Supplementary Fig. 1). CYP76AH1 was synthesized by Eurofins with 20 bp distance)39. Genomics GmbH as a codon-optimized sequence for yeast expression. Cloning into Golden Gate entry vector pL0-SC and yeast expression vector 1 was performed as previously described43. In further cloning steps, each Isolation and cloning of genes. Plasmids containing GGPPS, CPS, MS and ATR1 gene was fused to a synthetic galactose-inducible promoter and a terminator were kindly provided by colleagues within the institute12,40,41. The cDNA (Supplementary Table 2) in a yeast level 1 expression vector using BsaI sequences of CYP76AH1 (accession number JX422213.1) and CYP76AH4 were (Fermentas) and T4-Ligase in a 50 cycle restriction–ligation reaction. Different obtained from the literature14,15. Total RNA from R. officinalis leaves gene combinations were finally assembled into one yeast expression level M was isolated and transcribed to cDNA using Spectrum Plant Total RNA Kit vector by a 50 cycle restriction–ligation reaction with BpiI and T4-Ligase. (Sigma-Aldrich) and RevertAid Premium First Strand cDNA Synthesis Kit (Fermentas), respectively. Full-length coding sequences for CYP76AK6-8 were PCR amplified from R. officinalis or S. fruticosa cDNAs using specific Phylogenetic analysis. Amino acid sequences (Supplementary Table 7) primers (Supplementary Table 1) and AccuPrime High Fidelity Taq Polymerase were aligned and trimmed using MAFFT version 7 with default settings (Invitrogen). The amplified fragments were gel-purified and subcloned into (http://mafft.cbrc.jp/alignment/server/) and TrimAl version 1.3 with the method pENTR-D-TOPO vector (Invitrogen). The cDNAs, together with p35S and tOCS, Authomated1 (http://phylemon2.bioinfo.cipf.es/utilities.html), respectively. The were transferred into pEarleyGate100 binary vector42 using Gateway LR Clonase II phylogenetic tree was generated using Mega6 (ref. 44) with the maximum 8 NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 ARTICLE likelihood method based on the JTT/ þ G model (five categories)45 and a bootstrap evaluated using PROCHECK56 and ProSA57,58. Despite relatively low sequence of 1,000 replicates. identities of the enzymes with their template, all models were of excellent quality with 492% of the residues being in the most favoured region of the Ramachandran plot. The ProSA energy graphs were all in negative range and the Site-directed mutagenesis. Site-directed mutagenesis of Golden Gate entry calculated z-scores (for CYP76AH1 ¼  10.84 with 478 amino acid residues in the vector pL0-SC containing cDNAs for CYP76AH22 or CYP76AH1 was performed model and for CYP76AH22 ¼  11.04 with 481 amino acid residues) are in the using the QuikChange II Site-Directed Mutagenesis Kit (Agilent Technologies). range of natively folded proteins. Used primers are given in Supplementary Table 1. One microlitre of the reaction mixture was transformed into E. coli cells strain DH10B and plated onto agar plates containing appropriate antibiotics. Mutagenized plasmids were subsequently Production and quantification of diterpenes in yeast. Yeast expression vectors cloned as described above in yeast expression vectors. were transformed into S. cerevisiae strain INVSc1 (genotype: MATa his 3D1 leu2 trp1-289 ura3-52; Thermo Fisher Scientific) and plated out onto uracil-free selec- tion medium (1 g l  1 Yeast Synthetic Drop-out Medium Supplements without Yeast microsome isolation. Microsomal preparation was done according to the uracil (Sigma-Aldrich), 6.7 g l  1 Yeast Nitrogen Base With Amino Acids (Sigma- literature40 with the following adaptations. The plasmids with the P450 encoding Aldrich) and 20 g l  1 Micro Agar (Duchefa Biochemie)). Positively transformed gene and ATR1 were transformed in yeast strain INVSc1 as described above. A colonies were inoculated into 5 ml YPD medium (20 g l  1 tryptone and 10 g l  1 single positive colony was used to inoculate 5 ml of Ura-medium with 2% glucose yeast extract) containing 2% of glucose and grown for 24 h with shaking at 30 °C. and grown for 24 h at 30 °C with shaking. The culture was then used to inoculate To induce protein expression, the cell pellet was resuspended in fresh YPD medium 100 ml of fresh Ura-medium with 2% glucose in a 500 ml flask, which was then containing 2% galactose. After another 24 h of cultivation, whole cultures were further agitated at 30 °C for 24 h. The cells were then collected by centrifugation, extracted with 2 ml n-hexane. Quantification of shake-flask expression was done resuspended in 100 ml yeast extract-peptone-dextrose (YPD) medium with 2% according to the literature with the following adaptions15. Two single yeast colonies galactose to induce expression and grown with shaking for 24 h at 30 °C. All steps carrying the plasmids were inoculated into 5 ml selection medium (1 g l  1 Yeast were then carried out at 4 °C. The cells were recovered by centrifugation and Synthetic Drop-out Medium Supplements without uracil (Sigma-Aldrich) and resuspended in 30 ml of pre-chilled TEK buffer (50 mM Tris-HCl pH 7.5, 1 mM 6.7 g l  1 Yeast Nitrogen Base With Amino Acids (Sigma-Aldrich)) each. After EDTA, 100 mM KCL), centrifuged again and resuspended in 2 ml TES buffer cultivation for 24 h, they were transferred to 50 ml selection medium containing 2% (50 mM Tris-HCl pH 7.5, 600 mM sorbitol, 10 g l  1 BSA, 1.5 mM b- glucose in a 250 ml flask and grown for another 24 h. Expression was induced the mercaptoethanol) and transferred to a 50 ml tube. Acid-washed autoclaved next day by resuspending the cell pellet in 50 ml YPD medium containing 2% 450–600 mm diameter glass beads were then added until they reached the surface of galactose. After cultivation for 24 h the optical density (OD600) was determined and the cell suspension. The suspension was shaken vigorously by hand for 1 min and a volume with comparable cell number between all samples was used to extract the returned to ice for 1 min. This was repeated four times. To wash the glass beads, formed products with 45 ml hexane. An aliquot was used for analysis by GC–MS 5 ml TES buffer was added, after careful inversion the supernatant was collected and LC–MS as described below. The extracts were then evaporated to complete and transferred to a new tube. The procedure was repeated three times. The tube dryness and dissolved in 800 ml C6D6 containing hexamethyldisiloxane (HMDS) at was then centrifuged at 7,500 g for 10 min, the supernatant transferred to a concentration of 0.0235 mmol l  1 as internal standard for chemical shift ultracentrifugation tubes, which were centrifuged for 2 h at 100 000 g. The referencing (d1H HMDS: 0.112 p.p.m.) as well as quantification.1H NMR spectra supernatant was discarded and gently washed successively with 5 ml TES and were recorded on a Varian/Agilent VNMRS 600 NMR spectrometer operating at a 2.5 ml TEG buffer (50 mM Tris-HCl pH 7.5, 1 mM EDTA and 30% glycerol). The proton NMR frequency of 599.83 MHz using a 5 mm inverse detection cryoprobe pellet was then collected with a spatula to a Potter homogenizer resuspended in with the following parameters: pulse width ¼ 6.25 ms (90° flip angle), relaxation 2 ml TEG buffer and carefully homogenized. One hundred microlitre aliquots were delay ¼ 27.3 s, acquisition time ¼ 2.7 s, number of transients ¼ 40 and digital transferred to 1.5 ml microtubes and stored at  80 °C until used. resolution 0.367 Hz per point. An exponential multiplication of the free induction decays were performed using a line broadening factor lb ¼ 0.4 Hz. A zero filling by In vitro CYP assays. In a 2.0 ml microtube, in a total volume of 600 ml, 40 ml of the a factor of 2 was used before Fourier transformation. All spectra were manually microsome preparation were mixed with 100 mM substrate, 1 mM NADPH, 50 mM phase corrected and baseline corrected. For the quantification of metabolites, the sodium phosphate pH and incubated at 30 °C for 2 h with gentle shaking. The peak area of selected proton signals belonging to the target compounds and the reaction was then extracted with 1 ml hexane under strong agitation (vortex). After peak area of the internal standard (HMDS) were integrated manually. Peak centrifugation, the organic phase was transferred to fresh tubes, dried under a N2 assignments were done according to chemical shifts of reference compounds based stream and resuspended in 50 ml hexane for GC–MS analysis or 150 ml methanol on commercial standards or characterization of metabolites performed in this for LC–MS analysis. work. Transient expression in N. benthamiana. The cDNAs for CPS, MS, ATR1 and GC–MS analysis. Plant surface extracts (R. officinalis, S. fruticosa and infiltrated CYP76AK7 were cloned into T-DNA vectors (binary vector pL1F-1) under the N. benthamiana) were prepared by shaking three young leaves (R. officinalis and control of the 35S promoter and flanked by the Ocs terminator46. The resulting S. fruticosa) or ten leaf discs (7 mm diameter in case of N. benthamiana) with 1 ml T-DNA plasmids were transformed into Agrobacterium tumefaciens strain n-hexane for 2 min at room temperature. The extract was evaporated to complete GV3101::pMP90 (resistances to gentamycin and rifampicin) and streaked onto LB dryness and resuspended in 200 ml n-hexane. The analysis of yeast and plant agar plates with appropriate antibiotics. Positively transformed colonies were extracts were carried out using a Trace GC Ultra gas chromatograph (Thermo inoculated into LB medium with selective antibiotics and grown for 18 h at 28 °C. Scientific) coupled to ATAS Optic 3 injector and an ISQ single quadrupole mass To co-infiltrate several genes, appropriate volumes required to reach a final OD600 spectrometer (Thermo Scientific) with electron impact ionization. Chromato- of 0.5 each were combined, centrifuged and resuspended in one-fourth volume graphic separation was performed on a ZB-5ms capillary column (30 m  0.32 mm, fresh LB medium, one-fourth volume sterile water, half volume twofold Phenomenex) using splitless injection and an injection volume of 1 ml. The injec- concentrated infiltration medium (0.3 M sucrose, 20 mM glucose, 8.6 g l  1 tion temperature rose from 60 °C to 250 °C with 10 °C s  1 and the flow rate of Murashige and Skoog medium (Duchefa Biochemie), adjusted to pH 5.6) and helium was 1 ml min  1. The GC oven temperature ramp was as follows: 50 °C for 20 mM acetosyringone (diluted in dimethyl sulfoxide). The suspension was 1 min, 50 to 300 °C with 7 °C min  1, 300–330 °C with 20 °C min  1 and 330 °C for infiltrated into the abaxial leaf side of 4-week-old N. benthamiana plants using a 5 min. Mass spectrometry was performed at 70 eV, in a full scan mode with m/z syringe without needle. After treatment, the plants were cultivated under regular from 50 to 450. Data analysis was done with the device specific software Xcalibur growth conditions for 5 days. (Thermo Scientific). Homology modelling. Protein homology modelling of CYP76AH1, CYP76AH4, RP–UPLC–ESI–MS/MS analysis (LC–MS). Hexane leaf surface extracts of CYP76AH22, CYP76AH23 and CYP76AH24 was performed with YASARA47. R. officinalis and S. fruticosa leaves were completely evaporated under a nitrogen After search for templates in the protein database48 for each protein, 100 models stream, dissolved in 150 ml methanol and subjected to reversed phase–ultra- were created based on alternative sequence alignments including secondary performance LC–electrospray ionization (ESI)–MS/MS analysis. Metabolites were structure predictions and comparisons with ten found appropriate X-ray templates separated using a Nucleoshell RP18 column (2  150 mm, particle size 2.7 mm, of different CYPs49. The best suited template was the X-ray structure of 2HI4 Macherey-Nagel) and a ACQUITY UPLC System (Waters), including an (refs 50,51). Models based on some other templates such as 2NNJ52, 3T3Q53, 2PG5 ACQUITY Binary Solvent Manager and an ACQUITY Sample Manager (10 ml (ref. 54) and 4NKY55 are almost identical, except in some loop conformations. The sample loop, partial loop injection mode, 5 ml injection volume). For elution, substrates were docked using the molecular modelling environment programme aqueous 0.3 mmol l  1 NH4HCOO (adjusted to pH 3.5 with formic acid) (A) and MOE 2014.09 (https://www.chemcomp.com/). All models with docked substrates acetonitrile (B) were used. The elution conditions were as follows: isocratic from 0 in appropriate position for an oxidation reaction in vicinity to the iron ion to 2 min at 5% eluent B, from 2 to 19 min linear from 5 to 95%, from 19 to 22 min complexed by haem were refined with a 20 cycle simulated annealing procedure isocratically at 95%, from 22 to 22.01 min linear from 95 to 5% and from 22.01 to (md-refine.mcr) in YASARA. For creating the figures, the reactive oxygen atom 24 min isocratically at 5% eluent B. During the separation, the flow rate was set to bound to the iron ion was added manually in MOE corresponding to the proposed 400 ml min  1 and the column temperature was maintained at 40 °C. A TripleToF catalytic mechanism50. The quality of the resulting homology models were 5600-1 mass spectrometer (AB Sciex) was used to detect the metabolites, which was NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 equipped with an ESI-Duo-TurboIon-Spray interface (it operated in negative ion 7. Ulubelen, A., Öksüz, S., Topcu, G., Gören, A. C. & Voelter, W. Antibacterial mode) and was controlled by Analyst 1.6 TF software (AB Sciex). The LC–ESI diterpenes from the roots of Salvia blepharochlaena. J. Nat. Prod. 64, 549–551 source operation parameters were as follows: ion spray voltage:  4,500 V, (2001). nebulizing gas: 60 p.s.i., source temperature: 600 °C, drying gas: 70 p.s.i., curtain 8. Satoh, T. et al. Carnosic acid, a catechol-type electrophilic compound, protects gas: 35 p.s.i. Data acquisition was performed in the MS1-ToF mode59, scanned from neurons both in vitro and in vivo through activation of the Keap1/Nrf2 250 to 500 Da with an accumulation time of 50 ms, and the MS2-SWATH mode, pathway via S-alkylation of targeted cysteines on Keap1. J. Neurochem. 104, divided into 5 Da segments of 20 ms accumulation time. Fifty-six separate scan 1116–1131 (2008). experiments were carried out covering the mass range from 65 to 500 Da. In this 9. Fischedick, J. T., Standiford, M., Johnson, D. A. & Johnson, J. A. Structure process, a declustering potential of 35 V (  35 V) was applied and collision activity relationship of phenolic diterpenes from Salvia officinalis as activators energies were set to 55 V with a collision energy spread of ±45 V. of the nuclear factor E2-related factor 2 pathway. Bioorg. Med. Chem. 21, 2618–2622 (2013). 10. Birtić, S., Dussort, P., Pierre, F. X., Bily, A. C. & Roller, M. Carnosic acid. Isolation of new PDs and structure confirmation by NMR. For isolation of Phytochemistry 115, 9–19 (2015). 11-hydroxyferruginol, a single yeast colony expressing GGPPS, CPS, MS, ATR1 11. Zhang, Y. et al. Tanshinones: sources, pharmacokinetics and anti-cancer and CYP76AH22 was inoculated into 5 ml YPD medium containing 2% glucose and grown at 28 °C with shaking. After 24 h, the suspension was transferred into activities. Int. J. Mol. Sci. 13, 13621–13666 (2012). 400 ml culture medium for shake-flask expression and grown for another 24 h. The 12. Brückner, K. et al. Characterization of two genes for the biosynthesis of expression was induced by resuspending the cell pellet in 400 ml YPD with 2% abietane-type diterpenes in rosemary (Rosmarinus officinalis) glandular galactose and grown as mentioned. The diterpenes were extracted from the yeast trichomes. Phytochemistry 101, 52–64 (2014). culture by adding 300 ml n-hexane and shaking thoroughly. Miltiradien-20-al was 13. Brückner, K. & Tissier, A. High-level diterpene production by transient isolated from 84 g infiltrated leaf material of N. benthamiana expressing GGPPS, expression in Nicotiana benthamiana. Plant Methods 9, 46 (2013). CPS, MS and the C20Ox CYP76AK6, which were homogenized and extracted with 14. Zi, J. & Peters, R. J. Characterization of CYP76AH4 clarifies phenolic 800 ml n-hexane. The extract was separated from leaf material by centrifugation, diterpenoid biosynthesis in the Lamiaceae. Org. Biomol. Chem. 11, 7650–7652 evaporated and dissolved in 50 ml n-hexane. To initially separate miltiradien-20-al (2013). from other leaf compounds, the extract was applied to a SiOH column (25 g). 15. Guo, J. et al. CYP76AH1 catalyzes turnover of miltiradiene in tanshinones Elution steps of 99:1, 98:2, 97:3 and 96:4 n-hexane:ethyl acetate were collected biosynthesis and enables heterologous production of ferruginol in yeasts. Proc. and the last fraction containing miltiradien-20-al was used for preparative high Natl Acad. Sci. USA 110, 12108–12113 (2013). performance LC. Carnosaldehyde was extracted twice from 100 young leaves of 16. Božić, D. et al. Towards elucidating carnosic acid biosynthesis in lamiaceae: R. officinalis with 80 ml n-hexane each time. The combined extracts were functional characterization of the three first steps of the pathway in Salvia evaporated, dissolved in 40 ml 95:5 n-hexane:ethyl acetate and applied to a fruticosa and Rosmarinus officinalis. PLoS ONE 10, e0124106 (2015). SiOH column (15 g), which was equilibrate with 95:5 n-hexane:ethyl acetate. 17. Mizutani, M. & Ohta, D. Two isoforms of NADPH:cytochrome P450 reductase Carnosaldehyde eluted in the 94:6 n-hexane:ethyl acetate fraction which was in Arabidopsis thaliana: gene structure, heterologous expression in insect cells, used for further purification. and differential regulation. Plant Physiol. 116, 357–367 (1998). All enriched fractions were applied to a rotary evaporator and the dried 18. Tada, M., Kurabe, J., Yoshida, T., Ohkanda, T. & Matsumoto, Y. Syntheses and products were resuspended in 1 ml of methanol. Semi-preparative separation of antibacterial activities of diterpene catechol derivatives with abietane, totarane 11-hydroxyferruginol and carnosaldehyde was done by LC-UV (Waters 2695 with and podocarpane skeletons against methicillin-resistant Staphylococcus aureus 2996 diode array detector, Waters GmbH) equipped with a fraction collector and Propionibacterium acnes. Chem. Pharm. Bull. 58, 818–824 (2010). (Waters, FC III). Between the column and fraction collector, a capillary splitter 19. Cuvelier, M. E., Berset, C. & Richard, H. Antioxidant constituents in sage (1:100) was installed which allowed ultraviolet–visible analysis at 254 nm and (Salvia officinalis). J. Agr. Food Chem. 42, 665–669 (1994). fraction collection. Using an XTerra Prep C18 column (7.8 mm  150 mm  5 mm, 20. Richheimer, S., Bernart, M., King, G., Kent, M. & Beiley, D. Antioxidant activity Waters GmbH) and a flow rate of 6 ml min  1, the same LC gradient was applied as for reverse phase–ultra-performance LC–ESI–MS/MS analysis. One hundred of lipid-soluble phenolic diterpenes from rosemary. J. Am. Oil Chem. Soc. 73, microlitres of each enriched fraction were injected repeatedly (four times) and a 507–514 (1996). fraction eluting between 17 and 18 min (11-hydroxyferruginol) or between 16.1 21. Bicchi, C., Binello, A. & Rubiolo, P. Determination of phenolic diterpene and 16.5 min (carnosaldehyde) was repeatedly collected. The fractions were antioxidants in rosemary (Rosmarinus officinalis L.) with different methods of combined, dried down in a nitrogen stream and resuspended in deuterated solvents extraction and analysis. Phytochem. Anal. 11, 236–242 (2000). for NMR analysis. 1H and two-dimensional spectra were recorded on an Agilent 22. Ho, C.-T., Wang, M., Wei, G.-J., Huang, T.-C. & Huang, M.-T. VNMRS 600 NMR spectrometer operating at 599.832 MHz using a 5 mm inverse Chemistry and antioxidative factors in rosemary and sage. Biofactors 13, detection cryoprobe. Chemical shifts were referenced to internal tetramethylsilane 161–166 (2000). (d ¼ 0 p.p.m., 1H) or internal C6D11H (d ¼ 1.38 p.p.m., 1H) and CDCl3 23. Wenkert, E., Fuchs, A. & McChesney, J. D. Chemical artifacts from the family (d ¼ 77.0 p.p.m., 13C), or C6D12 (d ¼ 26.4 p.p.m., 13C) respectively. labiatae. J. Org. Chem. 30, 2931–2934 (1965). 24. Guo, J. et al. Cytochrome P450 promiscuity leads to a bifurcating biosynthetic pathway for tanshinones. New Phytol. 18, 525–534 (2015). Data availability. Data supporting the findings of this study are available within 25. Bak, S. et al. Cytochromes P450. Arabidopsis Book 9, e0144 (2011). the article and its Supplementary Information files, and from the corresponding 26. Ro, D.-K., Arimura, G.-I., Lau, S. Y. W., Piers, E. & Bohlmann, J. Loblolly pine author upon reasonable request. The sequences of the CYP76AK6, CYP76AK7 and abietadienol/abietadienal oxidase PtAO (CYP720B1) is a multifunctional, CYP76AK8 enzymes are available under GenBank accessions KX431218, multisubstrate cytochrome P450 monooxygenase. Proc. Natl Acad. Sci. USA KX431219 and KX431220, respectively. 102, 8060–8065 (2005). 27. Teoh, K. H., Polichuk, D. R., Reed, D. W., Nowak, G. & Covello, P. S. Artemisia annua L. (Asteraceae) trichome-specific cDNAs reveal CYP71AV1, a References cytochrome P450 with a key role in the biosynthesis of the antimalarial 1. Kingston, D. G. I. Taxol, a molecule for all seasons. Chem. Commun. 10, sesquiterpene lactone artemisinin. FEBS Lett. 580, 1411–1416 (2006). 867–880 (2001). 28. Ignea, C. et al. Carnosic acid biosynthesis elucidated by a synthetic biology 2. Watts, G. Nobel awarded to discoverers of ivermectin and artemisinin. BMJ platform. Proc. Natl Acad. Sci. USA 113, 3681–3686 (2016). 351, h5352 (2015). 29. Zhang, Y. et al. Degradation study of carnosic acid, carnosol, rosmarinic acid, 3. Jordán, M. J., Lax, V., Rota, M. C., Lorán, S. & Sotomayor, J. A. Relevance of and rosemary extract (Rosmarinus officinalis L.) assessed using HPLC. J. Agric. carnosic acid, carnosol, and rosmarinic acid concentrations in the in vitro Food Chem. 60, 9305–9314 (2012). antioxidant and antimicrobial activities of Rosmarinus officinalis (L.) 30. Schwarz, K., Ternes, W. & Schmauderer, E. Antioxidative constituents of methanolic extracts. J. Agr. Food Chem. 60, 9603–9608 (2012). Rosmarinus officinalis and Salvia officinalis. III. Stability of phenolic diterpenes 4. Genena, A. K., Hense, H., Smânia Junior, A. & Souza, S. M. d. Rosemary of rosemary extracts under thermal stress as required for technological (Rosmarinus officinalis): a study of the composition, antioxidant and processes. Z. Lebensm. Unters. Forch. 195, 104–107 (1992). antimicrobial activities of extracts obtained with supercritical carbon dioxide. 31. Fraga, B. M., Diaz, C. E., Guadano, A. & Gonzalez-Coloma, A. Diterpenes from Ciência e Tecnologia de Alimentos 28, 463–469 (2008). Salvia broussonetii transformed roots and their insecticidal activity. J. Agric. 5. Poeckel, D. et al. Carnosic acid and carnosol potently inhibit human Food Chem. 53, 5200–5206 (2005). 5-lipoxygenase and suppress pro-inflammatory responses of stimulated human 32. Hofer, R. et al. Dual function of the cytochrome P450 CYP76 family from polymorphonuclear leukocytes. Biochem. Pharm. 76, 91–97 (2008). Arabidopsis thaliana in the metabolism of monoterpenols and phenylurea 6. Johnson, J. et al. Carnosol, a dietary diterpene, displays growth inhibitory herbicides. Plant Physiol. 166, 1149–1161 (2014). effects in human prostate cancer PC3 cells leading to G2-phase cell cycle arrest 33. Hamberger, B., Møller, B. L., Pateraki, E., Andersen-Ranberg, J. & Jensen, N. B. and targets the 50 -AMP-activated protein kinase (AMPK) pathway. Pharm. Res. Biosynthesis of forskolin and related compounds. International patent 25, 2125–2134 (2008). application number WO2015/113569 A1 (2015). 10 NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12942 ARTICLE 34. Jorgensen, L. et al. 14-step synthesis of ( þ )-ingenol from ( þ )-3-carene. 55. Petrunak, E. M., DeVore, N. M., Porubsky, P. R. & Scott, E. E. Structures of Science 341, 878–882 (2013). human steroidogenic cytochrome P450 17A1 with substrates. J. Biol. Chem. 35. Yerger, E. H. et al. A rapid method for isolating glandular trichomes. Plant 289, 32952–32964 (2014). Physiol. 99, 1–7 (1992). 56. Laskowski, R. A., MacArthur, M. W., Moss, D. S. & Thornton, J. M. 36. Livak, K. J. & Schmittgen, T. D. Analysis of relative gene expression data using PROCHECK: a program to check the stereochemical quality of protein real-time quantitative PCR and the 2  DDCT method. Methods 25, 402–408 structures. J. Appl. Cryst. 26, 283–291 (1993). (2001). 57. Sippl, M. J. Calculation of conformational ensembles from potentials of mean 37. Engler, C. & Marillonnet, S. Golden gate cloning. Methods Mol. Biol. 1116, force. An approach to the knowledge-based prediction of local structures in 119–131 (2014). globular proteins. J. Mol. Biol. 213, 859–883 (1990). 38. Werner, S., Engler, C., Weber, E., Gruetzner, R. & Marillonnet, S. Fast track 58. Sippl, M. J. Recognition of errors in three-dimensional structures of proteins. assembly of multigene constructs using Golden Gate cloning and the MoClo Proteins 17, 355–362 (1993). system. Bioeng. Bugs 3, 38–43 (2012). 59. Hopfgartner, G., Tonoli, D. & Varesio, E. High-resolution mass spectrometry 39. Teo, W. S. & Chang, M. W. Development and characterization of AND-gate for integrated qualitative and quantitative analysis of pharmaceuticals in dynamic controllers with a modular synthetic GAL1 core promoter in biological matrices. Anal. Bioanal. Chem. 402, 2587–2596 (2012). Saccharomyces cerevisiae. Biotechnol. Bioeng. 111, 144–151 (2014). 40. Urban, P., Mignotte, C., Kazmaier, M., Delorme, F. & Pompon, D. Cloning, yeast expression, and characterization of the coupling of two distantly related Acknowledgements We thank Jürgen Schmidt for assistance in the analysis of mass spectral data and Arabidopsis thaliana NADPH-cytochrome P450 reductases with P450 Benedikt Athmer for his support concerning phylogenetic analysis. We also thank CYP73A5. J. Biol. Chem. 272, 19176–19186 (1997). Arianne Schnabel and Sarah Vorndran for technical assistance. The work was supported 41. Orlova, I. et al. The small subunit of snapdragon geranyl diphosphate synthase in part by the TERPMED project (EC grant number 227448) and by the Leibniz Institute modifies the chain length specificity of tobacco geranylgeranyl diphosphate of Plant Biochemistry in Halle, Germany. synthase in planta. Plant Cell 21, 4002–4017 (2009). 42. Earley, K. W. et al. Gateway-compatible vectors for plant functional genomics and proteomics. Plant J. 45, 616–629 (2006). Author contributions 43. Engler, C., Kandzia, R. & Marillonnet, S. A one pot, one step, precision cloning A.T., A.F. and A.K.K. conceived the project. A.T. conceived and designed the method with high throughput capability. PLoS ONE 3, e3647 (2008). experiments. U.S., K.R., D.M., D.B., D.P., G.U.B., A.H., A.P. and W.B. performed the 44. Tamura, K., Stecher, G., Peterson, D., Filipski, A. & Kumar, S. MEGA6: experiments. S.M. conceived and designed the yeast Golden gate cloning vectors. S.L. molecular evolutionary genetics analysis version 6.0. Mol. Biol. Evol. 30, conceived, designed and made the synthetic yeast GAL-inducible promoters. U.S., A.P. 2725–2729 (2013). and A.T. analysed the data. U.S. and A.T. wrote the paper. 45. Jones, D. T., Taylor, W. R. & Thornton, J. M. The rapid generation of mutation data matrices from protein sequences. Comput. Appl. Biosci. 8, 275–282 (1992). Additional information 46. Weber, E., Gruetzner, R., Werner, S., Engler, C. & Marillonnet, S. Assembly Supplementary Information accompanies this paper at http://www.nature.com/ of designer TAL effectors by golden gate cloning. PLoS ONE 6, e19722 (2011). naturecommunications 47. Krieger, E. et al. Improving physical realism, stereochemistry, and side-chain accuracy in homology modeling: Four approaches that performed well in Competing financial interests: The authors declare no competing financial CASP8. Proteins 9, 114–122 (2009). interests. 48. Berman, H. M. et al. The Protein Data Bank. Nucleic Acids Res. 28, 235–242 (2000). Reprints and permission information is available online at http://npg.nature.com/ 49. Jones, D. T. Protein secondary structure prediction based on position-specific reprintsandpermissions/ scoring matrices. J. Mol. Biol. 292, 195–202 (1999). How to cite this article: Scheler, U. et al. Elucidation of the biosynthesis of carnosic acid 50. Meunier, B., de Visser, S. P. & Shaik, S. Mechanism of oxidation reactions and its reconstitution in yeast. Nat. Commun. 7, 12942 doi: 10.1038/ncomms12942 catalyzed by cytochrome p450 enzymes. Chem. Rev. 104, 3947–3980 (2004). (2016). 51. Sansen, S. et al. Adaptations for the oxidation of polycyclic aromatic hydrocarbons exhibited by the structure of human P450 1A2. J. Biol. Chem. 282, 14348–14355 (2007). This work is licensed under a Creative Commons Attribution 4.0 52. Schoch, G. A. et al. Determinants of cytochrome P450 2C8 substrate binding: International License. The images or other third party material in this structures of complexes with montelukast, troglitazone, felodipine, and 9-cis- article are included in the article’s Creative Commons license, unless indicated otherwise retinoic acid. J. Biol. Chem. 283, 17227–17237 (2008). in the credit line; if the material is not included under the Creative Commons license, 53. DeVore, N. M. et al. Structural comparison of cytochromes P450 2A6, 2A13, users will need to obtain permission from the license holder to reproduce the material. and 2E1 with pilocarpine. FEBS J. 279, 1621–1631 (2012). To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ 54. Sansen, S., Hsu, M. H., Stout, C. D. & Johnson, E. F. Structural insight into the altered substrate specificity of human cytochrome P450 2A6 mutants. Arch. Biochem. Biophys. 464, 197–206 (2007). r The Author(s) 2016 NATURE COMMUNICATIONS | 7:12942 | DOI: 10.1038/ncomms12942 | www.nature.com/naturecommunications 11