Three cytochrome P450 enzymes consecutively catalyze the biosynthesis of furanoclerodane precursors in Salvia species

Abstract

Salvia species native to the Americas are rich in valuable bioactive furanoclerodanes like the psychoactive salvinorin A found in Salvia divinorum, which is used in the treatment of opioid addiction. However, relatively little is known about their biosynthesis. To address this, we investigated the biosynthesis of salviarin, the most abundant furanoclerodane structure in the ornamental sage Salvia splendens. Using a self-organizing map and mutual rank analysis of RNA-seq co-expression data, we identified three cytochrome P450 enzymes responsible for the consecutive conversion of kolavenol into the salviarin precursors: annonene, hardwickiic acid, and hautriwaic acid. Annonene and hardwickiic acid have been proposed as intermediates in the biosynthesis of salvinorin A, and we therefore tested for a common evolutionary origin of the furanoclerodane pathway in these Salvia species by searching for homologous genes in available data for S. divinorum. The enzymes encoded by orthologous genes from S. divinorum displayed kolavenol synthase, annonene synthase, and hardwickiic acid synthase activity, respectively, supporting the view that these are intermediate steps in the biosynthesis of salvinorin A. We further investigated the origin of annonene synthase and the role of gene duplication in the evolution of this specific activity. Our work shows how S. splendens can serve as a model species for the study of furanoclerodane biosynthesis in Salvia species, contributes to understanding the evolution of specialized metabolism in plants, and provides new tools for the production of salvinorin A in biotechnological chassis.

Neotropical Salvia species produce valuable bioactive diterpenoids, such as salvinorin A, which is used to treat opioid addiction. However, the biosynthetic pathways of these metabolites remain unknown. By using the ornamental species Salvia splendens as a model, this study identifies key enzymes involved in the conversion of kolavenol to precursors of the furanoclerodane salviarin. Additionally, it reveals two orthologs in Salvia divinorum that function in the salvinorin A biosynthetic pathway.

Introduction

Salvia is the largest and most diverse genus in the family Lamiaceae, with three distinct biogeographic phylogenetic clades in East Asia, the Mediterranean basin and Central Asia, and the Americas (Central and South America) (Walker et al., 2004). Many Salvia species, commonly referred to as sages, are well known for their medicinal properties as well as their culinary and cosmetics applications. The high value of sages has been attributed to their production of bioactive phytochemicals, including phenylpropanoids (Exarchou et al., 2002) and terpenoids (Ulubelen, 2003). Several diterpenoids produced by Salvia species have commercial value. For example, the antioxidant food additives carnosic acid and carnosol are abietane diterpenoids produced by culinary European Salvia species (Božić et al., 2015; Ignea et al., 2016b; Scheler et al., 2016), the fragrance ingredient sclareol is a labdane diterpenoid produced by Salvia sclarea (Schalk et al., 2012) native to the Mediterranean, and the tanshinones with cardiovascular-protective properties are a group of abietane diterpenoids extracted from the roots of the Chinese herb Salvia miltiorrhiza (Ma et al., 2021; Wang and Peters, 2022). Half of all Salvia species belong to the subgenus Calosphace, which is endemic in the neotropics (Central and South America) (Ortiz-Mendoza et al., 2022). The potent bioactive substances in this subgenus are furanoclerodane diterpenoids (Wu et al., 2012; Li et al., 2016; Ortiz-Mendoza et al., 2022). Clerodanes are a class of diterpenoids with a rearranged labdane carbon skeleton involving the 1,2-shift of two methyl groups. In Lamiaceae, the clerodane scaffold is formed by the paired enzymatic transformation of geranylgeranyl diphosphate (GGPP) to kolavenol, catalyzed by class II and class I clerodane synthases in plastids (Li et al., 2023). Following export from the plastid, kolavenol is subject to several oxidation steps generally catalyzed by cytochrome P450 enzymes, with one example reported in the Lamiaceae to date (Kwon et al., 2022).

Clerodanes from the neotropical sages share many structural features, such as a furan ring between carbon atoms C13–C16 (Figure 1, highlighted in blue) and a carboxyl moiety at C18 (Figure 1, highlighted in orange). Examples of these furanoclerodanes are salviarins (Fontana et al., 2006a) and salvisplendins (Fontana et al., 2006b) from the ornamental species Salvia splendens (scarlet sage) (Dong et al., 2018; Jia et al., 2021); hispanins (Fan et al., 2019) and salvihispins (Fan et al., 2020) from the highly nutritional pseudocereal Salvia hispanica (chia) (Fan et al., 2019; Wang et al., 2022a); and salvinorins and divinatorins from the Mexican magic sage Salvia divinorum (Siebert, 1994) (supplemental Figure 1). These furanoclerodanes differ mainly in their oxidation sites. The best-known furanoclerodanes are the hallucinogenic salvinorin A and its derivatives from Salvia divinorum, which are used in indigenous American folk medicine for pain relief, owing to their psychoactive properties (Tlacomulco-Flores et al., 2020; Ortiz-Mendoza et al., 2022). Salvinorin A is rather unique among psychedelic substances, as it has been identified as a therapeutic for opioid addiction and drug abuse (Kivell et al., 2014). However, our understanding of the biosynthesis of these furanoclerodanes is limited, and in places contradictory (Chen et al., 2017; Pelot et al., 2017; Ngo, 2019; Kwon et al., 2022; Ngo et al., 2024).

Figure 1.

Clerodane metabolism in Salvia species.

(A) Examples of furanoclerodanes bearing the furane ring (highlighted in blue) and the oxidized C18 and C19 atoms (highlighted in orange) from neotropical sages: the ornamental scarlet sage (Salvia splendens), the psychoactive magic sage (Salvia divinorum), and the pseudocereal chia (Salvia hispanica).

(B) The postulated biosynthesis of furanoclerodanes in Salvia spp. (Chen et al., 2017): from geranylgeranyl diphosphate (GGPP) to hardwickiic acid (precursor of salvinorin A biosynthesis in S. divinorum) and to salviarin biosynthesis in S. splendens through hautriwaic acid as an intermediate. The depicted enzymatic transformations are catalyzed by the class II diterpene synthase kolavenyl diphosphate synthase (KPS) (Chen et al., 2017; Pelot et al., 2017; Li et al., 2023), the class I diterpene synthase kolavenol synthase (KLS) (Li et al., 2023), and the cytochrome P450 enzymes annonene synthase (ANS), hardwickiic acid synthase (HDAS), and hautriwaic acid synthase (HTAS) (this study).

Here, we report the discovery and characterization of genes encoding three cytochrome P450 enzymes that act in the biosynthetic pathway of salviarin in S. splendens (Figure 1), based on a combinatorial gene discovery strategy. The combination of two independent gene expression profiling methods with distinct emphasis on gene co-expression patterns streamlined the discovery of genes encoding enzymes involved in specialized metabolism. One intermediate, hardwickiic acid, has been identified and isolated from S. divinorum and is believed to be involved in salvinorin A biosynthesis (Bigham et al., 2003). To test the hypothesis that genes orthologous to those encoding CYP450 enzymes S. splendens act in the biosynthesis of salvinorin A in S. divinorum, we used sequence homology to search transcriptomic data of S. divinorum (Chen et al., 2017; Pelot et al., 2017; Kwon et al., 2022) and identified enzymes with the same catalytic activity. The biochemical and phylogenetic data indicate that evolution of the clerodane pathway in S. splendens involved recruitment of genes encoding enzymes that function in the synthesis of other terpenoids. This is demonstrated by the emerging new functions of genes encoding cytochrome P450s with ferruginol synthase activity in other Lamiaceae species to genes encoding annonene synthase operating within the furanoclerodane pathway in both S. splendens and S. divinorum. This finding aligns with our earlier work on the evolution of genes encoding class II and class I diterpene synthases in the clerodane pathway of S. splendens (Li et al., 2023). We show that comparison of closely related species with common biosynthetic pathways can contribute to understanding the production of rare, high-value fine phytochemicals in plants, opening up new options for their bioproduction.

Results

Gene discovery in S. splendens

In earlier work, we identified genes encoding class II and class I diterpene synthases that synthesize kolavenol in S. splendens (Li et al., 2023). To identify genes encoding the subsequent enzymatic steps in furanoclerodane metabolism in S. splendens, we performed gene discovery based on the principle that genes associated with the same biological processes, such as those encoding enzymes in the same metabolic pathway, are often co-expressed. We used the clerodane synthase genes from S. splendens as query sequences to select candidate genes from the available S. splendens transcriptomic data (Dong et al., 2018) (NCBI accessions SRX3476236–SRX3476269). This dataset consisted of three biological replicate samples of roots, stems, leaves, flower calyxes, and corollas from two varieties with red and purple flowers, for a total of 30 different samples. As the first step in transcriptomic analysis, we used a self-organizing map (SOM) to cluster genes that exhibited similar expression patterns across the different samples into nodes (Figure 2A) using an unsupervised neural network (Dang et al., 2017; Wang et al., 2022b). The formation of clerodane scaffolds is catalyzed by the class II diterpene synthase with kolavenyl diphosphate synthase (KPS) activity encoded by the gene SspdiTP2.1 (Li et al., 2023). The conversion of kolavenyl diphosphate to kolavenol is the second step in clerodane biosynthesis in S. splendens and is catalyzed by class I diterpene synthases (diTPSs). These class I diTPSs are either specific for kolavenol synthesis (SspdiTPS1.5) or also possess kaurene synthase activity (KS; genes SspdiTPS1.1 andSspdiTPS1.2) or miltiradiene synthase activity (MS; gene SspdiTPS1.3) in vitro (Li et al., 2023).

Figure 2.

Analysis of transcriptomic data from S. splendens.

(A) Self-organizing map of S. splendens transcriptomic data. Each hexagonal node represents at least 70 genes with the most similar expression profiles. Darker colors indicate higher node quality. The placement of previously identified diterpene synthase genes (Li et al., 2023) in nodes is highlighted with light (class II diTPS) and dark (class I diTPS) green dots and that of cytochrome P450 enzymes from the CYP76AH subfamily with blue dots. The enzymatic activities are shown in parentheses. The class II clerodane synthase gene SsKPS was placed in node 225 (highlighted by an orange perimeter), together with the annonene synthase gene SsANS (CYP76AH-like), the hardwickiic acid synthase gene SsHDAS (CYP728D-like), and the hautriwaic acid synthase gene SsHTAS (CYP728D-like).

(B) Expression profiles of genes encoding enzymes with SsKPS, SsKLS, SsANS, SsHDAS, and SsHTAS activities across 30 tissue samples.

(C) Heatmap of the co-expression analysis of SsKPS, SsANS, SsHDAS, and SsHTAS in S. splendens transcriptomic data. A low MR ranking indicates a relatively similar expression profile.

(D) qRT–PCR analysis of SsKPS, SsANS, SsHDAS, and SsHTAS expression in root, stem, flower, and leaf tissues of S. splendens. Data were obtained from three independent biological replicates. Transcript levels were normalized to that of actin (n = 3). Bars represent SD.

As a starting point, we used the KPS gene (SspdiTPS2.1/SsKPS Saspl_043012) (Li et al., 2023) as our “bait.” This gene encodes the first step in the formation of the clerodane skeleton and is part of node 225 (Figure 2A), which comprises 78 genes. The average expression level of genes in node 225 was higher in leaves and reproductive tissues (flower corolla and calyx) and lower in roots (Figure 2B and supplemental Figure 2). Among these genes, 12 were functionally annotated as encoding cytochrome P450 enzymes. To prioritize the choice of cytochrome P450 genes for cloning, expression, and activity assays, we used mutual rank-based co-expression analysis. Mutual ranking analysis (MRA) of the transcriptome data involved the pairwise comparison of gene expression levels across different samples (Obayashi and Kinoshita, 2009). MRA ranked the input genes on the basis of similarities in their co-expression patterns against the query genes and provided rankings (Ding et al., 2019).

Discovery and characterization of annonene synthase (SsANS)

To screen for candidate P450 genes involved in clerodane biosynthesis, we assembled the genes encoding the enzymes that provide GGPP substrate for the kolavenol biosynthetic pathway (Figure 3A) (Li et al., 2023) in baker’s yeast (Saccharomyces cerevisiae). The geranylgeranyl diphosphate synthase gene (ERG20::F96C) (Ignea et al., 2014) and a fused class II (SsKPS; SspdiTPS2.1) and class I (SsKLS; SspdiTPS1.5) clerodane synthase gene (Zhou et al., 2012; Li et al., 2023) were cloned into the pESC-His vector for recombinant expression in yeast. Using both the SOM and MRA rankings, we focused on cytochrome P450 genes that were part of SOM node 225 and were ranked highly by MRA in relation to SsKPS. Our top candidate gene, Saspl_037721, was placed in SOM node 225 and was ranked as the 12th most highly co-expressed gene with SsKPS (Saspl_043012) among all genes in the MRA (Figure 2C). Saspl_037721 is located on genomic scaffold 129 (Dong et al., 2018) and is functionally annotated as a ferruginol synthase-like (CYP76AH) gene. We cloned the full-length transcript of Saspl_037721 into the pESC-Leu vector together with a cytochrome P450 reductase gene (CrCPR2) from Catharanthus roseus (Wang et al., 2022b, 2022c) and transformed it into yeast strain AM119 with the engineered pESC-His vector. After galactose induction, the recombinant yeast culture was extracted with ethyl acetate and analyzed by GC–MS. A new peak appeared at 29.3 min (Figure 3B and supplemental Figure 3A) under the extracted ion chromatogram (EIC) mode at m/z 271. Using the mass spectrum of the new peak to search the NIST17 database, we found that annonene was the best match (75.2% probability). Comparison of the GC–MS chromatogram and EI–MS spectra of the product of the Saspl_037721 enzyme with those of synthetic annonene (Müller et al., 2015) (Figure 3B and 3C and supplemental Figure 3B–3E) confirmed the transformation of kolavenol to annonene. Because Saspl_037721 encoded a protein (CYP76AH54) with annonene synthase activity, we will refer to it as annonene synthase (SsANS) hereafter.

Figure 3.

Characterization of enzymes in the furanoclerodane diterpenoid biosynthetic pathway from kolavenol to hautriwaic acid in S. splendens.

(A) The biosynthetic pathway from kolavenol to hautriwaic acid.

(B) GC–MS analysis (selected m/z signal of 271) of yeast strains expressing GGPPS (Erg20::F96C), SsKPS, SsKLS, SsANS or SdANS, and CrCPR in blue and synthetic annonene (Müller et al., 2015) in orange. Negative control (in gray): yeast strain expressing GGPPS (Erg20::F96C), SsKPS, SsKLS, and CrCPR.

(C) Comparison of EI mass spectrum of the GC–MS peak eluting at 29.3 min (B, product of SdANS) with that of the synthetic annonene standard.

(D) LC–MS analysis (selected m/z signal of 315 in negative mode) of yeast strains expressing GGPPS (Erg20::F96C), SsKPS, SsKLS, SsANS, CrCPR, and combinations of SsHDAS or SdHDAS and SsHTAS.

(E) Comparison of electrospray ionization (ESI) mass spectrum of the LC–MS peak eluting at 9.6 min (D) with that of the hardwickiic acid standard.

(F) Genomic region (scaffold 24) of the S. splendens genome showing that SsHDAS (Saspl_010504) and SsHTAS (Saspl_010503) genes are the result of tandem duplication.

(G) LC–MS analysis (selected m/z signal of 331 in negative mode) of yeast strains expressing GGPPS (Erg20::F96C), SsKPS, SsKLS, SsANS or SdANS, CrCPR, and combinations of SsHDAS or SdHDAS and SsHTAS.

(H) Comparison of ESI mass spectrum of the LC–MS peak eluting at 6.7 min (G) with that of the hautriwaic acid standard.

Discovery and characterization of hardwickiic acid synthase (SsHDAS) and hautriwaic acid synthase (SsHTAS)

We performed MRA with reference to SsANS to select cytochrome P450 gene candidates that might encode proteins acting on annonene. The Saspl_010504 gene was annotated as encoding a cytochrome P450 assigned to the 728D subfamily, ranked 15th by MRA in relation to SsANS, and clustered in SOM node 225 like SsKPS and SsANS. We cloned Saspl_010504 into the pESC-Ura vector and tested its activity in vivo after galactose induction in recombinant yeast. The ethyl acetate extract of yeast was analyzed by GC–MS, and a new peak at EIC m/z 250 was detected, eluting at 36.9 min (supplemental Figure 4A and 4B). A search of the mass spectra in the NIST17 database suggested that the new peak had a top match with hardwickiic acid (79.3% probability, supplemental Figure 4C). Because hardwickiic acid bears a carboxylic acid moiety, we shifted our analysis to LC–MS. The hardwickiic acid standard eluted at 9.2 min with m/z 315 in negative mode, which aligned with the peak of the Saspl_010504 (CYP728D62) enzyme product in the assay (Figure 3D) and agreed with the MS² spectrum (Figure 3E and supplemental Figure 5). We therefore concluded that the cytochrome P450 encoded by Saspl_010504 acts on annonene and functions as a hardwickiic acid synthase (SsHDAS).

The Saspl_010503 gene is tandemly arrayed with Saspl_010504/SspHDAS (Figure 3F), likely as the result of a gene duplication event. The encoded proteins share high sequence similarity at the amino acid level (83.4%), and both are annotated as CYP728D cytochrome P450 enzymes. Saspl_010503 was also placed in SOM node 225 with a similar expression profile to SspdiTPS2.1 (SsKPS), SsANS, and SsHDAS (Figure 2A). Despite its low ranking in the MRA (Figure 2C), we tested whether Saspl_010503 encodes a protein acting in the pathway. When we co-expressed the cytochrome P450 (CYP728D63) encoded by Saspl_010503 together with SsKPS, SsKLS, and SsANS, no hardwickiic acid or any other product was formed with annonene, as assessed by LC–MS analysis (Figure 3D and supplemental Figure 5). When we co-expressed the cytochrome P450 encoded by Saspl_010503 with SsKPS, SsKLS, SsANS, and SsHDAS, a new peak appeared in the LC–MS at 6.7 min with an m/z 331 (negative mode) (Figure 3G and supplemental Figure 6). The new product was identified as hautriwaic acid by comparing its retention time and MS² spectrum with those of an authentic standard (Figure 3G and 3H and supplemental Figure 6), and Saspl_010503 was therefore named hautriwaic acid synthase (SsHTAS).

We performed qRT–PCR to assess the expression levels of all genes associated with hautriwaic acid biosynthesis in SOM node 225. The measurements covered various plant tissues, including roots, stems, leaves, and flowers (Figure 2D). The furanoclerodane biosynthetic genes exhibited significantly higher expression levels in flowers and leaves, consistent with metabolomic data showing the predominant accumulation of S. splendens clerodanes, such as salviarin, in these tissues (supplemental Figure 7).

Key clerodane pathway intermediates accumulate in abaxial peltate trichomes

In S. divinorum, salvinorin A and other furanoclerodanes accumulate specifically in peltate trichomes on the abaxial sides of leaves (Siebert, 2004a; Chen et al., 2017). S. splendens leaves have both capitate and peltate glandular trichomes (Agustin et al., 2022; Ghonam et al., 2014; Perveen). SEM confirmed that S. splendens leaves predominantly have peltate glandular trichomes up to 50 μm in diameter on their abaxial sides (Figure 4 and supplemental Figure 8A). Metabolite analysis of these trichomes, picked from the abaxial sides of S. splendens leaves, showed the presence of hardwickiic acid, hautriwaic acid, and salviarin (Figure 4 and supplemental Figure 8B). No signal for crotonolide G (a proposed intermediate in salvinorin A biosynthesis in S. divinorum) was detected (supplemental Figure 8B). These results suggested that the furanoclerodanes in S. splendens are produced within glandular trichomes, similarly to salvinorin A in S. divinorum. In addition, we performed matrix-assisted laser desorption/ionization (MALDI) mass spectrometry imaging (MSI) on the abaxial sides of S. splendens leaves (Figure 4A and 4B) to identify the site of salviarin accumulation (Figure 4). MALDI–MSI revealed the co-occurrence of peaks corresponding to the mass [M-H)^– of hautriwaic acid (Figure 4D) and the mass of salviarin (Figure 4E) at the same positions as the peltate trichomes (Figure 4F). The mass signals for annonene and hardwickiic acid were too weak to be distinguished clearly from the background (Figure 4B and 4C) but suggested a punctate distribution across the leaf, consistent with localization in leaf trichomes.

Figure 4.

Identification of intermediates in the salviarin synthesis pathway on the abaxial leaf surface of S. splendens by MALDI imaging.

(A) Photograph of the abaxial surface of an S. splendens leaf. The red arrows indicate a pattern of three medium-sized peltate trichomes (shown in detail in the SEM in supplemental Figure 8A), which can also be observed in the MALDI images below. Scale bars correspond to 590 μm (all images, A–G).

(B) MALDI image of annonene (mass 285 in negative ion mode; intensity scale 0–0.0007).

(C) MALDI image of hardwikiic acid (mass 315 in negative ion mode, intensity scale 0–0.0006).

(D) MALDI image of hautriwaic acid (mass 331 in negative ion mode, intensity scale 0–0.004).

(E) MALDI image of salviarin (mass 341 in negative ion mode, intensity scale 0–0.004). Note that blurring of the salviarin signal suggests that it may have been released from the trichomes and spread over the leaf surface.

(F) Overlay of (D) onto (A), showing overlap of the hautriwaic acid signal with the peltate trichomes on the abaxial leaf surface. The green arrows in (D–F) highlight the same trichomes shown in (A) in red.

Identification and characterization of enzymes with kolavenol synthase (KLS), annonene synthase (ANS) and hardwickiic acid synthase (HDAS) activities in S. divinorum

Convergence in plant specialized metabolism is common between genera. Many phytochemicals have identical or very similar chemical structures; however, it cannot be assumed that they are synthesized in the same way. In work by Chen et al. (2017), a scheme for the biosynthesis of salvinorin A and other clerodanes was proposed, with annonene and hardwickiic acid as intermediates between kolavenyl diphosphate (the product of the class II clerodane synthase) and salvinorin A. Detection and quantification of hardwickiic acid by LC–MS in peltate trichomes from S. divinorum (Siebert, 2004b) also supported its role as a potential intermediate in furanoclerodane metabolism. Because both S. splendens and S. divinorum belong to the Calosphace subgenus, originate from the same geographic region, and produce highly similar furanoclerodanes, we tested whether early steps in furanoclerodane biosynthetic pathways share functionally equivalent enzymes in these two species. In previous work on S. divinorum, the class II clerodane synthase KPS was characterized based on its expression in peltate trichomes and transcriptomic data from peltate trichomes on the abaxial leaf surface (Chen et al., 2017; Pelot et al., 2017). However, attempts to identify a functional class I diTPS with kolavenol synthase activity were unsuccessful (Chen et al., 2017; Pelot et al., 2017).

To examine the relevance of genes encoding cytochrome P450 enzymes in furanoclerodane metabolism from S. splendens to S. divinorum, we searched transcriptomic data from peltate trichomes of S. divinorum (Pelot et al., 2017) (NCBI accession number SRX1875790) for candidate genes encoding proteins involved in salvinorin A biosynthesis by sequence similarity. We searched first for genes encoding class I diterpene synthases closely related to those characterized functionally as having kolavenol synthase activity in S. splendens: SspdiTPS1.5, which acts specifically as a kolavenol synthase; SspdiTPS1.1 and SspdiTPS1.2, which also have kaurene synthase activity; and SspdiTPS1.3, which also has miltiradiene synthase activity. Three transcripts encoding three class I diTPSs (Chen et al., 2017; Pelot et al., 2017) were the top results. The first enzyme, named SdKSL1, aligned most closely to SspdiTPS1.3 (supplemental Figure 9); a second enzyme, SdKSL2, was most closely aligned to SspdiTPS1.5 (supplemental Figure 9), and a third, SdKSL3 (Pelot et al., 2017), aligned closely with SspdiTPS1.1 and SspdiTPS1.2 (supplemental Figure 9). Both SdKSL2 and SdKSL3 were highly expressed in peltate trichomes of S. divinorum (Chen et al., 2017; Pelot et al., 2017). Pelot et al. reported that neither of these class I diTPS genes had kolavenol synthase activity in in vivo assays dependent on a class II KPS activity (SdCPS2). Given the limitations of in vivo assays, resulting from the activity of endogenous phosphatases in N. benthamiana, it seemed likely that, by analogy to S. splendens, SdKSL2 and SdKSL3 might provide the class I diterpene synthase activity for salvinorin A biosynthesis.

The genes SdKSL1 and SdKSL2 were synthesized to encode truncated proteins without transit peptides and were expressed in E. coli (Li et al., 2023). To characterize the activity of SdKSL1 and SdKSL2 toward kolavenyl diphosphate, we performed a coupled in vitro enzyme assay with GGPP and SsKPS (Li et al., 2023). Both SdKSL1 and SdKSL2 were able to convert kolavenyl diphosphate (product of SsKPS) to kolavenol and could potentially act as kolavenol synthases in furanoclerodane biosynthesis of S. divinorum (Figure 5). Our searches of the S. divinorum transcriptome from peltate trichomes (Pelot et al., 2017) identified a transcript, Sd_DN33_c0_g1_i12, whose encoded protein was most closely related to SsANS (supplemental Figure 11). Sd_DN33_c0_g1_i12 is identical to the SdCS gene that encodes the cytochrome P450 CYP76AH39 with reported crotonolide G synthase activity, and it shares 82.3% amino acid sequence similarity with S. splendens SsANS (Kwon et al., 2022). In S. splendens, we could not detect the presence of crotonolide G in peltate trichomes on the abaxial side of the leaf (supplemental Figure 8B). We synthesized the Sd_DN33_c0_g1_i12 transcript, expressed it, and tested its enzymatic activity together with SsKPS and SsKLS in yeast. GC–MS analysis of the in vivo assay showed a chromatographic peak in EIC m/z 271 eluting at 29.3 min (Figure 3B), the same as the annonene produced by SsANS and synthetic annonene (Figure 3C). To confirm the annonene synthase activity of Sd_DN33_c0_g1_i12, we paired Sd_DN33_c0_g1_i12 with SsHDAS in yeast. LC–MS analysis of the in vivo assay products showed elution of a peak at 9.2 min with MS and MS² identical to those of the hardwickiic acid standard (Figure 3D and supplemental Figure 5), and we therefore refer to Sd_DN33_c0_g1_i12 as SdANS.

Figure 5.

Characterization of enzymes with kolavenol synthase (KLS) activity and biosynthesis of furanclerodane in S. divinorum.

(A) Enzymatic steps in biosynthesis of hardwickic acid in S. divinorum starting from geranyl-geranyl diphosphate (GGPP). The boxes with a colored background depict the enzymatic activities, with bold red fonts indicating the enzymatic activities characterized in the present study. KPS: kolavenyl diphosphate synthase activity performed by SdKPS1 (Chen et al., 2017; Pelot et al., 2017); KLS: kolavenol synthase activity performed by SdKSL1-2 (genes reported but not functionally characterized as kolavenol synthase KLS by Chen et al. [2017] and Pelot et al. [2017]); ANS: annonene synthase (Figure 3B) encoded by the CYP76AH39 gene (reported but not functionally characterized as ANS by Kwon et al. [2022]); HDAS: hardwickiic acid synthase (Figure 3D) encoded by the CYP728D26 gene (reported but not functionally characterized as HDAS by Ngo [2019] and Kwon et al. [2022]).

(B) GC–MS analysis (selected m/z signals at 189) of extracts from enzyme assays against GGPP of the class II diterpene synthase SsKPS with combinations of the class I diterpene synthases SdKSL1 and SdKSL2. Both class II and class I diterpene synthase enzymes were heterologously expressed and purified from E. coli. The enzyme assay of SsKPS without addition of class I enzymes or alkaline phosphatase was used as a negative control.

(C) EI–MS spectra of GC–MS peaks (from the corresponding GC–MS chromatograms shown in (B) eluting at 35.4 min and corresponding to kolavenol.

Using SsHDAS as a search template, we identified a 1130-bp partial transcript sequence, Sd_DN6570_c0_g1_i3, in the S. divinorum trichome transcriptome; the full length of this transcript had been deposited by Ngo (2019), Kwon et al. (2022), and Ngo et al. (2024) under the name CYP728D26 (NCBI accession QMS79245.1). Another top hit from the search results, Sd_DN990_c0_g1_i18, was a transcript that partially aligned to the sequences of CYP728D25 and CYP728D27 (NCBI accessions QMS79244.1 and QMS79246.1) deposited by the same authors (supplemental Figure 12). These three genes were reported to encode candidate enzymes involved in salvinorin A biosynthesis, on the basis of their presence in the S. divinorum peltate-trichome-specific transcriptomic data (Pelot et al., 2017). CYP728D25, CYP728D26, and CYP728D27 have nucleotide sequence similarities of 83.6%, 85.9%, and 83.3% to SsHDAS and amino acid sequence similarities of 71.5%, 75.6%, and 71.5% to SsHDAS (supplemental Figure 12). We synthesized these three genes and expressed them together with S. splendens SsKPS, SsKLS, and S. divinorum SdANS to assess their in vivo catalytic activity. Analysis of ethyl acetate extracts of yeast by LC–MS showed that only CYP728D26 exhibited HDAS activity, as demonstrated by the appearance of a new peak at EIC m/z 331 (Figure 3D and supplemental Figure 5), which aligned with that of the hardwickiic acid standard. We therefore refer to CYP728D26 as SdHDAS.

The gene encoding annonene synthase evolved from ferruginol synthase

In the latest chromosome-level S. splendens genome assembly, the tandemly duplicated genes Saspl_037721 and Saspl_037719 are found on chromosome 19 in proximity to SspdiTPS1.3 and the non-functional class II diterpene synthase gene SspdiTPS2.8 (supplemental Table 1) (Li et al., 2023). The genome of S. splendens has been shaped by two relatively recent whole-genome duplication (WGD) events. The older WGD is shared by S. hispanica and S. splendens, whereas the more recent one is unique to S. splendens (Wang et al., 2022a). Thus, in an intraspecies syntenic analysis, the genomic region containing SsANS has three other syntenic regions on chromosomes 10, 20, and 21 in S. splendens. The syntenic relationships between class II diterpene synthases SspdiTPS2.8 and SspdiTPS2.4 from chromosome 10, SspdiTPS2.5 from chromosome 21, and SspdiTPS2.2 from chromosome 20 are highlighted by green ribbons in Figure 6. Both SspdiTPS2.4 and SspdiTPS2.5 have been characterized as copalyl diphosphate synthases (Li et al., 2023). The dark green ribbon shows the synteny of genes encoding class I diterpene synthases: SspdiTPS1.3 from chromosome 19 and SspdiTPS1.4 from chromosome 20. SsANS and its tandem duplicate Saspl_037719 exhibit a syntenic relationship (highlighted by the navy blue ribbon, Figure 6B) with the cytochrome P450 genes CYP76AH36a and CYP76AH36b encoding ferruginol synthase activity. In addition, the characterized ferruginol synthase gene CYP76AH87b has a syntenic relationship only to the CYP76AH87a gene (highlighted by the pale blue ribbon) (Li et al., 2023). We have reported a ferruginol biosynthetic gene cluster (BGC) in the two largest subfamilies of Lamiaceae, Nepetoideae and Scutellaroideae, with SspdiTPS2.4, SspdiTPS2.5, SspdiTPS1.3, SspdiTPS1.4, CYP76AH36a/b, and CYP76AH87a/b all being part of the ferruginol BGC in S. splendens (Li et al., 2023). We used an engineered yeast strain producing miltiradiene (Li et al., 2023) to assay the activity of SsANS and Saspl_037719 toward miltiradiene, but no ferruginol synthase activity was observed (Figure 6C). When we assayed CYP76AH36a/b and CYP76AH87a/b with kolavenol, annonene synthase activity was detected for both enzymes (Figure 6A). The phylogeny of syntenic cytochrome P450s from Lamiaceae (supplemental Figure 13A) shows that the gene encoding SsANS diverged from that encoding CYP76H87 around 17.8 million years ago (MYA), and the divergence of the gene encoding SsANS from the syntenic CYP76AH36 genes with ferruginol synthase activity occurred around 10.6 MYA (supplemental Figure 13B). These enzymatic activities and phylogenetic and syntenic relationships suggest that the emergence of SsANS was the result of neofunctionalization of an ancestral gene encoding ferruginol synthase, and this shift in activities is parsimonious with other reported examples of neofunctionalization in plant specialized metabolism (Weng, 2014).

Figure 6.

Emergence of the annonene synthase SsANS from ferruginol synthase.

(A) GC–MS analysis (selected ion chromatogram at m/z 271) of yeast extracts expressing cytochrome P450 enzymes together with class II (kolavenyl diphosphate synthase [KPS]) and class I (kolavenol synthase [KLS]) clerodane synthases from S. splendens.

(B) Intraspecies syntenic analysis of S. splendens genome regions containing genes encoding SsANS, ferruginol synthase, and class II and class I diTPSs. The light green ribbon highlights the syntenic relationships among class II diTPSs, the dark green ribbon highlights the class I diTPSs, the light blue ribbon highlights the CYP76AH87 P450s, and the dark blue ribbon highlights the CYP76AH36 P450s.

(C) GC–MS analysis (selected ion chromatogram at m/z 286) of yeast extracts expressing cytochrome P450 enzymes together with copalyl diphosphate synthase (CPS) and miltiradiene synthase (MS).

Biosynthesis of furanoclerodanes in the subgenus Calosphace likely evolved after subgenus divergence but before speciation

The class II and class I diterpene synthases operating in the clerodane pathway in Lamiaceae were likely recruited from the gibberellin and abietane biosynthetic pathways, respectively (Li et al., 2023). Phylogenomic analysis supported by biochemical data has demonstrated the polyphyletic origin of clerodane biosynthetic pathways in Scutellaria barbata and S. splendens through repeated evolution, with the latter having evolved more recently. The estimated time of divergence is around 20 MYA for SsKPS and SdKPS and about 15 MYA for SsKLS and SdKLS (Li et al., 2023). The divergence time of S. splendens from its Eurasian relatives is estimated to have been 23.8 MYA (supplemental Figure 14), whereas the speciation time is estimated to have been 6.4 MYA between S. splendens and S. hispanica (supplemental Figure 14) and 10 MYA between S. splendens and S. divinorum (Ford et al., 2024). The lack of any syntenic relationships between the genes encoding CYP76AH36 cytochrome P450 enzymes and the CYP genes from S. miltiorrhiza may reflect the absence of orthologs in Eurasian Salvias, whereas the syntenic counterparts in S. hispanica are likely encoded by genes orthologous to those in S. splendens (supplemental Figure 13A).

Phylogenomic analyses indicate that SsHDAS and SsHTAS in S. splendens have orthologous counterparts in S. hispanica (supplemental Figures 12 and 15). Estimation of divergence times based on Bayesian phylogeny suggests that the divergence between SsHDAS and SsHTAS occurred around 9.4 MYA (supplemental Figure 16), indicating the emergence of enzymes with HDAS and HTAS activities before the speciation of S. splendens and S. hispanica. Therefore, the early steps (ANS and HDAS) of furanoclerodane biosynthesis presumably utilized orthologous genes encoding enzymes that catalyzed the same enzymatic reaction, with minimal or no alternative routes. A recent publication on the genome sequence of S. divinorum estimates the divergence of S. divinorum from the shared ancestor of S. splendens and S. hispanica at 10 MYA (Ford et al., 2024). Given that Calosphace is a monophyletic subgenus, all species within this subgenus are closely related (Walker et al., 2004). The identification of SdANS and SdHDAS using homology-based searches supports the hypothesis that the emergence of ANS and HDAS activities was common to the Calosphace.

Discussion

To better understand furanoclerodane biosynthesis in S. splendens, we initially used an unbiased and unsupervised neural network SOM to cluster all the genes from S. splendens into different nodes on the basis of their expression profiles. Both class I and class II clerodane synthases were used as guides to focus on specific nodes, although genes encoding proteins with kolavenol synthase activity (class I) were not precisely co-expressed with KPS or the three cytochrome P450 genes that act in the furanoclerodane pathway in different tissues, possibly reflecting the observed catalytic redundancy of kolavenol synthase (Li et al., 2023). The bifunctional class I diterpene synthases SspdiTPS1.1 and SspdiTPS1.2 (Li et al., 2023) with both kaurene and kolavenol synthase activity were expressed in all plant tissues analyzed (Figure 1) and likely fulfill the role of kolavenol synthase, although their expression profiles were not congruent with those of SsKPS or SsANS and SsHDAS or SsHTAS. The clerodane-specific KLS gene (SspdiTPS1.5) showed very low expression in all tissues, although this could reflect expression in very specialized cell types such as leaf trichomes. Hypothesizing that cytochrome P450 enzymes oxidize the clerodane skeleton, we used MRA to prioritize the selection of genes encoding cytochrome P450 enzymes for enzyme assays. This strategy led to the discovery of three cytochrome P450 enzymes involved in furanoclerodane biosynthesis in S. splendens; they catalyzed the formation of annonene, hardwickiic acid, and hautriwaic acid, which we propose are intermediates in salviarin biosynthesis. (−)-Hardwickiic and hautriwaic acid have been reported as constituents of neotropical Salvia wagneriana (Bisio et al., 2004) and the desert hybrid Salvia × jamensis (Bisio et al., 2009). Both (−)-hardwickiic and hautriwaic acid have been suggested as drug-design probes for therapies targeting pathologies of the central nervous system (Pittaluga et al., 2013), including neurological pain and visceral hypersensitivity (Cai et al., 2019). (−)-Hardwickiic acid is also used commercially as an antagonist of bitter taste receptors to mask or decrease the perception of bitterness and/or enhance the perception of sweetness (Slack et al., 2013).

The best-known example of a furanoclerodane from Salvia species is salvinorin A from S. divinorum, which is known for its use in the treatment of psychiatric diseases (Wu et al., 2012; Li et al., 2016; Ortiz-Mendoza et al., 2022). Because furanoclerodanes from S. splendens share many of the chemical features of salvinorin A that are essential for its anti-opioid activity, such as the furan ring and the carboxyl group at C18 on the clerodane scaffold (Prisinzano and Rothman, 2008; Roach and Shenvi, 2018), S. splendens is likely to share the same furanocleordane biosynthetic enzymes with its close relative S. divinorum. Our divergence time analysis performed on syntenic P450s between S. splendens and another Calosphace species, S. hispanica, indicated that the initial steps of clerodane biosynthesis are probably the same and share a common ancestor within this subgenus. Indeed, hardwickiic acid has been isolated from S. divinorum (Bigham et al., 2003; Chen et al., 2017) and shown to be localized in peltate glandular trichomes on the abaxial leaf surfaces of S. divinorum (Chen et al., 2017), a result that led to the proposal of a biosynthetic route for salvinorin A with annonene and hardwickiic acid intermediates (Chen et al., 2017). However, difficulties in identifying class I diTPS activity in S. divinorum led to the suggestion that crotonolide G was the first committed intermediate in furanoclerodane biosynthesis, and the identification of cytochrome P450 enzymes with decorating activity on crotonolide G was reported (Ngo, 2019; Kwon et al., 2022; Ngo et al., 2024). We used the class I diTPSs and cytochrome P450 enzymes from S. splendens to search for transcripts encoding orthologs in S. divinorum through homology alignment of the encoded proteins. Genes encoding orthologs of the class I diTPSs SsANS and SsHDAS from S. splendens were identified using transcriptomic data from peltate trichomes producing salvinorin A in S. divinorum and were characterized by in vitro and in vivo enzyme assays. The identified SdKSL1/2, SdANS, and SdHDAS enzymes from S. divinorum displayed the same activities as their orthologs in S. splendens. We used transit-peptide cleavage sites (supplemental Figure 10) different from those in earlier studies to express the proteins (Chen et al., 2017; Pelot et al., 2017) and identified the activity of SdKSL1 and SdKSL2. This difference may explain why no kolavenol synthase enzymatic activity was detected in earlier studies. Despite the fact that SdANS (CYP76AH39) has been reported to act in clerodane metabolism as a crotonolide G synthase, our in vivo assays suggested that it produces annonene in accordance with the pathway proposed by Chen et al. (2017). This difference may have been due to the use of different cytochrome P450 reductases to support P450 activity in yeast or to differences in the assay environments, in particular, the performance of P450 assays without a co-expressed class I diTPS. The coupled in vivo assay of SdANS (CYP76AH39) with SdHDAS (CYP728D26) showed the oxidation of annonene to hardwickiic acid, as postulated for salvinorin A biosynthesis by Chen et al. (2017). Although it is possible that crotonolide G is also an intermediate in salvinorin A biosynthesis, given the reported presence of hardwickiic acid in S. divinorum (Bigham et al., 2003; Chen et al., 2017) and the lack of reports of clerodanes with a dihydrofuran ring to date, we suggest that furanoclerodane formation in S. divinorum relies predominantly on the annonene synthase activity of CYP76AH39.

The presence of subfunctionalized enzymes in S. splendens, capable of catalyzing reactions in both clerodane and other diterpenoid pathways, suggests that the rise of the furanoclerodane pathway in Salvia is a relatively recent event. This phylogenetic development likely preceded the diversification of multiple clerodane-rich species. The enzymatic activities of SsHDAS, SsHTAS, and SdHDAS highlight their key roles in furanoclerodane biosynthesis in the subgenus Calosphace. These cytochrome P450 enzymes belong to the CYP728 family, and syntenic analysis showed the presence of multiple gene copies across different Lamiaceae species and lineages. Their activity in the furanoclerodane biosynthetic pathway may have precluded the chemical diversity of diterpenoids in the family Lamiaceae.

In earlier work, we showed that class I clerodane synthases subfunctionalized after the duplication of enzyme-encoding genes located in a conserved ferruginol-related BGC in the family Lamiaceae (Li et al., 2023). The emergence of annonene synthase was the result of neofunctionalization of a cytochrome P450 enzyme with ferruginol synthase activity, most likely following a WGD (Figure 6). Our findings support and expand upon the hypothesis that BGCs can serve as “metabolic toolboxes,” providing new biocatalysts for evolving metabolic pathways through gene duplication, although the biosynthetic genes are not arranged together in a BGC. Cytochrome P450 enzymes play an important role in the enormous chemo-diversity of diterpenoid metabolism in Lamiaceae, and the roles of HDAS and HTAS in furanoclerodane biosynthesis provide new examples of P450 biocatalysts in Lamiaceae diterpenoid metabolism.

In conclusion, we used a combination of bioinformatics analyses and discovered three new cytochrome P450 enzymes that act in biosynthesis of the clerodanes annonene, hardwickiic acid, and hautriwaic acid in S. splendens. We also found orthologs of SsANS and SsHDAS in S. divinorum that have the same catalytic activities, supporting the idea that S. splendens can serve as an excellent model for studying furanoclerodane biosynthesis in Salvia and that microevolutionary genomics can facilitate gene discovery and characterization in the biosynthesis of high-value clerodanes such as salvinorin A in the genus Salvia.

Materials and methods

Materials

Plant material and chemicals

Seeds of S. splendens Ker-Gawler cv. Olympic Flame were provided by Prof. Ai-Xiang Dong at Beijing Forestry University. Whole plant tissues, including flowers, leaves, stems, and roots from flowering S. splendens, were snap-frozen in liquid nitrogen. The frozen tissue was lysed using a tissue lyser, and RNA was extracted using the RNeasy Plant Mini Kit (QIAGEN, Germany). The extracted RNA was purified with the TURBO DNA-free kit (ThermoFisher) to remove gDNA. Subsequently, the RNA was reverse-transcribed into cDNA using SuperScript IV VILO Master Mix (Thermo Fisher).

(−)-Hardwickiic acid and hautriwaic acid were purchased from BioBioPha (Kunming, China). Annonene was synthesized by Matthew Furry and Christopher D. Vanderwal (University of California, Irvine) as in Müller et al. (2015).

Methods

Generation of SOM

We obtained two sets of RNA-seq data from NCBI SRA project PRJNA422035. Each set consisted of triplicate samples of RNA-seq data obtained from different tissues (roots, stems, leaves, calyx, and corolla). These RNA-seq data were aligned to the scaffold-level S. splendens reference genome using HISAT2 v2.1.0 (Kim et al., 2019) and converted into FPKM values using SAMtools v1.11 (Danecek et al., 2021) and Cuffdiff (Trapnell et al., 2012). The SOM was generated as described by Payne et al. (2017) using the Kohonen package 3.0.11 (Wehrens and Buydens, 2007). To enable a more extensive gene search, the map size was determined to allow at least 70 genes to cluster in each node. Multiple seed numbers were evaluated to explore the various clustering patterns, and a representative map was selected on the basis of its inclusion of multiple P450 genes that ranked highly in the mutual ranking profile with SsKPS (SspdiTPS2.1) and were located within the same node as SsKPS.

Mutual rank-based co-expression analysis

The FPKM values obtained above served as input for the MutRank program, which calculated the average ranking of genes on the basis of their correlation with a reference gene (Obayashi and Kinoshita, 2009; Poretsky and Huffaker, 2020).

Syntenic analysis of CYP76AH and CYP728D genes

Genomic data for Callicarpa americana (Hamilton et al., 2020), S. barbata (Li et al., 2023), S. splendens (Jia et al., 2021), S. miltiorrhiza (Song et al., 2020), Salvia bowleyana (Zheng et al., 2021), Salvia rosmarinus (Han et al., 2023), and S. hispanica (Wang et al., 2022a) were downloaded from an online repository. MCScan from the JCVI package (Tang et al., 2008) was used with default settings to search for interspecies collinear blocks. The cytochrome P450 genes and related diterpene synthases were color coded according to their identified gene functions (Li et al., 2023), Refseq functional annotations, or in vivo assay results from this study.

De novo assembly of the S. divinorum trichome transcriptome

Illumina RNA-seq reads from S. divinorum peltate trichomes were downloaded from NCBI (SRR3716680) and assembled into transcripts with Trinity 2.9.1 (Grabherr et al., 2011). S. divinorum homologs of SspdiTPS1.3, SspdiTS1.5, SsANS, and SsHDAS were identified by blast searches of the trichome transcriptome.

Phylogenetic locations of S. divinorum class I diTPS and CYP genes

SdKSL1, SdKSL2, and SdKSL3 (NCBI accessions KX268507–KX268509 [Chen et al., 2017] and KY057342–Y057344 [Pelot et al., 2017]) were added to the phylogenetic tree modified from Figure 5A in Li et al. (2023); MAFFT (Katoh and Standley, 2013) was used for the alignment, and iqtree-1.6.10 (Nguyen et al., 2015) was used to build the tree. The phylogenetic tree for the two studied CYP families was constructed using the same pipeline. The sequences included were derived from genes syntenic to those characterized as CYP76AH or HDAS genes in S. splendens and from unigenes found in the trichome transcriptome of S. divinorum.

Species divergence time estimation

OrthoFinder 2.5.4 (Emms and Kelly, 2019) was used to construct a species phylogenetic tree. The genomes of Oryza sativa ssp. japonica (IRGSP1.0, EnsemblPlants), Solanum lycopersicum (SL3.0, EnsemblPlants), Antirrhinum majus (https://doi.org/10.1038/s41477-018-0349-9), C. americana (NCBI: PRJNA529675), S. barbata (NCBI: PRJNA649842), S. rosmarinus (https://doi.org/10.6084/m9.figshare.21443223.v1), S. bowleyana (National Genomics Data Center: PRJCA003734), S. miltiorrhiza (National Genomics Data Center: PRJCA003150), S. hispanica (China National GeneBank DataBase: CNA0047366), and S. splendens (https://doi.org/10.1093/gigascience/giy068) were used to estimate speciation times. The longest protein sequence isoforms from each genome were supplied to OrthoFinder to produce a species tree. Speciation times were estimated with MCMCTree from the PAML package (Yang, 2007), using the phylogenetic tree generated by OrthoFinder and the species divergence times estimated by TimeTree (Kumar et al., 2017) as calibration points (O. sativa and S. lycopersicum, S. lycopersicum and A. majus).

Bayesian analysis of syntenic cytochrome P450s from the CYP76AH and CYP728D subfamilies

Bayesian phylogenetic trees of proteins encoded by syntenic CYP76AH and CYP728D genes were constructed using BEAST 2 (Bouckaert et al., 2014). Translated CYP76AH or CYP728D gene sequences were first aligned using MAFFT (L-INS-i) (Katoh and Standley, 2013), and these alignments were converted to codon alignments using PAL2NAL (Suyama et al., 2006). Gene divergence times were estimated using the following priors: HKY85 substitution model, empirical frequencies, strict molecular clock, and a Calibrated Yule Model. The estimated species divergence times obtained from OrthoFinder and PAML served as the calibration points (Bouckaert et al., 2014). The MCMC chain length for posterior probability was set to 20 000 000. A 10% burn-in was applied to the tree using TreeAnnotator, and the tree was plotted using FigTree.

cDNA cloning

The coding sequences of P450 genes were isolated using the primers listed in supplemental Table 2. Initial amplification of candidate genes was performed with either KOD plus Neo (Toyobo, Japan) or Phusion Plus PCR Master Mix (Thermo Fisher). PCR products were isolated and ligated into pESC vectors using the In-Fusion cloning kit (Takara Bio, Japan), and the resulting vectors were transformed into DH5α competent cells. Transformed colonies were screened on carbenicillin selective medium (50 mg l⁻¹), and colonies that showed the correct insert size in colony PCR were picked for overnight incubation in 5 ml carbenicillin selective liquid Luria–Bertani medium. Plasmids were extracted with the TIANprep Mini Plasmid Kit (Tiangen) and sent for sequencing to confirm sequence identity.

pESC vectors have two multiple cloning sites, Gal1 and Gal10. All Gal1 sites were digested with BamHI–SalI, and all Gal10 sites were digested with SpeI. Thus, for genes inserted at the Gal1 site, the primers added a BamHI site at the 5′ end and a SalI site at the 3′ end. For genes inserted at the Gal10 site, the primers added an SpeI site on both the 5′ and 3′ ends.

Yeast vector construction

pESC-His vector for kolavenol production

To introduce a kolavenol pathway into yeast, we constructed a pESC-His vector carrying a GGPP synthase gene and a fused diterpene synthase gene. Erg20 is a farnesyl pyrophosphate synthase endogenous to yeast, but its F96C mutant functions as a GGPP synthase (Ignea et al., 2015). The ERG20 gene was cloned from yeast cDNA and subjected to site-directed mutagenesis using the primers listed in supplemental Table 2 to produce ERG20::F96C. The mutated ERG20::F96C was introduced into the Gal1 site of the pESC-His vector. Kolavenol production was achieved by introducing a truncated kolavenol synthase (SspdiTPS1.5Δ50) fused to a kolavnyl pyrophosphate synthase (SspdiTPS2.1). Both class I and class II diterpene synthases are plastid-localized enzymes. As is typical for expression of diterpene synthases in yeast, we used truncated versions of these proteins (lacking their transit peptides) to avoid problems in expressing soluble proteins or recovering active proteins. We used the same truncated versions described in Li et al. (2023). The approach of fusing a class I diTPS with a class II diTPS has been shown to be sufficient for producing the desired diterpenoid scaffold upon expression in yeast (Zhou et al., 2012). In our case, SspdiTPS1.5Δ50 and SspdiTPS2.1 were linked by a GGGS linker (5′ GGT GGT GGT TCT 3′), with the stop codon of SspdiTPS1.5Δ50 replaced by the GGGS linker to enable complete translation of the fused protein. A start codon was also added in front of SspdiTPS1.5Δ50.

pESC-His vector for miltiradiene production

The miltiradiene pathway was introduced into yeast using the same procedure described above for the kolavenol pathway. The gene inserted into the Gal10 site was replaced by a fusion of truncated miltiradiene synthase SspdiTPS1.3Δ61 with full-length copalyl diphosphate synthase SspdiTPS2.4.

pESC-Leu vector for CYP76 screening

Full functionalization of CYP required the presence of cytochrome P450 oxidoreductase (CPR), and CrCPR2 from Catharanthus roseus was therefore introduced into the Gal1 site of the pESC-Leu vector (Meijer et al., 1993). CYP76 genes were inserted into the Gal10 site.

pESC-Ura for CYP728D screening

In the initial screening stage, all three CYP728D genes were ligated into the Gal1 site of the pESC-Ura vector. Later, when Saspl_010504 was identified as the hardwickiic acid synthase, Saspl_010503 and Saspl_010505 were subcloned onto the Gal10 site of the pESC-Ura vector harboring HDAS in the Gal1 site.

Heterologous expression in yeast and in vitro activity assay

Yeast strain AM119 was used as the host strain for CYP screening (Ignea et al., 2016a). Depending on the screening purpose, different combinations of pESC vectors were co-transformed into the host strain (see supplemental Table 3 for details). For heterologous expression, single yeast colonies were picked from fresh transformants and incubated in 2% glucose-supplemented SD dropout medium for 12 h. The seed cultures were then added to 20 ml fresh glucose-supplemented SD dropout medium, grown to an OD₆₀₀ of 0.6–0.8, washed, pelleted, and added to 20 ml SD dropout medium supplied with 2% galactose. Induced cultures were incubated on a 30°C shaker for 48 h. The cultures were then centrifuged, and the pellets were lysed in a tissue lyser. The supernatant and lysed pellets were pooled back together and extracted with an equal volume of ethyl acetate three times. Ethyl acetate extracts were fully evaporated on a rotary evaporator, then resuspended in 50 μl of hexane for GC–MS analysis or 150 μl of methanol for LC–MS analysis.

The lengths of chloroplast signal peptides in the diterpene synthases were predicted using ChloroP (https://services.healthtech.dtu.dk/services/ChloroP-1.1/). The truncated versions of SdKSL1 and SdKSL2 (NCBI accession numbers KX268507 and KX268508) were synthesized as SdKSL1Δ77 and SdKSL2Δ73, with the cleavage sites (indicated by the green arrows in supplemental Figure 10) chosen by alignment with the SbbKLS genes derived from S. barbata (supplemental Figure 10). These truncated genes were inserted into the POPIN-F vector by In-Fusion cloning using ClonExpress II One Step Cloning (Vazyme, Nanjing, China) for expression in E. coli. Colonies were screened by colony PCR to check the size of the inserted gene, and those with the correct band size were sent out for sequencing. pOPINF or pOPINM expression vectors carrying N-terminally truncated versions of the diterpene synthases were transformed into E. coli (BL21) for protein expression and purification.

Selected recombinant E. coli colonies were inoculated onto LB medium containing selective antibiotics and cultured overnight at 220 rpm and 37°C. This overnight culture was used to inoculate Terrific Broth (TB) medium containing the same selective antibiotics at a 1:50 dilution. The TB culture was incubated at 37°C and shaken at 220 rpm until the OD₆₀₀ reached 0.8–1.0. Cultures were cooled to 16°C for 30 min and induced with 1 mM isopropyl thio-β-galactoside, then incubated at 16°C for 16 h before centrifugation. The pellets were resuspended in chilled lysis buffer (50 mM Tris–HCl, 50 mM glycine, 5% v/v glycerol, 0.5 M NaCl, 20 mM imidazole, 0.2 mg ml⁻¹ lysozyme, and 1 mM phenylmethylsulfonyl fluoride or Complete Protease Inhibitor Cocktail [Roche] [pH 8]) and kept on ice for 30 min. The lysate was disrupted by sonication (Scientz JY92-IIN) and centrifuged at 4°C (35 000 rpm, 20 min). A slurry of 200–250 μl Ni NTA beads (Smart LifeSciences) was added to the collected supernatants to pull down the His-tagged protein. After gentle shaking on a rocking platform at 4°C for 1.5 h, the mixture was centrifuged and washed twice with ice-cold lysis buffer (without protease inhibitor or lysozyme). Proteins were eluted with 600 μl elution buffer (50 mM Tris–HCl [pH 8], 50 mM glycine, 5% v/v glycerol, 0.5 M NaCl, 0.5 M imidazole), and the eluate was applied to a 4-ml Amicon Centrifugal 30k NMWL tube for dialysis, in which the buffer content was exchanged to phosphate-buffered saline (0.137 M NaCl, 0.0027 M KCl, 0.01 M phosphate buffer [pH 7.4]). Protein concentration was measured by the A280 value using a NanoDrop spectrophotometer, and samples were diluted to approximately 1 mg ml^–1 to avoid precipitation. Purified enzymes were divided into aliquots, snap frozen in liquid N₂, and stored at −80°C.

In vitro enzyme assays for class I diterpene synthases

The enzyme assay buffer consisted of 50 mM Tris–HCl (pH 7.2), 100 mM KCl, 7.5 mM MgCl₂, 5 mM DTT, and 5% glycerol; GGPP was added to 100 μM in 250-μl assays. Approximately 100 μg class II diterpene synthase SsKPS, alone or in combination with 80 μg class I diterpene synthases SdKSL1 and SdKSL2, was used in the reaction system. Assays were incubated at 37°C overnight in the dark. As a positive control for the SsKPS enzyme assay, the assay mixture was further incubated with 2 μl alkaline phosphatase (CIAP Takara Bio) for 2 h. Enzyme products were extracted three times with hexane, then dried and resuspended with 50 μl hexane before GC–QTOF analysis.

GC–MS analysis

Ethyl acetate extracts of yeast products were analyzed using an Agilent DB-5HT column (30 m × 0.25 mm × 0.1 μm) on a GC7890B-MS7200B QTOF. Samples (2 μl) were injected in split mode (split ratio 5:1) at an inlet temperature of 250°C. The gradient program was as follows: hold at 50°C for 2 min; increase to 110°C at 4°C min⁻¹; increase to 250°C at 8°C min⁻¹; increase to 310°C at 10°C min⁻¹; and hold for 5 min. The ion source temperature of the mass spectrometer was set to 230°C, and the spectra were recorded from m/z 50 to m/z 350.

LC–MS analysis

Ethyl acetate extracts of yeast products were dried with a rotary evaporator, then dissolved in methanol and measured by LC–MS (Q-Exactive plus) using an Agilent BEH C18 column (100 × 3.1 mm) with a 2.6-μm particle size. The aqueous eluent phase A was 2 mM ammonium formate + 0.01% formic acid, and the organic eluent phase B was acetonitrile. The conditions were as follows: increase the concentration of B from 40% to 99% from 0 to 10 min, and hold at 99% for 2 min.

qRT–PCR

Roots, stems, flowers, and leaves of S. splendens were collected at the flowering stage, and each tissue was extracted in three biological replicates. RNA was extracted using the CTAB method. Subsequent experiments were performed with the TB Green Premix DimerEraser reagent (Perfect Real Time, Takara) on an ABI StepOne Plus system, with SspqACTIN1 as the reference gene. The qRT–PCR conditions were initial denaturation at 95°C for 30 s, followed by 40 cycles of 95°C for 3 s, 55°C for 30 s, and 72°C for 30 s. Gene expression was calculated as described in our previous work (Li et al., 2023). The primers for qRT–PCR are shown in supplemental Table 2.

S. splendens tissue extracts

Root, leaf, flower, and stem tissues of S. splendens were collected from the greenhouse, snap frozen, ground into powder, transferred to 2-ml tubes, and extracted with acetone at room temperature overnight in a shaker at 200 rpm. The tubes were then sonicated for 10 min twice and centrifuged at 10 000 rpm for 10 min. The supernatant was dried in a rotary evaporator and re-dissolved in 150 μl methanol before LC–MS (Q-Exactive) examination. Clerodanes were tentatively identified from mass spectra on the basis of molecular weight and exact mass.

LC–MS analysis of peltate trichome metabolites

Peltate trichomes (about 50 μm in diameter) from the abaxial side of an S. splendens leaf were collected manually using a scalpel blade tip under a dissection microscope. About 100–200 trichomes were pooled and extracted with MeOH. The trichome extracts were evaporated down to 50 μl in volume, and 2 μl of the extract was examined using an Agilent Q Exactive equipped with an electrospray ionization source under positive and negative ion mode with the collision energy set to 40 hcd. Metabolites were analyzed using an Agilent Kinetex C19 column (2.6 μm, 2.1 × 100 mm). The mobile phase with a flow rate of 0.4 ml min⁻¹ consisted of solvent A (2 mM ammonium formate with 0.1% formic acid) and solvent B (ACN). The binary gradient elution was performed as follows: linear gradient 40%–99% B from 0 to 10.0 min, 99% B from 10.0 to 12.0 min, returned to 40% B at 12.1 min, and maintained until 14.0 min for column equilibration. Clerodanes were identified from mass spectra on the basis of molecular weight and exact mass and compared with standards.

MALDI–MSI analysis

Mature leaves of S. splendens were clamped flat between two glass slides and vacuum dried in a desiccator. Optical images of the abaxial side of each leaf were obtained using a Canon 5D Mark IV camera with a Canon MP-E 65 mm f/2.8 1–5× Macro Photo lens. Leaf samples were covered with 2,5-dihydroxybenzoic acid matrix (DHB, Sigma-Aldrich) using a SunCollect MALDI Sprayer (SunChrome) with a DHB solution of 10 mg ml⁻¹ in 80% methanol to a density of approximately 3 μg mm⁻². MALDI imaging was performed using a Synapt G2-Si mass spectrometer with a MALDI source (Waters) equipped with a 2.5-kHz Nd:YAG (neodymium-doped yttrium aluminum garnet) laser operated at 355 nm. Red phosphorous clusters were used for instrument calibration.

The slides with the leaves were fixed in the instrument metal holder and were scanned using a flat-bed scanner (Canon). The images were used to generate pattern files and acquisition methods in HDImaging software version 1.4 (Waters). An area of approximately 6.5 × 4 mm was selected and scanned with the 60-μm laser beam using steps of 45 μm. The instrument was set to MALDI–MS-negative sensitivity mode, m/z 100–1200, scan time 0.5 s, laser repetition rate 1 kHz, and laser energy 220.

The raw MS files were processed in HDI1.4 with the following parameters: detection of the 2000 most abundant peaks, m/z window 0.05, and MS resolution 10 000. A list with the masses of interest was loaded as a target mass file. The processed data were loaded into HDI1.4 and normalized by total ion content. The MALDI images were generated using the HotMetal2 color scale and overlaid with the original optical images to show the location of the MALDI signals. To improve the resolution of the imaging, the range for visibility for all 5 masses was fine-tuned on the images, which enhanced many signals, especially weak ones. Image smoothing was also applied. The size of the pixels is determined by the acquisition parameters, including the laser spot diameter and step size, and a step size of 45 μm was used for these images. A correlation search for the 5 masses was run against m/z 331 (salviarin). This returned 341 (hautriwaic acid) with a correlation value of R = 0.73. However, images for the lower correlation values shown in the correlation table (supplemental dataset) did not show significant similarities in the distribution of specific masses.

Data availability

Gene sequence data for this study have been deposited and are publicly available at the National Center for Biotechnology Information (NCBI) database OR837137–OR837141 and at GenBase of the China National Genomics Data Center (https://ngdc.cncb.ac.cn) under accession numbers C_AA104630.1 to C_AA104651.1. Datasets are available at Science Data Bank (ScienceDB, https://doi.org/10.57760/sciencedb.20358).

Funding

The work was supported by the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB27020204 awarded to E.C.T.); the International Partnership Program of the Chinese Academy of Sciences (153D31KYSB20160074 awarded to E.C.T.); and the National Natural Science Foundation of China, Research Fund for International Excellent Young Scientists (RFIS-II) (grant 32150610477, awarded to E.C.T.). H.L. acknowledges support from the Youth Fund of the National Natural Science Foundation of China (32200313). R.L. and C.M. gratefully acknowledge support from the BBSRC ISP Grant “Harnessing Biosynthesis for Sustainable Food and Health (HBio)” (BB/X01097X/1). C.M. and E.C.T. gratefully acknowledge the Royal Society for the Newton Advanced Fellowship awarded to E.C.T. (NAF∖R2∖192001).

Acknowledgments

The authors thank the staff and management of the CEMPS Core Facility Center for their excellent support in metabolomics (LC–MS, NMR) and microscopy services, the personnel at the CEMPS glasshouse facilities, and Matt Downie from the John Innes Center for conducting SEM imaging work. The authors also thank Prof. A.M. Makris (INAB-CERTH) for providing the AM119 yeast strain. R.L., H.L., and E.C.T. have filed a patent application.

Author contributions

E.C.T. and C. Martin initiated and conceived the project and designed the experimental strategy. R.L. assembled the S. splendens transcriptomic data, performed the bioinformatic analysis (co-expression analysis MR and SOM; syntenic analysis, phylogenetic trees), cloned and expressed the S. splendens cytochrome P450, assembled the pathway in yeast, and analyzed the yeast assays. H.L. assembled the S. divinorum transcriptomic data, designed and expressed the synthetic genes from S. divinorum in E. coli (class I diterpene synthases) and in yeast (cytochrome P450s), and analyzed the enzyme assays. R.L., C. Martins, and G.S. performed the MALDI–MSI analysis. M.F. and C.D.V. supplied a sample of annonene. Y.X., Z.W., and L.L. provided valuable help in cloning, protein expression in E. coli, plant extraction, and trichome isolation. R.L., H.L., C. Martin, and E.C.T. wrote the manuscript with valuable input from all authors.

Published: February 18, 2025