Identification of candidate genes affecting Δ⁹-tetrahydrocannabinol biosynthesis in Cannabis sativa

Abstract

RNA isolated from the glands of a Δ⁹-tetrahydrocannabinolic acid (THCA)-producing strain of Cannabis sativa was used to generate a cDNA library containing over 100 000 expressed sequence tags (ESTs). Sequencing of over 2000 clones from the library resulted in the identification of over 1000 unigenes. Candidate genes for almost every step in the biochemical pathways leading from primary metabolites to THCA were identified. Quantitative PCR analysis suggested that many of the pathway genes are preferentially expressed in the glands. Hexanoyl-CoA, one of the metabolites required for THCA synthesis, could be made via either de novo fatty acids synthesis or via the breakdown of existing lipids. qPCR analysis supported the de novo pathway. Many of the ESTs encode transcription factors and two putative MYB genes were identified that were preferentially expressed in glands. Given the similarity of the Cannabis MYB genes to those in other species with known functions, these Cannabis MYBs may play roles in regulating gland development and THCA synthesis. Three candidates for the polyketide synthase (PKS) gene responsible for the first committed step in the pathway to THCA were characterized in more detail. One of these was identical to a previously reported chalcone synthase (CHS) and was found to have CHS activity. All three could use malonyl-CoA and hexanoyl-CoA as substrates, including the CHS, but reaction conditions were not identified that allowed for the production of olivetolic acid (the proposed product of the PKS activity needed for THCA synthesis). One of the PKS candidates was highly and specifically expressed in glands (relative to whole leaves) and, on the basis of these expression data, it is proposed to be the most likely PKS responsible for olivetolic acid synthesis in Cannabis glands.

Chalcone synthase, glandular trichome, hemp, hop, Humulus lupulus, marijuana, polyketide synthase, trichomes

Introduction

Cannabis sativa has a long history of cultivation for a variety of uses including food, fibre, medicine, and recreational drugs (Measham et al., 1994; Ware and Tawfik, 2005; Kostic et al., 2008). Cannabis produces many different secondary compounds such as cannabinoids, flavonoids, stilbenoids, alkaloids, lignanamides, and phenolic amides (Flores-Sanchez and Verpoorte, 2008b). Δ⁹-Tetrahydrocannabinolic acid (THCA), a product of the cannabinoid class, is the primary psychoactive agent. This compound is produced as an acid in the glandular trichomes of inflorescence bracts and undergoes decarboxylation with age or heating to Δ⁹-tetrahydrocannabinol (THC) (Mechoulam, 1970; Turner et al., 1980; Pertwee, 2006). Cannabis cultivars differ substantially in economic traits that range from marijuana, arguably the most widespread illicit drug, to hemp fibre derived from the stems of the plant. Marijuana consists of the dried female inflorescences in which the quantity of THC exceeds that of cannabidiol (CBD, produced initially as cannabidiolic acid (CBDA)), and potency varies among cultivars by several orders of magnitude (ElSohly et al., 2000). Marijuana cultivars are known to have THC levels exceeding 2–24% of inflorescence dry weight whereas hemp cultivars produce substantially less THC but rather high levels of CBD (Hillig and Mahlberg, 2004). THCA and CBDA share the same biosynthetic pathway except for the last step in which THCA synthase and CBDA synthase produce THCA or CBDA, respectively (Taura et al., 2007). Recent evidence suggests that the genes encoding the two synthases are allelic (de Meijer et al., 2003; Pacifico et al., 2006). CBD and THC are enatiomers, but only THC elicits psychotropic effects, whereas CBD may mediate anti-psychotropic effects (Long et al., 2006; Zuardi et al., 2006), a difference highlighting the stereo-selectivity of receptors in the human body that bind these compounds.

Although classified as a drug without therapeutic value in the United States, ingestion of THC is widely regarded as having effects including pain relief and appetite stimulation, that may, among other things, increase the tolerance of cancer patients to chemotherapy (Baker et al., 2003). Dronabinol, a synthetic analogue of THC, is approved for use as an appetite stimulant in the United States as a Schedule III drug (marketed as Marinol by Unimed Pharmaceutical, a subsidiary of Solvay Pharmaceuticals, Marietta, GA). Cesamet (Nobiline; Valeant Pharmaceuticals, Costa Mesa, CA), another synthetic analogue, is used as an anti-emetic for patients undergoing cancer therapy. The natural product Sativex (GW Pharmaceuticals, Salisbury, UK) is approved for use in the UK and is derived from Cannabis cultivars containing both THC and CBD, and is used to treat pain symptoms associated with multiple sclerosis.

Compounds from Cannabis sativa are of undeniable medical interest, and subtle differences in the chemical nature of these compounds can greatly influence their pharmacological properties. For these reasons, a better understanding of the secondary metabolic pathways that lead to the synthesis of bioactive natural products in Cannabis is needed (Mechoulam, 2005). Knowledge of genetics underlying cannabinoid biosynthesis is also needed to engineer drug-free and distinctive Cannabis varieties capable of supplying hemp fibre and oilseed. In this report, RNA from mature glands isolated from the bracts of female inflorescences was converted into cDNA and cloned to produce a cDNA library. DNA from over 2000 clones has been sequenced and characterized. Candidate genes for almost all of the enzymes required to convert primary metabolites into THCA have been identified. Expression levels of many of the candidate genes for the pathways were compared between isolated glands and intact inflorescence leaves.

Materials and methods

Plant growth and gland isolation

Seeds from the marijuana cultivar Skunk no. 1 were provided by HortaPharm BV (Amsterdam, The Netherlands) and imported under a US Drug Enforcement Administration (DEA) permit to a registered controlled substance research facility. Plants were grown under hydroponic conditions in a secure growth chamber yielding cannabinoid levels in mature plants as reported in Datwyler and Weiblen (2006). Approximately 5 g of tissue was harvested from mature female inflorescences 8 weeks after the onset of flowering. Tissue was equally distributed into four 50 ml tubes (50 ml PP-Tube sterile, Greiner Bio-one, Kremsmünster, Austria) containing 20 ml phosphate buffered saline (PBS) as described by Sambrook et al. (1989), but made with all potassium salts and mixed at maximum speed with a Vortex 2 Genie (Scientific Industries, Bohemia, NY) for four repetitions of 30 s mixing followed by 30 s rest on ice, for a total of 2 min of mixing. Material was sieved through four layers of 1×1 mm plastic mesh and the flow-through was split into two 50 ml tubes and spun in a centrifuge for 30 s at 500 rpm. Supernatants were decanted and pellets were resuspended in PBS. The suspensions were combined into one tube and pelleted as before. The resulting pellet was diluted into 100 μl of PBS. Five μl were used for cell counting with a haemocytometer, and the total suspension was estimated to contain 70 000 intact glands. Plant residue was incinerated by a DEA-registered reverse distributor (Return Logistics, Savannah, GA).

RNA isolation, cDNA library construction, EST sequencing and analysis

Total RNA was isolated from the glands using the RNAeasy Plant Mini Kit (Qiagen Inc.-USA, Valencia, CA), and 120 ng of total RNA was used to make a cDNA library with the Creator SMART cDNA Library Construction Kit (Clontech, Mountain View, CA). This kit allows directional cloning of cDNA inserts into the pDNR_Lib plasmid vector. All procedures were followed as described in the kit except that the final cDNA product was size-selected on a 0.8% agarose gel and gel purified with the Qiaex II kit from Qiagen. Over 100 000 clones were generated. Plasmid DNA was isolated from 2112 bacterial clones, and sequenced using an M13 forward primer that reads into the 5′ end of the oriented cDNA inserts. Sequences can be found in GenBank with accession numbers GR220588-222152.

Cannabis polyketide synthase cloning and analysis

Full-length cDNAs from CAN24, 383, and 1069 contigs encoding putative polyketide synthases were used as templates for PCR reactions. Sequences identical to CAN24 and 1069 are available in GenBank with accession numbers AB164375 and AAL92879, respectively. The GenBank accession for CAN383 is GQ222379. The PCR reactions were designed to add 5′ NcoI and 3′ BamHI restriction enzyme sites to the ends of each sequence. After digestion with NcoI and BamHI, the inserts were cloned into the corresponding sites of pHIS8 (Jez et al., 2000), which adds eight histidine residues to the N-terminus of the encoded protein. Clones corresponding to each PKS-related gene were sequenced. The Lasergene (Madison, WI) MegAlign program using the Clustal W algorithm was used to generate the alignment of PKS genes. PKS gene constructs were transformed into E. coli BL21-(DE). For protein isolation, cultures were grown to an OD of approximately 0.6 at 37 °C when transcription of the cDNA inserts was induced with IPTG and cultures were grown for an additional 10–12 h at 28 °C. His-tagged protein was isolated from the bacteria using the MagneHis Protein Purification System (Promega, Madison, WI).

PKS enzyme assays (40 μl) contained 4 μg protein, 100 mM KPO₄ (pH 7.0), 5 mM malonyl-CoA, and either 250 μM hexanoyl-CoA or 67.5 μM 4-coumaroyl-CoA as substrates. Boiled protein was assayed in parallel with all reactions. All negative controls showed a lack of product formation. Reactions were incubated for 1 h at 30 °C, dried in a Speed Vac, resuspended in 40 μl methanol, and applied to Agilent Tech 1200 HPLC (Agilent Technologies, Santa Clara, CA) with a Spherisorb 6 μ ODS2 (Waters, Milano, Italy) separation column. Products and reactants were resolved across a gradient of 1% H₃PO₄ to 100% acetonitrile with molecular weights determined by LC-MS.

Quantitative PCR

Quantitative (qPCR) reactions were performed as described previously (Marks et al., 2007) using primers listed in Supplementary Table 4B at JXB online. Equivalent quantities of RNA isolated from glands and inflorescence-associated leaves were used to generate the respective single stranded cDNAs. qPCR reactions containing equal quantities of gland or leaf cDNA were run in duplicate along with reactions containing standards consisting of 100-fold sequential dilutions of isolated target fragments, on a Lightcyler qPCR machine (Roche, Indianapolis, IN). Lightcycler software was used to generate standard curves covering a range of 10⁶ to which gland and leaf data were compared. Two biological replicates were used to generate the means and standard deviations shown in Supplementary Table 4A at JXB online. These values were used to compute the gland over leaf ratios and P-values shown in Supplementary Table 4A at JXB online. Raw relative expression data, means, standard deviations, P-values from gland versus leaf t tests, qPCR primer sequences, and representative real-time qPCR tracings are shown in Supplementary Table 4A at JXB online.

Results

Identification of candidate unigenes required for THCA production

Anatomical study revealed that glands located on mature floral bracts of female plants are the site of enhanced secondary metabolism leading to the production of THCA and other compounds in Cannabis sativa (Kim and Mahlberg, 1997) (Fig. 1A, C, D). These glands are located on multicellular stalks and typically are composed of eight cells (Hammond and Mahlberg, 1977, 1978). The outer gland surface is composed of a smooth capsule covered by a membrane. The capsule contains exudates derived from the gland cells (Kim and Mahlberg, 2003). The weakly attached glands can easily be separated from the bracts and purified as shown in Fig. 1E and F. An EST library was constructed using RNA isolated from purified glands. Over 100 000 ESTs were cloned. Plasmid DNA was isolated and sequenced from over 2000 clones. Because of the directed orientation of cDNA insertion, sequences are expected to represent the coding strand. After the removal of vector-only, poor quality sequences, and sequences obviously originating from organelles or ribosomal RNA, the remaining sequences were clustered into 1075 unigenes (see Suplementary Tables 1 and 2 at JXB online). Overall, 111 of the unigenes were contigs containing two or more closely related ESTs (see Supplementary Table 2 at JXB online). Only 14 contigs lacked a similar sequence in the NCBI database. Nine hundred and sixty four of the ESTs were only found once and of these 710 were similar to sequences in the NCBI database (see Supplementary Fig. 2 at JXB online).

Table 1.

NCBI hits to most abundant Cannabis gland unigenes

NCBI accession	NCBI description	Unigene	No. of ESTs
ABM21763.1	Metallothionein-like protein MT1A	CAN4	31
BAB60848.1	RD22-like BURP domain-containing protein	CAN5	22
AAL30422.1	Hevein-like protein	CAN7	19
AAF73006.1	NADP-dependent malic protein	CAN6	18
1Q53	Hypothetical Arabidopsis thaliana protein At3g17210	CAN14	9
ABB29926.1	Fructose-bisphosphate aldolase-like	CAN19	9
NP_194153.1	VEP1 (VEIN PATTERNING)	CAN15	8
AAX11454.1	Chalcone synthase	CAN24	8
O24248	Microsomal oleic acid desaturase	CAN20	8
AAD22104.1	Major allergen Pru av 1 (Pru a 1)	CAN21	6
ABA12220.1	Translation elongation factor 1A-4	CAN26	6
AAC50014.1	B12D protein	CAN22	5
AAO33357.1	Non-specific lipid transfer protein 1	CAN25	5
AAO37754.1	Delta-12 oleate desaturase	CAN27	5
ABA27052.1	TO71-3	CAN30	5
AAG21984.1	LYTB-like protein precursor	CAN31	5
AAD56020.1	Elongation factor-1 alpha 3	CAN33	4
O22342	ADP,ATP carrier protein 1, mitochondrial precursor	CAN38	4
AAF85975.1	Cytosolic phosphoglycerate kinase	CAN40	4
AAS79351.1	Terpenoid synthase	CAN41	4
AAF35186.1	Lipid transfer protein precursor	CAN42	4
AAA33401.1	Ubiquitin	CAN44	4
BAB47196.1	Aspartate aminotransferase glyoxysomal isozyme AAT1 precursor	CAN23	3
Q96423	Trans-cinnamate 4-mono-oxygenase (cytochrome P450 73)	CAN32	3
BAB09969.1	Cytochrome P450	CAN35	3
XP_476547.1	Putative succinate dehydrogenase flavoprotein alpha subunit	CAN47	3
NP_172092.1	Kinase/ribokinase	CAN54	3
AAS66357.1	Terpene synthase	CAN55	3

Table 2.

Real-time qPCR analysis of selected unigenes

Unigene	Predicted protein	Gland versus leaf	P-value
MEP pathway
CAN548	DOXP synthase	21.5	0.0002
CAN6	Malic enzyme	70.5	0.0001
de novo fatty acid pathway
CAN498	Acyl carrier protein	148	0.0001
CAN82	3-ketoacyl-ACP reductase	735	0.0005
CAN915	3-ketoacyl-ACP reductase	95.3	0.0001
Fatty acid breakdown pathway
CAN162	Lipoxygenase	0.0754	0.0001
CAN295	Lipoxygenase	1.03	0.87
CAN413	Hydroperoxide lyase	9.15	0.0001
CAN94	3-ketoacyl-CoA thiolase; acetyl-CoA acyltransferase	6.79	0.0001
THCA synthesis
CAN70	Acyl-CoA synthetase	22.5	0.0001
CAN24	Putative olivetolic acid synthase	1670	0.0003
CAN1069	CAN CHS	220	0.0001
CAN383	CHS-like protein	1.4	0.564
CAN720	THCA synthase	437	0.0009
Control genes
CAN1084	Beta tubulin	3.45	0.0397
CAN986	Histone H2A	4.24	0.0002
CAN219	Chlorophyll a/b binding protein	0.0246	0.0019
Putative transcription factors
CAN833	AtMYB78&112	954	0.0001
CAN738	AtMYB12&111	586	0.0001
CAN483	MYB related	7.09	0.0004
CAN792	MYB related	8.51	0.0012
CAN747	COP9 complex subunit	11.9	0.0002
CAN227	Homeodomain HAT4	3.48	0.0024
CAN618	CONSTANS-LIKE3	0.00531	0.1903

See Supplementary Table 4 at JXB online for raw data, means, and standard deviations and qPCR primers. Means and standard deviations were used in t tests to calculate the P values shown in the last column.

Fig. 1.

Isolation of glandular trichomes from mature female bracts. (A) Female inflorescence 8 weeks after germination. (B) Leaves associated with the female inflorescence (note arrow). (C) Glands on the bracts of pistillate florets, each bearing a pale bifid style (note arrow indicating glands. (D) Scanning electron micrograph of capitate glands. (E) Micrograph of bract after removal of glands. (F) Micrograph of isolated glands. Bar in (D) and (F)=80 μm.

The top three unigenes representing the greatest number of ESTs encoded proteins related to metallothionein, RD22-like BURP domain-containing proteins, and chitin binding hevein-like proteins (Table 1). All three of these proteins have functions related to biotic or abiotic stress responses (Cobbett and Goldsbrough, 2002; Granger et al., 2002; Van den Bergh et al., 2004). Gene Ontology (GO) analysis was performed on the sequence dataset (Ashburner et al., 2000). An analysis of biological function indicates that 27% of the unigenes encode proteins with metabolic activity. Unigenes with NCBI matches encoding proteins with unknown function comprise 14% of the total and another 28% are predicted to be involved in various cellular processes such as protein synthesis and protein degradation.

Candidate genes in THCA biosynthesis

The specific biochemical steps leading to THCA are proposed to begin with a reaction involving a type III PKS enzyme that catalyses the synthesis of olivetolic acid from hexanoyl-CoA and three molecules of malonyl-CoA (Fig. 2A; Fellermeier et al., 2001). Malonyl-CoA is derived from the carboxylation of acetyl-CoA. ESTs encoding acetyl-CoA carboxylase were identified. Hexanoyl-CoA could be produced by more than one pathway in the trichomes. One route to produce hexanoyl-CoA would involve the early termination of the fatty acid biosynthetic pathway, yielding hexanoyl-ACP (Fig. 2B). The hexanoyl moiety would then be transferred to CoA by the action of an ACP-CoA transacylase or it would be cleaved by the action of a thioesterase, yielding n-hexanol, which would then be converted into n-hexanoyl-CoA by the action of acyl-CoA synthase. Most of the enzymes needed for this route are represented in the EST database, except for the transacylase and 2,3-trans-enoyl-ACP reductase (Fig. 2B). A second route to hexanoyl-CoA would involve the production of hexanol from the breakdown of the fatty acid linoleic acid via the lipoxygenase (LOX) pathway (Fig. 2C; Hatanaka, 1999). A survey of the sequenced ESTs revealed candidate genes encoding the enzymes needed to synthesize linoleic acid from acetyl-CoA by the typical fatty acid biosynthetic pathway in plastids followed by the production of hexanol from linoleic acid via the LOX pathway. An third pathway related to the biosynthesis of branched chain amino acids has been proposed to be involved in the production of short-chain and medium-chain fatty acids (Kroumova et al., 1994). However, the enzymes in this pathway [2-isopropylmalate synthase, 3-isopropylmalate dehydratase, 3-isopropylmalate dehydrogenase, and 2-oxoisovalerate dehydrogenase (acylating)] were not represented in the Cannabis trichome EST library.

Fig. 2.

Biochemical pathways leading from primary metabolites to THCA showing candidate Cannabis unigenes encoding enzymes in the pathway. (A) Production of THCA. (B) De novo fatty acid pathway leading to the formation of hexanol. (C) Breakdown of fatty acids leading to the formation of hexanol. (D) MEP pathway leading to geranyl pyrophosphate. Candidate Cannabis unigenes are shown, as well as the relative gland over leaf expression ratios for a subset of the candidates (see Table 2).

After the formation of olivetolic acid, a prenyltransferase is proposed to add a prenyl group derived from geranyl diphosphate (GPP) to create cannabigerolic acid. GPP is derived from the fusion of two isoprene units (Fig. 2D). Two different biochemical pathways support the synthesis of isoprenoids in plants (Eisenreich et al., 1998; Rohmer, 1999). Within the list of unigenes all but one of the enzymatic activities needed to convert pyruvate and glyceraldehyde-3-phosphate into isopentenyl and dimethylallyl diphosphate via the methylerythritol 4-phosphate (MEP) pathway were represented (Fig. 2D). This finding is consistent with isotopic studies showing that the GPP cannabinoid precursors are synthesized via this pathway (Fellermeier et al., 2001). The formation of GPP is mediated by GPP synthase. Several unigenes related to GPP synthase were identified (with an average of 10% identity), however, they were more closely related to other terpene synthases. In particular, CAN36 and CAN55, which possibly were derived from the same gene, and the closely related CAN37, are most similar to hop (Humulus lupulus) sesquiterpene synthases HISTS1 and HISTS2 (NCBI accessions EU760350 and EU760351, respectively), with an average identity of 56% over the first 160 amino acid residues (Wang et al., 2008). CAN41 is most similar to hop monoterpene synthase HIMTS2 (NCBI accession EU760349; 66% identity over the first 160 amino acid residues; Wang et al., 2008).

The nature of the prenyltransferase is unknown. However, previous studies identified a soluble aromatic geranylpyrophosphate:olivetolate geranyltransferase in the extract of young leaves with the appropriate activity (Fellermeier and Zenk, 1998). The only EST encoding a predicted prenyltransferase was CAN121. However, the encoded protein is more closely related to members of the membrane-bound chloroplast-localized family of prenyltransferases than to soluble prenyltransferases (Heide, 2009). The final step in the pathway is mediated by THCA synthase, which mediates the conversion of cannabigerolic acid to THCA (Fig. 2A). Two ESTs (CAN296 and 720) with sequences identical to the previous reportedly THCA synthase were identified (Sirikantaramas et al., 2004).

Characterization of Cannabis PKS genes

Whereas the nature of the prenyltransferase responsible for the synthesis of cannabigerolic acid is unknown, three unigenes, CAN24, CAN383, and CAN1069, comprising eight, one, and two ESTs, respectively, could encode the PKS activity needed to synthesize olivetolic acid. These were therefore characterized in more detail. All three unigenes were represented by individual ESTs encoding complete PKS polypetides. These were sequenced and compared to related PKS sequences (Fig. 3; see Supplementary Table 3 at JXB online). CAN1069 was identical to a previously identified Cannabis gene encoding a chalcone synthase, and is the most closely related of the PKS sequences to other known chalcone synthases from hop and Arabidopsis (Raharjo et al., 2004b). The relationships of hop phlorisovalerophenone synthase (VPS), which mediates the conversion of malonyl-CoA and isovaleryl-CoA to phlorisovalerophenone, to CAN24 and CAN383 is less clear (Okada and Ito, 2001). CAN24 and CAN383 show 64.6% identity and are nearly equally similar to hop VPS at 71.2% and 72.0%, respectively.

Fig. 3.

Comparison of PKS-related unigenes from Cannabis sativa glandular trichomes. Accession numbers: Arabidopsis thaliana CHS, NP_196897; hop (Humulus lupulus) VPS, BAB121202; hop CHS, CAK19319; Cannabis CHS, AAL92879. The alignment was generated as described in the Materials and methods. The distance relationships are described in the main text.

The enzymatic activities encoded by CAN24 and CAN1069 were explored in detail. The coding regions of the two genes were inserted into the pHis8 vector in frame with a His8 tag. The tagged proteins were purified on a nickel-containing magnetic bead matrix and were assayed for chalcone and olivetol/olivetolic acid synthase activities (Fig. 4). Recombinant protein from CAN1069, but not CAN24, produced reaction products when incubated with coumaroyl-CoA and malonyl-CoA (Fig. 4C, peak 2). The reaction products were analysed by LC-MS and peak 2 was found to have a molecular mass and absorption spectrum consistent with naringenin (Fig. 4E), the major product of chalcone synthases. Both CAN24 and CAN1069 were capable of using malonyl-CoA and hexanoyl-CoA as reaction substrates (Fig. 4C, peak 1; D, peak 3) and LC-MS indicated that products of these enzymes were the same, but neither molecular mass nor the absorption spectrum of this product matched olivetol or olivetolic acid (Fig. 4F). Results similar to CAN24 were obtained using protein purified from CAN383 (data not shown).

Fig. 4.

Analysis of PKS activity for two candidate gene products expressed in Cannabis glands. (A) SDS-PAGE analysis of PKS proteins encoded by CAN1069 and CAN24 expressed in E. coli harbouring the corresponding gene expression plasmids. (B) HPLC separation of the reactants malonyl-CoA (mal), coumaroyl-CoA (cou), and hexanol-CoA (hex), and the potential product olivetol. Separation by HPLC of reaction products obtained using CAN1069 protein (C) or CAN24 protein (D) with either hex or cou as substrates. Numbered peaks represent products not seen in control reactions containing boiled protein. (E) Absorption spectrum of product 2 (solid line) compared to naringenin (dashed line). The molecular mass of the most abundant ion species is shown in the insert. (F) Absorption spectrum of product 3 (solid line) compared to olivetol (dashed line). The molecular mass of the most abundant ion species is shown in the insert. Product 1 yielded the same mass and spectrum as product 3 shown in (F).

qPCR of candidate genes in THCA biosynthesis

Genes required for THCA production are probably more highly expressed in glands of pistillate inflorescences because this is where THCA is most highly concentrated. To test this hypothesis, the relative expression levels in isolated glands versus young inflorescence-associated leaves of selected unigenes were compared using real-time qPCR. The identity of the genes assayed and the differences in relative expression levels are listed in Table 2 and in Supplementary Table 4A at JXB online. Consistent with this hypothesis, THCA synthase (CAN720) expression was 437 times higher in isolated glands than in leaves. CAN24 was expressed 1600 times higher in glands of the inflorescence than in associated leaves. CAN1069 encoding CHS was also more highly expressed in glands than leaves (220-fold). The expression of a third PKS, CAN383, was expressed at similar levels in glands and leaves (Pval=0.564). These results are not explained by poor RNA isolation from leaves as unigene CAN219 encoding chlorophyll A/B binding protein showed elevated leaf expression levels (Table 2; See Supplementary Table 4A, C at JXB online). The activities of several housekeeping genes were also tested. A relatively modest increase in levels of histone H2A (CAN986) and beta tubulin (CAN1084) expression in glands compared to leaves was detected. The increase in expression levels of these latter two genes might reflect a combination of the heightened metabolic activity and the unique cellular structure of glandular trichomes.

Two different pathways could provide the hexanol required for olivetolic acid synthesis, as shown in Fig. 2. Expression levels provide support for the de novo pathway as a primary source, given that CAN498, CAN82, and CAN915 were much more highly expressed in glands than leaves (Fig. 2; Table 2; see Supplementary Table 4A at JXB online), whereas the relative expression of genes encoding enzymes in the lipid breakdown pathway were depressed or modestly elevated in glands.

Identification of potential transcription factors expressed in Cannabis glands

Cannabis unigenes were compared to known Arabidopsis thaliana transcription factors (Guo et al., 2005). Eighty Cannabis unigenes were similar to transcription factors found in Arabidopsis and 11 contain MYB DNA binding domains (see Supplementary Table 5 at JXB online for P-scores). Expression of four MYB genes in isolated glands and leaves was compared by real-time qPCR (Table 2; see Supplementary Table 4A at JXB online). CAN833 and CAN738 exhibited 954-fold and 586-fold higher expression in glands, respectively, whereas CAN483 and 792 showed more modest induction in glands. None of the other putative transcription factors that were assayed showed the same degree of differential expression as CAN833 and CAN738 (Table 2).

Discussion

Characterization of Cannabis gland ESTs

The identities of the most abundant ESTs derived from the glandular trichomes of Cannabis sativa are consistent with the protective function of plant glands. For example, the most abundant ESTs encoded a protein closely related to type II metallothioneins. These proteins bind heavy metals such as Cd, Zn, and Cu, and their proposed primary function is the maintenance of Cu tolerance (Cobbett and Goldsbrough, 2002). The second most abundant class of ESTs encoded an RD22-like BURP domain containing protein. This class of proteins contains a hydrophobic N-terminal signal peptide, and an N-terminal conserved region followed by a series of small repeats (Granger et al., 2002). The BURP domain of approximately 230 amino acids is located in the C-terminal region. The function of RD22-like proteins is unknown but some members of this class of genes are induced by dehydration (Yamaguchi-Shinozaki and Shinozaki, 1993). The third most abundant ESTs encoded a protein containing a hevein domain. Hevein domains contain a conserved 43-amino acid motif that binds chitin and members of this protein class are known for antifungal activity (Van den Bergh et al., 2004). The unique secondary metabolism in Cannabis may also play a role in plant defence. Synthesis of THCA is extracellular and results in hydrogen peroxide production, which has general antimicrobial properties (Sirikantaramas et al., 2005), and a recent report further indicates that THCA may directly inhibit microbial growth (Appendino et al., 2008).

The analysis of gland-derived ESTs has identified nearly all the candidate genes required for THCA synthesis from primary metabolic products. These findings differ from a proteomic study that aimed to identify genes expressed in Cannabis glands but failed to associate any highly expressed proteins with THCA synthesis (Raharjo et al., 2004c). This difference reflects the much greater volume of genomic data enabling more robust identification of DNA sequences when compared to proteomics approaches based on the molecular weights of fragmented polypeptides. This is especially true for species such as Cannabis sativa for which there is little amino acid sequence data available to compare with peptide profiles.

The present study highlights the utility of using isolated glands as starting material for making EST libraries to study gland metabolism, as was the case in other plant species (Lange et al., 2000; Gang et al., 2001; Fridman et al., 2005). In this study more than 50% of the ESTs with NCBI matches were involved in metabolism or cellular activities such as transport and protein translation. Many other cannabinoids, in addition to THCA, have been identified in Cannabis (Mechoulam, 1970), and it is likely that many of the genes identified in Supplementary Table 2 at JXB online are involved in the production of these other compounds. In addition to cannabinoids, many other classes of secondary compounds have been found in Cannabis (Flores-Sanchez and Verpoorte, 2008b). For example, both monoterpenes and sesquiterpenes have been identified and candidate ESTs encoding activities to produce these compounds have been identified.

Characterization of Cannabis PKSs and the difficulty in the identification of olivetolic acid synthase

Synthesis of olivetolic acid from malonyl-CoA and hexanoyl-CoA represents the first committed step toward the synthesis of THCA. Olivetolic acid synthesis is predicted to be mediated by a member of the type III PKS family through a series of three condensation reactions producing a triketide (Fig. 7B; Fellermeier et al., 2001). CAN24, represented by eight ESTs and one of the most highly expressed unigenes in our analysis, encodes a member of the PKS family. This gene was expressed 1600-fold higher in glands than in leaves. CAN1068, another PKS member represented by two ESTs, corresponds to a previously identified Cannabis CHS gene (Raharjo et al., 2004b). A third PKS represented by a single EST, CAN383, was also identified. Analyses of PKS crystal structures indicate that the type III PKS enzymes are composed of a dimer with conserved reaction centres and a hollow reaction cavity (Austin and Noel, 2003). All three Cannabis PKS genes encoded polypeptides containing the conserved amino acids, Cys 167, His 307, and Asn 340, that are believed to constitute the reaction centre (see Supplementary Table 3 at JXB online; Jez et al., 2000). In addition, two of the three amino acids that are important for defining the size of the reaction cavity in chalcone synthases (Jez et al., 2000; Abe et al., 2006) are conserved. The third amino acid, Thr at position 300 that is conserved in all chalcone synthases, was missing in CAN24 and CAN383 (see Supplementary Table 3 at JXB online). Instead CAN24 and CAN383 contained Leu and Iso, respectively, at position 300. Such differences might alter substrate specificity.

It has been proposed that either olivetol or olivetolic acid are products of polyketide synthase in the THCA pathway (Raharjo et al., 2004b). However assays of plant extracts found that olivetol, the decarboxylation product of olivetolic acid, was not a substrate in the pathway (Fellermeier and Zenk, 1998). Products of the three PKS genes identified in this study were tested for olivetolic acid synthesis in vitro. CAN24 and CAN383 yielded identical products according to HPLC analysis (data not shown). Because CAN24 was more abundant, this PKS gene was analysed in detail, along with CAN1069, which had CHS activity as shown in Fig. 4. The size of the product produced by the CAN24-encoded enzyme was smaller than olivetolic acid (Fig. 4F). Further, the absorption spectrum did not match olivetol. A sequence identical to CAN24 has been deposited in the NCBI database (accession AB164375) and was annotated as having olivetol synthase activity but without supporting data. The product generated in our analyses possibly represents a derailment product in which the enzyme catalyses two decarboxylative condensations instead of three (Fig. 5). The failure of in vitro conditions to support the complete synthesis of native PKS products is well-documented (Springob et al., 2007). The artificial nature of in vitro reaction products is reflected by the failure to find these products in vivo (Austin and Noel, 2003). This may indicate that reaction conditions were insufficient or that accessory proteins may be needed to produce olivetolic acid. Similar results were obtained in a study that characterized a gene identical to CAN1069 where the enzyme could use hexanoyl-CoA as a starter molecule, but only yielded a possible derailment product (Raharjo et al., 2004b). That plant extracts used in in vitro assays have also not yielded olivetolic acid suggests that in vivo-like reaction conditions have yet to be imitated (Raharjo et al., 2004a; Flores-Sanchez and Verpoorte, 2008a).

Fig. 5.

Comparison of potential products produced by two or three decarboxylative condensation reactions. (A) Prediction of pyrone production from two decarboxylative condensations. (B) Expected product from three decarboxylative condensations.

The assay results with CAN1069 also highlight the permissiveness of substrate use by PKSs. CAN1069 clearly had CHS activity in that it could use coumaroyl-CoA as a substrate to produce naringenin. Given that CAN1069 was preferentially expressed in the glands and can act on hexanoyl-CoA, this PKS also may contribute to THCA production.

Insights on THCA biosynthesis may be gained by comparison with the secondary metabolism of glandular trichomes in hop, the closest relative to Cannabis. Production of the bitter acid humulone in hop inflorescence glands requires a PKS called VPS. In two independent studies, ESTs representing VPS were among the most abundant in the collections and were at least 5-fold more abundant than other PKS encoding ESTs (Nagel et al., 2008; Wang et al., 2008). Accordingly, the role of CAN24 as olivetolic acid synthase remains tentative pending further biochemical support, however, it is the best candidate based on expression data.

Identification of MYBs with potential roles in gland chemistry

A large number of Cannabis gland unigenes encoded proteins similar to transcription factors (see Supplementary Table 5 at JXB online). Several of these were analysed by qPCR and two were found to be preferentially expressed in glands. This is potentially significant as studies have shown that all but one of the 12 transcription factors required for Arabidopsis trichome formation are preferentially expressed in trichomes compared to whole leaves (Marks et al., 2009). In this study it was found that two R2R3-type Cannabis MYBs, encoded by CAN833 and CAN738, were preferentially expressed in isolated glands compared to leaves by 954-fold and 586-fold, respectively. CAN833 and CAN738 are most similar to the Arabidopsis MYBs related to AtMYB112 and AtMYB12, respectively (see Supplementary Table 5 at JXB online; Stracke et al., 2001). AtMYB112 corresponds to the BOTRYTIS SUSCEPTIBLE1 (BOS1) gene. bos1 mutants are more susceptible to pathogens such as Botrytis cinerea and Alternaria brassicicola, and have impaired tolerance to oxidative stress (Mengiste et al., 2003). A role for CAN883 in tolerance to oxidative stress in Cannabis glandular trichomes is logical, as the last enzymatic reaction in THCA synthesis releases hydrogen peroxide (Sirikantaramas et al., 2005). AtMYB12 controls the synthesis of flavonol secondary metabolites in Arabidopsis and can induce the synthesis of similar compounds in tobacco (Mehrtens et al., 2005; Luo et al., 2008). Flavonoids have been isolated from Cannabis leaves and flowers, but evidence is lacking for gland flavonoid production (Flores-Sanchez and Verpoorte, 2008b). Since flavonols are not predominant in Cannabis glands, it is possible that CAN738 instead plays a role in controlling the expression of genes required for other secondary metabolites in Cannabis such as THCA.

Potential utility of comparing gland ESTs from hop and Cannabis

The PKSs and many other genes identified in this study are closely related to those from hop (Humulus lupulus). Humulus and Cannabis are monotypic sister genera in the family Cannabaceae (Datwyler and Weiblen, 2004). Glandular trichomes located on the inflorescence bracts of both Humulus and Cannabis are the location of unique PKS-derived secondary metabolism (Nagel et al., 2008; Wang et al., 2008). Hop glands produce the bitter acid humulone, which is important for beer flavour, and the prenylated chalcone xanthohumol, which has several potential health beneficial properties (Stevens and Page, 2004). The biochemical pathways leading to THCA, xanthohumol, and humulone have common steps that include polyketide synthases and prenyltransferases. It is probable that these plants share other homologous biochemical pathways given their close ancestry. Information from Cannabis ESTs has the potential to improve the understanding of hop biochemical pathways as well.

This work was funded by a sabbatical supplement from the University of Minnesota College of Biological Sciences (MDM), by the University of Minnesota Life Sciences Summer Undergraduate Research Programs (SNO and WS-F), the Samuel Roberts Noble Foundation (LT and RAD), the Minnesota Agricultural Experiment Station (GDW), and a Packard Fellowship in Science and Engineering (GDW). We acknowledge Mahmoud El Sohly and Zlatko Mehmedic at the University of Mississippi for previously published analyses of cannabinoid content under the National Institute on Drug Abuse contract number N01DA-5-7746.

Author notes

†

Present address: Department of Plant Sciences, University of California Davis, One Shield Avenue, Davis, CA 95616, USA.