Origin and early evolution of the plant terpene synthase family

To determine the distribution of TPS genes in green plants, genome sequences of 31 species of green algae (SI Appendix, Table S1), 14 species of land plants (SI Appendix, Table S2), and transcriptomes of 1,178 species of green plants that include 1,109 species of the OneKP dataset (8) and 69 species of ferns (19) (SI Appendix, SI Material and Methods) were mined for TPS genes using a hidden Markov model (HMM) search. TPS genes were detected in the transcriptomes and genomes of all major lineages of land plants (Fig. 2): angiosperms, gymnosperms, ferns, lycophytes, and all three lineages of bryophytes (hornworts, liverworts, and mosses). They were identified from every genome investigated, but not in each transcriptome. TPS genes were detected in 85 to 100% of the transcriptomes from angiosperms, gymnosperms, lycophytes, and liverworts, but the detection rate in the transcriptomes of other plant lineages was only 63% for ferns, 54% for mosses, and 50% for hornworts. In contrast to land plants, no TPS genes were detected in green algae after searching 158 transcriptomes (47 charophytes and 111 chlorophytes) and 31 genomes. While several hits with significant e-values were identified in the genome of the charophyte Klebsormidium flaccidum, none had the conserved features of TPS proteins when the sequences were examined in detail. The absence of TPS genes in red algae has been previously noted (20). The lack of TPS genes in a wide range of green algae, particularly in charophytes, presumed to be the closest relatives of land plants, contrasts with their ubiquitous occurrence in land plants. Accordingly, it can be hypothesized that the TPS family originated in ancestral land plants after their divergence from green algae.

To understand the early evolution of TPSs in land plants, we created a dataset of TPS sequences enriched in those from nonseed plants. This included all TPSs identified from the transcriptomes of nonseed plants in the OneKP database and the additional 69 fern transcriptomes. In addition, TPSs from eight nonseed plants with sequenced genomes (SI Appendix, Table S3) were included. Since the evolutionary relationship of the TPSs from seed plants is relatively well understood, only TPSs from three angiosperms (Arabidopsis thaliana, Oryza sativa, and Amborella trichopoda) and four gymnosperms (Ginkgo biloba, Pseudotsuga menziesii, Picea abies, and Pinus lambertiana) were included, plus several other sequences to anchor the various TPS subfamilies (SI Appendix, Table S2). To ensure the quality of phylogenetic reconstructions, our dataset was confined to sequences with a length longer than 350 amino acids.

The unrooted TPS phylogenetic tree made from our TPS dataset (Fig. 3 and SI Appendix, Fig. S1) has a number of deep divisions that could be separated in various ways, assuming no early gene loss. However, given the distribution of the TPS subfamilies in the major lineages of land plants combined with their biochemical and physiological function, three divisions were defined that correspond to the TPS-c subfamily, the TPS-e/f subfamily, and the rest of the TPS subfamilies (h/d/a/b/g). The separation of the rest of the subfamilies from the TPS-c and -e/f subfamilies has strong bootstrap support (96%), while the separation of TPS-c and -e/f is based on their previous definitions (i.e., groups of monofunctional class I TPSs are placed into TPS-e/f, while those with class II activity fall within TPS-c). To gain additional evidence for the observed phylogeny, two additional phylogenetic trees were reconstructed with longer TPS sequences. As shown in SI Appendix, Fig. S2 (made with TPSs longer than 475 amino acids) and SI Appendix, Fig. S3 (made with TPSs longer than 500 amino acids), the topologies of the phylogenetic trees are largely consistent. As such, our interpretation and discussion about TPS evolution in this work is mainly based on the phylogenetic tree presented in Fig. 3.

The TPS-c subfamily is uniquely present in all groups of land plants (Fig. 3). By definition, the TPS-c subfamily contains the extant examples of bifunctional CPSKSs, which are functionally analogous to the ancestral TPS required for biosynthesis of gibberellins and related phytohormones (9, 16). In addition to the bifunctional CPSKSs, the TPS-c subfamily also includes large numbers of TPSs from nonseed plants, which must have radiated out from the ancestral CPSKSs in this lineage. The majority of the newly identified TPSs in this study fall within this group. While the presence of a fern TPS (MON_CQPW_PTPS4) in a branch otherwise composed of all hornwort TPSs (Fig. 3) is somewhat surprising, the significance of the placement of this isolated sequence remains to be determined.

The TPS-e/f subfamily is anchored by the KSs from phytohormone biosynthesis. While these TPSs almost all still retain the ancestral γβα tridomain architecture, they no longer exhibit class II diterpene cyclase activity and the associated DxDD motif. There is a deep division in the TPS-e/f subfamily between two distinct branches, which leaves the origins of the subfamily uncertain. The two branches are largely divided between seed and nonseed plants, indicative of parallel subfunctionalization within each lineage. However, inclusion of one TPS from hornworts in the branch otherwise containing only those from seed plants argues against this interpretation (Fig. 3). Indeed, the TPS-e/f subfamily in the phylogenetic tree made with TPSs longer than 500 amino acids (SI Appendix, Fig. S3) no longer exhibits such a deep division. The observed separation may stem from relaxed selection in the γβ didomains given the loss of the associated class II (CPS) activity in this subfamily. Regardless, TPSs from this subfamily do not seem to be present in either the mosses or ferns. This absence is correlated with the presence of bifunctional CPSKSs in these plant lineages [e.g., the moss PpCPSKS (17) and fern LjCPSKS (21)], which would preclude the need for a separate (monofunctional) KS. Yet in some plant lineages, monofunctional KSs of the TPS-e/f subfamily and bifunctional CPSKSs are both present, implying metabolic redundancy. A separate KS might indeed have occurred alongside a bifunctional CPSKS for an extended time prior to subfunctionalization of the CPSKS to a monofunctional CPS due to the ability of KS to react with ent-CPP released by the CPSKS from its class II diterpene cyclase active site. Analogous reasoning has been used to support the existence of monofunctional class I diterpene synthases alongside bifunctional class I and II enzymes in gymnosperm resin acid biosynthesis (22), where the release of the CPP intermediate has been shown (23).

The remaining TPS subfamilies form a distinct group termed here TPS-h/d/a/b/g. As none of these genes are known to be involved in phytohormone biosynthesis, they are all apparently dedicated to secondary metabolism. The TPS-h/d/a/b/g subfamilies have members in all plant lineages except hornworts, suggesting that the common ancestor of this TPS lineage arose from early gene duplication and neo-functionalization. This group also contains bifunctional diterpene synthases, but these are not involved in phytohormone biosynthesis. Nevertheless, since the bifunctional enzymes of the TPS-h/d/a/b/g subfamilies are present near the root of this group, it is assumed that they are derived from gene duplication of the ancestral bifunctional CPSKS with neofunctionalization to more specialized metabolism.

The TPS phylogenetic tree presented here clarifies the origin of the βα didomain architecture dominant in angiosperm TPSs, which results from the loss of the γ domain from the ancestral γβα TPS. While it had been suggested that βα TPSs (which all have only class I activity) arose from the TPS-e/f subfamily, which also has only class I activity (2), analysis of the domain composition of the TPSs in the phylogenetic tree clearly indicates that this loss of the γ domain occurred in the lineage leading to the TPS-h/d/a/b/g subfamilies. In particular, this loss seems to have occurred early in the seed plant lineage, giving rise to the TPS-d1 group in gymnosperms and the TPS-a/b/g subfamilies in angiosperms. The loss of class II diterpene cyclase activity seems to have preceded γ domain loss since there is no class II activity in the TPS-a, b, g, d1, and d2 subfamilies (9). A branch of Amborella TPSs, which was previously postulated as the tentative TPS-x subfamily and placed between the TPS-h and TPS-d subfamilies (24), is here placed between the TPS-d3 and TPS-d2 groups (Fig. 3). This indicates that additional TPS sequences, especially those from basal flowering plants, are needed to resolve the early evolution of TPSs in seed plants.

Loss of the γ domain has independently occurred at least twice in the TPS-e/f subfamily (25) and also in the TPS-c subfamily. Though this loss frequently follows the loss of class II TPS activity, a parallel loss of the α domain following the loss of class I activity has not been observed, presumably due to the previously reported mutual structural interdependence of the α and β domains as well as their associated class I and class II activities (13).

Building on the hypothesis that all extant plant TPS genes were derived from the three ancestral lineages in the common ancestor of land plants, the functional evolution of TPSs within and among the three groups was then investigated, with a total of 31 genes from nonseed plants selected for biochemical characterization (SI Appendix, Table S5). Given the expansion observed in the TPS-c subfamily, the majority of those selected here for characterization were from this subfamily, along with substantial numbers from the putative bifunctional TPSs in the TPS-h subfamily, enabling analysis of whether loss of class I or class II activity in these subfamilies is correlated with the loss of the relevant DDxxD and NSE/DTE or DxDD motifs.

Beyond the motifs required for essential TPS catalytic activity, additional motifs hypothesized to be specifically conserved in the CPSs and KSs involved in the biosynthesis of gibberellins and related ent-kaurene–derived phytohormones were also monitored. In the case of the CPSs, it has been shown that the catalytic base terminating bicyclization is a water molecule ligated in part by a histidine-asparagine dyad, conserved as LHS and PNV motifs (26, 27). In addition, it has been suggested that these motifs are susceptible to synergistic inhibition by GGPP and Mg²⁺ (28), which also serves as a cofactor for CPSs (29), as mediated by the presence of histidine at a particular position (30). In the case of KSs, it has been shown that formation of ent-kaurene is dependent on a particular isoleucine that is conserved in such TPSs (31, 32). Moreover, the KSs specifically involved in phytohormone biosynthesis are marked by a pair of threonines just upstream of the first of the two Mg²⁺-binding motifs, which is then TTxxDDxxD, although the functional importance of these residues is unclear (33). Here, the correlation of these additional motifs with CPS and KS activity was also examined (SI Appendix, Table S5). For this purpose, the histidine hypothesized to impose synergistic GGPP/Mg²⁺ inhibition on CPS was defined as falling within a FEHxW motif, while the key isoleucine for ent-kaurene production by KS was defined as falling within a PIx motif.

The TPSs chosen for study were functionally characterized by use of a previously reported modular metabolic engineering system (34) that enables facile coexpression with a GGPP synthase in Escherichia coli to provide the necessary substrate. For TPSs that exhibit only class I activity, the modular system can supply each of the three known stereoisomers of CPP (35). In total, 20 of the 31 TPS genes investigated here were demonstrated to encode functional diterpene synthases. The results of these studies are summarized in Table 1, with the underlying data reported as Supporting Information (SI Appendix, Figs. S4–S20). Note that class II diterpene cyclase products are detected as their corresponding dephosphorylated derivatives resulting from the activity of endogenous (E. coli) phosphatases. When the TPSs investigated here are indicated to produce the primary alcohol derivative of the phosphorylated intermediate, the dephosphorylation itself is not directly demonstrated and is only based on the presence of the Mg²⁺-binding motifs required for class I activity (36) (Table 1, observed activity indicated with a number sign). Nevertheless, these TPSs are still considered to be bifunctional here. The 20 newly characterized diterpene synthases in this study, together with 13 previously characterized diterpene synthases from nonseed plants (SI Appendix, Table S6), are plotted to the TPS phylogenetic tree with their catalytic activity, domain architecture, and catalytic motifs (Fig. 4).

The members of the TPS-c subfamily investigated here were found to include bifunctional CPSKSs (Fig. 4), whose stereospecific production of ent-kaurene was verified by inactivation of their class I (KS) activity (via alanine substitution of the first two aspartates of the DDxxD motif) and coupling of the remaining CPS activity to a stereospecific KS (37). Other TPSs from this subfamily were found to have subfunctionalized their ancestral CPSKS activities (Fig. 4). For example, subfunctionalization via loss of class I (KS) activity to yield a monofunctional CPS occurred early in the evolution of vascular plants (as evidenced here by the lycophyte TPS SwCPS), and this appears to have occurred independently in the plant lineages, leading to the extant liverworts (MpCPS, LcTPS and MpDTPS3) and hornworts (PcCPS) as well as ferns (VaCPS), indicating parallel evolution within this subfamily. In each case, these TPSs were found to stereoselectively produce ent-CPP, shown as previously described (38). Each of these CPS activities also retained the specific catalytic base dyad motif of the CPSKS ancestor and so are hypothesized to have kept the ancestral CPS activity. The histidine and associated motif for synergistic GGPP/Mg²⁺substrate inhibition is also conserved in these newly identified nonseed CPSs, but this histidine residue is not present in the liverwort CPSKS identified here (PlCPSKS), consistent with two other previously characterized CPS(KS)s as well, the gymnosperm PgCPS and moss PpCPSKS (17, 39). Thus, the wider import of this histidine remains uncertain, and the implications of its retention or loss for evolutionary derivation remains unclear. Regardless, all the CPSKSs identified here contain the ancestral class II catalytic base dyad motifs (i.e., LHS and PNV) identified in all previously characterized CPS(KS)s to date (26).

The subfunctionalized CPSs in ferns apparently arose independently of those found in lycophytes, gymnosperms, and angiosperms (Fig. 3), which is incongruent with the accepted evolutionary history of these plant taxa (Fig. 2). This may reflect the retention of a bifunctional CPSKS in ferns, as noted above. The subfunctionalization of CPS requires a KS activity, which could be provided by the bifunctional CPSKS. Alternatively, this might be provided by an independently arising KS. For example, Asu_TPS1 is a KS whose sequence indicates that it falls within the TPS-c subfamily instead of the TPS-e/f subfamily, where all other plant KSs are found. Nevertheless, consistent with a retained function in phytohormone biosynthesis, this KS, along with all the bifunctional CPSKSs identified here, contains the KS specific motifs (i.e., PIx and the TTxx extension of DDxxD).

Some of the bifunctional TPSs of the TPS-c subfamily from liverworts and hornworts characterized here do not act as CPSKSs, suggesting that neofunctionalization occurred in this subfamily following the evolutionary event that founded the lineage leading to the TPS-h/d/a/b/g subfamilies. In most of these cases, the class II active site catalyzes rearrangement of the initially formed bicycle to yield kolavenyl diphosphate (KPP) of various stereochemical configurations. There is also an example of a monofunctional KPP synthase (OpKOS) that no longer contains the class I Mg²⁺-binding motifs. It can be hypothesized that the monofunctional KPP synthase (OpKOS) forms ent-KPP, as this contains a tyrosine in place of the histidine of the ancestral CPS catalytic base dyad, and such substitution has been previously shown to be sufficient to alter product outcome from ent-CPP to ent-KPP (40). Indeed, the presence of an aromatic residue at this position has been found to be predictive, as demonstrated by reverse engineering of both known ent-KPP synthases to produce ent-CPP instead (41, 42).

Another neofunctionalized TPS-c family member from the mosses (OlIAS) retains bifunctionality and produces isoabienol (Fig. 4). This requires production of a hydroxylated derivative of CPP by its class II active site, which retains the histidine but contains a serine in place of the asparagine from the ancestral CPS catalytic base dyad. Such substitution has been previously shown to be sufficient to alter product outcome from ent-CPP to 8β-hydroxy–ent-CPP (26), suggesting that this enzyme may produce ent-isoabienol.

TPSs of the TPS-h/d/a/b/g subfamilies investigated here were all found to yield products other than the ent-CPP or ent-kaurene required for phytohormone biosynthesis, consistent with dedication of the TPS-h/d/a/b/g subfamilies to secondary metabolism. For example, two bifunctional levopimaradiene synthases were found in lycophytes, HsLS and PdLS, that exhibit intriguing parallels to the functionally analogous TPS-d3 group members from conifers involved in resin acid biosynthesis containing similar features. The catalytic base in the class II diterpene cyclase active site of gymnosperm resin acid TPSs is formed by direct hydrogen-bonding between the side chains of a tyrosine and histidine, where the tyrosine is from a LYS motif that replaces the ancestral (CPS) LHS, while the histidine is from a PCH motif that similarly replaces the ancestral (CPS) PNV (26, 43, 44). Both motifs are also present in these lycophyte levopimaradiene synthases. Moreover, a key alanine in the class I active site of the gymnosperm TPS-d3 subfamily members that plays an equivalent role to the key isoleucine in the ancestral KS but is found four residues upstream (45), is also conserved in these lycophyte levopimaradiene synthases, along with surrounding residues that form a VSIAL motif. Thus, it is tempting to hypothesize that levopimaradiene synthase activity evolved early in the vascular plant lineage. Alternatively, the same arrangement of key residues and surrounding motifs independently evolved in both the class II and class I active sites in TPSs from the lycophyte and gymnosperm lineages respectively.

Also notable among the TPS-h/d/a/b/g subfamilies members characterized here are two TPSs from a fern and a liverwort TPS, both of which have lost class I activity, consistent with their loss of the associated Mg²⁺-binding motifs. Both then act as monofunctional class II diterpene cyclases, with the fern enzyme (OsCPS) producing 8-endo-CPP, while the liverwort enzyme (MpDTPS5) produces rearranged KPP. These represent monofunctional class II diterpene cyclases outside the TPS-c subfamily, demonstrating that parallel evolutionary loss of class I activity has occurred.

Other TPS-h/d/a/b/g subfamilies members have lost class II activity and now function as monofunctional class I diterpene synthases (46). For example, a fern syn-manool synthase was characterized from the TPS-h subfamily that reacts with syn-CPP to produce syn-manool. Interestingly, this monofunctional class I TPS is not closely related to the previously described lycophyte 16α-hydroxykaurene synthase, also a class I TPS from the TPS-h subfamily, and may reflect independent loss of class II activity. Such an evolutionary event in the lineage leading to the TPS-h/d/a/b/g subfamilies parallels the loss of class II activity from CPSKS that initiated the lineage leading to the TPS-e/f subfamily, as well as the equivalent losses of class II activity that occurred in the transition from TPS-d3 to TPS-d2 and within the TPS-d3 groups (22). Thus, the loss of class II activity is a repeated theme in TPS evolution, which then enables loss of the γ domain involved in such activity, as has now been observed in all three main groups of TPSs.

The OneKP transcriptome dataset was described previously (8). The additional transcriptomes of 69 fern species also have been previously reported (19). The sources of genomes analyzed in this study are listed in SI Appendix, Table S1 (green algae) and SI Appendix, Table S2 (land plants). A detailed description of various databases is provided in SI Appendix, SI Materials and Methods. The proteomes for all the datasets were searched against the Pfam-A database locally using HMMER 3.0 hmmsearch (50) with an E value of 1e-5. Only sequences with best hits from the following two HMM profiles were considered as putative TPSs: Terpene_synth_C (PF03936) and Terpene synthase N-terminal domain (PF01397). For sequences from the same species that have 100% identity, only the longest one was kept as the representative sequence to reduce redundancy. All the putative TPS sequences were subjected to BLASTP (51) search against the National Center for Biotechnology Information’s nonredundant database using default parameters.

Sequences were aligned using MAFFT (einsi) with 1,000 iterations of improvement. ProtTest (52) was used to select the most appropriate protein evolution model for alignment under the Akaike information criterion. For the maximum likelihood analyses, RAxML (53) was used with 1,000 bootstrap replicates under the best substitution model (JTT+G+F).

For each TPS gene selected for biochemical characterization, the encoded protein was first analyzed using TargetP (https://services.healthtech.dtu.dk/service.php?TargetP-2.0) to predict the putative transit peptide. Then, with the sequence for putative transit peptide removed and a codon for Met added to the beginning of the coding sequence, a complementary DNA for each pseudomature TPS was synthesized and cloned into pEXP-5-CT/TOPO. The resulting expression construct was incorporated into a previously described modular metabolic engineering system (39) for expression of recombinant TPSs in E. coli and for assays of class I, class II, or class I/class II bifunctional diterpene synthase activities. Terpene products were identified using gas chromatography–mass spectrometry. The detailed procedure for TPS enzyme assays is provided in SI Appendix, SI Materials and Methods.