A flavin-dependent monooxygenase catalyzes the initial step in cyanogenic glycoside synthesis in ferns

Cyanogenic glycosides form part of a binary plant defense system that, upon catabolism, detonates a toxic hydrogen cyanide bomb. In seed plants, the initial step of cyanogenic glycoside biosynthesis—the conversion of an amino acid to the corresponding aldoxime—is catalyzed by a cytochrome P450 from the CYP79 family. An evolutionary conundrum arises, as no CYP79s have been identified in ferns, despite cyanogenic glycoside occurrence in several fern species. Here, we report that a flavin-dependent monooxygenase (fern oxime synthase; FOS1), catalyzes the first step of cyanogenic glycoside biosynthesis in two fern species (Phlebodium aureum and Pteridium aquilinum), demonstrating convergent evolution of biosynthesis across the plant kingdom. The FOS1 sequence from the two species is near identical (98%), despite diversifying 140 MYA. Recombinant FOS1 was isolated as a catalytic active dimer, and in planta, catalyzes formation of an N-hydroxylated primary amino acid; a class of metabolite not previously observed in plants.

P lants produce a plethora of natural products (phytochemicals or specialized metabolites) enabling interactions with their biotic and abiotic environment. Cyanogenic glycosides are one such class of amino acid-derived natural products present in more than 3000 plant species, including ferns, gymnosperms, and angiosperms [1][2][3] . For example, the cyanogenic glycosides prunasin and amygdalin are responsible for the bitterness of wild almond (Prunus dulcis) 4,5 . Upon tissue disruption, cyanogenic glycosides are hydrolyzed by specific β-glucosidases resulting in detonation of a hydrogen cyanide bomb as an immediate toxic chemical response e.g. towards chewing herbivores 2 . More recently, cyanogenic glycosides have been shown to possess alternative functions as remobilizable storage molecules of reduced nitrogen, controllers of bud break and flower induction, and as quenchers of reactive oxygen species [6][7][8][9] .
In higher plants (gymnosperms and angiosperms), the biosynthesis of cyanogenic glycosides is catalyzed by cytochromes P450 (CYPs) and UDP-glucosyltransferases (UGTs) 2,10,11 . In all cases, initial conversion of the parent amino acid to the corresponding E-oxime is catalyzed by a functionally conserved CYP79 family enzyme. Independently evolved CYP71, CYP706, or CYP736 enzymes convert the oxime into an αhydroxynitrile 10,12,13 that is glycosylated by a UGT85 or UGT94 family member to produce the cyanogenic mono-or diglycosides 2,4,14 (Fig. 1).
The CYP79-catalyzed reaction proceeds via two N-hydroxylations, a decarboxylation, and a dehydration reaction in a single catalytic site [15][16][17] . No other enzymes across the plant kingdom are known to catalyze the conversion of an α-amino acid into an oxime. In addition to their involvement as intermediates in cyanogenic glycoside biosynthesis, CYP79-formed oximes are important metabolites in general and specialized metabolism 18 exemplified by indole-3-acetaldoxime, which is a shared precursor for the phytohormone auxin (indole-3-acetic acid, IAA), the phytoalexin camalexin and tryptophan-derived glucosinolates [19][20][21] . An evolutionary conundrum arises as no gene sequences encoding CYP79s have been found in fern transcriptomes nor genomes 22,23 despite the occurrence of cyanogenic glycosides in ferns 1,24 . Due to their significant phylogenetic position, ferns represent an important lineage for studying the evolution of land plants. Fern research has been hampered by the scarcity of genome information 22 . In total, 11 orders of ferns are known of which four are extant: Psilotopsida, Equisetopsida, Marattiopsida, and Polypodiopsida 25 . Polypodiopsida are termed modern ferns, and~3% of the species in this order have been reported as cyanogenic 1,24,26 .
Here, we investigate the cyanogenic glycoside biosynthetic pathway in modern ferns by a differentially expression survey of de novo assembled transcriptomes from Phlebodium aureum and Pteridium aquilinum. We report biochemical and biological evidence that ferns harbor an N-hydroxylating flavin-dependent monooxygenase that converts phenylalanine to a corresponding oxime via N-hydroxyphenylalanine. This demonstrates convergent evolution at the biochemical pathway level and resolves how ferns produce cyanogenic glucosides in the absence of CYP79 encoding genes.

Results
Metabolite-guided pathway discovery in ferns. The two distantly related modern ferns, Phlebodium aureum (Polypodiaceae) and Pteridium aquilinum (Dennstaedtiaceae) (Fig. 2), produce the phenylalanine-derived cyanogenic monoglucoside prunasin (D-mandelonitrile-β-D-glucopyranoside) 1,27 and the diglycoside vicianin (6-O-arabinopyranosylglucopyranoside) 28 , respectively. Targeted metabolite profiling of a population of 25 field-collected P. aquilinum (Paq) identified individuals with high and low cyanogenic glycoside containing pinnae ranging from 0.2 to 0.6 mg prunasin g −1 fw (Fig. 2d). Two individuals were selected based on their metabolite content and the quality of RNA extracted. Similarly, analysis of different tissues within a single Phlebodium aureum (Pa) fern identified variable levels of vicianin in the different tissue types from negligible levels in the spores, to 5 mg vicianin g −1 fw in the emerging fiddlehead (Fig. 2c, Supplementary Fig. 1). For Pa, fiddlehead and young pinna were selected for transcriptome analysis (Fig. 2c). mRNA was isolated from these four tissues to obtain biosynthetic gene candidates using a comparative transcriptomic approach.
BLAST searches of the de novo assembled transcriptomes from Paq and Pa identified 139 CYPs in Paq compared to 120 in Pa (Supplementary Table 1). Neither genes encoding CYP79s nor CYPs with the CYP79 signature F/H substitution in the PERF region were identified 29,30 . Additional searches of the gene sequences deposited in the OneKP database 31 confirmed the previously reported absence of CYP79 encoding gene sequences in ferns (Fig. 3) 22,23 . In the absence of CYP candidates, we extended the search to other gene families encoding monooxygenases with focus on genes showing interesting differential expression patterns of gene homologs within and between the two fern species.
The contig Pa22758 encoded a flavin-dependent monooxygenase ORF of 543 aa in accordance with a predicted full-length sequence 32 . Deep-mining of the transcriptome reads revealed that Pa contained variants of Pa22758 ( Supplementary Fig. 2). The transcript level of this gene was 24-fold higher (−logFDR of 4.32) in Pa fiddlehead compared to the young pinnae (Fig. 2e). A reciprocal BLAST search between the predicted flavin-dependent monooxygenases of Paq and Pa identified a contig Paq18302 in Paq harboring a full-length nucleotide sequence encoding a flavin-dependent monooxygenase identical to the sequence of Pa22758, except for a single amino acid substitution. The expression level of Paq18302 showed a 23% increase (−logFDR of 0.50) in the individual containing higher levels of prunasin (Fig. 2e). The deduced amino acid sequences of the Pa and Paq flavin-dependent monooxygenases revealed that they contain conserved motifs specific to Class B flavin-dependent monooxygenases (Fig. 2f, Supplementary Fig. 2) 33 .
A BLAST search in the CATH: Protein Structure Classification Database (cathdb.info, version 4.2) with the Paq18302 protein sequence, classified the fern sequences as belonging to the Class B, flavin-dependent monooxygenases. These enzymes are characterized by being single-component FAD-binding enzymes harboring In ferns, the conversion of the parent amino acid into an oxime is catalyzed by a multifunctional FMO, whereas in all higher seed plant species analyzed, the reaction is catalyzed by a cytochrome P450 from the CYP79 family 2,4,10,12,14,61,73-79 . *Uncharacterized pathway partners.
Biochemical characterization of fern oxime synthase 1 (FOS1) in planta. Functional characterization of the FMO enzyme Paq18302 (Fig. 4a) was obtained by Agrobacterium tumefaciensmediated transient expression of the encoding gene in Nicotiana benthamiana. Leaf discs of agro-infiltrated tissue were harvested after 4 days and subjected to metabolite profiling. Expression of the Paq18302 FMO afforded production of two constituents of m/z 136 eluting at rt = 8.17 and 8.70 min corresponding to the [M+H] + adduct of (E)-and (Z)-phenylacetaldoxime, respectively, as verified by co-elution with an authentic standard (Fig. 4a). Two additional constituents were identified as a glucoside of phenylacetaldoxime (m/z 320, [M+Na] + at rt = 6.7 min) and as a phenylacetaldoxime glucoside-malonic acid conjugate (m/z 406, [M+Na] + at rt = 7.9 min) based on the diagnostic fragments in the MS/MS spectra (Fig. 4c, Supplementary  Fig. 3) [36][37][38] . Identical oxime derivatives have previously been COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-020-01224-5 ARTICLE COMMUNICATIONS BIOLOGY | (2020) 3:507 | https://doi.org/10.1038/s42003-020-01224-5 | www.nature.com/commsbio observed in N. benthamiana in response to expression of CYP79 enzymes 10 . Guided by Pa22758, the FMO encoding sequence from Pa was isolated from cDNA and shown by transient expression to be functionally equivalent to the Paq18302 FMO (Fig. 4, Supplementary Fig. 4). We designate the orthologous FMO proteins as "FOS1". Transient expression of FOS1 in N. benthamiana did not give rise to formation of other oximes or additional products, demonstrating that FOS1 has phenylalanine as its specific amino acid substrate.
Independent functional characterization of the FOS1 genes was obtained in N. benthamiana by co-expression with CYP71AN24 and UGT85A19, encoding the last two enzymes in the prunasin biosynthetic pathway in almond (Prunus dulcis) 4,14 . This resulted in production of prunasin as demonstrated by the formation of a constituent comigrating with an authentic standard with an extracted ion chromatogram (EIC) of m/z 318 corresponding to the [M+Na] + adduct of prunasin (Fig. 4b, Supplementary Fig. 4). Neither phenylacetaldoxime nor prunasin are endogenous constituents of N. benthamiana.
In vitro assays revealed existence of N-hydroxyphenylalanine. Of the limited number of FMOs characterized from plants, only YUCCA6 from A. thaliana and AsFMO from garlic (Allium sativum) have been successfully expressed and purified in enough quantity for downstream biochemical analyses 33,39,40 . Here, we expressed the PaqFOS1 protein with an N-terminal 6xHIS tag using Escherichia coli as a heterologous host and isolated Paq-FOS1 by immobilized metal affinity chromatography followed by size exclusion chromatography (SEC) with a yield of 0.7 mg/L culture (Fig. 5). The isolated PaqFOS1 protein binds the cofactors FAD and NADPH as demonstrated by absorbance spectrometry, and migrates with an apparent molecular mass of 60 kDa on SDS-PAGE (Fig. 5) in agreement with a calculated molecular mass of 62.5 kDa. Upon SEC, PaqFOS1 eluted with a mass of 120 kDa suggesting that the native protein is a homodimer. In vitro assays of the isolated PaqFOS1 protein followed by LC-MS analyses confirmed that FOS1 can catalyze the conversion of phenylalanine to phenylacetaldoxime (Fig. 6). Concomitant production of N-hydroxyphenylalanine was also observed and verified using a chemically synthesized standard ( Supplementary Fig. 8). Surprisingly, targeted LC-MS analysis of fiddleheads of Pa as well as of N. benthamiana leaves transiently expressing FOS1 demonstrated the in vivo presence of N-hydroxyphenylalanine in these tissues ( Fig. 6, Supplementary Fig. 5).

Discussion
In this report, we identify and functionally characterize a flavindependent monooxygenase designated FOS1 from two modern fern species, Pa and Paq. Metabolite profiling of N. benthamiana leaves following transient expression of PaqFOS1 revealed the production of phenylacetaldoxime, which was also confirmed by the targeted in vitro experiments. When jointly expressed with CYP71AN24 and UGT85A19 from almond (Prunus dulcis), the entire prunasin pathway was established.
The FOS1 enzyme belongs to FMO class B, which are singlecomponent FAD-binding enzymes that also harbor binding sites for the hydride electron donor NAD(P)H and molecular oxygen 34,41 . PaqFOS1 was isolated and purified as a functionally active homodimer. In vitro assays demonstrated that FOS1 is able to convert its substrate phenylalanine into two products, Nhydroxyphenylalanine and phenylacetaldoxime. The first product is obtained by a single N-hydroxylation reaction, the second by two consecutive N-hydroxylations followed by decarboxylation and dehydration reactions. In plants, the conversion of an amino acid to the corresponding aldoxime has only been reported as catalyzed by P450s from the CYP79 family. It is to be noted that an FMO from the actinomycete fungus Streptomyces coelicolor A3 can convert tryptophan and C5 prenylated tryptophan into their Fig. 2 Cyanogenic glycoside content and FMO transcript abundances in tissues from the two modern fern species Pteridium aquilinum and Phlebodium aureum. a Two phenylalanine-derived cyanogenic glycosides have been reported from ferns: the monoglucoside prunasin (D-mandelonitrile-β-Dglucopyranoside) and the diglycoside vicianin (6-O-arabinopyranosylglucopyranoside). b Phylogenetic relationship between Pteridium aquilinum (Dennstaedtiaceae) and Phlebodium aureum (Polypodiaceae) showing that these two modern ferns species diversified 140 million years ago (tree adapted from 43 ). c Content of vicianin across different tissue types of P. aureum (see also Supplementary Fig. 1), with the blue bars indicating the tissue selected for downstream transcriptomic analysis. d Content of prunasin in the pinnae of a population of 25 field-collected P. aquilinum, with the green bars indicating individuals selected for transcriptomic analysis. e Transcript abundance of predicted flavin monooxygenases (FMOs) in tissues of P. aquilinum and P. aureum containing high (gray bars) or low (black bars) cyanogenic glycoside levels. Arrows indicate candidate genes. TPM transcripts per million mapped reads. f Schematic illustration of identity and motifs in the transcriptome-deduced amino acid sequences of Paq18302 (PaqFOS1) and Pa22578. Differences between Pa22578 and the isolated PaFOS1 are indicated by blue lines. The position of putative binding motifs for FAD and NADPH, the FATGY and FMOidentifier motifs conserved across plant FMOs are highlighted. Supporting alignment is shown in Supplementary Fig. 1. The diagram is based on analysis of the OneKP database 31 , which includes transcriptomes from 74 ferns. The presence of 8930 cytochromes P450 fern sequences were predicted and sorted into families in accordance with nomenclature. The analyzed fern transcriptomes harbor at least 81 different P450 families of which 49 (60%) are novel fern-specific families. Approximately half of the CYP families present in higher plants are also found in ferns. The CYP79 family is present from gymnosperm to eudicots, but based on the >40% sequence identity, the CYP79 is absent in ferns. This also applies to the other known CYP families involved in cyanogenic glucoside biosynthesis in plants: the CYP71, CYP706, and CYP736 families. corresponding aldoximes 42 . Free N-hydroxyphenylalanine was also present in the fern tissue of Pa. N-hydroxylated protein amino acids have to our knowledge not previously been detected in biological tissues, and their functional roles remains unknown.
The 98% conservation of the FOS1 encoding gene sequences of Pa and Paq is remarkable. Among the total number of 55,000 and 63,000 gene sequences present in the transcriptomes of Pa and Paq, respectively, only 2% (1297 transcripts) share a sequence identity higher than 95%. Modern ferns diversified into the Pteridoids (including Paq) and Eupolypoids (including Pa) 140 million years ago (Fig. 2b) 43 . A blast search of the FOS1 gene sequence against the 68 fern transcriptomes available in the oneKP identified the presence of an identical transcript in Phlebodium pseudoaureum, a close relative to Pa. Deep mining of the transcriptomic resources documented the presence of homologous sequences across the evolutionary gap between these species (Supplementary Tables 2 and 3). If a common ancestral FOS1 sequence was present in a progenitor to the derived ferns, the sequence has been under a remarkably high selection pressure. The four-electron oxidative decarboxylation reaction catalyzed by FOS1 is complex and may impose such a selection pressure. However, FMOs are "loaded guns" with the energy to drive an oxygenation reaction stored in the enzyme without precise docking of the substrate 33,39 . This can explain why Nhydroxyphenylalanine escapes the entire enzymatic reaction sequence and is present as an in vivo metabolite in Pa. Nhydroxyphenylalanine may have yet unrecognized functional roles in addition to being the initial intermediate in cyanogenic glycoside synthesis. The CYP79 family enzymes catalyzing the same set of reactions in higher plants show less sequence conservation. This may be because electron donation is provided to CYP enzymes by a separate NADPH-dependent cytochrome P450-oxidoreductase 3,44 . In CYP79-catalyzed reactions, all intermediates are bound within the active site as demonstrated by stable isotope experiments preventing release of the N-hydroxy amino acid intermediate 45 . Alternatively, the high sequence similarity between FOS1 from the two distantly related fern species may also reflect horizontal gene transfer. This phenomenon has been observed several times in fern species 46,47 . Ferns and other seed-free plants may be more prone to horizontal gene transfer due to the weaker protection of the gametophytic eggs and sperm to the external environment, enabling transfer of genetic material 48 . Identification of the oxime-producing step in the cyanogenic fern species that phylogenetically lie within the 140 million years gap between Pa and Paq would establish the evolutionary relationships of these two ortholog genes.
Class B flavin-dependent monooxygenases are found in all kingdoms of life 35 . All plant and animal flavin-dependent monooxygenases belong to this Class B FMO type, with a single exception of a Baeyer-Villiger monooxygense found in moss (Physcomitrella patens) 33,49 . The Arabidopsis thaliana genome contains 29 FMO genes 32,33 and the human genome contains a gene cluster encoding five FMOs (FMO1-FMO5) 50 . The Pfam 31.0 database (accessed May 2020) lists a total of 1861 predicted Class B FMO sequences from 85 plant species (pfam.xfam.org 51 ). To investigate the evolutionary diversity of plant FMOs, a representative set of FMO sequences from species representing land plant evolution, including the predicted full-length FMOs from fern transcriptomes (six from Pa and five from Paq, (Supplementary Fig. 6) and conifer (Chamaecyparis hodginsii and Picea abies) were used to build a phylogenetic tree (Fig. 7, Supplementary Table 4). The phylogenetic analysis includes representative functionally characterized FMOs: AsFMO1 from garlic (A. sativum) catalyzing S-oxygenation of allyl-mercaptan 40 , the A. thaliana YUCCAs involved in the biosynthesis of the phytohormone auxin 52 , the AtFMO GS-OX1-5 performing Soxygenation of methylthioalkyl glucosinolates 53 , and A. thaliana AtFMO1 that catalyzes N-hydroxylation of pipecolic acid to form N-hydroxypipecolic acid, the critical signaling molecule in systemic acquired resistance [54][55][56] . The analysis shows that plant FMOs cluster in three phylogenetically distinct groups (Fig. 7), with each harboring members from all evolutionary distinct species from Selaginella (moss) to A. thaliana (angiosperm). This suggests an evolutionary split of the FMOs prior to emergence of the early land plants.
By adding fern FMOs to a plant FMO-specific phylogenetic tree, the role of the other putative fern monooxygenases can be hypothesized (Fig. 2e, Supplementary Fig. 6). The analysis identified six transcripts, three orthologues from each fern species, to cluster in the YUCCA clade. To date all characterized YUCCAs are involved in auxin biosynthesis, as they are proposed to catalyze a decarboxylation of indole-3-pyruvate acid to form IAA. An additional role of thiol reductase activity has been linked to these enzymes 33,57 . The establishment of a YUCCA-like mediated auxin function for fern FMOs would contribute to the evolutionary perspectives of hormone biosynthesis and signaling. PaFOS1 and PaqFOS1 group together with two additional FMO contigs (Pa22435 and Paq33416; Fig. 2e, Supplementary Fig. 6). This group of FMOs also encapsulates the AtFMO1 catalyzing Nhydroxylation of pipecolic acid 54 (Fig. 7). Based on the phylogeny and functional characterization, we suggest that the group encapsulating PaFOS1, PaqFOS1, and AtFMO1 catalyze Nhydroxylation reactions. The PaqFOS1 and the AtFMO1 share 40% amino acid sequence identity (209/518) with a similarity score of 61% (318/518). As suggested by the SEC-elution profile, PaqFOS1 elutes solely as a homodimer. In parallel to PaqFOS1, we also expressed and isolated AtFMO1 and demonstrated that it also elutes as a homodimer (Supplementary Fig. 7). A dimeric FMO protein has previously been isolated and crystallized from the methylotropic bacterium Methylophaga 58 . Most recently, the crystal structures and proposed dimeric arrangement of class B FMOs from multicellular organisms were reported for the pyrrolizidine alkaloid N-oxygenase (ZvPNO) from the Locust grasshopper (Zonocerus variegatus) 59 and for ancestral reconstructed mammalian FMOs 60 . Based on the sequence similarities and crystal structures obtained, and in agreement with the isolation of PaqFOS1 and AtFMO1 as stable homodimers, the tertiary structure of the class B FMOs are predicted to be conserved across the kingdoms of life.
Here, we show that an FMO catalyzing N-hydroxylation of an α-amino acid plays a key role in the convergent evolution of cyanogenic glycoside biosynthesis in ferns and higher plants. In all currently investigated gymnosperms and angiosperms, oxime formation from amino acids is catalyzed by a cytochrome P450 from the CYP79 family 18 . Our study demonstrates that the introduction of cyanogenic glycoside biosynthesis in ferns was based on independent recruitment of a unique class B FMO protein.
In addition to lacking a CYP79, ferns also lack the other cyanogenic glycoside-related CYP families: CYP71, CYP736, and CYP706 (Fig. 3). These families are all members of the large 71 clan, and selected members of these families catalyze conversion of oximes into the cyanohydrin intermediate in seed plants (Fig. 1). This opens speculation on the possible identity of the remaining biosynthetic pathway members in ferns. Based on less than 40% amino acid sequence identity criteria, many of the CYPs identified in the Pa and Paq transcriptomes did not correspond to any previously named and characterized P450 families. This study therefore unmasks a treasure trove of what is to our knowledge new CYP families and possible pathway candidates. Interestingly, other families of the 71 clan are also highly abundant in fern species 61 , and would be possible targets for gene discovery. The diversity of previously unnamed CYP families accentuates convergent evolution of cyanogenic glycoside biosynthesis as well as other metabolic pathways in ferns. Our comparative transcriptomic strategy to identify FOS1 was highly successful, and would provide a robust approach to identify the remaining pathway members.
The evolution of the cyanogenic glycoside pathway is quite dynamic 12 . Recently, the classical three-step pathway was revised, with the discovery that sugar gum (Eucalyptus cladocalyx) harbors four genes that catalyze the conversion of amino acid to  10 . It is therefore also a possibility that more (or less) biosynthetic steps and intermediates might be present in ferns. Further, the cyanogenic glycoside biosynthetic pathway is shown to act as a dynamic metabolon, ensuring channeling of intermediates 62 . Here, Sorghum bicolor metabolons encounter the soluble UGT into tight organization. A likewise orchestration of FOS1 into a membrane-bound complex could be plausible, and indeed FMO from humans are associated with the membrane 50,60,63 . The identification of FOS1 in cyanogenic ferns alters traditional perceptions of the origin of cyanogenic glycoside biosynthesis, and highlights the importance of N-hydroxy amino acids, oximes, and cyanogenic glycosides throughout plant evolution.
At present, only a few plant FMOs have been functionally characterized. These FMOs have been shown to catalyze unique and crucial oxygenation reactions in plant hormone metabolism, pathogen resistance, signaling and chemical defense 32  Metabolite profiling of fern tissue. Plant tissue (~30 mg) was weighed and boiled in 85% methanol (v/v, 300 μL) for 5 min. The vial was transferred to an ice bath and the material macerated with a small pestle. The supernatant obtained after centrifugation (13,000 × g, 1 min) was filtered (0.45 μm low-binding Durapore membrane) and diluted 1:5 in water prior to LC-MS analysis.
RNA isolation and transcriptome mining. Total RNA was prepared from 30 to 50 mg of plant tissue using the Spectrum Plant Total RNA Kit (Sigma-Aldrich, US). Transcriptomes were prepared from mRNA isolated from Pa fiddlehead, Pa young frond, and from two Paq frond tips containing high and low cyanogenic glycoside levels. Transcriptome sequencing was carried out by Macrogen (Seoul, South Korea) using an Illumina HiSeq2000 sequencer (Illumina, San Diego, CA) to generate paired-end libraries. The reads were de novo assembled and relative transcript abundance estimated by Sequentia Biotech (www.sequentiabiotech.com) using the Trinity pipeline 64 .
The expression level of all transcripts was quantified. Identification of gene families was performed using the OrthoMCL pipeline (http://orthomcl.org/ orthomcl/about.do). Differential gene expression analyses based on the Pa and Paq transcriptomes were carried out using eXpress 65 and the data transferred into R and analyzed with the package NOISeq 66 .
Isolation and transient expression of FMO candidate genes. The full-length sequence of the predicted ORF Paq18302 encoding gene (most upstream methionine) without codon optimization was fitted with attb1 and attb2 Gateway cloning sites: attB1: ggggacaagtttgtacaaaaaagcaggct, attB2: ggggaccactttgtacaagaaagctgggt and synthesized by GenScript. Guided by Pa22758, we isolated the full length of the FMO sequence from Pa cDNA using gene specific primers flanked by gateway sites (Forward: ggggaccactttgtacaagaaagctgggtctattcatctttgtagtccatgtta, Reverse: ggggaccactttgtacaagaaagctgggtctattcatctttgtagtccatgtta) and by PCR attached the gateway sites. The fragment was cloned into pUC57. Both constructs were subcloned by Gateway recombination from the pUC57 vector into the expression vector pEAQ3-HT-DEST 67 .
Generation of expression plasmids for paqFOS1 and AtFMO1. The nucleotide sequence of contig Paq18305 and of AtFMO1 from A. thaliana (Q9LMA1) was codon optimized for expression in E. coli fitted with an upstream 6xHis encoding tag and inserted into the expression vector pET-30a(+). To express the recombinant proteins, both pET-30a-PaqFOS1 and pET-30a-AtFMO1 were transformed into Phage-resistant BL21(DE3)-R3 strain (Structural Genomics Consortium Oxford).
Expression and isolation of recombinant PaqFOS1 in E. coli. A single colony from a fresh plate (1.5% LB-agar containing 50 µg/mL kanamycin and 25 µg/mL chloramphenicol) was grown O/N at 37°C with shaking (250 rpm) in 250 mL LB medium supplemented with kanamycin and chloramphenicol. A 4 L expression batch was set up by inoculation each liter of LB medium (supplemented with kanamycin and chloramphenicol) with 20 mL of bacterial culture into four 2.5 L Ultra Yield flasks fitted with AirOtop enhanced seals (Thomson Instrument, Germany). Preinduction cultures were incubated at 37°C and 225 rpm to reach OD 1.2-1.5 before induction with IPTG (final concentration 0.5 mM). The expression culture was grown O/N at 18°C and the E. coli cells sedimented (5000 × g, 10 min) and stored at −20°C until used.
E. coli cells were thawed and resuspended in lysis buffer (1 g cells per 5 mL buffer; 100 mM HEPES pH 7.5, 500 mM NaCl, 1 mM MgSO 4 , 0.5 mM TCEP, Benzonase (25 U/mL; Sigma-Aldrich)) and lysed in a high-pressure homogenizer (Avestin EmulsiFlex D20, 40 psi). The lysate was clarified by centrifugation (11,000 rpm, 40 min), filtered (0.22 μM) and applied to two 5-mL His-Trap FF columns (GE Healthcare) connected in line and preequilibrated with Binding buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 10 mM Imidazole, 0.5 mM TCEP). Columns were washed with ten column volumes of wash buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 30 mM Imidazole, 0.5 mM TCEP) and protein eluted using a ten column volume 0-100% gradient of elution Buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 500 mM Imidazole, 0.5 mM TCEP). The eluate was collected in 1.5 mL fractions in 96-wells plates. Fractions of interest were analyzed by SDS-PAGE using 4-12% NuPAGE gradient protein gels (Thermo Scientific). Fractions containing the target protein were concentrated to a final total volume of 5 mL by centrifugation using preequilibrated membrane filters (30 kDa cut off; Thermo scientific) and applied to a HiLoad 16/60 Superdex 200 column (120 mL, GE Healthcare). Before usage, columns were calibrated using the High molecular weight calibration kit (ranging from 43 to 669 kDa; GE Healthcare) and preequilibrated using gel filtration (GF) buffer (50 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM TCEP), which was also used for eluting the protein. All buffers, except the GF buffer, contained one tablet cOmplete protease inhibitor cocktail pr. Fifty milliliters of buffer (cOmplete Inhibitor, EDTA-free, Roche). The isolated protein was frozen in liquid nitrogen and stored at −80°C.
Recombinant enzyme assays. The activity of the isolated recombinant FOS1 protein was determined in vitro in assay mixtures (total volume: 50 µL) containing 30 µL of diluted FOS1 protein (0.345 mg mL −1 ) reconstituted with FAD (10 µM final concentration) and phenylalanine (50 µM final concentration). Enzyme reaction was initiated by addition of NADPH (final concentration 6 mM). Following incubation (1 h, 30°C, 300 rpm), the reaction was stopped by addition of 100 µL MeOH. Assays without enzyme, substrate, or cofactor served as controls. All assays were carried out in triplicates.
Analysis of recombinant PaqFOS1 and Pa tissue. Enzyme reaction mixtures (50 µL aliquots) were diluted with 50 µL Milli-Q grade water and filtered (Durapore ® 0.22 μm PVDF filter plates, Merck Millipore, Tullagreen, Ireland) together with filtered and diluted MeOH extracts of Pa fiddlehead and young pinna (see above). Samples were chromatographically separated using an Advance UHPLC system (Bruker, Bremen, Germany) fitted with a Zorbax Eclipse XDB-C18 column (100 × 3.0 mm, 1.8 µm, Agilent Technologies, Germany) with a column temperature maintained at 40°C. The mobile phases were: (a) HCOOH (0.05%, v/v); and (b) MeCN in 0.05% (v/v) HCOOH. The gradient elution profile was: 0-0.5 min, isocratic phase 3% B; 0.5-3.8 min, linear gradient 3-70% B; 3.8-4.4 min. linear gradient 70-100% B; 4.4-4.9 min, isocratic phase 100% B, 4.9-5.0 min, linear gradient 100-3% B; 5.0-6.0 min, isocratic phase 3% B using a flow rate of 0.5 mL min −1 . The EVOQ Elite Triplequadrupole mass spectrometer (Bruker, Bremen, Germany) was equipped with an ESI operated in positive mode. The instrument parameters were optimized by infusion experiments with pure standards. The ion spray voltage was maintained at +4000 V, cone temperature was set to 350°C and cone gas to 20 psi. Heated probe temperature was set to 400°C and probe gas flow to 50 psi. Nebulizing gas was set to 60 psi and collision gas to 1.6 mTorr. Nitrogen was used as probe and nebulizing gas, and argon as collision gas. Active exhaust was constantly on. Multiple reaction monitoring (MRM) was used to monitor analyte parent ion → product ion transitions. MRM transitions and collision energies were optimized by direct infusion experiments into the MS (Supplementary Table 4). Both Q1 and Q3 quadrupoles were maintained at unit resolution. Bruker MS Workstation software (Version 8.2.1, Bruker, Bremen, Germany) was used for data acquisition and processing.
Phylogenetic analysis. FMO sequences from angiosperms and mosses were accessed from Phytozome (phytozome.jgi.doe.gov, version 12.1.6). The following eight species were selected to span the evolution of higher plants: Malus domestica, Solanum lycopersicum, Arabidopsis thaliana, Citrus sinensis, Populus trichocarpa, Sorghum bicolor, Oryza sativa, and Selaginella moellendorffii. Their FMO sequences were obtained using BLASTp 2.2.26+ with the A. thaliana FMOs as query sequences resulting in a total of 169 hits. Fern sequences were accessed from the transcriptomes of Pa and Paq reported in the present study (Paq-or PaTRINITY). To obtain robust phylogenetics, only full-length sequences were included. In all cases the ORF and initial methionine were chosen. FMO sequences from conifers were obtained from the databases OneKP (ualberta.ca/onekp) and Congenie (congenie.org) by blast-searching Chamaecyparis hodginsii (four sequences) and Picea abies (four sequences) with Arabidopsis FMOs as query sequences. The functionally characterized S-oxygenating FMO (AsFMO1) from garlic (Allium sativum) was also included 40,70 . Sequence analyses were conducted using MEGA 7.0. All amino acid sequences were manually inspected before being aligned using ClustalW. The phylogenetic relationship was inferred using the maximum likelihood method based on the JTT matrix-based model and n = 100 replicates for bootstrapping. The phylogenetic tree is drawn to scale, not rooted, and with branch lengths measured in the number of substitutions per site. The analysis involved 189 amino acid sequences. The sequence IDs employed in the phylogenetic analysis can be found in Supplementary Data 2.
Small read archive (SRA) searches for FOS1 orthologues. Data from the fern whole genome duplication study at Fudan University were retrieved from NCBI (accession PRJNA422112, date 12 December 2017). The SRA searches were carried out using PaqFOS1 as query. An FMO hit was obtained from all but four of the 119 fern transcriptomes, using a conserved N-terminal fragment as query.
Chemicals. [UL-14 C]-L-phenylalanine (0.25 μCi, specific activity 487 mCi mmol −1 ) was purchased from Perkin-Elmer. N-hydroxyphenylalanine, (E)-and (Z)-phenylacetaldoxime and prunasin were chemically synthesized as previously reported [71][72][73] . Specifically, chemical synthesis of L-(N-hydroxy)phenylalanine is outlined in Supplementary Fig. 8. Vicianin was obtained from a methanol extract of Vicia sativa seeds. Compound validation was based on UV absorption and accurate mass upon LC-MS analysis. Quantification of vicianin was carried out using a dilution series of amygdalin as reference compound, as the diglycosides are structurally comparable and expected to behave similarly with respect to their degree of ionization and UV absorption.
Statistics and reproducibility. The fern species were chosen based on published literature reports of cyanogenic glycoside content, and confirmed by LC-MS analysis. RNA was extracted from identified tissue at least twice, with the highest quality RNA used for downstream transcriptomic analysis. Functional characterization by agroinfiltration in Nicotiana benthamiana plants was repeated in three independent experiments, using two biological N. benthamiana replicates, and three technical replicates each time. Similarly, in vitro enzyme assays in E. coli were repeated in three independent experiments with three technical replicates each time. FOS1 constructs expressed in A. tumefaciens and E. coli were confirmed by sequencing.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The OneKP database is accessible at (ualberta.ca/onekp). The data from 120 SRA experiments are residing at NCBI as the fern whole genome duplication study with accession PRJNA422112. All other data are available in the main text or in the supplementary materials ( Supplementary Data 1-3). Sequence data from this article can be found in the EMBL/GenBank data libraries under accession numbers MT856954 and MT856955 for the fern oxime synthase Paq18302 and Pa22578.

Code availability
Details about software and algorithms used in this study are given in the "Methods" section. No customized code or algorithm deemed central to the conclusion was used.