Discovery of an isoflavone biosynthetic pathway in Iridaceae plants Authors: Yucheng Zhao, Xinyu Shi, Xiang Zhang, Ming-Ju Amy Lyu, Mei Tian, Shuai Li, Lanlan Sun, Yixuan Zhong, Yiqun Dai, Huihui Zeng, Yan Zhu, Guoyong Xie, Jianfang Bai, Xiaojing Ma, Lian-Wen Qi, Minjian Qin Published: 2026 ABSTRACT AbstractIsoflavones, characterized by their distinctive C6-C3-C6 skeleton, represent an important class of plant secondary metabolites with considerable pharmacological and physiological relevance. Although their biosynthetic pathways have been extensively elucidated in Fabaceae species, the molecular mechanisms governing isoflavone production in other plant families remain largely unexplored. Herein, we uncovered an isoflavone biosynthetic pathway in Iridaceae plants. Notably, it involves an uncharacterized CYP736D family isoflavone synthase, a non-CYP450 (iso)flavone 6-hydroxylase, two cation-dependent type I O-methyltransferases, and a newly identified CYP75F isoflavone 3′-hydroxylase. Evolutionary analyses revealed that Iridaceae have retained an ancient and conserved isoflavone biosynthetic pathway, potentially representing a prototypical route in land plants. In contrast, soybean utilizes a specialized pathway unique to Fabaceae. Our findings not only resolve a long-standing gap in our understanding of isoflavone biosynthesis in Iridaceae plants, but also provide insights into the independent evolutionary trajectory of isoflavone biosynthetic mechanisms in Iridaceae and Fabaceae.Subject terms: Secondary metabolism, Plant evolution, Enzymes Introduction Isoflavones constitute an important class of plant secondary metabolites that play dual roles in plant defense and human health1–3. While sharing the fundamental C6-C3-C6 skeleton with other flavonoids (e.g., flavones, flavanols, and flavanones), isoflavones are structurally distinguished by the migration of the B-ring from C2 to C3 of the central C-ring (Supplementary Fig. 1). Through individual or combinatorial structural modifications of rings A, B, and C, plants biosynthesize a wide spectrum of isoflavone derivatives2,4. This structural diversity underlies their multifaceted bioactivities, including antibacterial, anti-inflammatory, antioxidant, antidiabetic, and antitumor properties, as well as estrogenic effects2. Despite their biological significance, research on isoflavone biosynthesis has overwhelmingly focused on Fabaceae, the best-known producers of these compounds5–7. In contrast, Iridaceae, the other major isoflavone-accumulating plant family, has received comparatively little attention, even though it is renowned for its distinctive 6-O-methylated isoflavones and their glycosides (Supplementary Figs. 2, 3)8–10. To date, no reports have elucidated the biosynthesis mechanisms of the isoflavone backbone in Iridaceae, let alone those other non-Fabaceae lineages with low isoflavone contents11–13. Indeed, isoflavone production outside Fabaceae has often been regarded as rare or exceptional, raising the possibility that their biosynthetic logic may diverge fundamentally from the Fabaceae paradigm8. The biosynthesis of isoflavones has been well illustrated in Fabaceae14,15. It shares the same upstream biosynthesis pathway with flavonoids, deriving from the phenylpropanoid pathway. Isotopic labeling experiments in the mid-20th century revealed that isoflavone formation involves oxidation of flavanone, accompanied by aromatic B-ring migration from C2 to C3 (Fig. 1, Supplementary Fig. 1)16–19. By the end of the 20th century, NADPH- and dioxygen-dependent CYP93C family genes known as isoflavone synthase (IFS), or 2-hydroxyisoflavanone synthase (2-HIS), were cloned from liquorice and soybean. These enzymes mark the initiation of carbon flux in the phenylpropanoid pathway from flavone to isoflavone20,21. Over subsequent decades, more than 30 IFSs have been identified, the vast majority of which originate from the Fabaceae family. While two IFS-like genes have been reported in sugar beet, their incomplete open reading frames and unverified functions leave their roles ambiguous14,22–24. The recent main breakthrough is the resolution of the crystal structures of 2-HIS and 2-hydroxyisoflavanone dehydratase (2-HID), which have illuminated the molecular basis of B-ring migration25. Additionally, the discovery of TaCYP71F53, a wheat-specific IFS, provided the first functional evidence of isoflavone biosynthesis in a non-Fabaceae species, suggesting the existence of alternative evolutionary solutions to isoflavone production13. Summary of the predicted isoflavone biosynthetic mechanism in Iridaceae and Fabaceae plants. Isoflavone biosynthesis in Iridaceae and Fabaceae plants shares common upstream steps but diverges starting from the intermediate naringenin. In Fabaceae, IFS belonging to the CYP93C family catalyzes the conversion of naringenin or carthamidin into genistein or 6-hydroxygenistein, respectively. In contrast, Iridaceae utilizes a CYP736D family IFS that specifically converts naringenin to genistein. Red highlights denote the chemical group modified in the reactions. Abbreviations: PAL phenylalanine ammonia lyase; C4H cinnamate 4-hydroxylase; 4CL 4-coumarate: coenzyme A ligase; CHS chalcone synthase; CHI chalcone isomerase; 2-HID 2-hydroxyisoflavanone dehydratase; IFS isoflavone synthase; FAD flavin adenine dinucleotide (FAD)-binding monooxygenase-like proteins; F6H flavonoid 6-hydroxylase; IF6H flavonoid/isoflavone 6-hydroxylase; I6OMT isoflavone 6-O-methyltransferase; I3′H isoflavone 3′-hydroxylase; I7UGT isoflavone 7-UDP-dependent glucosyltransferase; I4′OMT isoflavone 4′-O-methyltransferase; I3′OMT isoflavone 3′-O-methyltransferase; I2′H isoflavone 2′-hydroxylase. Beyond skeleton formation, diversification of isoflavones and flavonoids is usually mediated by hydroxylation, methylation, and glycosylation (Fig. 1). The corresponding enzymes often exhibit broad substrate promiscuity, enabling them to modify multiple isoflavone or flavonoid scaffolds and thereby contribute to metabolic diversity14,23. Although enzymes involved in flavonoid modifications have been extensively characterized, those dedicated specifically to isoflavones remain poorly explored2. For example, no authentic isoflavone 6-hydroxylase (IF6H) enzymes had been identified, and until recently, only a single isoflavone 6-O-methyltransferase (I6OMT) had been functionally validated in soybean14,21,22. Additionally, all identified isoflavone 2′,3′-hydroxylases (IF2′/3′H) belong to the CYP81E family and are restricted to Fabaceae26,27. Most reported enzymes with activity toward isoflavones appear to stem from the broad substrate specificity of general flavone-modifying enzymes rather than true isoflavone specialization28–30. In non-Fabaceae plants, only a handful of tailoring enzymes have been characterized, including a few hydroxylases and O-methyltransferases from wheat, and one UGT706L family glycosyltransferase from Belamcanda chinensis (also known as Iris domestica)11,13,31. Consequently, the molecular mechanisms of isoflavone biosynthesis beyond Fabaceae remain incompletely understood. Guided by the established isoflavone biosynthetic logic in Fabaceae and the predominant chemical profiles of isoflavone in Iridaceae, we hypothesized that Iridaceae employ a distinct biosynthetic strategy, differing not only from Fabaceae but also from Poaceae (Fig. 1)9,14. Integrated with transcriptome mining and in vitro functional verification, we have fully elucidated an isoflavone biosynthetic pathway in Iridaceae. This pathway diverges fundamentally from that previously reported in Fabaceae: it involves an uncharacterized CYP736D family isoflavone synthase, a non-CYP450 (iso)flavone 6-hydroxylase, two cation-dependent type I O-methyltransferases, and a newly identified CYP75F isoflavone 3′-hydroxylase. Furthermore, we found genes encoding upstream enzymes producing the isoflavone precursor, i.e., naringenin, are broadly conserved across land plants, whereas those governing isoflavone skeleton formation and downstream modification show lineage-specific divergence. Evolutionary analyses further suggest that the Iridaceae pathway represents an ancient, possibly ancestral route to isoflavones, whereas the Fabaceae system is a derived specialization. This study thus fills a long-standing gap in understanding isoflavone biosynthesis in non-Fabaceae plants and provides a foundation for engineering routes for these compounds. Results Identification of an IFS converting flavanones to isoflavones To elucidate the biosynthetic pathway of isoflavones in Iridaceae, we first focused on identifying IFS, the key enzyme responsible for catalyzing the migration of the B-ring from the C2 to the C3 position of the phenylchroman skeleton, thereby forming the fundamental structure of isoflavones20. Despite an exhaustive search of genome and transcriptome datasets, we found no homologs of CYP93C or CYP71F, the gene families previously reported to encode IFS in Fabaceae and Poaceae, respectively11,13,20. Considering that isoflavonoids predominantly accumulate in the roots and rhizomes of B. chinensis (Fig. 2a) and that all known IFS enzymes belong to the CYP71 clan, we prioritized CYP71 family members that are highly expressed in underground tissues. From this set, 13 CYP71 candidate genes were initially considered, and 11 of them were successfully cloned into a eukaryotic expression vector (Fig. 2b, Supplementary Figs. 4–6 and Supplementary Table 1). In vitro functional verification indicated that only one candidate (Bch12G003021) converted naringenin to genistein (Fig. 2c, Supplementary Fig. 7). The activity was further confirmed using recombinant Bch12G003021 protein purified from Escherichia coli, with product identity validated by high-resolution mass spectrometry (HR-MS) (Supplementary Figs. 8, 9). Notably, Bch12G003021 also accepted hesperetin, liquiritigenin, and eriodictyol as substrates but showed no activity toward the 6-hydroxylated flavonoid carthamidin (Fig. 2d, e). The substrate preference suggests a distinct biosynthetic mechanism in Iridaceae, whereby the isoflavone backbone is formed prior to hydroxylation (Fig. 1, Supplementary Fig. 10)20. Moreover, no reaction intermediates or byproducts were detected, indicating that Bch12G003021 efficiently catalyzes the single-step conversion of flavanones to isoflavones20. In contrast, IFS enzymes in Fabaceae and Poaceae require a two-step process: 2-hydroxylation (with B-ring migration) to form 2-hydroxyisoflavanone, followed by dehydration catalyzed by 2-HID. (Supplementary Fig. 10). Bch12G003021 exhibited optimal activity at pH 8.5 and 37 °C and localized to the endoplasmic reticulum (ER) (Supplementary Figs. 11, 12). This enzyme contains a conserved heme-binding domain (FGXGRRXCXG) and an N-terminal proline-rich transmembrane region (PPXPXXXXP), both hallmarks of canonical CYP450 enzymes (Supplementary Fig. 13). However, phylogenetic analysis placed Bch12G003021 outside the CYP93C and CYP71F clades, instead assigning it to CYP736D (Supplementary Fig. 4)13,20. Hence, in accordance with the nomenclature guidelines of the Cytochrome P450 Nomenclature Committee (http://drnelson.utmem.edu/CytochromeP450.html, CYP NC), we designated the protein encoded by Bch12G003021 (a member of the CYP736D subfamily) as BcIFS. Functional characterization of BcIFS. a Relative contents of three major isoflavones (tectoridin, iristectorin B, and irigenin) in different tissues of B. chinensis. Data are presented as mean ± SD (error bars). For each tissue, three biologically independent plants were used (n = 3). b Heatmap showing the expression profiles of 13 highly expressed CYP450 genes from the CYP71 subfamily. c HPLC chromatograms from the functional characterization assay of BcIFS. d Panels show the chemical structures of the flavanone substrates tested by BcIFS (liquiritigenin, hesperetin, and eriodictyol). e HPLC chromatograms, MS/MS fragmentation spectra and deduced structures of the products 3b (m/z 279.1586), 4b (m/z 313.2369), and 5b (m/z 289.0700), generated from the catalysis of liquiritigenin, hesperetin, and eriodictyol by BcIFS, respectively. The abbreviations Rt, Rh, St, Lf, Fl, and Fr stand for root, rhizome, stem, leaf, flower, and fruit, respectively. Identification of a non-CYP450 isoflavone 6-hydroxylase Following the discovery of BcIFS, we sought to characterize the enzyme responsible for isoflavone 6-hydroxylation. Initially, we focused on CYP71D family members and other highly expressed CYP450s in roots and rhizomes, but none exhibited detectable IF6H activity (data not shown). Given the abundant accumulation of 6-O-hydroxylated isoflavonoids in underground tissues of B. chinensis (Fig. 2a, Supplementary Fig. 3), we hypothesized that the corresponding enzyme might be enriched in these organs. We therefore assayed the IF6H/F6H activity using crude protein extracted from roots and rhizomes of B. chinensis. As expected, the crude enzyme converted genistein and apigenin into their respective 6-hydroxylated products (Supplementary Fig. 14). Interestingly, enzymatic activity persisted when NADPH was replaced with NADH (Supplementary Fig. 15), and it remained unaffected by classic P450 inhibitors such as clotrimazole and ketoconazole (Supplementary Fig. 16). These results collectively indicated that the IF6H in B. chinensis is not a canonical CYP450 protein. We thus expanded our search to non-CYP450 hydroxylase families, including 2-oxoglutarate-dependent dioxygenases (2-OGDDs), pheophorbide a oxygenase (PAO), and flavin adenine dinucleotide (FAD)-binding monooxygenase-like proteins, classes previously implicated in plant oxidative metabolism32–34. NAD(P)H dependence ruled out 2-OGDDs, which require α-ketoglutarate, Fe²⁺, and ascorbate for activity32. Two PAO candidates (Bch11G001862 and Bch16G000119) were then heterologously expressed and purified, but neither catalyzed 6-hydroxylation of genistein (Supplementary Fig. 17). We therefore focused on FAD-binding monooxygenases. Addition of 0.1 mM FAD—a cofactor essential for electron transfer—to the rhizome extracts marginally increased 6-hydroxygenistein production (Supplementary Fig. 18), indicating the presence of active FAD-binding monooxygenases in the extracts34. From an initial set of 117 FAD-binding monooxygenase genes (Supplementary Fig. 19), we selected nine candidates showing strong co-expression with BcIFS (Pearson correlation coefficient > 0.7) for functional testing (Fig. 3a). Among these, only Bch05G001280 efficiently converted genistein to 6-hydroxygenistein (Fig. 3b, Supplementary Fig. 20). This activity was further confirmed using Bch05G001280 protein purified from E. coli, which catalyzed the near-complete conversion of both genistein and apigenin within 30 min (Fig. 3c, d, Supplementary Fig. 21 and Supplementary Table 2). The enzyme exhibited optimal activity at pH 8.0 and 37 °C, and shared ER subcellular localization with IFS (Supplementary Fig. 11, Supplementary Figs. 22–25). Although Bch05G001280 displayed broad substrate specificity, it showed a strong preference for isoflavones over flavones or flavanones, confirming its functional identity as a dedicated IF6H (Fig. 3e, Supplementary Figs. 26–36). We thus designated this enzyme BcIF6H as a non-CYP450 isoflavone 6-hydroxylase in plants. Functional characterization of BcIF6H. a Identification of candidate IF6H transcripts through correlation analysis with BcIFS expression. b Extracted ion chromatogram (EIC) of the nine successfully cloned candidate genes. c, d HPLC analysis of reaction products from assays using purified BcIF6H with genistein (c) or apigenin (d) as the substrate. e Substrate specificity of BcIF6H. Isoflavones, flavones, flavanones, and chalcones are shown in red, blue, orange, and green, respectively. Error bars represent mean ± SD of three independent samples (n = 3). All the structures of the tested compounds are shown on the right with corresponding colors. The abbreviations Rt, Rh, St, Lf, Fl, and Fr stand for root, rhizome, stem, leaf, flower, and fruit, respectively. We further investigated the cofactor requirements of BcIF6H35,36. The enzyme utilized both NADPH and NADH as electron donors but exhibited higher affinity for NADPH (Supplementary Table 2, Supplementary Fig. 15). Similarly, while ferredoxin (Fdx) and FAD supported the most efficient electron transfer, ferredoxin-NADP⁺ oxidoreductase (FNR) could also serve as a functional coenzyme (Supplementary Fig. 37). Interestingly, enzymatic activity of IF6H remained even in the absence of FAD (Supplementary Fig. 37). All these results highlight the distinctive biochemical properties of a non-P450 isoflavone 6-hydroxylase identified in plants—it belongs to the FAD-binding monooxygenase, it does not exhibit strict dependence on FAD—underscoring its functional divergence from canonical members of this enzyme class. Identification of the enzymes mediating structural modification of isoflavones In B. chinensis and some other Iridaceae plants, the majority of isoflavones are characterized by a methoxy group at the C-6 position, a defining chemical feature that distinguishes them from Fabaceae-derived isoflavones1,2. This modification is catalyzed by isoflavonoid 6-O-methyltransferase (I6OMT), a key tailoring enzyme in the pathway (Fig. 4a). To identify potential I6OMTs in B. chinensis, we first considered the previously reported soybean enzyme GmIOMT1 as a reference guidance. This enzyme belongs to the cation-dependent type I O-methyltransferases within the Methyltransf_3 (PF01596) family, a conserved domain found in many plant OMTs28. Using the Hidden Markov Model (HMM) profile for Methyltransf_3 (PF01596) from the Pfam database (https://pfam.xfam.org/browse), we identified 12 candidate OMT genes from the B. chinensis transcriptome (E-value < 10−5; Supplementary Table 3). Four highly expressed candidates, i.e., Bch15G001374, Bch13G000062, Bch11G000682, and Bch02G001419 (Supplementary Fig. 38, Supplementary Table 3), were cloned for functional validation. In vitro assays showed that Bch15G001374 (BcI6OMT1) and Bch13G000062 (BcI6OMT2) efficiently converted 6-hydroxygenistein to tectorigenin, with optimal activity at pH 7.0 (Fig. 4, Supplementary Figs. 39, 40). While BcI6OMT2 maintained robust activity across a broad temperature range (30–50 °C), BcI6OMT1 was more thermosensitive and performed optimally at 35 °C (Supplementary Fig. 40). Both enzymes also catalyzed the methylation of flavonoids, particularly those bearing a 6-hydroxyl group, demonstrating broad substrate promiscuity (Supplementary Figs. 41–45). But for isoflavones, the presence of a 6-hydroxyl group seems essential for their activities (Supplementary Fig. 41). Functional characterization of the enzymes involved in structural modification. a Proposed biosynthetic pathway from 6-hydroxygenistein to iristectorigenin A, sequentially catalyzed by I6OMT, I3’H, and I3’OMT. b HPLC analysis of reaction products from assays using BcI6OMT, BcI3’H, and BcI3’OMT with 6-hydroxygenistein, tectorigenin, and irilin D as their respective substrates. c MS/MS spectra of the enzymatic products generated by BcI6OMT, BcI3’H, and BcI3’OMT. Next, we investigated B-ring hydroxylation, a critical step that diversifies isoflavone substitution patterns. In Fabaceae, enzymes such as flavone 3′-hydroxylase (F3′H), flavone 3′,5′-hydroxylase (F3′5′H), and isoflavone 2′/3′-hydroxylase (I2′H/I3′H) have been well characterized and primarily assigned to the CYP81E and CYP75 families26,27,37–39. Guided by this precedent, we selected 12 CYP81 and CYP75 homologs from B. chinensis as I3′H candidates (Supplementary Table 4). Six highly expressed candidates were cloned for functional screening (Supplementary Table 4, Supplementary Fig. 46). Among these, Bch04G000598 utilized isoflavones (genistein, tectorigenin, and iristectorigenin) as substrates (Fig. 4, Supplementary Figs. 47–50), whereas Bch04G000599 only accepted flavanones and flavones (e.g., naringenin and apigenin; Supplementary Figs. 51–54). We thus designated Bch04G000598 as BcI3′H and Bch04G000599 as BcF3′H. Notably, while I3′Hs in Fabaceae belong to the CYP81E family26, BcI3′H from B. chinensis was classified into the CYP75F family by CYP NC, highlighting another divergence in Iridaceae isoflavone metabolism. The final steps in isoflavone biosynthesis involve O-methylation or glycosylation15,40. We previously identified six UGTs as (iso)flavone 7- and 3′-O-glycosyltransferases responsible for synthesizing major isoflavones such as tectoridin, irigenin, and iridin in B. chinensis11. To identify the isoflavone 3′-O-methyltransferase (I3′OMT), we employed a Pfam-based search for proteins containing three conserved domains: OMT (PF00891), OMT dimerization (PF08100), and AdoMet-dependent proline dimethyltransferase (PF05891) (Supplementary Fig. 55)41–43. This approach yielded nine transcripts in B. chinensis harboring all three motifs (Supplementary Table 5). Three highly expressed candidates were selected for functional validation using 3′-hydroxy tectorigenin as the substrate (Supplementary Fig. 56). Notably, Bch01G002312 (designated BcI3′OMT), demonstrated catalytic activity toward 3′-hydroxy tectorigenin, yielding a product with an [M-H]⁻ ion at m/z 329.0659 (Fig. 4). BcI3′OMT also methylated various flavonoids and flavanones, consistent with the typical substrate flexibility of OMTs (Supplementary Figs. 57–62)41–43. Interestingly, like BcF3′H, BcI3′OMT also showed low and non-tissue-specific expression patterns (Supplementary Figs. 63, 64), suggesting they may function under specific developmental or environmental cues rather than constitutive biosynthesis. To this end, we cloned five key genes—BcIFS, BcIF6H, BcI6OMT1, BcI3′H, and BcI3′OMT—into plant expression vector PEAQ-HT and transiently co-expressed them in Nicotiana benthamiana leaves via Agrobacterium-mediated infiltration. LC-MS analysis of the co-expressing leaves not only detected pathway intermediates (including genistein, 6-hydroxygenistein, etc.) but also clearly identified the final product iristectorigenin A (Supplementary Fig. 65). These results confirm that the five enzymes are sufficient to catalyze the complete cascade from naringenin to iristectorigenin A, achieving heterologous biosynthesis of this isoflavone in N. benthamiana. Phylogenetic analysis revealed a more ancient origin of the isoflavonoid biosynthetic pathway in Iridaceae compared to Fabaceae Functional analyses revealed striking divergence in isoflavonoid biosynthesis between Iridaceae and Fabaceae. To determine whether their flavanone precursors are synthesized via conserved or distinct upstream pathways, we investigated key enzymes of the general phenylpropanoid–flavonoid cascade: phenylalanine ammonia-lyase (PAL), cinnamate 4-hydroxylase (C4H), 4-coumarate-CoA ligase (4CL), chalcone synthase (CHS), and chalcone isomerase (CHI). Using the corresponding Fabaceae homologous genes as templates, we cloned a series of genes. Enzymatic activity assays confirmed the functionality of one PAL, two C4Hs, three 4CLs, one CHS, and one CHI (Supplementary Fig. 66). Notably, nearly all functional genes were highly expressed in underground tissues across six organs examined (Supplementary Fig. 66). These results demonstrate that Iridaceae and Fabaceae share a conserved biosynthetic pathway for the flavonoid backbone formation, with divergence occurring specifically at the stage of isoflavone formation. To trace the evolutionary origins of these isoflavone biosynthetic enzymes, we conducted a comparative phylogenomic analysis in 15 representative plant species spanning major evolutionary transitions, from aquatic algae to subaerial algae, bryophytes, gymnosperms and angiosperms. Our analysis revealed that genes encoding enzymes catalyzing the first five steps of phenylpropanoid metabolism, i.e., PAL, C4H, 4CL, CHS, and CHI, as well as one later structural modification enzyme (I3′OMT), are members of orthologous groups that shared between Iridaceae and Fabaceae (Supplementary Fig. 67). With the exception of I3′OMT, all of these genes originated in subaerial algae or liverworts, supporting the hypothesis that this core pathway has an ancient evolutionary origin. In contrast, genes encoding IFS, IF6H, I6OMT, and I3′H in B. chinensis and soybean are derived from distinct orthologous groups (Supplementary Fig. 68, Fig. 5b, c). Similarly, the recently characterized wheat IFS-like gene (TaCYP71F53) forms a distinct clade separate from soybean or B. chinensis IFS (Fig. 5d). Interestingly, the functional enzymes in B. chinensis exhibited more ancient evolutionary origins than those in soybean. In B. chinensis, IF6H traced back to the origin of subaerial algae approximately 580 million years ago (mya), and I6OMT to the last common ancestor of green plants around 1000 mya. In contrast, their soybean counterparts originated much more recently, i.e., in the recent common ancestor of eudicots and monocots (~150 mya; Fig. 5e). Moreover, gene copy numbers for these enzymes have remained relatively stable throughout the evolution of B. chinensis, while soybean displayed extensive lineage-specific expansion driven by recent tandem duplications (Supplementary Figs. 67, 68). Evolutionary analysis of enzymes in the isoflavone biosynthetic pathway. a Phylogenetic tree of genes encoding the enzyme PAL. Functional PAL genes from B. chinensis and soybean (marked with red stars) cluster within the same orthologous group, which originated in liverwort. b–d Phylogenetic tree of genes encoding enzyme IFS. Functional IFS genes from B. chinensis (indicated as red star in (b)), soybean (indicated as red star in (c)) and wheat (indicated as red star in (d)) belong to distinct orthologous groups. The B. chinensis IFS genes originated from the recent common ancestor of eudicots and monocots, whereas the soybean IFS is specific to the Fabaceae family (Supplementary Fig. 69a), and wheat IFS-like genes are specific to Pooideae (Supplementary Fig. 69b). All orthologous groups were predicted genome-wide using protein sequences, and phylogenetic trees of genes were constructed from protein sequence alignments. e Summary of the evolutionary origins of isoflavone pathway enzymes in B. chinensis, soybean and wheat across 15 representative plant species, spanning from aquatic algae (C. reinhardtii) to subaerial algae, bryophytes, gymnosperms, and angiosperms. Genes labeled in black and red besides the table indicate shared and divergent origins between B. chinensis and soybean, respectively. The phylogenetic relationships and divergent time for the 15 species were retrieved from the TimeTree database (https://timetree.org/home). Clades containing genes from different species in (a, b) are color-coded. For visual simplicity, only clades containing three or more genes from the same species are labeled with distinct colors. List of species: B. chinensis: Belamcanda chinensis, Garlic: Allium sativum, Asparagus: Asparagus officinalis, Rice: Oryza sativa, Wheat: Triticum aestivum, Cherry: Prunus avium, Ginkgo: Ginkgo biloba, Soybean: Glycine max, Arabidopsis: Arabidopsis thaliana, Amborella: Amborella trichopoda, Liverwort: Marchantia polymorpha, M. endlicherianum: Mesotaenium endlicherianum, C. reinhardtii: Chlamydomonas reinhardtii. Created in BioRender. Lyu, M. (2026) https://BioRender.com/sikx26q. Notably, no orthologs of soybean IFS or wheat IFS-like were detected in any of the other 14 species examined. To further determine the taxonomic distribution of both soybean IFS and wheat IFS-like genes, we performed homology searches against the NCBI non-redundant protein database (https://www.ncbi.nlm.nih.gov) using stringent criteria (E-value ≤ 1e−5, protein identity > 60%). The results confirmed that soybean IFS orthologs were widely present in other Fabaceae species, such as Vigna radiata (mung bean), Trifolium pratense (red clover), Pisum sativum (pea), and Medicago sativa (alfalfa), among others (Supplementary Fig. 69a), whereas few orthologs were identified outside the Fabaceae, indicating that the soybean IFS originated within the Fabaceae lineage around 100 mya (Fig. 5e). On the other hand, wheat IFS-like homologs are exclusively found within the Pooideae subfamily (including Triticum, Aegilops, and Hordeum species), with no orthologs detected in other Poaceae (e.g., rice and maize), Fabaceae or Iridaceae (Supplementary Fig. 69b), suggesting that wheat IFS-like originated within Pooideae around 70 mya (Fig. 5e). In contrast, B. chinensis utilizes more ancient IFS genes that originated in the recent common ancestor of eudicots and monocots around 150 mya. Therefore, our findings also provide evidence for the independent evolutionary recruitment of distinct enzymes to catalyze isoflavone biosynthesis in different plant lineages. Discussion Isoflavones are widely distributed in numerous taxonomically distant species and at least 60 plant families reported to contain these compounds8. However, the biosynthetic mechanisms of isoflavones have remained elusive for decades. Since the first IFS gene was cloned from soybean, functional characterization of the pathway has been largely confined to Fabaceae and, to a lesser extent, Poaceae13,20. Consequently, isoflavone biosynthesis in non-Fabaceae taxa has long been regarded as a rarity rather than a general rule8,13,20. Our findings challenge this prevailing assumption. Phylogenomic and functional evidence suggest that the Fabaceae-specific isoflavone biosynthetic pathway, which centers on CYP93C-type IFS, may represent a derived evolutionary innovation. In contrast, the pathway in Iridaceae appears to reflect an ancient mechanism. This raises the possibility that isoflavone biosynthesis may have originated earlier in land plant evolution than previously recognized (Fig. 5, Supplementary Figs. 67–69). However, its presence and underlying mechanisms in other lineages remain to be experimentally verified. This conclusion is supported by three observations: (i) all currently identified CYP93C-type IFS genes are Fabaceae-specific, with no orthologs detected in other species; (ii) most genes involved in isoflavone biosynthesis in Fabaceae originated more recently during land plant evolution compared to those utilized by Iridaceae; and (iii) Fabaceae exhibits extensive, lineage-specific expansion of isoflavone biosynthesis genes through recent tandem duplications, whereas Iridaceae maintains stable, conserved copy numbers across evolutionary time (Fig. 5, Supplementary Figs. 67–69). On the other hand, the recently reported wheat IFS-like gene is restricted to Pooideae, with no orthologs detected in other Poaceae species (Fig. 5, Supplementary Fig. 69). Collectively, these findings suggest that Iridaceae employs an isoflavone biosynthetic route that is distinct from the well-characterized Fabaceae pathway or recently reported wheat pathway, and that this route involves a CYP93C-family IFS enzyme closely related to canonical C4Hs. This raises the intriguing possibility that isoflavone biosynthesis may have evolved earlier in angiosperm history than previously assumed. However, without functional validation of IFS-like genes in additional non-Fabaceae lineages, it remains uncertain whether this represents an ancestral trait that was lost in most plant families or a case of convergent evolution. Further investigation of isoflavone biosynthesis across diverse taxa is essential to distinguish between these scenarios. Supporting this view, homology-based gene cloning revealed that key genes involved in isoflavone biosynthesis are highly conserved across three representative Iridaceae species: B. chinensis, Iris germanica and Iris tectorum (Supplementary Figs. 3, 70, Fig. 2c). When integrated with existing knowledge from Fabaceae, it becomes evident that the isoflavone biosynthetic pathways in Iridaceae and Fabaceae are not only biochemically distinct but also rooted in separate genetic and evolutionary trajectories20,21. However, several important questions remain: What is the situation in other families within Liliidae (which includes Iridaceae) and Rosidae (which includes Fabaceae)? How do these pathways differ between monocots and dicots, and do they exist at all in more ancient lineages such as Bryopsida and Pinopsida? Addressing these questions will require further clarification, particularly through the elucidation of isoflavone biosynthetic mechanisms across a broader phylogenetic spectrum, including orders such as Rosales, Bryales, Pinales, Juncales, and others8. In Fabaceae, IFS activity has long been attributed exclusively to members of the CYP93C subfamily32. In this study, however, we found that IFSs from three examined Iridaceae species all belong to a previously uncharacterized P450 family: CYP736D (Supplementary Fig. 70). This discovery exemplifies a recurring theme in plant specialized metabolism: the convergent evolution of distinct enzymes to catalyze identical biochemical transformations44,45. Surprisingly, CYP736D and CYP93C do not merely represent functional analogs; they appear to channel substrates through divergent biosynthetic routes toward isoflavone formation. For instance, CYP736D cannot utilize 6-hydroxylated flavonoids as substrates, whereas CYP93C accepts both hydroxylated and non-hydroxylated flavonoids (Figs. 1, 2, Supplementary Fig. 10)21,28. Furthermore, the two enzymes exhibit distinct substrate preferences: CYP736D shows the highest activity with naringenin, while CYP93C favors liquiritigenin (Fig. 2)20,21. These differences most likely stem from the distinct substrate specificities of the two enzymes, and potentially, from variations in their protein sequence, protein structure, and catalytic mechanisms16,18,46–48. As shown in Supplementary Fig. 13, BcIFS shares only 30.21% protein sequence identity with the IFS from T. aestivum (UPL51477.1) and approximately 23.45% identity with the IFS from Medicago truncatula (RHN62339.1). At the protein structure level, BcIFS also exhibits an exposed, distinct α helix compared to CYP93C and CYP71F (Supplementary Fig. 72a). In CYP93C, the catalytic mechanism has been well established for the detection of the key intermediate 2-hydroxyflavanone, the byproduct 3-hydroxyflavanone, and isotope labeling experiments16,46. It involves an oxidative rearrangement of flavanone to isoflavone, proceeding through 2-hydroxylation, 2,3-migration of the B ring, and a dehydration step (Supplementary Fig. 71)16. Moreover, mutagenesis and structural analyses identified Ser310 and Lys375 in CYP93C as active-site residues critical for the unusual aryl migration of the flavanone substrates46,47. However, we did not detect any intermediates or byproducts in the reactions catalyzed by BcIFS (CYP736D). Moreover, sequence alignment revealed no conserved serine or lysine residues corresponding to Ser310 and Lys375 of CYP93C in CYP736D enzyme (Supplementary Fig. 13). Instead, molecular docking simulations of CYP736D with naringenin implicated Lys222, Thr330, Ser456 as key residues for substrate binding; mutation of either residue completely abolished its activity (Supplementary Fig. 72b). Although high-resolution structural data or theoretical calculations are still needed to definitively resolve the catalytic trajectory of CYP736D, our results suggest that the catalytic mechanisms of CYP736D differ from CYP93C, at least in terms of key catalytic residues. Ongoing efforts to obtain crystal structures of BcIFS will be crucial to elucidate whether this enzyme employs a modified rearrangement pathway or an entirely distinct chemical strategy to achieve the same net transformation. To date, nearly all characterized F6Hs belong to the P450 family29,49. Here, we reported the discovery of a non-CYP IF6H in B. chinensis, which was characterized as the FAD family protein. Phylogenetically, BcIF6H clusters within the FAD-binding monooxygenase family but forms a distinct clade separate from other known flavonoid oxygenases (Supplementary Fig. 73). Although BcIF6H retains canonical FAD-binding motifs, including the classical N-terminal “GxGxxG”, as well as conserved “DG” and “GD” sequences, commonly found in FAD protein33,50,51, all cloned Iridaceae IF6H sequences carry a glycine-to-serine (G → S) substitution within the “GxGxxG” motif (Supplementary Fig. 74). We initially hypothesized that this substitution may underlie enzyme’s unexpected FAD-independent activity (Supplementary Fig. 37). However, site-directed mutagenesis reverting serine back to glycine (S → G) in BcIF6H did not restore strict FAD dependence (Supplementary Fig. 75), implying that structural or conformational features beyond the primary motif are responsible for the enzyme's FAD independence. An alternative explanation, proposed for the flavone 8-hydroxylase LjF8H in Lotus japonicus, is that the enzyme binds FAD tightly in yeast, making the exogenous FAD supplementation unnecessary33. To test this possibility, we expressed BcIF6H in E. coli and purified it using His-tag system, yet the purified protein still exhibited robust activity without adding FAD (Supplementary Fig. 37). Further, we used high-resolution mass spectrometry to analyze the above reaction mixture, and no FAD was detected in the group without FAD addition (Supplementary Fig. 76). This consistent lack of detection suggests that BcIF6H may not require FAD for its catalytic function, implying that the FAD-binding domain may play a structural or evolutionary role rather than a catalytic one. Additionally, BcIF6H was able to utilize FNR as an electron donor, though with lower efficiency compared to Fdx or FAD (Supplementary Fig. 37). This functional flexibility aligns with the established roles of Fdx and FNR as core components of photosynthetic electron transport chains in chloroplast36,52,53. This type of iron-sulfur (Fe-S) cycle may represent an ancient evolutionary adaptation to low oxygen conditions prevalent during early eukaryotic evolution52,54–56. Consistent with this, molecular dating places the origin of BcIF6H at approximately 1000 mya, a period coinciding with the rise of eukaryotic life under hypoxic environments (Fig. 5d)52. Moreover, primary sequence analyses reveal that BcIF6H shares greater similarity with prokaryotic FAD-binding proteins than with canonical plant oxygenases (Supplementary Figs. 73, 74), further supporting a deep evolutionary origin. Collectively, these findings imply that the origin of isoflavone biosynthesis may be much older than previously thought, and could have originated during the transition of plants from aquatic to terrestrial habitats. Isoflavones are predominantly associated with Fabaceae and Iridaceae, given the ecological significance of isoflavones in mediating plant-soil microbe interactions, particularly in plant invasions and nitrogen-fixing symbioses57–59. Future work should explore whether this ancient pathway was co-opted independently in different lineages to support niche specialization in terrestrial environments. In this study, we elucidate a complete isoflavone biosynthetic pathway in Iridaceae plants. This pathway is fundamentally distinct from that observed in Fabaceae; it includes an IFS belonging to the CYP736D family, rather than the CYP93C family; a non-CYP450 (iso)flavone 6-hydroxylase, which belongs to the FAD family but exhibits FAD-independent activity; a type I I3’OMT; and an I3’H assigned to the CYP75F family. Furthermore, we demonstrate that the isoflavone biosynthesis pathway in soybean is restricted to the Fabaceae lineage, whereas the pathway in Iridaceae traces back to a more ancient and conserved origin, potentially representing an ancestral mechanism present in early land plants. This work fills a long-standing knowledge gap regarding isoflavone biosynthesis in non-Fabaceae plants and suggests multiple evolutionary origins for isoflavones within the plant kingdom. Methods Plant materials and reagents B. chinensis and other Iridaceae species were collected from the Botanic Garden of China Pharmaceutical University, Nanjing, China (31°53′N, 118°54′E). Genomic DNA was extracted using the Plant Genomic DNA Kit (TIANGEN, Beijing, China), while total RNA was isolated with the RNAsimple Total RNA Kit (TIANGEN). Subsequently, a cDNA library was constructed using the HiScript III 1st Strand cDNA Synthesis Kit (Vazyme, Nanjing, China). All solvents used for extraction and biochemical analyses were of HPLC/MS grade. The compounds and antibiotics used in the enzymatic reaction and microbial culture were purchased from Herbest (Baoji, Shanxi, China) and Sigma-Aldrich (Shanghai, China). Enzymes and molecular biology kits were obtained from Vazyme, TransGen Biotech (Beijing, China), and Beyotime (Shanghai, China). Crude protein preparation and enzyme activity assays Crude proteins were extracted from the rhizome and root of B. chinensis separately. Briefly, the tissues were ground into a fine powder in liquid nitrogen and homogenized in 100 mM PBS (pH 8.0) using a buffer-to-tissue ratio of 5:1 (v/m). The homogenate was stirred at 4 °C for 4 h and then centrifuged at 3580 × g for 10 min to remove debris. The resulting supernatant was further centrifuged at 25,000 × g for 2 h at 4 °C. The pellets obtained were resuspended in ice-cold PBS for subsequent analysis. Protein concentration was determined using the BCA Protein Assay Kit (Beyotime, Shanghai, China). Enzyme assays were conducted in a 200 μL reaction mixture containing assay buffer (50 mM potassium phosphate, 200 mM NaCl, pH 7.5), 50 μg of crude protein, 2 mM NADPH, 1 mM DTT, and 100 μM of different substrates (e.g., naringenin, apigenin, or genistein). After incubation at 25 °C for 12 h, the reactions were stopped by adding 400 μL of ethyl acetate. The organic phase was collected, evaporated to dryness, and the residue was reconstituted in methanol for HPLC analysis. Reaction products were analyzed using an Agilent 1100 HPLC system equipped with an SB C18 column (5 μm, 250 × 4.6 mm; Agilent Technologies) and a UV detector set at 340 nm. The mobile phase consisted of 0.1% formic acid (solvent A) and methanol (solvent B). Elution was carried out with a linear gradient as follows: 20% B to 30% B over 5 min, 30% B to 50% B from 5 to 15 min, and 50% B to 60% B from 15 to 25 min, at a flow rate of 1 mL/min. The heat-inactivated enzyme reaction served as a negative control. Heterologous expression of candidate proteins in E. coli and in vitro activity assays The target genes were cloned into various prokaryotic expression vectors, as summarized in the table below. All recombinant plasmids were then transformed into E. coli BL21(DE3) for protein expression under optimized conditions. Using IF6H as an example, the engineered strains were first cultured at 37 °C for 3–5 h and subsequently induced with 0.1 mM isopropyl β-D-1-thiogalactoside (IPTG). Protein expression was then carried out at 18 °C for 24 h with shaking at 120 rpm. After cultivation, the cells were harvested by centrifugation, and the pellets were resuspended in PBS buffer for protein extraction via ultrasonication. The recombinant proteins were purified using Ni-NTA affinity chromatography, with protein purity and concentration assessed by SDS-PAGE and the BCA Protein Assay Kit (Beyotime, Shanghai, China), respectively. The purified proteins were subsequently subjected to activity analysis under the optimized reaction conditions as detailed in Supplementary Table 7. Unless otherwise specified, all enzymatic reactions were performed in 200 μL of PBS (pH 7.5) at 25 °C for 30 min. The reactions were quenched by adding 400 μL of ethyl acetate. The mixtures were extracted three times with ethyl acetate, and the combined organic layers were evaporated under reduced pressure. The resulting residues were redissolved in methanol and analyzed by high-performance liquid chromatography (HPLC) using an Agilent 1100 HPLC system equipped with an SB C18 column (5 μm, 250 × 4.6 mm; Agilent Technologies). Temperature Dependence: The optimal reaction temperature was assessed at 4 °C, 16 °C, 25 °C, 37 °C, 45 °C, 50 °C, and 65 °C. pH Dependence: The influence of pH on enzyme activity was evaluated across a broad range (pH 2.5, 3.5, 6.5, 7.0, 7.5, 8.0, 8.5, 9.0, 9.5, 10.0, and 11.0). Metal Ion Effects: The impact of various divalent and trivalent metal ions (Cu2+, Mg2+, Zn2+, Ca2+, Mn2+, Ni2+, Co2+, Ag+, Fe2+, and Fe3+) on catalytic activity was examined. Heterologous expression of CYP450 proteins in S. cerevisiae and in vitro activity assays Recombinant CYP450 proteins were heterologously expressed in S. cerevisiae WAT11. The yeast expression vector pYES2, harboring the candidate CYP450 genes, was first transformed into the yeast strain and selected on synthetic dropout medium (SD-URA) plates containing 20 g/L glucose. Single colonies were inoculated into 5 mL of liquid SD-URA medium and cultured overnight at 28 °C. When the OD600 of the engineered strains reached 2–3, cells were harvested by centrifugation, and the pellets were washed twice with sterile water to remove residual glucose. For protein induction, the cells were resuspended in SG-URA medium supplemented with 20 g/L galactose to initiate whole-cell catalysis. Substrates (e.g., naringenin, apigenin, and genistein at 50 μM) were added to the induced cultures and incubated for 24 h. Subsequently, the culture was extracted twice with ethyl acetate, and the organic phase was evaporated to dryness under vacuum. The residue was then redissolved in 100 μL methanol for HPLC analysis. For specific activity assays: IFS activity was tested with 100 μM naringenin as the substrate, and product formation was monitored at 270 nm; In addition, to assess F6H activity, 10 μM genistein was used as the substrate, requiring BcFdx co-expression or FAD supplementation for its activity; I3’H activity was assessed using 10 μM of various substrates (naringenin, apigenin, genistein, iristectorigenin A, and tectorigenin). Gene expression and subcellular location analysis Total RNA was extracted from various organs of B. chinensis following our previously established protocol12. RNA quality and concentration were assessed using both a NanoDrop Spectrophotometer (Eppendorf, Germany) and 1% agarose gel electrophoresis. First-strand cDNA was synthesized from qualified RNA samples using HiScript II Q RT SuperMix (Vazyme, Nanjing, China) for subsequent quantitative real-time PCR (qRT-PCR) analysis. qPCR reactions were performed in a 20 μL reaction volume containing ChamQ SYBR qPCR Master Mix (Vazyme, Nanjing, China) under optimized cycling conditions: initial denaturation at 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 30 s. Gene expression levels were calculated using the comparative ΔΔCT method60. All primer pairs (Supplementary Table 6) were validated by melting curve analysis and efficiency testing through standard curve construction (slope analysis). For subcellular localization studies, BcIFS and BcIF6H were cloned into the plant expression vector pAN580-GFP, which contains a strong GFP tag. Transient expression in plant cells was achieved through PEG-mediated transformation, allowing visualization of fluorescent protein localization61. Transcriptome sequencing and transcript assembly of Iridaceae species Transcriptome analysis was performed on three Iridaceae species: B. chinensis, I. tectorum, and I. germanica. Total RNA was extracted from plant tissues and used to prepare sequencing libraries with the NEBNext® Ultra™ RNA Library Prep Kit for Illumina® (NEB, USA). High-throughput sequencing was conducted on the Illumina HiSeq 2000 platform (Illumina, San Diego, CA, USA) to generate 150 bp paired-end reads. Raw sequencing data underwent rigorous quality control processing, including removal of adapter sequences, trimming of poly-N sequences, and elimination of low-quality reads (Q-score < 20). The resulting high-quality clean reads were de novo assembled into contigs using Trinity software (v2.8.4) with default parameters62. Functional annotation of the assembled transcripts was performed through comprehensive database searches against NR (Non-Redundant Protein Database), Pfam (Protein Family Database), KOG/COG (Clusters of Orthologous Groups), eggNOG (Evolutionary Genealogy of Genes), Swiss-Prot (Curated Protein Sequence Database), KEGG (Kyoto Encyclopedia of Genes and Genomes) and GO (Gene Ontology) database. Evolutionary analysis of isoflavone biosynthetic pathway enzymes To trace the evolutionary trajectory of isoflavone biosynthetic enzymes across green plants, we conducted comparative genomic analyses spanning from aquatic algae to angiosperms. Our study encompassed 13 species with available genome sequences, representing key evolutionary transitions in plant territorialization and diversification. Specifically, the protein sequences of Chlamydomonas reinhardtii (aquatic green algae) (v5.5), Marchantia polymorpha (liverwort, v1), Amborella trichopoda (AMTR1.0), Arabidopsis thaliana (Tair10), Glycine max (soybean, v2.1), and Asparagus officinalis (v1) were downloaded from Ensembl plant (https://plants.ensembl.org/index.html), that of Mesotaenium endlicherianum (subaerial algae) and Allium sativum (garlic, v1) was downloaded from figshare (https://figshare.com/articles/dataset/Genomes_of_subaerial_Zygnematophyceae_provide_insights_into_land_plant_evolution/9911876/1)63 and (https://figshare.com/articles/dataset/garlic_genome_and_transcriptome_resources/12570947/1)64, respectively, that of Oryza sativa (rice, v2.1) was downloaded from Phytozome 13 (https://phytozome-next.jgi.doe.gov), and that of Prunus avium (cherry, v1.0.a1) was downloaded from GDR (https://www.rosaceae.org). For the three Iridaceae species (B. chinensis, I. tectorum, and I. germanica), we predicted protein sequences from de novo assembled transcripts using OrfPredictor65. When multiple alternatively spliced transcripts were present, we retained the longest isoform as the representative protein-coding sequence for each gene. Orthologous groups among the 13 species on a genome-wide scale were predicted by applying Orthofinder (v2.3.11)66 with default parameters. The protein sequences for orthologous groups were aligned by applying MUSCLE (v3.8.31)67. Gene trees were constructed applying RAxML (v7.9.3)68 with the PROTGAMMAILG model. Bootstrap scores were obtained based on 1000 times of sampling. The trees were visualized using Figtree (v1.4). To examine whether soybean IFS is specifically present in soybean, we searched the orthologous genes of soybean IFS in NCBI non-redundant protein database (v2022) applying blast (v2.2.31+) with an E-value threshold of 1e−5 69. LC-MS analysis The reaction products were filtered through a 0.22 μm membrane filter and transferred to an HPLC vial for LC-MS analysis. LC-MS analysis was performed on an Agilent Infinity II 6545B Q-TOF LC/MS system equipped with an electrospray ionization source. Separation was achieved on a Syncronis™ C18 column (2.1 × 100 mm, 1.7 μm; Thermo Scientific, USA) at a flow rate of 0.500 mL/min. The solvent gradient utilized two solutions: solvent A consisted of 0.1% formic acid in water, while solvent B was acetonitrile. The gradient program was as follows: 0–6 min, 95–5% A (5–95% B); held at 5% A until 8 min; returned to 95% A at 9 min; and re-equilibrated to 11 min. Mass spectrometry was conducted in negative ion mode with a spray voltage of 3.7 kV, sheath gas 40 Arb, auxiliary gas 15 Arb, and capillary temperature 350 °C. Data were acquired in centroid mode across a mass range of m/z 100–1700 using a time-segmented method (MS from 0–1 min, then Auto MS/MS). The acquisition rate was 1 spectrum/s (1000 ms/spectrum) with 8124 transients/spectrum for MS/MS, and the system was operated under standard conditions (3200) with a 2 GHz extended dynamic range. Raw data were processed using Agilent Qualitative Analysis 10.0. Compounds were identified by comparing retention times and MS/MS fragmentation patterns with authentic standards, with a mass accuracy threshold of <5ppm. No statistical tests were applied to the MS data as the analysis was qualitative. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Source data Source Data Acknowledgments We would like to thank Professor Anne Osbourn from the John Innes Center for the valuable comments provided during the writing of the manuscript. Yucheng Zhao acknowledges the funding support from the China Scholarship Council (CSC), which enabled a one-year visit to the John Innes Center and the completion of this work. Author contributions Conceptualization: Y.Zhao, L.Qi, and M.Qin; Methodology: Y.Zhao, X.Shi, X.Zhang, M.Lv, and M.Tian; Investigation: Y.Zhao, X.Shi, X.Zhang, G.Xie, and Y.Zhu; Visualization: Y.Dai, H.Zeng, Y.Zhong, S.Li, and L.Sun; Funding acquisition: Y.Zhao and M.Qin; Project administration: X.Ma and J.Bai; Supervision: L.Qi and M.Qin; Writing-original draft: Y.Zhao and X.Zhang; Writing – review & editing: Y.Zhao and M.Lv. Peer review Peer review information Nature Communications thanks Ancheng Huang and Satya Swathi Nadakuduti for their contribution to the peer review of this work. A peer review file is available. Funding This work was supported by the National Natural Science Foundation of China (81872958, 32570471), the ability establishment of sustainable use for valuable Chinese medicine resources (2060302) and the open fund of Key Laboratory of Standardization of Chinese Medicine (Chengdu University of Traditional Chinese Medicine), Ministry of Education (MOECDU2025-06). It was also supported by the State Key Laboratory of Natural Medicines, China Pharmaceutical University (SKLNMZZ2024JS37) and the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB0630101). Data availability All data generated in this study are publicly available through the indicated repositories. The transcriptome sequence of different samples had been submitted to the NCBI Sequence Read Archive (SRA) database under the accession numbers PRJNA739545 (I. germanica), PRJNA739549 (I. tectorum), and PRJNA690976 (I. dichotoma or B. chinensis). DNA sequences characterized in this study are deposited in the National Center for Biotechnology Information (NCBI) GenBank database with the following accessions: PAL969 OP413811, C4H074 OP413812, C4H260 OP413813, 4CL987 OP413814, 4CL278 OP413815, 4CL206 OP413816, CHS OP413817, CHI OP413818, IFS OP413819, IF6H OP413820, I3’H OP413821, F3’H OP413822, I6OMT1 OP413823, I6OMT2 OP413824, OMT201 OP413825, OMT755 OP413826. Source data are provided with this paper. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Yucheng Zhao, Xinyu Shi, Xiang Zhang, Ming-Ju Amy Lyu. Contributor Information Yucheng Zhao, Email: zhaoyucheng1986@126.com. Lian-Wen Qi, Email: Qilw@cpu.edu.cn. Minjian Qin, Email: minjianqin@163.com. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-74916-x. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Source Data Data Availability Statement All data generated in this study are publicly available through the indicated repositories. The transcriptome sequence of different samples had been submitted to the NCBI Sequence Read Archive (SRA) database under the accession numbers PRJNA739545 (I. germanica), PRJNA739549 (I. tectorum), and PRJNA690976 (I. dichotoma or B. chinensis). DNA sequences characterized in this study are deposited in the National Center for Biotechnology Information (NCBI) GenBank database with the following accessions: PAL969 OP413811, C4H074 OP413812, C4H260 OP413813, 4CL987 OP413814, 4CL278 OP413815, 4CL206 OP413816, CHS OP413817, CHI OP413818, IFS OP413819, IF6H OP413820, I3’H OP413821, F3’H OP413822, I6OMT1 OP413823, I6OMT2 OP413824, OMT201 OP413825, OMT755 OP413826. Source data are provided with this paper.