Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Transcriptome Profiling of Khat (Catha edulis) and Ephedra sinica Reveals Gene Candidates Potentially Involved in Amphetamine-Type Alkaloid Biosynthesis

  • Ryan A. Groves,

    Affiliation Department of Biological Sciences, University of Calgary, Calgary, Alberta, Canada

  • Jillian M. Hagel,

    Affiliation Department of Biological Sciences, University of Calgary, Calgary, Alberta, Canada

  • Ye Zhang,

    Affiliation Department of Biochemistry and Molecular Biology, University of Calgary, Calgary, Alberta, Canada

  • Korey Kilpatrick,

    Affiliations Department of Biology, Western University, London, Ontario, Canada, Southern Crop Protection and Food Research Centre, Agriculture and Agri-Food, London, Ontario, Canada

  • Asaf Levy,

    Affiliations The Robert H. Smith Institute of Plant Science and Genetics in Agriculture, The Hebrew University of Jerusalem, Rehovot, 76100, Israel, Department of Aromatic, Medicinal and Spice Crops, Newe Ya’ar Research Center, Agricultural Research Organization, P.O. Box 1021, Ramat Yishay, 30095, Israel

  • Frédéric Marsolais,

    Affiliations Department of Biology, Western University, London, Ontario, Canada, Southern Crop Protection and Food Research Centre, Agriculture and Agri-Food, London, Ontario, Canada

  • Efraim Lewinsohn,

    Affiliation Department of Aromatic, Medicinal and Spice Crops, Newe Ya’ar Research Center, Agricultural Research Organization, P.O. Box 1021, Ramat Yishay, 30095, Israel

  • Christoph W. Sensen,

    Affiliation Department of Biochemistry and Molecular Biology, University of Calgary, Calgary, Alberta, Canada

  • Peter J. Facchini

    Affiliation Department of Biological Sciences, University of Calgary, Calgary, Alberta, Canada


Amphetamine analogues are produced by plants in the genus Ephedra and by khat (Catha edulis), and include the widely used decongestants and appetite suppressants (1S,2S)-pseudoephedrine and (1R,2S)-ephedrine. The production of these metabolites, which derive from L-phenylalanine, involves a multi-step pathway partially mapped out at the biochemical level using knowledge of benzoic acid metabolism established in other plants, and direct evidence using khat and Ephedra species as model systems. Despite the commercial importance of amphetamine-type alkaloids, only a single step in their biosynthesis has been elucidated at the molecular level. We have employed Illumina next-generation sequencing technology, paired with Trinity and Velvet-Oases assembly platforms, to establish data-mining frameworks for Ephedra sinica and khat plants. Sequence libraries representing a combined 200,000 unigenes were subjected to an annotation pipeline involving direct searches against public databases. Annotations included the assignment of Gene Ontology (GO) terms used to allocate unigenes to functional categories. As part of our functional genomics program aimed at novel gene discovery, the databases were mined for enzyme candidates putatively involved in alkaloid biosynthesis. Queries used for mining included enzymes with established roles in benzoic acid metabolism, as well as enzymes catalyzing reactions similar to those predicted for amphetamine alkaloid metabolism. Gene candidates were evaluated based on phylogenetic relationships, FPKM-based expression data, and mechanistic considerations. Establishment of expansive sequence resources is a critical step toward pathway characterization, a goal with both academic and industrial implications.


Plants produce a wide variety of specialized nitrogenous metabolites, including a large number of pharmacologically important alkaloids. Ephedrine alkaloids, also termed phenylpropylamino alkaloids or substituted amphetamines, are a diverse compound class featuring a phenethylamine backbone with a methyl group at the α-position relative to the nitrogen [1]. Myriad functional group substitutions have yielded a series of synthetic drugs with diverse pharmacological properties as stimulants, empathogens, and hallucinogens [2]. Although most analogues are synthetic, certain plants have evolved a capacity for their biosynthesis. Two prominent examples include khat (Catha edulis) and Ephedra sinica, each of which have been cultivated for centuries owing to their mild stimulant and medicinal properties. The chewing of khat leaves as a social activity dates back at least a thousand years [3], a practice that might even predate the use of coffee [4]. Khat chewing remains an important tradition in parts of East Africa and the Middle East, although possession of khat is illegal in many Western countries. (S)-Cathinone (Fig. 1) is the principle active neurostimulant in khat, although other alkaloids such as (1S,2S)-pseudonorephedrine (cathine) and (1R,2S)-norephedrine also occur. Alkaloid biosynthesis in E. sinica extends beyond that of khat to include N-methylated enantiomers (1S,2S)-pseudoephedrine and (1R,2S)-ephedrine (Fig. 1). N,N-Dimethylated versions of these alkaloids also occur, although in lesser quantities [5]. E. sinica, known as Ma Huang in Chinese medicine, also has a long history of traditional use, particularly in the treatment of asthma, coughs, congestion, headaches and edema [6]. Today, (1S,2S)-pseudoephedrine is a common ingredient in cold and allergy formulations, whereas (1R,2S)-ephedrine is marketed as decongestant, diet aid and sports-enhancing drug.

Fig 1. Proposed biosynthetic routes leading from L-phenylalanine to amphetamine-type alkaloids in khat and Ephedra sinica.

A CoA-independent, non-β-oxidative pathway of L-phenylalanine side chain-shortening is shown in blue, whereas a CoA-dependent, β-oxidative route is shown in purple. Benzaldehyde, benzoic acid and/or benzoyl-CoA undergo condensation with pyruvate, a reaction putatively catalyzed by a ThDP-dependent carboligase. 1-Phenylpropane-1,2-dione undergoes transamination to yield (S)-cathinone, which is reduced to cathine and (1R,2S)-norephedrine. N-Methylation is restricted to Ephedra spp. and does not occur in khat. Activity has been detected for enzymes highlighted in yellow, and corresponding genes are available for enzymes highlighted in green. Enzymes highlighted in red have not been isolated, although database mining revealed numerous potential candidates (Tables 2 and 3). Abbreviations: CoA, Coenzyme A; NAD(H), nicotinamide adenine dinucleotide; NADP(H), nicotinamide adenine dinucleotide phosphate. PAL, phenylalanine ammonia lyase; ThDP, thiamine diphosphate.

Despite their occurrence in plants, the modern manufacture of (1S,2S)-pseudoephedrine and (1R,2S)-ephedrine relies predominantly on a process involving fermentation coupled with synthetic chemistry [1]. The fact that industrial production of these compounds is at least partially completed through fermentation raises the intriguing possibility of employing synthetic biology for the production of these alkaloids. Such an endeavor requires thorough knowledge of the enzyme catalysts responsible for alkaloid biosynthesis in plants. The first step in the pathway is catalyzed by L-phenylalanine ammonia lyase (PAL), a well-characterized enzyme in many species including E. sinica [7] (Fig. 1). Recent work in the field of floral scent biosynthesis has largely elucidated benzoic acid metabolism, a process occurring either in the peroxisomes as part of the core β-oxidative CoA-dependent pathway, or in the cytosol as part of the non-β-oxidative, CoA-independent pathway [8]. Alternatively, benzaldehyde may be formed via phenylpyruvate, a transamination product of L-phenylalanine (Phe). This route occurs in lactic acid bacteria [9] but has not been confirmed in plants. Whereas Phe serves as the initial precursor for the aromatic C6-C1 component of ephedrine alkaloids, the C2-C3 portion derives from pyruvate through a carboligation mechanism, likely catalyzed by a ThDP-dependent enzyme (Fig. 1). It was previously suggested that either benzoic acid or benzoyl-CoA serves as the C6-C1 carboligation co-substrate [10], although recent evidence supports the involvement of benzaldehyde [11], at least in Ephedra species. Stem extracts of E. sinica and E. foeminea catalyzed the conversion of benzaldehyde and pyruvate to five distinct carboligation products, including 1-phenylpropane-1,2-dione. Although the gene encoding the benzaldehyde carboligase (BCL) was not identified, the in vitro turnover of diverse C6-C3 backbone structures prompted a reconsideration of possible route(s) leading to alkaloid formation in planta. It has long been assumed that 1-phenylpropane-1,2-dione, a naturally occurring metabolite of both khat [12,13] and E. sinica [5], is a key intermediate. Transamination of this diketone yields (S)-cathinone, which undergoes stereospecific reduction to either cathine or (1R,2S)-norephedrine. N-Methylation of these diastereomers completes the pathway in E. sinica. An alternative route circumventing 1-phenylpropane-1,2-dione is possible, although the natural occurrence of alternative transaminase substrates (Fig. 1) has not been confirmed.

Despite the commercial importance of amphetamine analogues and the potential for synthetic biology applications [1], PAL remains the only step characterized at the molecular level in plants accumulating these compounds [7]. We recently reported a Sanger-based, expressed sequence tag (EST) library from khat containing 3,293 unigenes [14]. Here we report a dramatic expansion of our functional genomics platform to include over 200,000 unigenes, derived from Illumina-based, next-generation sequencing (NGS, or RNA-seq) of both khat and E. sinica plants. Illumina NGS technology remains the most popular platform owing to its ever-growing read lengths and overall number of reads per run, allowing both transcriptome assembly in the absence of a reference genome (de novo assembly) and quantitative gene expression analysis [15]. The results of this study provide an unprecedented opportunity to identify novel alkaloid biosynthetic genes and quantify gene expression. Further, we evaluate the performance of two different assembly programs, Trinity de novo RNA-seq assembler [16] and Velvet-Oases v0.1.16 [17].

Materials and Methods

Plant materials

Khat shrubs (Catha edulis, Forsk.) cv “Mahanaim” were grown in open field conditions under commercial growing practices, including drip irrigation and fertilization, at the Newe Ya’ar Research Center in Northern Israel. The shrubs were approximately 10 years old at the time of harvest. The Ephedra sinica plants used in this study were germinated from seeds acquired from wild, openly pollinated varieties originating in Northern China (Horizon Herbs, OR, USA). Cultivation was carried out in Canada at the Southern Crop Protection and Food Research Centre (SCPFRC; London, Ontario), where plants were grown under standard greenhouse conditions in pots containing a 50:50 blend of sand and commercial cactus soil mix.

Poly(A)+ RNA purification, cDNA library preparation and Illumina GA sequencing

Young khat leaves, approximately 1–3 cm in length, were harvested during daylight hours, and total RNA was isolated using an RNeasy Midi kit (Qiagen). For E. sinica, freshly emerging, light green “young” stems up to 5 cm in length were harvested for total RNA extraction using essentially the same procedure as for the khat tissue. Poly(A)+ RNA purification, cDNA library preparation, emulsion-based PCR (emPCR) and sequencing were performed at the McGill University and Génome Québec Innovation Center (Montréal, Canada). RNA quality was assessed using an RNA 6000 Nano chip on a BioAnalyzer 2100 (Agilent Technologies) to ensure an RNA Integrity Number (RIN) of > 7.5. Poly(A)+ RNA was prepared using a Dynabead mRNA Purification kit (Invitrogen). cDNA library preparation and Illumina GA sequencing were performed as described [18].

De novo transcriptome assembly

Sequence quality control and cleaning were performed as described [18]. Short-read sequence data corresponding to khat cultivated at the Newe Ya’ar Research Center were assembled using the Trinity de novo RNA-seq assembler [16]. In contrast, data corresponding to E. sinica plants cultivated at SCPFRC were assembled using both Trinity de novo RNA-seq assembler and Velvet-Oases v0.1.16 [17], respectively, generating two distinct libraries. The parameters used for Trinity-based assemblies (module “Butterfly”) were as follows: graph compaction option: edge-thr = 0.26; path extension mode = compatible_path_extention; min_contig_length = 300; paired_fragment_length = 270 (50 bp + estimated median fragment size of readset). Standard settings were used otherwise. The Velvet-based assembly was performed as described previously [18].

Functional annotation and GO analysis

Annotation of the three assembled transcriptome datasets was performed using the Magpie Automated Genomics Project Investigation Environment (MAGPIE) as described [18,19]. Briefly, MAGPIE automates sequence similarity searches against major public and local target databases. The TimeLogic Tera-BLAST algorithm ( was used to compare transcripts to the NCBI database NR (non-redundant) and the viridiplantae subset of RefSeq [20]. An expected e-value of 1e-3 and a minimum alignment length of 30 bp were used. Information regarding sequence motifs was collected using accelerated Hidden Markov Model (HMM) searches against local instances of Interpro HMM libraries at an e-value of 1e-10. The NCBI Conserved Domain Database (CDD) was also queried for further structural information. Finally, a functional description was assigned to each contig, based on a weighted summary of all search results. GO annotations were compiled from GIDs extracted from “level 1” evidence (i.e. annotations based on BLAST matches > 1e-35, HMM matches > 1e-20, and sequence similarities > 65%). GO terms assigned to each contig were used to sort the contigs into one of 13 functional categories, including “secondary metabolism” (ID GO:0019748) and an “unknown” category comprised of contigs assigned to the GO term “Biological Process Unknown” (ID GO:0008150). In addition contigs that did not match to any GO term were grouped separately. All the raw transcriptome data have been deposited in the NCBI Short Read Archive (SRA).

Identification and expression analyses of gene candidates

Khat and E. sinica databases (CED-Trinity and ESI-Velvet, respectively) were searched for contigs representing genes potentially involved in phenylpropylamino alkaloid biosynthesis. ESI-Trinity was not queried in the first round, since contigs in this database were relatively short compared with those found in ESI-Velvet. Full-length query sequences representing functionally characterized enzymes were used for tBLASTn searches. When available, gymnosperm sequences were used to query ESI-Velvet, whereas only angiosperm sequences were used to query CED-Trinity. Of the 40 gene candidates identified from CED-Trinity, two were comprised of more than one contig, which were assembled manually based on obvious regions of identity. All E. sinica gene candidates corresponded to auto-assembled contigs or singlets. To assign FPKM values to E. sinica gene candidates (available only through Trinity-based assembly) a second round of tBLASTn was performed using gene candidates identified in ESI-Velvet to query ESI-Trinity. In cases where a single query revealed more than one identical match in ESI-Trinity, a single FPKM was re-calculated representing all contigs within the group.

Phylogenetic analysis

Amino acid alignments were performed using ClustalW [21], and phylogenetic trees were built from bootstrap values generated using Geneious software package (Biomatters, Newark, NJ). GenBank accession numbers for sequences used to construct the trees are as follows: BDH outgroup, NP_035853.2; AmBDH Antirrhinum majus, ACM89738.1; AtBDH Arabidopsis thaliana, NP_563711; BL outgroup, Q5SKN9.1; BL Arabidopsis thaliana, NP_176763.1; 4CL outgroup, Q336M7.3; 4CL Arabidopsis thaliana, Q42524.1; 4CL Nicotiana tabacum, O24145.1; 4CL Petroselinum crispum, P14912.1; 4CL Pinus taeda, AAA92669.1; 4CL Solanum tuberosum, P31684.1; KAT outgroup, P45362.2; 2WUA; KAT Arabidopsis thaliana, 2WU9; KAT Helianthus annuus, KAT Petunia x hybrida, ACV70032.1; NMT outgroup, Q5C9L6.1; NMT Arabidopsis thaliana 1, NP_196113; NMT Arabidopsis thaliana 2, NP_199713; NMT Atropa belladonna, BAA82264; NMT Coffea arabica, BAC75663; NMT Papaver somniferum, AAY79177; NMT Solanum lycopersicon, AAG59894; PAL outgroup, P10248.2; PAL Arabidopsis thaliana, P35510.3; PAL Cunninghamia lanceolata, AFX98070.1; PAL Ephedra sinica, AB300199.1; PAL Eucalyptus robusta, BAL49995.1; PAL Larix kaempferi, AHA44840.1; PAL Petroselinium crispum, P45729.1; PAL Pinus taeda, P52777.1; PAL Selaginella moellendorffii, XP_002982907.1; RED outgroup, P39640.2; RED Datura stramonium TRI, TRI_L20473; RED Datura stramonium TRII, RED Eschscholzia californica, ADE41047.1; TRII_L20474; RED Hyoscyamus niger TRI, P50164.1; RED Papaver somniferum, AAF13739.1; TA outgroup, Q4J8X2.1; CmTA Cucumis melo, ADC45389.1; TA Papaver somniferum, ADC33123.1; PxhTA Petunia x hybrida, E9L7A5.1; PpTA Pinus pinaster, CAF31327.1; ThDPC outgroup, P27868.1; AtAHAS Arabidopsis thaliana, ABJ80681.1; AtPDC Arabidopsis thaliana, NP_200307.1; BnAHAS Brassica napus, P27818.1; NtPDC Nicotiana tabacum, P51846.1; PsPDC Pisum sativum, P51850.1; ZmPDC Zymomonas mobilis, 1ZPD.

Results and Discussion

Tissue selection for enrichment of biosynthetic genes

In khat, pathway intermediates 1-phenylpropane-1,2-dione and (S)-cathinone, and end products cathine and (1R,2S)-norephedrine, accumulate mainly in young leaves and flowers with lesser quantities in young stems [12,13]. In contrast, mature leaves lack (S)-cathinone and accumulate only cathine and (1R,2S)-norephedrine suggesting that alkaloid biosynthetic gene expression is highest in young tissues. For this reason, young khat leaves were selected for RNA extraction. Precursors 1-phenylpropane-1,2-dione and (S)-cathinone are present in young E. sinica stems, but not mature stems or roots [5]. In contrast, downstream metabolites including cathine, (1R,2S)-norephedrine, N-methylated or N,N-dimethylated products occur in both young and mature stems, although accumulation is much greater in mature stems. This pattern suggests that alkaloid biosynthesis is carried out predominantly in younger tissue, resulting in the presence of end products in older tissue as the stems elongate. We targeted a transciptome presumably enriched in alkaloid biosynthetic gene transcripts by choosing young E. sinica stems for RNA extraction. Marked diversity in alkaloid stereochemical composition has been noted for different E. sinica accessions. For example, some varieties accumulate only 1R isomers, whereas others accumulate both 1S and 1R forms [5]. Assaying for (S)-cathinone reductase activity in extracts of these E. sinica accessions revealed that the proportion of (1S,2S)-cathine and/or (1R,2S)-norephedrine product corresponded directly with plant accumulation profiles, supporting the existence of stereospecific reductases. The E. sinica used in this study originates with seeds harvested from wild populations native to the steppes of north and northwestern China, and accumulates both isomers. Therefore, if stereospecific reactions are involved, representation for enzymes favoring both 1S and 1R epimers is expected.

Transcriptome sequencing, assembly and annotation

RNA was screened for sufficient quality (S1 Fig.) prior to Illumina sequencing to generate 272,586,558 and 191,352,154 short sequence reads for khat and E. sinica, respectively (Table 1). Quality filtering these data yielded “clean” reads representing 79% (khat) and 86% (E. sinica) of the original number of raw reads. We initially opted to use the newer Trinity de novo RNA-seq software [16] instead of an older package, Velvet-Oases v0.1.16 [17], to assemble khat and E. sinica transcriptomes. Comparison of Trinity and Velvet-Oases platforms has found that the Trinity assembler is superior at resolving splice alternates, and produces less duplicates or assembly chimeras, which are often introduced in a Velvet-Oases multi-run merging stage, hence generating a lower total number of long reads than Velvet-Oases [22]. Furthermore, Velvet-Oases v.0.1.16 [17] does not allow calculation of FPKM (fragments mapped per kilobase of exon per million reads mapped), a normalizing statistic measuring gene expression while accounting for variation in gene length [23]. It is noteworthy that newer versions of Velvet-Oases [24] (, which were not available at the time of assembly, are routinely used to calculate normalizing statistics. Trinity-based assembly of khat and E. sinica libraries yielded 77,290 and 63,344 unigenes respectively, proportionately in line with the number of reads (CED-Trinity and ESI-Trinity; Table 1). However, it was discovered during full-length coding sequence (CDS) analysis that unlike CED-Trinity where half (50%) of the unigenes represented full-length clones, less than one third (29%) of ESI-Trinity unigenes contained a complete CDS. BLAST searches of CED-Trinity for gene candidates using full-length homologues from various plant species yielded a modest number of hits per query (<10) (Table 2), but searches of ESI-Trinity yielded large lists of hits (>50), many of which were identical except for 1 or 2 base pairs (data not shown). A possible reason for the ability of Trinity to assemble full-length clones for khat data but not E. sinica data is the occurrence of single nucleotide polymorphisms (SNPs) within E. sinica transcripts. Unlike the khat tissue, which derived from a single commercial cultivar with a long history of breeding, the E. sinica plants used in this study were derived from wild populations possibly representing more than one accession. Nonetheless, a marked lack of fully assembled, full-length contigs in ESI-Trinity impeded our ability to reliably identify candidate biosynthetic genes. Therefore, Velvet-Oases was used to assemble a third library, ESI-Velvet, which contained 59,448 unigenes representing a larger number (41,671) of CDSs (Table 1). BLAST performed with ESI-Velvet assemblies showed a greater number of full-length contigs and fewer unassembled singlets compared with ESI-Trinity (Table 3), and were thus used for gene candidate searches.

Table 1. Summary of the construction and assembly for three Illumina NGS libraries.

Table 2. Khat unigenes representing enzymes putatively involved in ephedrine alkaloid biosynthesis.

Table 3. Ephedra sinica unigenes representing enzymes putatively involved in ephedrine alkaloid biosynthesis.

The MAGPIE-based pipeline used to annotate the three libraries included gene- and domain-level searches against NCBI and RefSeq [20] databases, in addition to Hidden Markov Models searches [18]. All search results were summarized and weighted to arrive at a single annotation per unigene. Additionally, GO (Gene Ontology) annotations were assigned to individual contigs, based on continually updated criteria ( Fig. 2 illustrates the GO annotation results for CED-Trinity and ESI-Trinity databases. The GO pie chart for ESI-Velvet is found in S2 Fig. GO annotation analysis yielded very similar results for all three libraries, with primary metabolism comprising the largest category of unigenes (Fig. 2, S2 Fig.). The GO analysis results for CED-Trinity are very similar to those reported for a Sanger sequencing-based khat cDNA library [14]. This finding is not surprising, as source tissue was the same in both cases.

Fig 2. Functional category analysis based on Gene Ontology (GO) annotations of CED-Trinity (upper panel) and ESI-Trinity (lower panel) unigenes.

Results for ESI-Velvet are found in S2 Fig.

Evidence for benzoic acid metabolism in khat and Ephedra sinica

Toward the goal of identifying genes putatively involved in ephedrine alkaloid metabolism, khat and E. sinica libraries were queried using amino acid sequences corresponding to enzymes of known function. Search results are listed in Table 2 (khat) and Table 3 (E. sinica) with accompanying phylogenetic analyses presented in Fig. 3 and S3 Fig. Full sequences of these search results are listed in S1 Dataset and S2 Dataset for khat and E. sinica, respectively. Complete sequences of the queries used for data mining of khat and E. sinica libraries are found in S3 Dataset and S4 Dataset, respectively. As illustrated in Fig. 1, many steps of benzoic acid metabolism have been characterized at the molecular genetic level from such plants as Arabidopsis thaliana, snapdragon (Antirrhinum majus) and petunia (Petunia x hybrida). This enabled the use of queries with established roles in the pathway, at least in source plant species, to find putative homologues in khat and E. sinica. In some cases, more than one enzyme is known to catalyze a particular step; for example, dehydrogenation of benzaldehyde to benzoic acid may occur in Arabidopsis seeds through aldehyde oxidase-4 (AtAAO4) [25] or via benzaldehyde dehydrogenase in snapdragon (AmBALDH) [26]. Previous analysis of a modest (3,293 unigenes) khat transcriptomic library indicated that only a AmBALDH homologue was present, suggesting a pathway more similar to snapdragon petals than seeds of Arabidopsis [14]. The more extensive sequencing performed in this study revealed the presence of both AtAAO4 (CeBDH1–1) and AmBALDH (CeBDH2–1, 2–3, 2–4) homologues in CED-Trinity (Table 2). Phylogenetic analysis of dehydrogenase candidates distinguished these khat sequences in separate clades (Fig. 3A). Nonetheless, CeBDH1–1 identity with AtAA04 was low (51%) compared with that between AmBALDH and the top khat homologue CeBDH2–1 (78%). Relative expression of CeBDH2–1 was > 80-fold greater than that of CeBDH1–1 (Fig. 4), perhaps underscoring an important role for BALDH activity in khat leaves. In E. sinica, a similar pattern emerged where a close homologue was found for AmBALDH (EsBDH2–1, 80% identity), but not for AtAA04 (Table 3).

Fig 3. Phylogenetic analysis of gene candidates.

Abbreviations: benzaldehyde dehydrognase (BDH) (A), thiamin diphosphate-dependent carboligase (ThDPC) (B) and transaminase (TA) (C). Similar analyses for remaining candidates are found in S3 Fig. Sequences were aligned and analyzed for phylogenetic relationships using the neighbor-joining algorithm. Numbers at each node represent bootstrap values calculated using 1000 iterations. Accession numbers are found under Experimental section, and abbreviations are defined in Tables 2 and 3.

Fig 4. Relative expression of gene candidates identified in khat (CED-Trinity).

FPKM (fragments mapped per kilobase of exon per million reads mapped) is a normalizing statistic measuring gene expression while accounting for variation in gene length [23]. Abbreviations are defined in Tables 2 and 3.

Khat sequences closely related (~80% identity) to petunia cinnamoyl-CoA hydratase-dehydrogenase (PhCHD) and 3-ketoacyl-CoA thioliase (PhKAT1) were identified (Table 2) suggesting that a β-oxidative, CoA-dependent pathway is also operative in this species. Several putative PhKAT1 homologues were identified in E. sinica, ranging in identity up to 78% (Table 3). Less identity (<60%) was observed between E. sinica candidates and PhCHD, possibly reflecting an alternative metabolism occurring in this plant. For example, trans-cinnamoyl-CoA may be converted to 3-oxo-3-phenylpropionyl-CoA by two separate enzymes rather than a single bi-functional enzyme. However, sequence divergence could also reflect evolutionary distance between E. sinica, a gymnosperm [27,28], and flowering plants.

Carboligase candidates belonging to AHAS and PDC enzyme families

While benzoic acid metabolism is largely elucidated at the molecular genetic level, enzymes catalyzing steps predicted to occur beyond this pathway have not been cloned. Therefore, plant enzymes catalyzing mechanistically similar reactions were used as queries to identify khat and E. sinica genes potentially involved in amphetamine analogue biosynthesis. The first key step in the formation of these alkaloids likely involves carboligation between a C6-C1 molecule (e.g. benzaldehyde, benzoic acid, or benzoyl-CoA) and pyruvate [29] (Fig. 1). Two distantly related ThPD-dependent enzymes isolated from microbes, acetohydroxyacid synthase (AHAS) and pyruvate decarboxylase (PDC) are known to convert benzaldehyde and pyruvate to (R)-phenylacetylcarbinol (R-PAC), an intermediate in the industrial, semi-synthetic production of ephedrine alkaloids. AHAS from diverse species normally catalyzes carboligation of two α-keto acids toward the biosynthesis of branched-chain amino acids. While AHAS from Escherichia coli was found to accept benzaldehyde as an alternative substrate [30] it is not known whether plant AHAS isoforms display this substrate flexibility. In contrast with AHAS, carboligation is only a minor side reaction for PDC; the latter enzyme largely performs pyruvate decarboxylation to form acetaldehyde and CO2. However, mutational analysis of a PDC from Zymomonas mobilis revealed that a single amino acid substitution effectively converted the enzyme from a decarboxylase to a carboligase, thereby creating a variant capable of R-PAC biosynthesis [31]. It is conceivable that PDCs from ephedrine alkaloid-accumulating plants could have acquired carboligation capacity, enabling the production of R-PAC and other C6-C3 compounds. Recently, extracts of E. sinica were shown to convert benzaldehyde to R-PAC, in addition to other products [11]. To identify putative ThDP-dependent enzymes responsible for this activity, functionally characterized Arabidopsis thaliana AHAS [32] and PDC [33] were used to query khat and E. sinica libraries. Candidate sequences ranged in identity from 40% to 83% to either query (Tables 2 and 3) and grouped in one of two clades based on similarity to AHAS or PDC (Fig. 3B). FPKM analysis revealed notable differences in expression levels between different ThDP-dependent carboligase candidates in E. sinica. For example, a 36-fold difference was observed between AHAS homologues EsThDPC1–1 and EsThDPC1–2, and a nearly 10-fold difference was noted between PDS homologues EsThDPC2–1 and EsThDPC2–2 (Fig. 5). Candidates identified in the khat library did not display marked differences in relative expression levels (Fig. 4).

Fig 5. Relative expression of gene candidates identified in Ephedra sinica (ESI-Trinity).

FPKM (fragments mapped per kilobase of exon per million reads mapped) is a normalizing statistic measuring gene expression while accounting for variation in gene length [23]. Abbreviations are defined in Tables 2 and 3.

Evaluation of transaminase candidates

Ephedrine alkaloid biosynthesis occurring through 1-phenylpropane-1,2-dione intermediate would require transamination yielding (S)-cathinone (Fig. 1). The in vitro production of alternative C6-C3 structures such as R-PAC in E. sinica extract [11] raises the question of whether hydroxyketones can themselves undergo transamination, or whether they must first be oxidized to the diketone 1-phenylpropane-1,2-dione. Plant transaminases aren’t known to accept hydroxyketone substrates; in this case, such activity would yield cathine and (1R,2S)-norephedrine without the need for a reductase. In contrast, amino transfer reactions involving hydroxyketone substrates and aminoalcohol products have been reported in microorganisms [34,35]. Phylogenetic analysis revealed a distinct clade comprised of khat and E. sinica transaminase candidates closely related to aromatic amino acid transaminases, such as those from opium poppy and melon (Cucumis melo) (Fig. 3C). Opium poppy tyrosine aminotransferase, which also acts on phenylalanine and tryptophan, contributes to benzylisoquinoline alkaloid biosynthesis [36]. In melon (Cucumis melo), phenylalanine / tyrosine aminotransferase plays a role in the catabolism of amino acids to aroma volatiles [37]. A second distinct clade grouped khat and E. sinica sequences with prokaryotic-type amino transferases from pine (Pinus pinaster) and petunia (Fig. 3C). Prephenate [38] and aspartate [39] transaminases are two prominent members of the class 1b aspartate aminotransferase family, and both enzymes act on non-aromatic substrates. Based solely on the aromatic nature of C6-C3 intermediates, it is conceivable that an enzyme more similar to phenylalanine / tyrosine transaminases is involved in ephedrine alkaloid biosynthesis.

Mining reductase candidates

The involvement of a reductase in amphetamine alkaloid metabolism is supported by the detection of (S)-cathinone reductase activity in both khat [12] and E. sinica [5] extracts. Varieties of E. sinica accumulating predominantly isomers with a (1R,2S) configuration exhibited stereoselective (S)-cathinone reductase activity yielding (1R,2S)-norephedrine. On the other hand, reductase activity in varieties of E. sinica featuring alkaloids with both 1R and 1S configurations yielded mixed isomer products (i.e. both cathine and (1R,2S)-norephedrine) (Fig. 1). These results support the existence of at least two stereospecific reductases, a feature observed in other systems such as menthol biosynthesis in peppermint (Mentha x piperita) [40] and tropane alkaloid biosynthesis in several Solanaceae members, including henbane (Hyoscyamus niger) [41] and Jimson weed (Datura stramonium) [42]. Two different tropinone reductases (TRI and TRII) form one stereoisomeric product each, either tropine for esterified alkaloids or pseudotropine as a precursor to calystegines. Stereospecific reduction of salutaridine and codeine is observed in opium poppy within the context of morphine biosynthesis [43]. A common feature in all these pathways, including ephedrine alkaloid biosynthesis, is the reduction of a keto group, creating a chiral center featuring a hydroxyl moiety. Queries, including TRI, TRII and codeinone reductase (COR), were drawn from a variety of plants to search khat and E. sinica databases for stereospecific reductases, with top hits shown in Tables 2 and 3, respectively. Overall, percent identities were low (<60%) compared with results obtained for genes of benzoic acid metabolism and upstream carboligase and transaminase steps (Tables 2, 3). This result was not surprising. Although evidence suggests that benzoic acid biosynthesis takes place in khat and E. sinica (substrates are expected to be similar, if not identical, to those of query enzymes), diverse alkaloid types have not been reported in these plants. (S)-Cathinone is structurally quite different from tropinone, menthone and morphinan alkaloids. Database mining for reductase candidates (and N-methyltransferase candidates) was expected to yield many more “false leads” compared with more conserved upstream steps. Further, potentially successful leads cannot be evaluated strictly on the basis of sequence identity to queries.

For comparison, sanguinarine reductase (SanR) from California poppy (Eschscholzia californica) was used to query khat and E. sinica libraries. SanR substrates (e.g. sanguinarine, chelerythrine) are charged, planar, highly aromatic molecules lacking a keto group, and reduction takes place through an alkanolamine intermediate yielding a non-chiral center [44]. Interestingly, querying with SanR revealed closer matches (top hits of 64% and 60% respectively, for khat and E. sinica) (Tables 2, 3). However, neither the nature of SanR substrates nor the SanR reaction mechanism supports involvement of homologues in ephedrine metabolism.

Mining N-methyltransferase candidates

Alkaloid metabolism in E. sinica proceeds beyond reduction to include N-methylated products, such as diastereomers (1S,2S)-pseudoephedrine and (1R,2S)-ephedrine (Fig. 1). N,N-Dimethylated versions of these alkaloids are also present in E. sinica [5], suggesting that either a single enzyme performs consecutive N-methylations, or additional enzyme(s) are necessary for subsequent steps. Although S-adenosylmethionine (SAM)-dependent N-methyltransferase (NMT) activity was detected in E. sinica stem extracts [5], N-methylation of ephedrine alkaloids has not been characterized at the molecular level. SAM-dependent N-methylation is common to several plant natural product biosynthetic pathways. For example, N-methylation of benzylisoquinoline alkaloids by enzymes such as opium poppy (S)-tetrahydroprotoberberine N-methyltransferase (TNMT) [45] yields tertiary or quaternary amines. Caffeine biosynthesis in coffee (Coffea arabica) requires the participation of three closely related NMTs [46]. An early step toward the formation of several plant alkaloids, including nicotine, cocaine, calystegines, atropine and scopolamine, involves N-methylation by putrescine N-methyltransferase (PMT) [47]. Queries representing TNMT, PMT and caffeine synthase (CS) were used to search the E. sinica database for homologues, with the top hits shown in Table 3. Sequence identity between queries and hits was generally low, due in part to the evolutionary distance between E. sinica and the angiosperm species from which queries were drawn. Other reasons include those discussed as part of the reductase mining strategy.

We also considered the possibility that E. sinica could have recruited an N-methyltransferase unrelated to previously described enzymes of plant secondary metabolism. It is generally accepted that enzymes of plant specialized biochemistry arise through divergent evolution; for example, through the recruitment of “rogue” enzymes sustaining mutational changes compared with close homologues in primary metabolism [48]. A prominent example of N-methylation in primary metabolism is the mono-, di-, or trimethylation of lysine (Lys) or arginine (Arg) residues of histones and other proteins or peptides [49]. N-Methylation of histone Lys and Arg side chain amino groups forms a crucial part of the histone code. Proteins other than histones undergo similar modifications. For example, the ε-amino group of Lys-14 in the large subunit of Rubisco undergoes N-methylation via Rubisco LSMT (large subunit N-methyltransferase) [50]. Several homologues of protein Lys and Arg NMTs were identified in E. sinica using query sequences from A. thaliana (Table 3). High identity to query sequences could imply a conserved function in primary metabolism, as processes such as histone modification are expected to occur in E. sinica as with all other eukaryotes. Nonetheless, the involvement of an enzyme related to Arg / Lys NMTs in alkaloid metabolism should not be ruled out, as there exists some similarities substrate structure, such as availability of a terminal amino group attached to a flexible side chain. Arg / Lys NMTs are also known to perform successive methylations [49], which may occur in ephedrine alkaloid biosynthesis as well.


Humanity has long been aware of the pharmacological properties of amphetamine-type alkaloids, firmly establishing plants that accumulate them as traditional sources of medicine. Yet, the widespread use of amphetamine analogues in modern medicine has relied almost exclusively on semi-synthetic, or entirely synthetic production of these compounds [1]. The existing fermentation-based commercial preparation of (1S,2S)-pseudoephedrine and (1R,2S)-ephedrine could benefit immensely from the isolation of plant genes involved in the biosynthesis of these substituted amphetamines. The partially mapped ephedrine alkaloid pathway provides a logical basis for the prediction of enzyme types involved in producing these compounds. The establishment of extensive and annotated sequence resources for khat and E. sinica represents a critical step toward the elucidation of the pathway at a molecular level.

Supporting Information

S1 Dataset. Complete sequences of candidate genes identified in khat (CED-trinity) (see Table 2).


S2 Dataset. Complete sequences of candidate genes identified in Ephedra sinica (ESI-Velvet) (see Table 2).


S3 Dataset. Complete sequences of queries used to mine khat (CED-trinity) for candidate genes.


S4 Dataset. Complete sequences of queries used to mine Ephedra sinica (ESI-Velvet) for candidate genes.


S1 Fig. Agilent Bioanalyzer scan results obtained for khat (upper panel) and Ephedra sinica (lower panel) total RNA preparations.

Data reflect overall length and quantity of RNA molecules prior to preparation for Illumina GA sequencing.


S2 Fig. Functional category analysis based on Gene Ontology (GO) annotations of ESI-Velvet unigenes.


S3 Fig. Phylogenetic analysis of gene candidates.

Abbreviations: L-phenylalanine ammonia lyase (PAL) (A), 4-coumaroyl-CoA ligase (4CL) (B), 3-ketoacyl-CoA-thiolase (KAT) (C), cinnamoyl-CoA hydratase-dehydrogenase (CHD) (D), benzoate CoA-ligase (BL) (E), reductase (RED) (F), and N-methyltransferase (NMT) (G). Sequences were aligned and analyzed for phylogenetic relationships using the neighbor-joining algorithm. Numbers at each node represent bootstrap values calculated using 1000 iterations. Accession numbers are found in the Materials and Methods, and abbreviations are defined in Tables 2 and 3.



Sequencing was performed at the McGill University and Génome Québec Innovation Center. KK was co-supervised by Norman P.A. Hüner (Department of Biology, University of Western Ontario, Canada).

Author Contributions

Conceived and designed the experiments: PJF JMH EL FM. Performed the experiments: RAG KK AL. Analyzed the data: RAG KK AL YZ CWS. Contributed reagents/materials/analysis tools: YZ CWS. Wrote the paper: JMH.


  1. 1. Hagel JM, Krizevski R, Marsolais F, Lewinsohn E, Facchini PJ. Biosynthesis of amphetamine analogues in plants. Trends Plant Sci. 2012;17: 404–412. pmid:22502775
  2. 2. Iversen L. Speed, ecstasy, ritalin: The science of amphetamines. Oxford: Oxford University Press; 2006.
  3. 3. Klein A, Beckerleg S, Hailu D. Regulating khat—dilemmas and opportunities for the international drug control system. Intl J Drug Policy. 2009;20: 509–513. pmid:19535239
  4. 4. Balint EE, Falkay G, Balint GA. Khat—a controversial plant. Wien Klin Wochenschr 2009;121: 604–614. pmid:19921126
  5. 5. Krizevski R, Bar E, Shalit O, Sitrit Y, Ben-Shabat S, Lewinsohn E. Composition and stereochemistry of ephedrine alkaloids accumulation in Ephedra sinica Stapf. Phytochemistry. 2010;71: 895–903. pmid:20417943
  6. 6. Liu L, Liu Z. Essentials of Chinese medicine. Berlin: Springer–Verlag; 2009.
  7. 7. Okada T, Mikage M, Sekita S. Molecular characterization of the phenylalanine ammonia lyase from Ephedra sinica. Biol Pharm Bull. 2008;31: 2194–2199. pmid:19043198
  8. 8. Qualley AV, Widhalm JR, Abebesin F, Kish CM, Dudareva N. Completion of the core β-oxidative pathway of benzoic acid biosynthesis in plants. Proc Natl Acad Sci USA. 2012;109: 16383–16388. pmid:22988098
  9. 9. Nierop-Groot MN, de Bont JAM. Involvement of manganese in conversion of phenylalanine to benzaldehyde by lactic acid bacteria. Appl Environ Microbiol. 1999;65: 5590–5593. pmid:10584022
  10. 10. Grue-Sørensen G, Spenser ID. Biosynthetic route to the Ephedra alkaloids. Am Chem Soc. 1994;116: 6195–6200.
  11. 11. Krizevski R, Bar E, Shalit O, Levy A, Hagel JM, Kilpatrick K, et al. Benzaldehyde is a precursor of phenylpropylamino alkaloids as revealed by targeted metabolic profiling and comparative biochemical analyses in Ephedra spp. Phytochemistry. 2012;81: 71–79. pmid:22727117
  12. 12. Krizevski R, Dudai N, Bar E, Lewinsohn E. Developmental patterns of ephedrine alkaloids accumulation in khat (Catha edulis, Forsk., Celastraceae). J Ethnopharmacol. 2007;114: 432–438. pmid:17928181
  13. 13. Krizevski R, Dudai N, Bar E, Dessow I, Ravid U, Lewinsohn E. Quantitative stereoisomeric determination of phenylpropylamino alkaloids in khat (Catha edulis Forsk.). Isr J Plant Sci. 2008;56: 207–213.
  14. 14. Hagel JM, Krizevski R, Kilpatrick K, Sitrit Y, Marsolais F, Lewinsohn E, et al. Expressed sequence tag analysis of khat (Catha edulis) provides a putative molecular biochemical basis for the accumulation of phenylpropylamino alkaloids. Genet Mol Biol. 2011;34: 640–646. pmid:22215969
  15. 15. McGettigan PA. Transcriptomics in the RNA-seq era. Curr Opin Chem Biol. 2013;17: 4–11. pmid:23290152
  16. 16. Grabherr MG, Haas BJ, Yassour M, Levin JZ, Thompson DA, Amit I, et al. Full-length transcriptome assembly from RNA-seq data without a reference genome. Nature Biotechnol. 2011;15: 644–652.
  17. 17. Zerbino DR, Birney E. Velvet: algorithms for de novo short read assembly using de Bruijn graphs. Genome Res. 2008;18: 821–829. pmid:18349386
  18. 18. Xiao M, Zhang Y, Chen X, Lee E- J, Barber CJS, Chakrabarty R, et al. Transcriptome analysis based on next-generation sequencing of non-model plants producing specialized metabolites of biotechnological interest. J Biotechnol. 2013;166: 122–134. pmid:23602801
  19. 19. Gaasterland T, Sensen CW. MAGPIE: automated genome interpretation. Trends Genet. 1996;12: 76–78. pmid:8851977
  20. 20. Pruitt KD, Tatusova T, Maglott DR. NCBI reference sequences (RefSeq): a curated non-redundant sequence database of genomes, transcripts and proteins. Nucleic Acids Res. 2007;35: 61–65.
  21. 21. Larkin MA, Blackshields G, Brown NP, Chenna R, McGettigan PA, McWilliam H, et al. Clustal W and Clustal X version 2.0. Bioinformatics. 2007;23: 2947–2948 pmid:17846036
  22. 22. Drew DP, Dueholm B, Weitzel C, Zhang Y, Sensen C, Simonsen HT. Transcriptome analysis of Thapsia laciniata Rouy provides insights into terpenoid biosynthesis and diversity in Apiaceae. Int J Mol Sci. 2013;14: 9080–9098. pmid:23698765
  23. 23. Garber M, Grabherr MG, Guttman M, Trapnell C. Computational methods for transcriptome annotation and quantification using RNA-seq. Nature Methods. 2011;8: 469–478. pmid:21623353
  24. 24. Schulz MH, Zerbino DR, Vingron M, Birney E. Oases: robust de novo RNA-seq assembly across the dynamic range of expression levels. Bioinformatics. 2012;28: 1086–1092. pmid:22368243
  25. 25. Ibdah M, Chen Y- T, Wilkerson CG, Pichersky, E. An aldehyde oxidase in developing seeds of Arabidopsis converts benzaldehyde to benzoic acid. Plant Physiol. 2009;150: 416–423. pmid:19297586
  26. 26. Long MC, Nagegowda DA, Kaminaga Y, Ho KK, Kish CM, Schnepp J, et al. Involvement of snapdragon benzaldehyde dehydrogenase in benzoic acid biosynthesis. Plant J. 2009;59: 256–265. pmid:19292760
  27. 27. Chaw S- M, Parkinson CL, Cheng Y, Vincent TM, Palmer JD. Seed plant phylogeny inferred from all three plant genomes: monophyly of extant gymnosperms and origin of Gnetales from conifers. Proc Natl Acad Sci USA. 2000;97: 4086–4091. pmid:10760277
  28. 28. Finet C, Timme RE, Delwiche CF, Marlétaz F. Multigene phylogeny of the green lineage reveals the origin and diversification of land plants. Curr Biol. 2010;20: 2217–2222. pmid:21145743
  29. 29. Grue-Sørensen G, Spenser ID. The biosynthesis of ephedrine. Can J Chem. 1989;67: 998–100.
  30. 30. Engel S, Vyazmensky M, Geresh S, Barak Z, Chipman DM. Acetohydroxyacid synthase: A new enzyme for chiral synthesis of R-phenylacetylcarbinol. Biotechnol Bioeng. 2003;83: 833–840. pmid:12889023
  31. 31. Meyer D, Walter L, Kolter G, Pohl M, Müller M, Tittmann K. Conversion of pyruvate decarboxylase into an enantioselective carboligase with biosynthetic potential. J Am Chem Soc. 2011;133: 3609–3616. pmid:21341803
  32. 32. Chang AK, Duggleby RG. Expression, purification and characterization of Arabidopsis thaliana acetohydroxyacid synthase. Biochem J. 1997;327: 161–169. pmid:9355748
  33. 33. Kürsteiner O, Dupuis I, Kuhlemeier C. The pyruvate decarboxylase 1 gene of Arabidopsis is required during anoxia but not other environmental stresses. Plant Physiol. 2003;132: 968–978. pmid:12805625
  34. 34. Ward J, Wohlgemuth R. High-yield biocatalytic amination reactions in organic synthesis. Curr Org Chem. 2010;14: 1914–1927.
  35. 35. Ingram CU, Bommer M, Smith ME, Dalby PA, Ward JM, Hailes HC, et al. One-pot synthesis of amino-alcohols using a de novo transketolase and beta-alanine:pyruvate transaminase pathway in Escherichia coli. Biotechnol Bioengineer. 2007;96: 559–569.
  36. 36. Lee E- J, Facchini PJ. Tyrosine aminotransferase contributes to benzylisoquinoline alkaloid biosynthesis in opium poppy. Plant Physiol. 2011;157: 1067–1078. pmid:21949209
  37. 37. Gonda I, Bar E, Portnoy V, Lev S, Burger J, Schaffer AA, et al. Branched-chain and aromatic amino acid catabolism into aroma volatiles in Cucumis melo L. fruit. J Exp Bot. 2010;61: 1111–1123. pmid:20065117
  38. 38. Maeda H, Yoo H, Dudareva N. Prephenate aminotransferase directs plant phenylalanine biosynthesis via arogenate. Nature Chem Biol. 2011;7: 19–21. pmid:21102469
  39. 39. de la Torre F, de Santis L, Suárez MF, Crespillo R, Cánovas FM. Identification and functional analysis of a prokaryotic-type aspartate aminotransferase: implications for plant amino acid metabolism. Plant J. 2006;46: 414–425. pmid:16623902
  40. 40. Davis EM, Ringer KL, McConkey ME, Croteau R. Monoterpene metabolism. Cloning, expression, and characterization of menthone reductases from peppermint. Plant Physiol. 2005;137: 873–881. pmid:15728344
  41. 41. Nakajima K, Hashimoto T, Yamada Y. Two tropinone reductases with different stereospecificity are short chain dehydrogenase/reductases evolved from a common ancestor. Proc Natl Acad Sci USA. 1993;90: 9591–9595. pmid:8415746
  42. 42. Portsteffen A, Dräger B, Nahrstedt A. The reduction of tropinone in Datura stramonium root cultures by two specific reductases. Phytochemistry. 1994;37: 391–400. pmid:7765621
  43. 43. Hagel JM. Facchini PJ. Benzylisoquinoline alkaloid metabolism: a century of discovery and a brave new world. Plant Cell Physiol. 2013;54: 647–672. pmid:23385146
  44. 44. Vogel M, Lawson M, Sippl W, Conrad U, Roos W. Structure and mechanism of sanguinarine reductase, an enzyme of alkaloid detoxification. J Biol Chem. 2010;285: 18397–18406. pmid:20378534
  45. 45. Liscombe DK, Facchini PJ. Molecular cloning and characterization of tetrahydroprotoberberine cis-N-methyltransferase, an enzyme involved in alkaloid biosynthesis in opium poppy. J Biol Chem. 2007;282: 14741–14751. pmid:17389594
  46. 46. Ashihara H, Sano H, Crozier A. Caffeine and related purine alkaloids: biosynthesis, catabolism, function and genetic engineering. Phytochemistry. 2008;69: 841–856. pmid:18068204
  47. 47. Biastoff S, Brandt W, Dräger B. Putrescine N-methyltransferase—the start for alkaloids. Phytochemistry. 2009;70: 1708–1718. pmid:19651420
  48. 48. Pichersky E, Lewinsohn E. Convergent evolution in plant specialized metabolism. Annu Rev Plant Biol. 2011;62: 1–18. pmid:21314429
  49. 49. Clarke SG. Protein methylation at the surface and buried deep: thinking outside the histone box. Trends Biochem Sci. 2013;38: 243–252. pmid:23490039
  50. 50. Klein RR, Houtz RL. Cloning and developmental expression of pea ribulose-1,5-bisphosphate carboxylase/oxygenase large subunit N-methyltransferase. Plant Mol Biol. 1995;27: 249–261. pmid:7888616