Δ⁹-Tetrahydrocannabinol (THC) and other cannabinoids are responsible for the psychoactive and medicinal properties of Cannabis sativa L. (marijuana). The first intermediate in the cannabinoid biosynthetic pathway is proposed to be olivetolic acid (OA), an alkylresorcinolic acid that forms the polyketide nucleus of the cannabinoids. OA has been postulated to be synthesized by a type III polyketide synthase (PKS) enzyme, but so far type III PKSs from cannabis have been shown to produce catalytic byproducts instead of OA. We analyzed the transcriptome of glandular trichomes from female cannabis flowers, which are the primary site of cannabinoid biosynthesis, and searched for polyketide cyclase-like enzymes that could assist in OA cyclization. Here, we show that a type III PKS (tetraketide synthase) from cannabis trichomes requires the presence of a polyketide cyclase enzyme, olivetolic acid cyclase (OAC), which catalyzes a C2–C7 intramolecular aldol condensation with carboxylate retention to form OA. OAC is a dimeric α+β barrel (DABB) protein that is structurally similar to polyketide cyclases from Streptomyces species. OAC transcript is present at high levels in glandular trichomes, an expression profile that parallels other cannabinoid pathway enzymes. Our identification of OAC both clarifies the cannabinoid pathway and demonstrates unexpected evolutionary parallels between polyketide biosynthesis in plants and bacteria. In addition, the widespread occurrence of DABB proteins in plants suggests that polyketide cyclases may play an overlooked role in generating plant chemical diversity.

Identification of Polyketide Cyclase Candidates in the Trichome Transcriptome.

Cannabinoid biosynthesis occurs primarily in glandular trichomes that develop on female flowers and, to a lesser extent, leaves. We extracted proteins from trichome secretory cells isolated from a hemp cultivar of cannabis and tested their ability to catalyze the formation of OA from hexanoyl-CoA and malonyl-CoA. Crude trichome protein formed OA in addition to PDAL, HTAL, and olivetol (Fig. 2A). Comparing this result with the inability of the recombinant TKS to synthesize OA, we hypothesized that an OA-forming accessory protein is present in trichomes. To identify this protein, we analyzed a trichome-specific EST catalog from a potent marijuana strain of cannabis that contains a high number of ESTs corresponding to cannabinoid biosynthetic enzymes (e.g., TKS and THCA synthase) (15).

Fig. 2.

OA formation requires the presence of the DABB protein, olivetolic acid cyclase (OAC). (A) TKS produces the by-products PDAL, HTAL, and olivetol, but crude protein from hemp trichomes catalyzes OA formation. Assays of TKS together with polyketide cyclase …

We reasoned that an OA-forming enzyme may be prominently represented in the trichome EST dataset and, therefore, searched for proteins with sequence or structural similarity to polyketide cyclases. This approach identified three candidates with high expression levels as determined by EST counts (Table 1). A chalcone isomerase (CHI)-like protein was selected based on the catalytic relationship of CHS with CHI (16, 17), which suggested that the cannabis CHI-like protein could partner with TKS to form OA, a possibility also discussed by Taura et al. (10). The second candidate was a member of the dimeric α+β barrel (DABB) protein superfamily with similarity to stress-responsive proteins from plants (18–20). The presence of this protein was intriguing because DABB proteins act as polyketide cyclases (e.g., TcmI cyclase) in Streptomycesspecies (21), although the bacterial cyclases show low sequence similarity to plant DABB proteins. The third candidate was a Betv1-like protein in the same protein family as the Streptomyces TcmN ARO/CYC polyketide cyclase (22). Several Betv1-protein family members function as enzymes in plant natural product biosynthesis (23). High numbers of ESTs corresponding to the DABB protein (9 ESTs) and the Betv1-like protein (6 ESTs) were found in the cannabis trichome EST dataset reported by Marks et al. (11).

Table 1.

Candidate polyketide cyclases identified in cannabis trichome EST dataset

Trichome-Expressed DABB Protein Forms OA.

We expressed the three cyclase candidates in Escherichia coli and tested the purified proteins separately in polyketide synthesis assays containing recombinant TKS, hexanoyl-CoA, and components for malonyl-CoA synthesis [malonyl-CoA synthetase (MCS), malonate, CoA and ATP]. MCS was used to produce malonyl-CoA in situ as has been reported for in vitro assays of type II PKSs (24). OA was present in assays containing the DABB protein but not with the Betv1-like or CHI-like proteins (Fig. 2A). This small protein (12 kDa, 101 amino acids), which we named olivetolic acid cyclase (OAC), had no intrinsic PKS activity, and only produces OA in the presence of TKS. OAC did not convert HTAL to OA, indicating that ring opening of a HTAL is not occurring. Production of the α-pyrones was similar whether OAC was present, but olivetol formation decreased in assays containing OAC (Fig. 2B).

Quantitative RT-PCR analysis of cannabis tissues and cell types from a hemp cultivar found that the highest transcript levels of OAC were in trichomes and, to a lesser extent, female flowers, which parallels the expression of the transcripts for TKS andCBDA synthase (Fig. S1). Using transient expression of fluorescent protein fusions inNicotiana benthamiana leaves, TKS and OAC, which lack predicted signal peptides, were both localized to the cytoplasm (Fig. S1).

Functional analysis of OAC via RNAi was not possible because cannabis transformation has not been achieved and our attempts at virus-induced gene silencing were unsuccessful. To demonstrate OAC activity in vivo, we reconstituted OA biosynthesis in yeast. Yeast cultures expressing TKS and OAC fed sodium hexanoate produced 0.48 mg/L OA (mean, n = 3) and olivetol in the medium but no α-pyrones (Fig. S2). Surprisingly, we detected a trace amount of OA in the medium of cultures expressing TKS only. This result may be because yeast has some endogenous cyclase activity or the intracellular environment alters the catalytic properties of TKS.

OAC Does Not Physically Interact with TKS.

One explanation for the production of OA in reactions containing TKS and OAC is that OAC functions as an enzyme that acts on an intermediate produced by TKS. Alternatively, OAC may alter the catalytic properties of TKS through allosteric regulation, which allows TKS to itself form OA. To test the importance of physical interactions for OA formation, we separated TKS and OAC in 100-μl dialysis chambers by using a 5-kDa cutoff membrane that allowed substrates, intermediates, and reaction products to diffuse but not proteins. We also performed reactions in which one chamber contained both TKS and OAC (positive control) or TKS only (negative control) while the other chamber contained no enzyme. As shown in Fig. 3A, OA was formed in the OAC-containing chamber that was separated from TKS by the membrane. Large amounts of OA were formed in the positive control reaction containing TKS and OAC in the same dialysis chamber (Fig. 3B); OA was absent from the TKS only negative control (Fig. 3C). The reduced amount of OA formed when TKS and OAC were separated (Fig. 3A) compared with the positive control (Fig. 3B) may be due to the loss of the intermediate through conversion to olivetol before it reaches the OAC in chamber 2. In all three cases, there was diffusion of the reaction products from the TKS-containing chamber to the opposite side of the membrane during the 2-h assay.

Fig. 3.

Dialysis experiments show that physical interaction of TKS and OAC is not required for OA formation. Recombinant TKS and OAC were assayed in dialysis chambers separated by a 5-kDa cutoff membrane. (A) Assays with TKS and OAC in separate chambers resulted …

These results allow us to conclude that TKS synthesizes a diffusible intermediate that is converted to OA by OAC. We can exclude allosteric regulation of TKS because OA is produced when the two proteins are physically separated. We performed yeast two-hybrid analysis and found no evidence for the interaction of TKS and OAC (Fig. S3). It remains formally possible that OAC plays a chaperone-like role in guiding the folding of the tetraketide intermediate. We think it is more likely that OAC acts as an enzyme based on its structural similarities with bacterial DABB-type polyketide cyclases and the fact that aldol condensations in polyketide biosynthesis are enzyme catalyzed.

On the Nature of the OAC Substrate.

The novelty of the reaction catalyzed by OAC, and the reactivity of poly-β-keto intermediates produced by TKS, presents difficulties in determining its substrate. HTAL and PDAL production by TKS in the polyketide synthesis assays was similar whether OAC was present or not; however, OA formation was accompanied by a decrease in olivetol production (Fig. 2 A and B). We propose this pattern of products results from two co-occurring catalytic processes in the TKS-OAC coupled in vitro assay (Fig. 1): (i) hydrolytic release of poly-β-keto triketide and tetraketide intermediates from TKS, which are unacceptable substrates for OAC and undergo spontaneous lactonization to PDAL and HTAL, respectively; and (ii) TKS-catalyzed synthesis of a linear tetraketide-CoA intermediate, which is the substrate for OAC. In the absence of OAC, this intermediate undergoes aldol cyclization with decarboxylation to yield olivetol. Because HTAL, PDAL, and olivetol are in vitro products not known to be present in cannabis, it is likely that TKS only functions to produce the substrate for OAC in planta. The unstable nature of the putative tetraketide-CoA intermediate precludes its in vitro assay with OAC and its use in determining the kinetic parameters of OAC. We concluded that a comparison of the kinetics of TKS alone compared with the TKS-OAC coupled reaction, which may shed light on their respective catalytic roles, was complicated by the multiple products from TKS (PDAL, HTAL, olivetol, and the putative tetraketide-CoA intermediate). In addition, our uncertainty about the decarboxylation reaction that forms olivetol makes such experiments difficult to interpret. There are precedents for the release of CoA-linked intermediates in plant polyketide biosynthesis. A type III PKS from Curcuma longa forms a CoA-bound diketide that is transferred to a second type III PKS for further extension (25). Chalcone reductase has been postulated to act on a CoA-bound polyketide (26).

Structural and Mechanistic Analysis of OAC.

To gain insight into the mechanism by which OAC catalyzes OA synthesis through an intramolecular C2–C7 aldol condensation, we compared its structure with DABB proteins from bacteria and plants that have sequence or structural similarity to OAC. The structures of three plant stress-responsive DABB proteins are known [Arabidopsis heat stable 1 (AtHS1), theArabidopsis At5g22580 gene product, and poplar stable protein 1 (SP1)] (27–29). The catalytic mechanisms of the bacterial DABB proteins TcmI cyclase, ActVA-Orf6 monooxygenase, and 4-methylmuconolactone methylisomerase (MLMI) have been investigated by using structural and biochemical approaches (21, 30, 31). These proteins possess a ferredoxin-like fold with an intermolecular β-barrel and a deep hydrophobic cleft in each monomer where the α₂– and α₃-helices arch over the β-sheets. This cleft forms the active site in TcmI, ActVA-Orf6, and MLMI and was suggested to be the putative active site in AtHS1 and the At5g22580 gene product.

OAC is predicted to have the characteristic β-α-β-β-α-α-β topology and possesses amino acid residues that are conserved in plant stress-responsive DABB proteins (Fig. 4A).We used homology modeling to generate the structure of OAC by comparison with AtHS1 (PDB ID code 1q53), which shares 48% identity with OAC. The OAC model exhibits the same overall structure as other DABBs (Fig. 4B), with a hydrophobic cleft that likely serves as the active site for the cyclization reaction.

Fig. 4.

Comparison of the OAC structure with other DABB proteins. (A) A schematic representation of the secondary structures of OAC and representative DABB proteins from plants and bacteria showing the characteristic β-α-β-β-α-α-β …

The catalytic mechanism and active-site residues involved in the OAC aldol condensation remain to be elucidated. OAC possesses three conserved lysines (Lys4, Lys12, Lys38; Fig. 4A) that could form Schiff bases in a type I aldolase reaction. However, Lys4Ala, Lys12Ala, and Lys38Ala mutants created by site-directed mutagenesis were active (Table S1). A type II aldolase mechanism involving a metal ion can be excluded because 10 mM EDTA was not inhibitory. A mechanism for OAC is suggested by the base-catalyzed aldol condensations of Streptomyces cyclases, which involve enolate intermediates (22, 32, 33). OAC has several conserved aspartate and histidine residues that could act as catalytic bases. Asp45Ala and Asp96Ala mutants were active, but single-residue mutants in which His5, His57, or His78 were replaced by Ala led to complete loss of activity; the His75Ala mutant had 1% of wild-type activity (Table S1). We are attempting to determine the structure of OAC to better understand its catalytic mechanism.

Evolution of OAC Function.

To investigate the evolution of OAC, we identified OAC homologs in diverse plant genomes by BLAST and keyword searching of the Phytozome database. We also identified DABB proteins in the cannabis genome and from hop (Humulus lupulus) ESTs (34). We included representatives from dicot and monocot lineages and the basal plants Selaginella moellendorffii and Physcomitrella patens. DABB proteins with sequence similarity to OAC are present in all of the plant genomes we analyzed including four DABB-encoding genes in Arabidopsis thaliana, 8 in cannabis, and 12 in Populus trichocarpa. In some cases, the OAC homologs code for proteins with a single DABB domain (e.g., OAC), whereas others have a duplicated DABB domain (e.g., AtDABB1, At1g51360). OAC homologs are also found in bacteria including Rhizobium leguminosarum and members of the enigmatic Planctomycetes-Verrucomicrobia-Chlamydieae superphylum. The function of these bacterial DABB proteins is unknown.

We constructed a phylogenetic tree of the DABB proteins by using the maximum-likelihood method (35), which we rooted on a single-domain DABB protein from R. leguminosarum (Fig. 5). A tree with similar topology was inferred by using the neighbor-joining method (Fig. S4). Phylogenetic analysis was made challenging by the small size of OAC and its homologs, most of which are ~100 amino acids, and some of the nodes have low bootstrap values. The single-domain and double-domain DABB proteins formed separate clades, with the single-domain proteins further divided into subclades 1a and 1b. OAC was positioned in subclade 1a, where it clustered with a diverse group of the single-domain DABBs from dicots, monocots, and basal plants. In many cases, the bootstrap values in clade 1a were <50% and the relationships must therefore be considered tentative. However, it is clear that there has been an expansion of DABB proteins in Cannabaceae, with OAC grouped with four other cannabis proteins and the single hop protein. It is worth noting that the biosynthesis of the major polyketides in hop (e.g., humulone) by type III PKSs does not appear to require the involvement of a polyketide cyclase, and our analysis of hop trichome EST datasets did not find cDNAs corresponding to DABB proteins to be highly abundant. Plant stress-responsive DABB proteins such as AtHS1 and poplar SP1 are also present in subclade 1a. Most of the taxa that we used for our analysis were represented in clade 1b, with one gene from each of Arabidopsis, cannabis, corn, rice and Brachypodium, and two from apple and poplar. This result suggests its members may have a conserved function that is distinct from the proteins in subclade 1a. Clade 2 contained the double-domain DABB proteins such as AtDABB1 (At1g51360).

Fig. 5.

A phylogenetic tree of DABB proteins from plants inferred using the maximum-likelihood method. OAC and proteins that have been structurally or functionally characterized are highlighted. Branch lengths are proportional to the number of amino acid substitutions …

The structural similarity of OAC with the bacterial DABB-type cyclases demonstrates that plants and bacteria exploit the same α+β barrel fold for polyketide cyclization. However, the low sequence similarity between OAC and the bacterial enzymes indicates that they are not homologous. Rather we suggest that OAC and the bacterial DABB-type cyclases are an example of convergent evolution with polyketide cyclizing activity arising independently in plants and bacteria. The ferredoxin-like fold is known to be suitable for ligand binding and as a structural framework for diverse catalytic functions (36), of which polyketide cyclization is but one. A likely evolutionary scenario is that OAC evolved from a plant DABB protein that was not involved in polyketide biosynthesis.

Role of Cyclase Enzymes in Plant Polyketide Pathways.

The current model of plant polyketide biosynthesis is that carbon scaffolds are assembled exclusively by the activity of type III PKS enzymes (37). The discovery of OAC indicates a variant catalytic route in plants in which cyclase enzymes function cooperatively with type III PKSs. It also demonstrates that polyketide cyclases, which until now were only known to partner with type II PKSs in bacterial polyketide pathways, are also found in biosynthetic pathways involving type III PKSs. Although cannabinoid biosynthesis may be unique in its requirement for a cyclase enzyme, we speculate that other plant DABB proteins may act as polyketide cyclases because some possess the hydrophobic cleft and conserved residues that we implicate in OAC function (Fig. 4 A and B). However, polyketide synthesis assays with TKS and recombinant AtHS1 show the latter possesses no OAC activity (Fig. S5), and gene expression databases show no evidence for interactions of the four Arabidopsis DABB proteins and their encoding genes with flavonoid/polyketide pathways (SI Methods and Materials). The role of such enzymes may be confined to the formation of polyketide products that retain a carboxylic acid moiety. Examples of plant metabolites formed by C2–C7 aldol condensation with carboxylate retention are the anacardic acids in cashew and gingko and stilbene carboxylates in Hydrangea species and liverworts (38, 39). It is worth noting that a rice type III PKS synthesizes long-chain alkylresorcinolic acids without the need for a cyclase enzyme (40). Another indication that some plant polyketide pathways may require cyclases is the formation of α-pyrones when recombinant type III PKSs enzymes are assayed in vitro e.g., Hypericum perforatum octaketide synthase (41).

Assay of Trichome Protein.

Trichome cells isolated from female flowers of the hemp cultivar ‘Finola’ using the Beadbeater method were homogenized in buffer, centrifuged to remove cell debris, desalted, and concentrated. Protein extracts were assayed for polyketide synthesis activity as described below.

Protein Expression and Assay.

TKS, OAC, Betv1-like, CHI-like, and MCS were amplified (primers in Table S2), cloned and expressed in E. coli, and proteins were purified with Talon resin (Clontech). Enzyme assays (50 μL) contained 20 mM Hepes at pH 7.0, 5 mM DTT, 0.2 mM hexanoyl-CoA, 12 μg of MCS, 0.2 mM CoA, 0.4 mM ATP, 2.5 mM MgCl₂, 8 mM sodium malonate, TKS and either OAC, Betv1-like, or CHI-like. Reactions were incubated at 20 °C for 60 min, and products were analyzed by HPLC-photodiode array (PDA)/MS.

Dialysis Assay of TKS and OAC.

Assays were performed by using Fast Micro-Equilibrium Dialyzers with 100-μL chambers separated by a 5-kDa MWCO cellulose acetate membrane (Harvard Apparatus). Each chamber contained 20 mM Hepes at pH 7.0, 5 mM DTT, 200 μM hexanoyl-CoA, and 600 μM malonyl-CoA. Reactions consisted of TKS and OAC in separate chambers or together, or a TKS-only control. Reactions were incubated at 10 °C for 2 h, and products were analyzed by HPLC-PDA/MS.

Structural Analysis.

The homology structure of OAC was obtained from the 1q53 template by using comparative modeling with SWISS-MODEL (42). The OAC model had a QMEAN4 score of 0.45 (z-score of −3.25).

OAC Site-Directed Mutagenesis.

Single amino acid changes in OAC were introduced by gene synthesis (DNA 2.0) or, in some cases, using site-directed mutagenesis. The constructs were cloned directly into the plasmid vector pJExpress 411 (DNA 2.0). OAC mutants were expressed, purified, and assayed as above.

Phylogenetic Analysis.

A tree was inferred by the maximum likelihood method with a WAG matrix-based model (35) and 1,000 bootstrap replicates by using MEGA5 software (43).

We thank R. Taschuk, S. Polvi, and N. Theaker for technical assistance; S. Whitfield for assaying AtHS1; B. Haug for R. leguminosarum DNA; and P. Covello and M. Loewen for critical comments on the manuscript. This research was supported by funding from the Natural Sciences and Engineering Research Council of Canada and the National Research Council of Canada.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The sequences reported in this paper have been deposited in the GenBank database (accession nos. OAC, JN679224; Betv1-like, JN679225; CHI-like, JN679226).

This article contains supporting information online atwww.pnas.org/lookup/suppl/doi:10.1073/pnas.1200330109/-/DCSupplemental.