Unearthing a sesterterpene biosynthetic repertoire in the Brassicaceae through genome mining reveals convergent evolution

Significance

Sesterterpenoids are a rare terpene class with largely unexplored chemical diversity and bioactivities, representing a potential new drug source. Sesterterpene synthases catalyze the first committed biosynthesis step and shape sesterterpene structural diversity. Here, using a customized algorithm for systematically surveying plant genomes, we unearth a sesterterpene synthase gene repertoire in the Brassicaceae that synthesizes fungal-type sesterterpenes with diverse scaffolds, thus fueling the drug-discovery pipeline. Mechanistic studies shed light on the origin of structural diversification and offer insights for expanding nature’s diversity by protein engineering. Our work also supports the concept of convergent evolution in natural product biosynthesis.

Abstract

Sesterterpenoids are a rare terpene class harboring untapped chemodiversity and bioactivities. Their structural diversity originates primarily from the scaffold-generating sesterterpene synthases (STSs). In fungi, all six known STSs are bifunctional, containing C-terminal trans-prenyltransferase (PT) and N-terminal terpene synthase (TPS) domains. In plants, two colocalized PT and TPS gene pairs from Arabidopsis thaliana were recently reported to synthesize sesterterpenes. However, the landscape of PT and TPS genes in plant genomes is unclear. Here, using a customized algorithm for systematically searching plant genomes, we reveal a suite of physically colocalized pairs of PT and TPS genes for the biosynthesis of a large sesterterpene repertoire in the wider Brassicaceae. Transient expression of seven TPSs from A. thaliana, Capsella rubella, and Brassica oleracea in Nicotiana benthamiana yielded fungal-type sesterterpenes with tri-, tetra-, and pentacyclic scaffolds, and notably (−)-ent-quiannulatene, an enantiomer of the fungal metabolite (+)-quiannulatene. Protein and structural modeling analysis identified an amino acid site implicated in structural diversification. Mutation of this site in one STS (AtTPS19) resulted in premature termination of carbocation intermediates and accumulation of bi-, tri-, and tetracyclic sesterterpenes, revealing the cyclization path for the pentacyclic sesterterpene (−)-retigeranin B. These structural and mechanistic insights, together with phylogenetic analysis, suggest convergent evolution of plant and fungal STSs, and also indicate that the colocalized PT–TPS gene pairs in the Brassicaceae may have originated from a common ancestral gene pair present before speciation. Our findings further provide opportunities for rapid discovery and production of sesterterpenes through metabolic and protein engineering.

Sesterterpenoids are a largely unexplored class of terpenes with only around 1,000 members isolated from nature so far, representing a mere <2% of the reported terpene family members (>70,000). These compounds are structurally diverse and have a wide spectrum of biological activities, ranging from antiinflammatory, anticancer, cytotoxic, and antimicrobial bioactivities to phytotoxicity and plant defense (1). Recent work on sesterterpenoids has suggested that this terpene class could be an important new source of anticancer drugs (2, 3). The majority of sesterterpenoids characterized to date are from marine sponges and terrestrial fungi, with only ∼60–70 being of plant origin. Many of these plant sesterterpenes were isolated from the mint family (Lamiaceae) and have been implicated in plant defense (4–7). Although large transcriptome datasets for the mint family have been generated (8), very limited genome sequences for the Lamiaceae are currently available (9, 10).

Like other classes of terpenes, the structural diversity of sesterterpenes largely originates from the very first scaffold-generating step, which in this case is catalyzed by sesterterpene synthases (STSs), a class of terpene synthase (TPS). In fungi, six STSs have recently been identified and shown to synthesize ophiobolin F (in Aspergillus clavatus) (11), sesterfisherol (in Neosartorya fischeri) (12), and stellata-2,6,19-triene, (2E)-α-cericerene, quiannulatene, and astellifadiene (in Emericella variecolor) (13–16) (Fig. 1A). These fungal enzymes are all bifunctional, containing a C-terminal trans-prenyltransferase (PT) and an N-terminal TPSs domain. The PT domain synthesizes the universal C25 sesterterpene precursor geranylfarnesyl diphosphate (GFPP, 1) from dimethyl allyl diphosphate (DMAPP) and isopentenyl diphosphate (IPP), which is then cyclized by the TPS domain to form diverse scaffolds. Some fungal diterpene synthases (DTSs), such as Phomopsis amygdali fusicoccadiene synthase (17, 18) and phomopsene synthase (19), are also bifunctional. Fusion of PT and TPS activities has been proposed to contribute to more efficient production of bioactive terpenes from precursor supply (18, 20). In most cases, however, the condensation and cyclization steps for diterpene biosynthesis in fungi are carried out by individual enzymes encoded by genes that are sometimes colocalized within fungal biosynthetic gene clusters (21).

Fig. 1.

Identification of candidate STS genes from plant genomes. (A) Chemical structures of diterpenes and sesterterpenes synthesized by fungal bifunctional DTSs and STSs. (B) Phylogenetic tree constructed using TPSs from 55 plant genomes together with fungal and bacterial TPSs. The TPSs for which the genes were found to be colocalized with PT genes with high stringency are shown in red, and those with low stringency in green. Bifunctional fungal TPSs are shown in dark blue. (C) A graphic representation of PT–TPS gene pairs from A. thaliana, C. rubella, and B. oleracea selected for functional validation in this work.

In plants, two TPSs from Arabidopsis thaliana were recently reported to synthesize the sesterterpenes arathanatriene (2) and retigeranin B (3) (22). Intriguingly, these two TPSs are colocalized with PTs that synthesize the precursor substrate GFPP. To our knowledge, bifunctional TPSs that contain both PT and TPS domains have not been reported from any plant species. This raises questions regarding the landscape of PT and TPS genes and their related functions in plant genomes. We have previously developed algorithms for mining plant genomes for genes encoding natural product biosynthetic pathways (23, 24). Here, we carried out a systematic search of 55 sequenced plant genomes using a customized version of the plantiSMASH genome mining algorithm (24) to identify genes predicted to encode PT or TPS domains. Although we did not find evidence for bifunctional genes encoding both PT and TPS domains, we identified a pool of colocalized pairs of PT and TPS genes using this algorithm. Phylogenetic analysis of all TPSs from the 55 plant genomes that were investigated identified a single clade containing those TPS genes that were physically clustered with PT genes, greatly expanding on previous observations (25). Transient expression of selected TPS genes from this group from A. thaliana (thalecress), Capsella rubella (pink shepherd's purse), and Brassica oleracea in heterologous host Nicotiana benthamiana revealed five additional TPSs that make a variety of previously unknown fungal-type sesterterpenes with diverse scaffolds, including 11/6/5 tricyclic (−)-caprutriene (7), 5/12/5 tricyclic (−)-variculatriene A (5), 5/8/6/5 tetracyclic (−)-aleurodiscalene A (9), 5/5/5/6/5 pentacyclic (−)-ent-quiannulatene (4) and (+)-boleracene (8), and 5/4/7/6/5 pentacyclic (+)-astellatene (6) (Fig. 2B, SI Appendix, Tables S1–S8, and Dataset S1). Among these scaffolds, (−)-ent-quiannulatene (4) is an enantiomer of the fungal metabolite (+)-quiannulatene. Homology modeling and sequence comparisons implicated a key amino acid as being likely to be important for scaffold diversification. Site-directed mutagenesis of this residue further expanded scaffold diversity and, coupled with quantum chemical calculations, shed light on the cyclization mechanisms leading to the formation of the pentacyclic sesterterpenes (−)-retigeranin B (3) and (+)-astellatene (6). The carbocation cascade sequences following protonation of the GFPP substrate in the active site of plant STSs likely mirror those of the distantly related fungal bifunctional STSs. However, phylogenetic analysis suggests the independent evolution of fungal and plant STSs, indicating that the plant and fungal enzymes have arisen by convergent evolution.

Fig. 2.

Functional analysis of candidate STSs using transient tobacco expression. (A) Comparative GC-MS total ion chromatograms (TICs) of extracts from N. benthamiana leaves transiently coexpressing different combinations of GFPPS–TPS gene pairs. A. tumefaciens LBA4404 carrying the green florescent protein gene in pEAQ-HT was used as a control. (B) Chemical structures of compounds 2–9 synthesized by the selected TPSs (AtTPS17, -18, -19, -25, -30, -Cr237, and Bo250). (C) Extracted ion chromatogram (m/z = 340) area percentages of the identified compounds in the total sesterterpene mixtures generated by the different TPSs. Sum area percentages represent the percentages of all characterized compounds in the total sesterterpenes made by each TPS. Note that the sum does not amount to 100% because some uncharacterized and very minor sesterterpenes are not included.

Although mining plant genomes and transcriptomes for individual PTs and TPSs has proven to be effective for uncovering terpene diversity (26–28), to our knowledge systematic genome mining for colocalized PT and TPS genes has not previously been attempted. Collectively, our work sheds light on the genomic organization of plant PT and STS genes and demonstrates an effective pipeline for plant sesterterpene discovery, from finding genes to production by engineering in the heterologous host N. benthamiana and further manipulation by protein engineering.

Results

Landscape of PT and TPS Genes in Plant Genomes and Identification of a Phylogenetic Clade of Candidate Plant STSs Using a Customized Genome Mining Algorithm.

To search for genes with predicted PT and TPS domains across 55 sequenced high-quality plant genomes (Dataset S2), we used the Polyprenyl_synt (PT) and Terpene_synth C (TPS) models from the Pfam database (29) in a version of plantiSMASH (24) with gene cluster detection logic customized for this specific purpose. The Terpene_synth_C pHMM (pfam.xfam.org/family/Terpene_synth_C#tabview=tab7) was seeded from plant, fungal, and bacterial TPSs and so is applicable across plants and microbes. Although we could not identify any candidate bifunctional TPS genes, we were able to identify 21 colocalized PT–TPS gene pairs using our default (high-stringency) parameters [cut-off 25 (±5) kb; deviation calculated using gene density/average intergenic distance for the neighboring 10 genes as a multiplier] (Dataset S2). An additional 75 candidate PT–TPS gene pairs were identified at lower stringency (allowing for distances between the PT and TPS domains that were up to five times larger) (Dataset S2). We next constructed a phylogenetic tree (Fig. 1B) using 2,846 TPS sequences extracted from the 55 plant genomes (Dataset S2), together with 724 fungal TPSs (including the known bifunctional TPSs). A total of 278 bacterial TPSs was included as an outgroup. This phylogenetic analysis enabled us to identify a single TPS clade in which the TPS genes that were colocalized with PT genes as identified using the high-stringency cut-off were heavily represented (Fig. 1B). Nine of the PTs in these PT–TPS gene pairs have indeed been previously shown to be functional plant GFPPSs (25, 30), and two of the TPSs (AtTPS18 and AtTPS19 from A. thaliana) were shown to biosynthesize the sesterterpenes (+)-arathanatriene and (−)-retigeranin B (22). However, two other TPSs from these pairs [AtTPS25 (At3g29410) and AtTPS30 (At3g32030) from A. thaliana] have recently been reported to have diterpene or sesquiterpene activities when expressed in Escherichia coli (31). A total of 18 TPS genes, including AtTPS25 and AtTPS30, that are colocalized with PT genes were grouped together in this clade (marked in red, Fig. 1B and SI Appendix, Fig. S1A), greatly outnumbering previous observations (25). Two outlier TPS genes (one from Vigna angularis and one from Brassica napus, marked as red dots in Fig. 1B) from the 21 PT–TPS gene pairs identified using high-stringency cut-off were located in other clades (Fig. 1B and SI Appendix, Fig. S1B). One TPS gene from these 21 PT–TPS gene pairs was filtered out during tree construction. The characterized fungal bifunctional TPSs (including STSs and DTSs) grouped together in two closely related clades that were clearly distinct from this plant TPS clade, suggesting that the plant TPS (red in Fig. 1B) and the fungal bifunctional STS clades (blue in Fig. 1B) have evolved independently. Other TPS genes that colocalized with PT genes over larger distances (low stringency) were spread across different clades (green dots in Fig. 1B). To understand the functions of the TPSs that emerged from our search and investigate the significance of this phylogenetic grouping, we chose to functionally test seven TPSs (five from A. thaliana, one from C. rubella, and one from B. oleracea) for which the genes are clustered with PT genes from the main clade (Fig. 1C and SI Appendix, Fig. S1A), along with two other A. thaliana TPSs [AtTPS22 (At1g33750) and AtTPS29 (At1g31950)] (SI Appendix, Fig. S1A) for which the genes are not clustered with a PT gene yet are part of the same monophyletic branch within the TPS phylogeny.

AtTPS17, -18, -19, -25, and -30 from A. thaliana, Cr237 from C. rubella, and Bo250 from B. oleracea Are all Functional STSs.

The selected TPSs, AtTPS17 (At3g14490), AtTPS18 (At3g14520), AtTPS19 (At3g14520), AtTPS25 (At3g29410), and AtTPS30 (At3g32030) from A. thaliana, Cr237 (CARUB_v10016237mg) from C. rubella, and Bo250 (LOC106343250) from B. oleracea all have corresponding clustered PT genes, namely AtGFPPS0 (At3g14510), AtGFPPS1 (At3g14530), AtGFPPS2 (At3g14550), AtGFPPS3 (At3g29430), AtGFPPS4 (At3g32040), CrGFPPS1 (CARUB_v10014047mg), and BoPT1 (LOC106293690), respectively (Fig. 1C). In contrast, AtTPS22 and AtTPS29 are not paired with any PT genes in the genome. AtGFPPS0 is a pseudo gene encoding a nonfunctional protein with a frame shift, whereas AtGFPPS1–4 and CrGFPPS1 are characterized GFPPSs (25, 30). BoPT1 has not been previously identified or characterized. N-terminal transit peptide (N-tp)–truncated recombinant AtTPS18 and AtTPS19 proteins were recently coexpressed with N-tp AtGFPPS1 and AtGFPPS2 in E. coli and reported to synthesize (+)-arathanatriene and (−)-retigeranin B, respectively (22). Recombinant proteins encoded by N-tp–truncated AtTPS22 and AtTPS25 were also recently expressed in E. coli but reported to have sesquiterpene synthase activity, making sesquiterpene mixtures when incubated with E,E-farnesyl diphosphate (E,E-FPP) as a substrate in vitro (31). Recombinant proteins encoded by N-tp–truncated AtTPS30 were found to have DTS activities when assayed with GGPP or ent-copalyl diphosphate (ent-CPP) as substrates in vitro, whereas AtTPS29 was not active in these assays (31). AtTPS17, Cr237, and Bo250 have not been investigated previously. We generated expression constructs for each of seven TPSs from the PT–TPS gene pairs (A. thaliana AtTPS17, -18, -19, -25, and 30; C. rubella Cr237; B. oleraceae Bo250), three AtGFPPSs (A. thaliana AtGFPPS1, -3, -4), and the two nonclustered A. thaliana TPS AtTPS22 and AtTPS29 genes using a vector designed for transient expression in N. benthamiana (pEAQ-HT) (32). These constructs, each harboring individual genes, were transformed into Agrobacterium tumefaciens. A. tumefaciens strains containing the different expression constructs were then used for transient expression in N. benthamiana leaves. This tobacco transient-expression platform allows high-level expression of genes of interest, especially those of plant origin, thus enabling us to rapidly test the biochemical functions of individual genes (33). Furthermore, different combinations of genes can be coexpressed simply by mixing A. tumefaciens strains containing the different expression constructs and coinfiltrating these into the leaves (33).

When we expressed all nine TPSs (AtTPS17, -18, -19, -25, -30, -22, -29, Cr237, and Bo250) individually in N. benthamiana, trace compounds 2, 3, 4, 6, 7, and 8 were detected from extracts of leaves expressing AtTPS18, -19, -25, -30, Cr237, and Bo250 (SI Appendix, Fig. S2B). However, we did not observe the formation of sesquiterpenes or diterpenes, even for AtTPS22, -25, -29, and -30, which had previously been reported to synthesize these types of terpenes in vitro (31). This outcome may be because of the low activity of the encoding enzymes, inadequate precursor supply, or the loss of the volatile/semivolatile products in N. benthamiana, especially if these products were produced in very low abundance. Expression of AtGFPPS1, -3, and -4 alone in N. benthamiana resulted in the accumulation of two minor new peaks (1c and 1d in SI Appendix, Fig. S2B), which were tentatively identified as β-geranylfarnesene and geranylfarnesol, based on their mass spectra (SI Appendix, Fig. S2C). These are likely to be hydrolyzed products of GFPP generated by endogenous N. benthamiana enzymes. However, when AtTPS17, -18, -19, -25, -30, Cr237, or Bo250 were coexpressed with a GFPPS (regardless of whether it was the corresponding paired one) in N. benthamiana, new peaks with a characteristic sesterterpene mass fragment at m/z = 340 (calculated for C₂₅H₄₀ = 340) were detected (Fig. 2A and SI Appendix, Figs. S2E and S3 A–H), indicating production of sesterterpenes by these TPSs. The electron-ionization (EI)-MS spectra of compounds 2 and 3 agreed with those of (+)-arathanatriene and (−)-retigeranin B characterized recently (22). We did not detect any sesterterpenes when AtTPS22 or AtTPS29 were coexpressed with a GFPPS. Apart from AtTPS17, which produced multiple products in very low yields, AtTPS18, -19, -25, -30, Cr237, and Bo250 all produced dominant peaks corresponding to compounds 2–8 (Fig. 2 A and C), suggesting that GFPP was limiting in N. benthamiana leaves in the absence of a coexpressed GFPPS. Compounds 2–8 were then isolated for structural elucidation.

AtTPS18, -19, -25, -30, Cr237, and Bo250 Synthesize the Fungal-Type Sesterterpenes (+)-Arathanatriene (2), (−)-Retigeranin B (3), (−)-Ent-Quiannulatene (4), (−)-Variculartriene A (5), (+)-Astellatene (6), (−)-Caprutriene (7), (+)-Boleracene (8), and (−)-Aleurodiscalene A (9).

To isolate compounds 2–9 for structural elucidation, we scaled up the transient expression of these different GFPPS–TPS pairs in N. benthamiana (34). Following extraction, repeated chromatography, and recrystallization, we obtained quantities of pure compounds 2 (102.0 mg), 3 (13.0 mg), 4 (3.7 mg), 5 (3.5 mg), 6 (4.2 mg), 7 (21.3 mg), 8 (17.3 mg), and 9 (2.1 mg) for collecting full 1D and 2D NMR datasets (Dataset S1). In addition to NMR data, we also used X-ray diffraction analysis to solve these complex structures unambiguously. We synthesized the epoxides 10, 12, and 13 of compounds 2, 4, and 6 and the enone 11 of compound 3 (Fig. 3) to obtain single crystals for X-ray diffraction analysis. We also succeeded in obtaining single crystals of 7, 8, and 9 for crystallography. The crystal structures of 7–13 enabled us to establish the relative structures of (+)-arathanatriene (2), (−)-retigeranin B (3), (−)-ent-quiannulatene (4), and (+)-astellatene (6), (−)-caprutriene (7), (+)-boleracene (8), and (−)-aleurodiscalene A (9) unambiguously, as depicted in Fig. 2B. The spectroscopic data of compounds 2 and 3 agreed with those reported in the literature (22). The most likely absolute structures of the compounds (2–4 and 6–9) were established using Bayesian statistics on the Bijovet differences (35) of crystal structures 7–13. (−)-ent-Quiannulatene (4) is an enantiomer of the fungal sesterterpene (+)-quiannulatene, sharing an identical relative structure, and its absolute structure was also confirmed by comparing its specific optical rotation (${\left[\alpha \right]}_{\text{D}}^{20}$ = −43.9 in benzene) with that of quiannulatene (${\left[\alpha \right]}_{\text{D}}^{28}$ = +41.7 in benzene-d₆) reported in the literature (15). The absolute structure of compound 5, which is a minor product of AtTPS25, was elucidated by 1D and 2D NMR analysis (SI Appendix, Table S4), as well as the fact that it is a product derived from one carbocation in the cyclization pathway toward (−)-ent-quiannulatene (4) (Fig. 4). This compound is named (−)-variculatriene A based on the 5/12/5 scaffold of variculanol, which was isolated from E. variecolor in 1991 (36). (−)-Retigeranin B (3) and (+)-astellatene (6) are direct precursors of fungal sesterterpenoids retigeranic acid B [isolated from lichens in 1972 (37–39)] and astellatol [isolated from E. variecolor in 1989 (40)], respectively (SI Appendix, Fig. S4). (−)-Aleurodiscalene A (9) shares the same scaffold with the antimicrobial sesterterpene glycoside aleurodiscal isolated from Aleurodiscus mirabilis in 1989 (41) (SI Appendix, Fig. S4).

Fig. 3.

Synthesis of sesterterpene derivatives (10–13) for X-ray diffraction analysis. Epoxides 10, 12, and 13 were synthesized by epoxidation with meta-chloroperoxybenzoic acid (m-CPBA). Enone 11 was synthesized by allylic oxidation with pyridinium dichromate (PDC) and tert-butyl hydroperoxide (TBHP) in dichloromethane (DCM). Crystal structures (10–13) are presented with displacement ellipsoids shown at 50% probability.

Fig. 4.

Proposed cyclization paths toward the formation of fungal-type sesterterpenes 2–9 and 14–16 by plant STSs. The universal sesterterpene precursor GFPP is cyclized to form the unified bicyclic C12 cation 1b (black box) following protonation in the active sites of plant STSs and mutated AtTPS19 (AtTPS19^428D). Cation 1b diverges to 5/12/5 and 11/6/5 tricyclic carbocations en route to the formation of (+)-arathanatriene (2), (−)-retigeranin B (3), (−)-ent-quiannulatene (4), (−)-variculatriene A (5), (+)-astellatene (6), (−)-caprutriene (7), (+)-boleracene (8), (−)-aleurodiscalene A (9), (−)-fusaproliferene (14), (−)-variculatriene B (15), and (+)-aleurodiscalene B (16). Compounds isolated and characterized are highlighted in colored boxes. Different colors indicate different cyclization paths. Crystal structures 7–9 are presented with displacement ellipsoids shown at 50% probability.

With these pure compounds in hand, we also detected compounds 2, 3, 4, and 6, (Fig. 5), which are products of A. thaliana TPSs in the plant of origin. We found that these four sesterterpenes are present across different A. thaliana tissues at various levels. (+)-Arathanatriene (2) and (−)-retigeranin B (3) were more abundant in the siliques and seeds, whereas (−)-ent-quiannulatene (4) and (+)-astellatene (6) were detected only in roots. The metabolite profiles are in good agreement with the expression profiles of the corresponding TPSs, suggesting that AtTPS18, -19, -25, and -30 are active and could function as STSs in planta (Fig. 5 and SI Appendix, Figs. S5 and S6).

Fig. 5.

Detection of compounds 2, 3, 4, and 6 from different tissues of A. thaliana accession Col-0. SD bars for three biological replicates are shown.

Divergence of Carbocation Cyclization Pathways Gives Rise to the Structural Diversity of Sesterterpenes 2–9.

The diverse fungal-type sesterterpene products of the plant STSs identified in this work arise from the divergence of carbocation cyclization pathways following the formation of a common ancestral bicyclic C12 carbocation 1b (Fig. 4). Feeding experiments with ¹³C-labeled precursors in fungi have suggested the formation of a quiannulatene scaffold via a 5/12/5 tricyclic carbocation and an intriguing cyclobutane intermediate (15, 42). With the isolation of tricyclic (5/12/5) and tetracyclic (5/8/6/5) carbocation-derived (−)-variculatriene A (5) and (−)-aleurodiscalene A (9) as minor products of AtTPS25 and Bo250, which produce mainly the quiannulatene scaffold sesterterpenes 4 and 8, respectively (Fig. 2C), it is very likely that cyclization of GFPP toward (−)-ent-quiannulatene (4) and (+)-boleracene (8) in plant STSs adopts the same path as (+)-quiannulatene in fungal bifunctional STS (EvQS), as shown in Fig. 4. Although the astellatene and retigeranin B scaffolds have recently been proposed to also be derived from 5/12/5 tricyclic carbocations (12), previous feeding experiments with ¹³C-labeled precursor (43) and our finding that the 11/6/5 scaffold (2) is the major product of AtTPS18 and is present as a very minor product of AtTPS19, -25, and -30 (Fig. 2C) suggests that the bicyclic C12 carbocation may be diverged into tricyclic 5/12/5 and 11/6/5 cations en route to the formation of pentacyclic sesterterpenes, such as (−)-retigeranin B (3) and (+)-astellatene (6). Such divergence is driven by inherent energetics of the carbocation reactions involved, the conformations allowed in STS active sites and interactions of carbocations with key amino acid residues in the active sites of STSs.

Mutation of Tyrosine428 to Aspartic Acid in AtTPS19 Reveals Intermediates in the Cyclization Pathway Toward the Formation of (−)-Retigeranin B (3).

Phylogenetic analysis of the seven STSs identified in this work showed that AtTPS18 is a relatively recently evolved homolog of AtTPS19 (SI Appendix, Fig. S1A). By comparing the protein sequences of these STSs (Fig. 6A and SI Appendix, Fig. S7), together with homology modeling of the encoding enzymes (Fig. 6B), we identified a conserved amino acid site present as an aromatic tyrosine (Tyr; Y) in the active sites of AtTPS17, -19, -25, -30, and Bo250. However, in AtTPS18, the corresponding residue is a negatively charged aspartic acid (Asp; D) in close proximity (approximately 2.75 Å) to the C12 of the GFPP substrate (Fig. 6B), suggesting that this site might be important for driving the formation of the 11/6/5 tricyclic scaffold of (+)-arathanatriene (2). We then carried out site-directed mutagenesis of the STSs from A. thaliana. We mutagenized the Tyr at the corresponding amino acid sites of AtTPS17, -19, -25, and -30 to Asp. This approach resulted in inactivation of AtTPS17, -25, and -30, as well as significant functional changes in AtTPS19, with (−)-retigeranin B dropping to 30% from 42% in wild-type AtTPS19 and the appearance of peak 16 and the increase of peaks 14 and 15 (Figs. 2C and 6C), whereas altering Asp to Tyr in AtTPS18 did not lead to metabolite profile change (Figs. 2C and 6C). The Tyr residue in AtTPS17, -25, and -30 therefore appears to be important for enzyme function and changing it to Asp could have possibly caused incorrect folding. Although AtTPS19 is a closer homolog to AtTPS18, sharing 87% protein sequence identity (Fig. 5A and SI Appendix, Fig. S8), the mutant form of AtTPS19, AtTPS19^428Asp, was still active after changing Tyr428 to Asp428. We then scaled-up coinfiltration of A. tumefaciens strains carrying pEAQ-HT/AtTPS19^428Asp and pEAQ-HT/AtGFPPS1 in N. benthamiana, isolated the compounds corresponding to peaks 14, 15, and 16, and established their structures as (−)-fusaproliferene (14), (−)-variculatriene (15), and (+)-aleurodiscalene (16), as depicted in Fig. 4 by NMR (SI Appendix, Tables S13–S15). Intriguingly, changing Tyr428 to Asp428 in AtTPS19 resulted in premature termination of the intermediate carbocations at different stages of the cyclization pathway toward (−)-retigeranin B (3) and led to the accumulation of the corresponding bi-, tri-, and tetracyclic products (−)-fusaproliferene (14), (−)-variculatriene B (15), (+)-arathanatriene (2), and (+)-aleurodiscalene B (16). We hypothesize that Tyr428 of AtTPS19 might play a role in stabilizing the carbocation intermediates in transition states via cation–π interactions (44), such that cyclization toward pentacyclic (−)-retigeranin B (3) can be completed. However, it is also possible that the enzyme activity of the AtTPS19^428Asp mutant is compromised such that not all of the intermediates could be cyclized to the final pentacyclic product. The fact that the composition of (+)-arathanatriene (2) increased from 5.5% of wild-type AtTPS19 products to 12% for the mutagenized AtTPS19^428Asp suggests that a common ancestral bicyclic C12 cation may have been diverted in the enzyme active sites into 5/12/5 and 11/6/5 tricyclic carbocations en route to the formation of pentacyclic sesterterpenes, such as (−)-retigeranin B (3), and that such divergence could occur within a single enzyme. Although changing Asp431 to Tyr431 in AtTPS18 did not alter its function and we have not yet identified the amino acid sites that drive the exclusive formation of the 11/6/5 tricyclic scaffold of 2, it is reasonable to hypothesize that AtTPS18 has evolved to be more specific and to have a more compact active site that would impose steric force on the bicyclic C12 cation (1b) to allow for the tight formation of the C2–C12 bond and subsequently the 11/6/5 tricyclic scaffold (2 and 7).

Fig. 6.

Protein sequence analysis, homology modeling and site-directed mutagenesis of A. thaliana STSs. (A) Sequence analysis of AtTPS17, -18, -19, -25, -30, Cr237, and Bo250 identified a conservation site implicated in structural diversification (protein sequences were aligned using MUSCLE). (B) Homology modeling of AtTPS18 with substrate GFPP (carbon backbone in cyan and phosphate in orange and red) docked in the active site showing the motif DDXXD and 431Asp suggested that 431Asp might be involved in making the 11/6/5 tricyclic scaffold of arathanatriene. The homology model of AtTPS18 was generated based on the crystal structure of 5-epi-aristolochene synthase on the Phyre² server (www.sbg.bio.ic.ac.uk/phyre2/html/page.cgi?id=index). (C) Comparative GC-MS total ion chromatograms of extracts of N. benthamiana leaves transiently coexpressing AtGFPPS1 with mutagenized AtTPS17^429Asp, AtTPS18^431Tyr, AtTPS19^428Asp, AtTPS25^431Asp, and AtTPS30^431Asp.

Discussion

The characterized bifunctional STSs from fungi and the plant STSs identified in this work are very distantly related (around 10% amino acid sequence identity) (SI Appendix, Fig. S8). However, both types of STSs are able to cyclize the universal precursor GFPP into similar or in some cases even identical scaffolds [e.g., (+)-quiannulatene by EvQS and (−)-ent-quiannlatene by AtTPS25] via virtually equivalent cyclization paths. This striking example of convergent evolution of terpenoids in plants and fungi has parallels with the well-known example of the diterpenoid gibberellin phytohormones (45). By delving into the carbocation cyclization mechanisms of the plant STSs that we have identified, we have demonstrated that cyclization of the acyclic GFPP precursor to pentacyclic sesterterpenes [e.g., (−)-retigeranin B and (−)-ent-quiannulatene] can diverge into 5/12/5 and 11/6/5 tricyclic scaffolds in one single enzyme after the formation of a common ancestral bicyclic C12 cation, supported experimentally by the isolation and characterization of intermediate cation derived compounds 2, 5, 9, and 14–16.

Quantum chemical calculations [mPW1PW91/6–31+G(d,p)//B3LYP/6–31+G(d,p)] (see Materials and Methods and Dataset S3 for details) were also performed on two reasonable cyclization pathways for the formation of (+)-astellatene (6) (Fig. 7). In path 1, a 11/6/5 ring system is formed first (carbocation D), whereas path 2 involves initial formation of a 5/12/5 ring system (carbocation K). On the basis of the computed energetics for these two rearrangements (Fig. 8), both pathways are energetically viable, but path 1 is inherently preferred, given that the highest energy transition-state structure along this pathway is predicted to be 6-kcal/mol lower than that along path 2. In both pathways, instances of the merging of chemical events (e.g., hydride shifts, carbocation-alkene cyclizations) into concerted processes are observed (e.g., E → G, G → I, L → N, N → I) (46). The results of these calculations indicate that several putative intermediates are unlikely to have lifetimes long enough for deprotonation to occur, allowing us to predict which carbocations are most likely to lead directly to sesterterpenes.

Fig. 7.

(A–O) Calculated structures for two possible cyclization paths en route to (+)-astellatene (6).

Fig. 8.

Computed energetics for two possible paths to (+)-astellatene (6). Energies computed with mPW1PW91/6–31+G(d,p)//B3LYP/6–31+G(d,p) are shown for the two pathways depicted in Fig. 7. Path 1 is energetically preferred. Numbers following carbocation letters indicate different conformations. See Dataset S3 for computed geometries.

By identifying a key amino acid residue Tyr428 in AtTPS19 important for the complete the cyclization toward (−)-retigeranin B (3), we have unveiled the bi-/tri-/tetracyclic structures 14, 15, 2, and 16 that are derived from likely carbocations in the cyclization path toward (−)-retigeranin B (3), demonstrating the capacity of plant STSs to generate these diverse scaffolds and showcasing the promising potential for directing the functional plasticity of STSs for creating chemical diversity via rational protein engineering (47). Solving the crystal structures of these enzymes will be key in guiding such approaches in the future. The fact that (−)-fusaproliferene (14) and (+)-aleurodiscalene B (16) are direct precursors of the known bioactive sesterterpenoids fusaproliferin (toxic) and aleurodiscal (antimicrobial) (SI Appendix, Fig. S4) illustrates the importance of harnessing the capacity to generate scaffolds of this type.

We present in this work the convergent evolution of STSs in fungi and plants as inferred from both metabolic and genomic data. Both fungi and plants appear to have independently evolved similar strategies for biosynthesizing structurally related sesterterpenes. Although, unlike the situation in fungi, plant STSs and GFPPSs are encoded by separate genes rather than bifunctional ones, it is intriguing that the genes encoding the STSs that we have characterized are colocalized in plant genomes with PT genes. These gene pairs are not obviously coexpressed (SI Appendix, Fig. S5) and the GFPPSs are interchangeable (Fig. 2A), suggesting that these gene pairs have not undergone intimate coevolution. The genomic colocalization of PT and TPS genes appears to be particularly prevalent in the Brassicaceae based on our analysis of currently available plant genomes. From the topology of the paired TPSs within the phylogenetic tree (Fig. 1B) and our functional analysis of selected TPSs from A. thaliana, C. rubella, and B. oleracea, it seems that the functions of closely related TPSs from these three species are quite conserved and that close homologs make sesterterpene analogs (Fig. 9). It is likely that these PT and TPS gene pairs may have originated from a common ancestral PT–TPS gene pair by duplication and functional divergence (23, 48–50). Although the biological roles of these sesterterpenes in plants are as yet unknown and inevitably many downstream products in the biosynthetic pathways remain to be elucidated, we expect that the biosynthetic and mechanistic insights provided by this work will facilitate the understanding and discovery of the biology and chemistry of this important class of compounds in plants.

Fig. 9.

Phylogenetic tree of the characterized plant and fungal STSs showing the phylogeny of these distantly related proteins. The chemical structures of analogs synthesized by plant STS homologs and the distantly related fungal STS EvQS are shown to aid in evolutionary interpretation. The tree was constructed using MEGA5.0 with 1,000 bootstrap resampling.

Materials and Methods

Bioinformatics Algorithms and Construction of Phylogenetic Trees.

A collection of 55 high-quality plant genomes (Dataset S2) was used as input for analysis with a customized version of plantiSMASH (plantismash.secondarymetabolites.org) (24). Terpene_synth_C (PF03936) and Polyprenyl_synt (PF00348) profile hidden Markov models (HMMs; pfam.xfam.org, database v30.0) were fitted into a custom gene cluster detection rule (Dataset S2).

Subsequently, the PF03936 profile HMM was used to collect all TPS-encoding genes from the 55 plant genomes, which resulted in the identification of 2,846 genes. A total of 724 fungal genes, including 17 bifunctional TPS genes (15), were also collected with the same method but queried against the reference proteome database with taxonomic filtering set to “Fungi.” Finally, 278 bacterial genes (same method, with filtering set to “Bacteria”) were included to act as an outgroup.

All sequences were aligned using the hmmalign program from the HMMer suite (51) with default parameters. Predicted repetitive subsequences were discarded. The remaining sequences were manually trimmed for tree reconstruction, which was performed using FastTree (v2.1) (52). The phylogenetic tree was visualized using iTOL software (53). Reliability tests were done using the FastTree-provided Shimodaira–Hasegawa test with 1,000× resamples without branch optimization.

Cloning and Transient Expression.

AtTPS19, AtTPS25, AtTPS30, AtGFPPS1, and AtGFPPS3 were amplified from a root cDNA library of A. thaliana accession Col-0. Coding sequences for AtTPS17, AtTPS18, AtTPS22, AtTPS29, and AtGFPPS4 were retrieved from The Arabidopsis Information Resource and those of Cr237 (CARUB_v10016237mg) and Bo250 (LOC106343250) from the National Center for Biotechnology Information and synthesized by Integrated DNA Technologies to include the 5′ and 3′ attB sites for Gateway cloning. These sequences were cloned into the pEAQ-HT expression vector for transient expression in N. benthamiana, as detailed in SI Appendix.

Metabolite Extraction and GC-MS Analysis.

N. benthamiana leaves expressing genes of interest were harvested 5 d postinfiltration, lyophilized, extracted with EtOAc, and analyzed directly by GC-MS, as described in SI Appendix. Qualitative and quantitative metabolite analysis were performed on an Agilent GC (7890B)-MSD (5977A) equipped with a robotic multipurpose autosampler (MPS) and a Zebra-5HT INFERO capillary column (35-m × 0.25-mm × 0.1-μm film thickness; phenomenex).

Isolation of Sesterterpenes Following Large-Scale Vacuum-Infiltration of N. benthamiana Leaves.

A. tumefaciens LBA4404 harboring expression constructs for the relevant GFPPSs and STSs were infiltrated into N. benthamiana leaves as described previously (34), with slight modifications using a custom-built vacuum infiltration system (SI Appendix, Materials and Methods). Leaves of N. benthamiana expressing desired constructs were harvested 5 d postinfiltration, lyophilized, and extracted with n-hexane. Pure compounds 2–9 and 14–16 were isolated from the extract as described in SI Appendix.

Chemical Synthesis of Sesterterpene Derivatives (10–13) for X-Ray Diffraction Analysis.

Compounds 10, 12, and 13 were synthesized by epoxidation of sesterterpenes 2, 4, and 5 with meta-chloroperoxybenzoic acid. Enone 11 was synthesized by allylic oxidation of compound 3 with pyridinium dichromate and tert-butyl hydroperoxide in dichloromethane (SI Appendix, Materials and Methods).

Structural Characterization of Sesterterpenes by NMR and Other Spectroscopic Techniques.

Standard 1D and 2D NMR spectra, including ¹H, ¹³C, DEPT135, COSY, HSQC, HMBC, and NOESY were acquired on a Bruker 400-MHz Topspin NMR spectrometer. All signals were acquired at 298 K. Samples were dissolved in CDCl₃ or benzene-d₆ for data acquisition and calibrated by referencing to either residual solvent ¹H and ¹³C signal or TMS. Detailed structural assignments and tabulated NMR data for compounds 2–16 are presented in SI Appendix, Tables S1–S15. Optical rotations of compounds 2–16 were measured in benzene using a PerkinElmer Polarimeter (Model 341) with a 100-mm path cell (1 mL) at 20 °C and converted to specific optical rotation using equation ${\left[\alpha \right]}_{\text{D}}^{20}$ = 100 × (measured optical rotation)/concentration (g/100 mL).

Single Crystal X-Ray Diffraction Analysis.

Single-crystal X-ray analysis was carried out for 7, 8, 9, 10, 11, 12, and 13 on a Bruker D8-QUEST instrument equipped with a PHOTON-100 area detector and Incoatec IμS Cu microsource (wavelength = 1.5418 Å, beam diameter at the crystal approximately 100 μm). Crystals were mounted on an X-ray transparent loop (Mitegen) using an inert oil and cooled to 180(2) K using an open-flow N₂ cryostat. Data were collected and processed using the APEX3 software package (Bruker). Structures were solved and refined using SHELXT and SHELXL (Bruker). In the absence of any significant anomalous scatterers, reliable indications of the absolute structure could not be obtained by conventional refinement of the Flack parameter. However, Bayesian statistical methods (35) indicate the following probabilities that the absolute structure is correct, under the assumption that the crystal is enantiopure [p2(true)]: 7 0.886, 8 0.980, 9 0.682, 10 1.000, 11 1.000, 12 1.000, and 13 0.745. See Datasets S4 and S5 for crystal structures and crystallographic raw data of compounds 7–13.

Sequence Comparisons and Homology Modeling.

Amino acid sequences of AtTPS17, -18, -19, -25, -30, Cr237, and Bo250 were aligned using Uniprot (www.uniprot.org/). Homology models of AtTPS18 were generated on modeling server Phyre² (www.sbg.bio.ic.ac.uk/phyre2/html/page.cgi?id=index) using the crystal structure of 5-epi-aristolochene synthase as template. Docking of STSs substrate GFPP was performed using Autodock4.0 and AutoDockTool developed by The Scripps Research Institute (54).

Mutagenesis of AtTPSs17, -18, -19, -25, and -30.

Site-directed mutagenesis was performed by PCR amplification using the entry vector pDONR207 harboring the wild-type genes as templates and the mutated complementary sequences as primers, as listed in the SI Appendix, Table S17. Mutagenized genes were Gateway-cloned into the pEAQ-HT expression vector, transformed into LBA4404 and coexpressed with LBA4404/GFPPS1 in our transient tobacco expression system, as described in Cloning and Transient Expression, above.

Quantum Chemical Calculations.

All quantum calculations were performed with the GAUSSIAN09 software suite (55). Geometries were optimized using the B3LYP density functional theory method and the 6–31+G(d,p) basis set (56). All stationary points were characterized as minima or transition-state structures using frequency calculations. All reported energies include zero-point energy corrections (unscaled) from these frequency calculations. mPW1PW91 single-point energies were calculated for all structures (57). These methods are well established for examining carbocation rearrangement reactions (46). Structures of intermediate carbocations, transition state structures, and corresponding coordinates are presented in Dataset S3.