A metabolic gene cluster in Lotus japonicus discloses novel enzyme functions and products in triterpene biosynthesis

Authors


Summary

  • Genes for triterpene biosynthetic pathways exist as metabolic gene clusters in oat and Arabidopsis thaliana plants. We characterized the presence of an analogous gene cluster in the model legume Lotus japonicus.
  • In the genomic regions flanking the oxidosqualene cyclase AMY2 gene, genes for two different classes of cytochrome P450 and a gene predicted to encode a reductase were identified. Functional characterization of the cluster genes was pursued by heterologous expression in Nicotiana benthamiana. The gene expression pattern was studied under different developmental and environmental conditions. The physiological role of the gene cluster in nodulation and plant development was studied in knockdown experiments.
  • A novel triterpene structure, dihydrolupeol, was produced by AMY2. A new plant cytochrome P450, CYP71D353, which catalyses the formation of 20-hydroxybetulinic acid in a sequential three-step oxidation of 20-hydroxylupeol was characterized. The genes within the cluster are highly co-expressed during root and nodule development, in hormone-treated plants and under various environmental stresses. A transcriptional gene silencing mechanism that appears to be involved in the regulation of the cluster genes was also revealed.
  • A tightly co-regulated cluster of functionally related genes is involved in legume triterpene biosynthesis, with a possible role in plant development.

Introduction

Triterpenes are a major subgroup of the terpene superfamily of plant secondary metabolites (Xu et al., 2004). The formation of the skeleton structure of these compounds is catalysed by oxidosqualene cyclases (OSCs; Iturbe-Ormaetxe et al., 2003) through the cyclization of 2,3-oxidosqualene. These OSC ‘signature’ enzymes (Osbourn, 2010) give rise to either tetracyclic sterols, catalysed by cycloartenol synthases (CS, E.C. 5.4.99.8) and lanosterol synthases (LS, E.C. 5.4.99.7), or to triterpenes, catalysed by enzymes such as β-amyrin synthase (β-AS, E.C. 5.4.99.B1), α-amyrin synthase (α-AS, E.C. 5.4.99) and lupeol synthase (LuS, E.C. 5.4.99.B3; Phillips et al., 2006; Ohyama et al., 2009). Subsequent modifications of the basic backbone of these precursor scaffold molecules are carried out by enzymes such as cytochrome P450s, acyltransferases, glycosyltransferases and methyltransferases, collectively known as ‘tailoring’ enzymes (Osbourn, 2010).

A number of genes and enzymes for triterpene synthesis have been characterized from plants (Osbourn, 2010; Augustin et al., 2011; Sawai & Saito, 2011). Interestingly, genes for certain triterpene biosynthetic pathways exist as metabolic gene clusters in oat (Avena strigosa) and Arabidopsis thaliana plants (Papadopoulou et al., 1999; Qi et al., 2004, 2006; Osbourn, 2010; Chu et al., 2011). Over the last few years, metabolic gene clusters for the synthesis of other classes of secondary metabolites have also been identified from diverse plant species. These include: the phytocassane and momilactone diterpenes in rice (Wilderman et al., 2004; Shimura et al., 2007; Swaminathan et al., 2009); cyanogenic glucosides in Lotus japonicus, cassava (Manihot esculenta) and sorghum (Sorghum bicolor) (Takos et al., 2011); and, recently, the alkaloid noscapine in poppy (Winzer et al., 2012). Clearly the occurrence of gene clusters in plants is a recurring phenomenon and, thus, implies some functional significance for this form of genetic organization.

Legumes produce a huge variety of secondary metabolites (Wink & Mohamed, 2003). Among these, the triterpenoid saponins have received considerable interest due to their antinutrient and, simultaneously, health-giving properties that can affect the quality of food and forage legumes (Rochfort & Panozzo, 2007). Similarly, the triterpenoid saponin content of the model legume Medicago truncatula has been studied in great detail (Huhman & Sumner, 2002; Pollier et al., 2011). Simple triterpene skeletons such as β-amyrin have been detected in the roots of legumes during the establishment of rhizobial and mycorrhizal symbioses (Baisted, 1971; Hernandez & Cooke, 1996; Grandmougin-Ferjani et al., 1999; Iturbe-Ormaetxe et al., 2003). Furthermore, we recently demonstrated that the simple triterpene lupeol is involved in negative regulation of early nodulation processes in L. japonicus (Delis et al., 2011). Interestingly, heterologous expression of a β-amyrin synthase from aster in M. truncatula resulted in nodulation enhancement (Confalonieri et al., 2009). As regards the biosynthesis of triterpenes in legumes, co-expression analysis has led to the identification of candidate genes involved in the production and modification of triterpenes in M. truncatula (Naoumkina et al., 2010). Similarly, in L. japonicus, in silico analysis has revealed several predicted OSC genes (Sawai et al., 2006). Amongst these, AMY1/OSC1 and AMY2/OSC8 have previously been characterized by heterologous expression in yeast; AMY1 was shown to be a β-amyrin synthase while AMY2 was reported to be a mixed-function OSC capable of synthesizing both β-amyrin and lupeol (Iturbe-Ormaetxe et al., 2003).

In this study, the genomic regions encompassing members of the OSC gene family in the model legume L. japonicus were investigated for candidate gene clusters for triterpene biosynthesis. This led us to identify two loci in which copies of the previously characterized AMY2 gene are flanked by genes encoding candidate tailoring enzymes. We then focused our investigations on one of these candidate clusters in L. japonicus, which contained the AMY2 gene along with genes for two different classes of cytochrome P450 (belonging to the CYP88D and CYP71D families) and a gene predicted to encode a reductase. The expression pattern of the genes within the cluster was studied under different developmental and environmental conditions. Using heterologous in planta approaches we then showed that this gene cluster probably represents a new pathway for the biosynthesis of a novel triterpene structure, dihydrolupeol, and its subsequent conversion to 20-hydroxybetulinic acid, catalyzed by the sequential activity of AMY2 and CYP71D353. Knockdown experiments suggested a role for this pathway in plant development. Finally, we unexpectedly identified a transcriptional gene silencing mechanism that appears to be involved in the regulation of the cluster genes.

Materials and Methods

Plant material and growth conditions

Lotus japonicus (Regel) K.Larsen (cultivar Gifu B-129 or MG20) plants uninoculated and inoculated with Mesorhizobium loti (strain R7A) were grown as described by Delis et al. (2011).

In situ hybridization

Sections (7 μm) of nodules from plants 14 and 28 d post-infection were hybridized with antisense and sense RNA probes, labeled with digoxigenin (DIG)-11-rUTP (Roche, Mannheim, Germany) according to Delis et al. (2006). Pairs of gene-specific primers were designed, AMY2isF/AMY2isR, LjCYP88D5isF/LjCYP88D5R, LjCYP71D353isF/LjCYP71D353isR (Supporting Information Table S1) and used for the in vitro transcription of RNA probes.

Real-time PCR experiments

Real-time PCR experiments were conducted as previously described (Delis et al., 2011). Relative transcript levels in different samples for the gene of interest were calculated as a ratio to the ubiquitin (UBQ) gene transcripts. Data were analyzed according to Pfaffl (2001) and the reaction efficiencies were estimated with LinRegPCR (Ramakers et al., 2003). For all samples a triplicate of PCR reaction was performed for each gene.

Roots, nodules and leaves at different developmental stages of plants grown as described were collected and ground in liquid nitrogen. Gene-specific primers were designed with Beacon designer v7.01 software (Premier Biosoft, Palo Alto, CA, USA) for the AMY2, LjCYP88D5 and LjCYP71D353 genes AMY2rtF, AMY2rtR, LjCYP88D5rtF, LjCYP88D5R, LjCYP71D353rtF, LjCYP71D353rtR, respectively (Table S1). For the different developmental stages and tissues, total RNA was isolated from organs of 50 plants for each sample using RNeasy extraction Kit (Qiagen). The experiment was repeated once.

Exogenous hormone and abiotic stress treatments

Seeds of L. japonicus (cultivar Gifu B-129) were pregerminated and seedlings were grown for 7 d on Petri dishes containing MS with 1% sucrose substrate at 22°C in a 16 h : 8 h, dark : light photoperiod. Then the seedlings were transferred on Petri dishes supplemented with 10 or 25 μM methyl jasmonate (MeJA; Duchefa, Haarlem, the Netherlands), 0.6 mg l−1 benzylaminopurine (BA), 0.15 mg l−1 2,4-dichlorophenoxyacetic acid (2,4-D) or 150 nM paraquat (all from Sigma) for 7 d. Control plants were grown on Petri dishes containing MS supplemented with the respective amount of ethanol for the MeJA and 2,4-D treatments. For salt stress treatments, plants were grown in MS with 1% sucrose supplemented with 25, 50 or 75 mM NaCl for 7 d. For the heat/cold stress, seedlings were grown for 11 d on Petri dishes containing MS substrate at 22°C in a 16 h : 8 h, dark : light photoperiod. Then the seedlings were exposed to 37°C for heat stress and to 4°C for cold stress for 24 h. Roots from 20 to 50 seedlings per treatment were collected and subjected to total RNA isolation using the QIAGEN Rneasy Mini Kit (Qiagen) and subjected to real-time PCR as described above. The experiments were repeated three times.

Heterologous expression into Nicotiana benthamiana leaves

Plasmids pBinPS1NT, pBinPS2NT and pBinPS2NT2AGFP, containing full-length copies of CPMV RNA-1 (35S-RNA-1), RNA-2 (35S-RNA-2) and RNA-2-GFP (35S-RNA-2-GFP), respectively, in the binary transformation vector pBINPLUS (van Engelen et al., 1995), have been described previously (Liu & Lomonossoff, 2002). The creation of deleted versions of RNA-2, based on the vector pN81S2NT containing the complete sequence of RNA-2, have been described before (Canizares et al., 2006). Full-length AMY2, LjCYP88D5 and LjCYP71D353 genes were obtained by appropriate digests and by using the two pairs of specific primers LjCYP88D5Fl-F/LjCYP88D5Fl-R and LjCYP71D353Fl-F/LjCYP71D353Fl-R, respectively, (Table S1) and cloned into pM81-FSC1. The pM81-FSC1 derived plasmids were digested with AscI/PacI and the fragments were transferred to vector pBINPLUS (van Engelen et al., 1995). The derivative plasmids were maintained in Agrobacterium tumefaciens strain LBA4404 and agroinfiltration into N. benthamiana leaves was carried out as previously described (Canizares et al., 2006). Cultures were co-infiltrated with an Agrobacterium culture carrying the pBIN61-P19 plasmid which encodes for the P19 silencing suppressor protein (Voinnet et al., 2003). For co-infiltration experiments the used cultures were mixed to an equal density. Leaf tissue was harvested and frozen in liquid nitrogen after 6 d.

Agrobacterium rhizogenes plant transformation

A polyubiquitin promoter-based binary vector, pUBI-GWS-GFP, which allows for GFP overproduction thus facilitating the detection of transgenic roots generated via the infection of A. rhizogenes (Maekawa et al., 2008), was used for silencing of the AMY2 and LjCYP88D5 genes. PCR amplicons were produced using cDNA from 14-d-old roots as template and pairs of specific primers AMY2-3F, AMY2-3R, AMY2-2F, AMY2-2R, LjCY-P88D5-1F LjCYP88D5-1R, LjCYP88D5-3F, LjCYP88D5-3R (Table S1). AMY2 and LjCYP88D5 ORF sections named AMY2-2, AMY2-3, LjCYP88D5-1 and LjCYP88D5-3 were cloned into the KpnI – XhoI restriction sites of pENTR4 (Invitrogen) replacing the ccdB gene of the original vector. The four new clones, pENTRY-AMY2.2/2.3 and pENTRY-LjCYP88D5-1/3 were then used in an LR Clonase reaction (Invitrogen) with destination vector pUBI-GWS-GFP in order to create the final expression vectors that were used in a plant binary transformation system. Hairy root transformation of L. japonicus (cultivar MG20) utilizing A. rhizogenes strain LBA 1334 and the binary vectors was performed according to Martirani et al. (1999). Control plant lines were obtained following the same procedure and A. rhizogenes transformed with empty pUBI-GWS-GFP vector. Following transformation, wild-type roots were removed and only one transgenic root was allowed to grow further to produce a new transformed root system per plant. The transformation procedure was repeated twice. Root tissue from all transformed plants, silenced for the two genes (AMY2 and LjCYP88D5), was retained for RNA extraction and the plants, together with fourteen control plants, were allowed to grow further for another 20 d as described before (Delis et al., 2011). Real-time PCR experiments were conducted as described above. Nodule numbers were counted at 20 and 40 d in all transformed plants in comparison to control plants.

Agrobacterium tumefaciens stable transformation

PCR amplicon was produced using cDNA from 14-d-old roots as template and a pair of specific primers AMY2-2F and AMY2-2R (Table S1). The amplification product was digested with XbaI/HindIII and XhoI/KpnI in order to obtain the antisense and sense direction, respectively, and ligated into pHannibal plasmid vector. The silencing construct was subcloned into pCambia 1300. The resultant binary vector plasmid was transferred into A. tumefaciens strain AGL1 by electroporation. Plants were transformed following procedures previously described by Lombari et al. (2003) and Barbulova & Chiurazzi (2005). Control plants were transformed with the empty vector. Five weeks after planting, roots were carefully washed and a small segment of plant roots was excised. Genomic DNA was isolated using the CTAB procedure. A PCR reaction was carried out, using a pair of primers, which amplify the 35S promoter and the hygromycin gene, 35S-F, 35S-R, Hyg-F and Hyg-R (Table S1), respectively. Total RNA was isolated from the four identified transformed plants using RNeasy extraction Kit (Qiagen). Real-time PCR experiments were conducted as described above.

DNA methylation assays-Bisulfite sequencing

Genomic DNA samples from wild-type and mutant roots were extracted using the Nucleospin Plant II kit (Macherey-Nagel, Düren, Germany). Sodium bisulfite treatment of the DNA was conducted using the EpiTect Bisulfite kit (Qiagen), following manufacturer's instructions. The target regions (all in coding sequences, 280–350 bp long) were amplified using the primer pairs described in Table S1. Amplified fragments were TA cloned using the pGEM-T-easy Vector System (Promega). Ten clones were sequenced for each amplicon to determine levels of methylation (percentage of all methylated deoxycytidine 5mdC in relation to the total deoxycytidine content in all 10 clones in mutant AMY2 stable transformants).

Metabolite extraction and GC-MS analysis

Nicotiania benthamiana leaf material was harvested, ground in liquid nitrogen, and lyophilized. The dry plant tissue (100–250 mg) was saponified in 10% KOH (w/v) in 80% EtOH (v/v) with 0.5 mg ml−1 butylated hydroxytoluene (Sigma-Aldrich) at 65°C for 2 h and extracted with hexane as previously described (Field & Osbourn, 2008). After hexane extraction, HCl was added to the aqueous solution to lower the pH c. 2.0 and another round of hexane extractions were performed to obtain an acid extraction fraction. The alkaline and acid hexane extracts were concentrated and derivatised with Tri-Sil Z (Pierce, Cambridge, UK) before GC-MS analysis. GC-MS analyses were conducted on an Agilent 5973 MSD (Agilent, Stockport, UK) coupled to an Agilent 6890 Gas Chromatograph. The GC was fitted with an Agilent DB-17 column (30 m × 0.25 mm internal diameter, 0.15 μm film). The injector port, source and transfer line temperatures were set at 230°C and an oven temperature program from 180°C (2.0 min) to 320°C (3.0 min) at 8°C min−1 was used. The flow rate of the helium carrier gas was set to a constant flow of 0.6 ml min−1 and mass spectral data were acquired for the duration of the GC program from m/z 50–800. Raw GC-MS data were analyzed with the AMDIS software package (http://chemdata.nist.gov/mass-spc/amdis/).

Phylogenetic analysis

The alignment and phylogenetic analysis of the cytochrome P450s and OSCs protein sequences were performed with the MEGA v5.05 software package (Tamura et al., 2011). The multiple alignment parameters were adjusted with gap cost 10 and gap extension 1. The phylogenetic trees were constructed using the neighbor-joining and maximum likelihood algorithm with bootstrap analysis of 1000 replicates.

The synteny of the specialized metabolic genes clusters of L. japonicus and A. thaliana was examined with CoGe (Lyons & Freeling, 2008) http://genomevolution.org/CoGe/index.pl). For OSCs protein sequences, tblastn analysis was performed against A. thaliana (ncbi unmasked v1), and L. japonicus (pseudomolecules v2.5) genome databases with expectation value 0.001. Selected nucleotide sequences were analyzed with Gevo algorithm for syntenic genomic regions.

Statistical analysis

All experiments were conducted at least twice and analyzed by ANOVA followed by Duncan multiple comparison tests (< 0.05). Standard errors were calculated for all mean values and t-tests were performed for pairwise comparisons of means at different time points ( 0.01).

Results

Analysis of the genomic regions encompassing OSC genes in legumes

We carried out in silico analysis using the public genome sequence databases for L. japonicus to investigate the genomic regions encompassing all predicted OSC genes present in the genomes of the model plant species (Sawai et al., 2006). Previously biochemically characterized and predicted OSC genes from L. japonicus (Iturbe-Ormaetxe et al., 2003; Sawai et al., 2006; Sato et al., 2008), M. truncatula (Naoumkina et al., 2010), A. thaliana and oat (Qi et al., 2004; Field & Osbourn, 2008; Field et al., 2011) were used as query sequences against the L. japonicus genome databases (all clones and contigs, L. japonicus genome assembly 1.0, http://www.kazusa.or.jp/lotus/). Sequences with expectation values of < 1 × e−10 were excluded from further analysis. A region of c. 300 kb flanking each side of the OSC genes was screened and analysed using FGENESH gene prediction software (http://linux1.softberry.com/berry.phtml?topic=fgenesh&group=programs&subgroup=gfind) and GeneScanwebserver (Burge & Karlin, 1998). For all predicted ORFs and translated amino acid sequences (Expasy translation tool http://web.expasy.org/translate/) blast searches were run against L. japonicus EST databases (http://compbio.dfci.harvard.edu/tgi/cgi-bin/tgi/gimain.pl?gudb=l_japonicus) and Expasy connected protein databases of UniProtKB/Swiss-Prot (http://web.expasy.org/blast/). Confirmed amino acid sequences were aligned to biochemically characterized proteins, and thus genes with potential roles in secondary metabolism were identified (Table 1).

Table 1. Predicted polypeptides encoded by the genes present in the two cluster identified in Lotus japonicus genome and similarities with characterized proteins and enzymes in legumes and other plants
Cluster OSC CYP Oxidoreductase
L. japonicus AMY2 clusterAMY2 (Iturbe-Ormaetxe et al., 2003)

LjCYP88D5 (72% similarity to GuCYP88D6; Seki et al., 2008)

LjCYP71D353 (50% similarity to GmCYP71D9; Latunde-Dada et al., 2001)

LjSDRt (82% similarity to HCF173; Schult et al., 2007)
L. japonicus AMY2-like clusterAMY2 interrupted (Iturbe-Ormaetxe et al., 2003)LjCYP88D4 (71% similarity to GuCYP88D6; Seki et al., 2008; 81% similarity to LjCYP88D5, this study) 

Two genomic regions were identified, in which genes potentially implicated in triterpenoid biosynthesis are assembled in cluster formation, located on chromosome 3 of L. japonicus (Fig. 1a). The first gene cluster consists of genes encoding the previously characterized AMY2 enzyme (Iturbe-Ormaetxe et al., 2003) and two cytochrome P450 enzymes, LjCYP88D5 and LjCYP71D353 (Fig. 1b). The two cytochrome P450 genes are adjacent to each other, 60 kb from AMY2, and probably share a 2 kb common regulatory region. A fourth gene (LjSDRt), is also located in this region, in between AMY2 and the two cytochrome P450 genes. LjSDRt is predicted to encode an enzyme with extensive similarity (82%) to an A. thaliana short chain dehydrogenase/reductase (SDR)-like protein, HCF173, (Q9FWQ6, At1G16720) (Fig. 1b). The second putative cluster in L. japonicus is located c. 200 kb from the first and consists of just two genes: one that is highly similar to AMY2 (99% similarity for the deduced amino acid sequences) and a cytochrome P450 gene LjCYP88D4 that is very similar to LjCYP88D5 (81% similarity for the coding amino acid sequences; Fig. 1b). Interestingly, this second AMY2-like gene copy appears to be interrupted at amino acid residue 228 and has an insertion of c. 5 kb of a gene coding for a predicted U-box protein (FGENESH). The distance between the AMY2-like gene and LjCYP88D4 is c. 70 kb. No other genes with predicted functions in secondary metabolism were identified in this region.

Figure 1.

Structure of genomic regions encompassing oxidosqualene cyclases (OSCs) flanked by genes putatively involved in triterpene metabolism in legumes. (a) Map of candidate gene clusters on chromosome 3 of Lotus japonicus, analysing LjT11L01 and the continuous LjT138B03-LjB16L08 genomic clones and (b) organization of genes in AMY2 flanking genomic regions.

The cytochrome P450 genes identified in the two clusters were subjected to phylogenetic analysis by comparison with cytochrome P450 genes that have been shown to be involved in triterpene biosynthesis, and/or have been reported to be co-expressed with OSC genes in transcriptome analysis of plants with known genome sequence (Qi et al., 2006; Ehlting et al., 2008; Field & Osbourn, 2008; Field et al., 2011; Figs 2a,b, S1). LjCYP88D4/5 belong to the Fabaceae-specific CYP88D subfamily (CYP85 clan), which has been implicated in triterpene biosynthesis, and exhibit 71% and 72% similarity, respectively, with GuCYP88D6, a licorice β-amyrin 11-oxidase (Nelson & Werck-Reichhart, 2011; Seki et al., 2011). LjCYP71D353 enzyme belongs to the CYP71 clan of P450s and is phylogenetically close to a CYP71A16 from A. thaliana, which is a member of the gene cluster for the marneral pathway (Field et al., 2011). A. thaliana CYP71A16 is a marneral oxidase.

Figure 2.

Phylogenetic trees of oxidosqualene cyclases (OSCs) (a), cytochrome P450s (b) and SDR-like proteins (c), constructed by the neighbour-joining method with 1000 bootstrap replicates. The scale bar indicates the number of amino acid substitutions per site. Cytochrome P450s from Arabidopsis thaliana, Medicago sativa, Lotus japonicus, Glycyrrhiza uralensis and Avena sativa adjacent to/highly coexpressed with OSCs or previously found to participate in triterpene biosynthesis were used for the phylogenetic analysis. SDR-like proteins from legumes L. japonicus, M. sativa, Glycine max, from A. thaliana, A. lyrata, Theobroma cacao as well as the green algae Micromonas pusilla are shown in (c). The open and black stars, open circles, open and black triangles indicate the cytochrome P450s clustered together with the BARS1, AMY2, THAS1, AsBAS1 and MRN1, respectively. The black box indicates the legume-specific cytochrome P450 subfamily.

Phylogenetic analysis was also performed for the predicted SDR-like LjSDRt (Figs 2c, S1). BLAST analyses (Altschul et al., 1994) indicated several LjSDRt/HCF173 homologs in many plants, including members of the green algae, although none of the homolog proteins in other plants has been characterized as yet.

The AMY2 cluster genes are co-expressed in roots and nodules of L. japonicus and in response to hormone and abiotic stress treatments

Previously we and others have shown that AMY2 is highly expressed in the root and nodules of L. japonicus plants that have been inoculated with the symbiotic bacterium M. loti (Iturbe-Ormaetxe et al., 2003; Sawai et al., 2006). We therefore investigated whether the genes that are clustered with AMY2 are similarly expressed in leaves and roots of L. japonicus plants and in M. loti-inoculated roots and nodules at various developmental stages (Fig. 3). These experiments revealed that the AMY2, LjCYP88D5 and LjCYP71D353 genes show similar expression patterns in all tissues and developmental stages examined (Figs 3, S2), indicating that the three genes are co-expressed. Thus, gene expression is higher in the roots of 7-d-old seedlings than in the roots of 14- and 28-d-old seedlings (Fig. 3a,c,e), while transcript levels were also detectable in the leaves at all three time points but at much lower levels than in roots. Transcript levels were highest in inoculated with M. loti roots 7 d post infection (dpi) (which include young nodules) and in nodules at 14 dpi and then decreased in mature nodules at 28 dpi (Fig. 3b,d,f). Accordingly, co-localization of AMY2, LjCYP88D5 and LjCYP71D353 gene transcripts was detected by in situ hybridization in the vascular bundles, parenchymatic cells (inner parenchyma) and also in uninoculated cells of the central tissue (Fig. S3). Differences in transcript levels for LjSDRt were less marked in the different tissues and developmental stages compared and there is no common expression pattern with the three other cluster genes (Fig. S4).

Figure 3.

AMY2 cluster gene expression in different developmental stages. Gene transcript levels of AMY2, LjCYP88D5 and LjCYP71D353, constituting the AMY2 gene cluster, were detected in both uninoculated (a,c,e) and inoculated with Mesorhizobium loti (b,d,f) Lotus japonicus roots, leaves and nodules. Uninoculated and inoculated plants are of the same age at the stages of 7 d old – 7 dpi (d post infection), 14 dpi and 28 dpi, respectively, but are analysed in different real-time PCR reactions. Total RNA was reverse transcribed, the concentration was normalized between samples and then real-time PCR was performed. Relative gene expression was measured with respect to UBQ transcripts. Mean values ± SD are shown (= 3).

We then investigated the effect of hormone treatment on the expression of the three cluster genes. In plants treated with three different types of hormones, that is 2,4-dichlorophenoxyacetic acid (2,4-D), benzylaminopurine (BA) and methyl jasmonic acid (MeJA), the transcript levels for all three genes were significantly increased compared to control mock-treated plants (Fig. 4a–c). We extended our investigations to abiotic stress treatments that included heat, cold, oxidative and salt stress. All three cluster genes were co-ordinately downregulated in response to cold, heat and oxidative stress (Fig. 4d–f), while no significant change was observed in the transcript levels of any of the genes in response to salt stress (Fig. S5). All together these data suggest that the AMY2, LjCYP88D5 and LjCYP71D353 genes comprise a tightly co-regulated cluster of functionally related genes.

Figure 4.

AMY2 cluster gene expression in Lotus japonicus root tissues subjected to various exogenously applied plant hormones and environmental cues. Gene transcript levels of AMY2, LjCYP88D5 and LjCYP71D353 in root tissues (14 d old) treated with various concentrations of 2,4D, BA and MeJA (a–c) and in root tissues treated with heat and cold stress (12 d old, treated 24 h) and paraquat (14 d old) (d–f). Total RNA from roots (20–50 seedlings per treatment) was reverse transcribed, the concentration was normalized between samples and then real-time PCR was performed. Relative gene expression was measured with respect to UBQ transcripts. Data from a single representative experiment are presented; three experimental repeats yielded similar results. Statistical comparisons were performed by Duncan tests (< 0.05). Indicator letters in common denote lack of significant difference. Bars indicate + SEM (= 3).

Functional analysis in Nicotiana benthamiana

By analogy with previously reported plant metabolic gene clusters, we hypothesized that the two cytochrome P450 enzymes that are encoded by the AMY2 cluster genes may act on the product of the signature enzyme AMY2 and therefore participate in the synthesis of triterpene secondary metabolites in L. japonicus. The full-length cDNAs of AMY2, LjCYP88D5 and LjCYP71D353 were cloned into the pBinP-NS-ER-GFP vector for transient expression in N. benthamiana leaves infiltrated with A. tumefaciens (Canizares et al., 2006). Following expression, metabolites were extracted and analysed by GC/MS. All candidate genes were introduced on their own and also in combination with each other in co-expression experiments.

Previously, AMY2 was assigned a multi-functional oxidosqualene cyclase function, because it produces both β-amyrin and lupeol when expressed in yeast (Iturbe-Ormaetxe et al., 2003). In N. benthamiana leaves, AMY2 produced β-amyrin (peak 2, Fig. 5) as expected and the mass spectrum of peak 2 matched that of the trimethylsilyl (TMS) ether derivative of a β-amyrin standard (Fig. 5b). Surprisingly, lupeol was not detected. Instead, a less polar compound was detected in leaf extracts of AMY2-expressing plants (peak 1, Fig. 5). Analysis of the mass spectrum of the trimethylsilyl (TMS) ether derivative of the less polar compound yielded a parent ion of m/z 500 (Fig. 5b), a number of fragment ion peaks characteristic for C-3 hydroxy lupanes (e.g. m/z 279, 207, 220), and two significant fragment ion peaks at m/z 457 and m/z 191, that suggested a saturated lupane structure. The fragment ion peak at m/z 457 is known to result from the loss of a propyl group, following a fragmentation favored only in saturated lupanes (Budzikiewicz et al., 1964); similarly, the fragment ion peak at m/z 191 corresponds to a fragment having rings D and E of a saturated lupane structure (Budzikiewicz et al., 1964; Dantanarayana et al., 1981). On the basis of these results, the less polar component (peak 1, Fig. 5) detected in the leaf extracts of LjAMY2-expressing plants is proposed to be dihydrolupeol.

Figure 5.

GC-MS analysis of saponified Nicotiana benthamiana leaf extracts after transient expression of AMY2, LjCYP71D353 and/or LjCYP88D5. (a) Total ion chromatograms (TIC) of derivatised samples from basic and acidic extracts after saponification of plant material are shown. AMY2 protein expression results in accumulation of dihydrolupeol (peak 1) and β-amyrin (peak 2). Co-expression of AMY2 and LjCYP71D353 leads to accumulation of 20-hydroxy-lupeol (peak 3; RT = 23.7 min in the basic extractions and peak 4, RT = 23.59 in the acid extractions) and 20-betulinic acid (peak 5, RT = 24.22 min, acid extraction). No activity for LjCYP88D5 was detectable. Other major peaks are plant sterols. Each column of chromatograms has the same scale (ion count, indicated in the top left corner). (b) Mass spectra of peaks 1–5 from the GC profiles shown in (a).

No new metabolites were observed when LjCYP88D5 or LjCYP71D353 were expressed alone in N. benthamiana. However, simultaneous expression of AMY2 and LjCYP71D353 did result in the production of two novel metabolites (peaks 3–5, Fig. 5a). We used basic and acid procedures to ensure extraction of a wide range of possible metabolites. Peaks 3 and 4 correspond to the same product, the TMS-derivative of 3,20-lupandiol (commonly referred to as 20-hydroxy-lupeol); both peaks showed the same fragmentation pattern (Fig. 5b), with a prominent parent ion peak at m/z 588 and characteristic fragment ion peaks at m/z 573 (M+–CH3), 498 (M+–TMS–H2O), and 408 (M+–2TMS–2H2O; Cole et al., 1991; Ulubelen et al., 1994). Peak 5, which was only detected in the acidic fraction, was identified as 3,20-dihydroxy-28-lupanoic acid (commonly referred to as 20-hydroxy-betulinic acid) after a detailed analysis of the mass spectrum of its TMS derivative, which showed fragment ion peaks at m/z 619, 513, and 408; these fragment ion peaks can be explained by the loss of a TMS protecting group following a McLafferty-type rearrangement from a protonated parent ion peak, the combined loss of a second TMS protecting group, a molecule of water and a methyl group, and by a fragment having the fully substituted rings D and E, respectively (Fig. S6; Budzikiewicz et al., 1964; Tsichritzis & Jakupovic, 1990). 20-hydroxylupeol and 20-hydroxybetulinic acid were detected only when LjCYP71D353 was expressed together with AMY2. This indicates that LjCYP71D353 catalyses the formation of 20-hydroxylupeol from dihydrolupeol in a single oxidation reaction. In addition, LjCYP71D353 catalyses the formation of 20-hydroxybetulinic acid in a three-step sequential oxidation at the C-28 position of 20-hydroxylupeol (Fig. 6). To further verify that LjCYP71D353 acts on dihydrolupeol and not on the β-amyrin produced by AMY2, we co-expressed LjCYP71D353 together with the oat β-amyrin synthase gene AsbAS1, which produces only β-amyrin. No new products were detected. Similarly, β-amyrin was not recognized as a substrate of LjCYP88D5 (Fig. S7). No further products were detected when AMY2, LjCYP71D353 and LjCYP88D5 were co-expressed together.

Figure 6.

Proposed enzymatic reactions catalysed by AMY2 and LjCYP71D353. The structures of compounds and possible biosynthetic intermediates produced by the Lotus japonicus AMY2 cluster are shown. AMY2 catalyzes the cyclization of 2,3-oxidosquealene to β-amyrin and dihydro-lupeol. LjCYP71D353 catalyzes the reaction to 20-hydroxy-lupeol and the formation of 20-hydroxybetulinic acid in a sequential three-step oxidation at C-28 of 20-hydroxylupeol.

Physiological significance of the AMY2 secondary metabolic gene cluster

Previously we have shown by gene silencing that lupeol has a role in suppression of nodule formation in L. japonicus (Delis et al., 2011). Other workers have reported that heterologous expression of a β-amyrin synthase from aster in M. truncatula resulted in enhanced nodulation (Confalonieri et al., 2009). Thus, simple triterpenes have different and opposing effects on nodule development. We employed a hairpin RNA gene-silencing strategy to investigate a putative role of the cluster genes in nodulation. Two silencing constructs were made for the AMY2 gene and transgenic roots were generated using an A. rhizogenes transformation protocol. Significantly decreased levels of AMY2 transcript were detected in almost all of these transgenic plants (three plants exhibiting different degree of silencing are shown in Fig. 7a). No obvious effects were observed with regard to the nodulation process (i.e. nodule number at 20 and 40 dpi). Interestingly, at 40 dpi a more severe retardation of the rate of hairy-root growth was recorded than at 20 dpi. We next produced, by using two different hairpin constructs, transgenic roots with reduced levels of the LjCYP88D5 expression (Fig. 7e). We found that silencing of LjCYP88D5 caused no obvious effects on plant development or nodulation.

Figure 7.

AMY2 cluster gene expression in Lotus japonicus hairy-root tissues. Plants, silenced either for AMY2 (a) or LjCYP88D5 (e) were generated by Agrobacterium rhizogenes mediated transformation. Expression levels of LjCYP88D5 (b), LjCYP71D353 (c) and LjSDRt (d) are downregulated in AMY2 silenced roots. Similarly, expression levels of AMY2 (f), LjCYP71D353 (g) and LjSDRt (h) are downregulated in LjCYP88D5 silenced roots. Three representative plants for each hairpin construct, with varying levels of silencing, all presenting statistically significant gene expression when compared to control plant lines (t-test, < 0.01), are shown. Error bars represent ± SE of means of technical repeats (= 3).

In order to further examine the physiological role of the AMY2 gene cluster, stably transformed lines silenced for AMY2 were also obtained via A. tumefaciens transformation. Four transformed plant lines with reduced levels of AMY2 transcript were obtained (Fig. 8a). These plants did not flower (and so we were unable to obtain seed) and consistently exhibited a short, stunted root phenotype, indicating that silencing of AMY2 in stably transformed plants has clear effects on growth and development.

Figure 8.

AMY2 cluster gene expression in Lotus japonicus AMY2 silenced stable transformants. Four plant lines (amy2-1 to 2-4) silenced for AMY2 (a) were generated by Agrobacterium tumefaciens mediated transformation. Expression levels of LjCYP88D5 (b), LjCYP71D353 (c) and LjSDRt (d) are variably down-regulated in AMY2 silenced roots. Statistical comparisons within plant lines were performed by Duncan tests (< 0.05). Indicator letters in common denote lack of significant difference. Bars indicate SEM (= 3).

AMY2 cluster genes are epigenetically regulated

Based on our results indicating strong co-ordinated regulation of the expression profile of the cluster genes, we investigated their expression profile in the silenced plant lines. Strikingly, we observed that the co-ordinate regulation of the cluster genes was manifested in a very unusual fashion. Thus, a decrease in AMY2 transcription levels due to A. rhizogenes-mediated gene silencing, in all plant lines examined, was accompanied by significant reductions in the transcript levels of other cluster genes, namely LjCYP88D5, LjCYP71D353 and LjSDRt (Fig. 7b, c, and d, respectively). Similarly, plant lines silenced for LjCYP88D5 had reduced transcript levels for AMY2 (Fig. 7f), LjCYP71D353 (Fig. 7g) and LjSDRt (Fig. 7h). These experiments were repeated for both genes over a period of > 2 yr and we always observed such a ‘spreading’ of transcription silencing in the cluster. Furthermore, we also observed the cis spreading of transcript level repression in AMY2 silenced stably transformed plants (Fig. 8b–d). The transcript levels of LjSDRt and LjCYP71D353 (but not of LjCYP88D5 in this case) were both significantly reduced in lines amy2-3 and amy2-4, while those of LjSDRt were also reduced in lines amy2-1 and amy2-2. To further investigate this phenomenon we used bisulfite sequencing to determine the DNA methylation levels of LjSDRt and LjCYP71D353 genes in the wild-type and in two AMY2 silenced plants, amy2-3 and amy2-4. We detected a significant increase in the degree of DNA methylation in both of the genes with reduced expression levels in the silenced plant lines compared to the methylation level of the genes in wild type plants (Table 2). Moreover, in the plant line amy2-4, in which the transcript levels of LjSDRt are not significantly reduced, the degree of DNA methylation was the same as in wild-type plants. Our cautious interpretation of these results is that the hairpin-derived siRNAs introduced by two different experimental approaches (i.e. hairy roots and stable transformation) induce RNA-directed DNA methylation (RdDM; Wassenegger et al., 1994; Dalakouras & Wassenegger, 2013), thus promoting transcriptional gene silencing (TGS). The spreading of RdDM into the adjacent regions by TGS transitivity would silence the adjacent genes.

Table 2. Degree of DNA methylation of LjSDRt and LjCYP71D353 in wild-type and silenced plant lines
Plant genotype% DNA methylation
Genes
LjSDRt LjCYP71D353
  1. a

    Statistical significant increase in DNA methylation degree between the silenced and wild-type genotypes (t-test, < 0.01).

Wild-type 14878
Wild-type 256.680.7
AMY2-379a100a
AMY2-440100a

Discussion

Triterpene gene clusters in legumes

In this paper we have mined the genome sequences of the model legume L. japonicus for triterpene biosynthesis gene clusters. We functionally validated the AMY2 gene cluster by the characterization and analysis of three of the genes comprising the cluster. A second L. japonicus cluster consists of a corrupted AMY2-like gene, flanked by a single cytochrome P450 gene (LjCYP88D4). Due to the corrupted AMY2 gene and the absence of a LjCYP71D353 homolog, we propose that this second cluster is either incomplete or in decay. Furthermore, the low degree of conservation in genome structure between the two regions suggests that the ancestral AMY2 was initially adjacent to a cytochrome P450 gene belonging to the CYP88D subfamily and that this region then underwent a tandem duplication followed by genome reorganization. This hypothesis is further supported by the presence of several transposable elements in these genomic regions. LjCYP71D353 is phylogenetically related to CYP71A16 that forms part of the marneral gene cluster in A. thaliana (Field et al., 2011). However, the low degree of synteny (Fig. S8) and the phylogenetic distance between the OSC genes in these two clusters (MRN1, in the marneral cluster, and AMY2 in the L. japonicus cluster) suggests that LjCYP71D353 and AMY2 are not derived from the tandem duplication of a common ancestral pair of OSC and cytochrome P450 genes and do not share a common origin with the marneral cluster. Thus, the gene clusters identified in L. japonicus further support the theory that specialized metabolic gene clusters in plants are likely to have arisen de novo within recent evolutionary history (Chu et al., 2011).

Co-ordinate regulation of cluster genes

Consistent with previous findings for the co-ordinated regulation of cluster activity (Osbourn & Field, 2009), the AMY2 cluster genes are co-ordinately expressed in response to developmental and environmental cues. Of the four genes comprising the AMY2 gene cluster, three were co-ordinately expressed and induced by abiotic stresses and rhizobium infection while one (LjSDRt) was not. The genes for several other plant metabolic gene clusters have also been reported to be induced by biotic or abiotic treatments. For example, some (but not all) genes of the momilactone diterpene gene cluster in rice are upregulated in response to treatment with a chitin oligosaccharide elicitor (Swaminathan et al., 2009). The AMY2 cluster genes are also developmentally co-regulated and their expression is co-ordinately induced by plant growth hormones. MeJA is a well-known elicitor for the production of many plant secondary metabolites, typically manifested when plants are under environmental stresses (Zhao et al., 2005). Regarding the induction of triterpene biosynthesis, MeJA induces the expression of the M. truncatula bAS gene (Suzuki et al., 2002). It is expected, through the hormonal ‘crosstalk’ networks that operate in plants, that other hormones are likely to participate in the amplitude and specificity of such elicitations (Pauwels et al., 2009). Auxin and cytokinins have been reported to alter alkaloid content when applied exogenously to Cantharanthus roseus cell cultures (Whitmer et al., 1998; Yahia et al., 1998). To our knowledge, auxin and cytokinin have not been previously reported to have functions in inducing triterpene production.

Mechanisms of gene cluster regulation

Very little is known about the mechanisms that control regulation of plant metabolic gene clusters. Common upstream cis elements that are recognized by a transcription factor required for regulation of diterpene biosynthesis in rice have been identified (Okada et al., 2009), but this transcription factor appears to be a global regulator of diterpenes and is not specific for the two characterized rice diterpene clusters. Transcriptional regulators for other plant gene clusters have not yet been defined. Additionally, the cluster for noscapine biosynthesis in poppy is co-ordinately regulated with regard to the homozygous/heterozygous state of the plant genotype, indicating an additional level of regulation beyond the transcriptional level (Winzer et al., 2012). There is also evidence that co-ordinate expression of plant secondary metabolic gene clusters is likely to be regulated at the level of chromatin modification/remodeling in oat and A. thaliana (Field & Osbourn, 2008; Wegel et al., 2009; Field et al., 2011). Nevertheless, systematic analysis of the type of chromatin modifications that are important for the expression of plant secondary metabolic gene clusters has not yet been reported. We report here such a modification, namely that DNA methylation appears to play a role in the AMY2 gene cluster regulation. We observed a ‘global-silencing’ phenomenon, in which hairpin-mediated gene silencing of any of the genes in the cluster induced DNA methylation and, thus, repression of gene expression in the adjacent genes within the cluster. It should be noted, though, that our conditions involved a transgenic RNAi-mediating silencing procedure which may not directly represent a natural regulatory event.

During post-transcriptional gene silencing, spreading of RdDM has been reported and is associated with the production of c. 21–22 nt secondary siRNAs that require RDR6 (Vaistij et al., 2002; Eamens et al., 2008; Voinnet, 2008). Both bidirectional and unidirectional spreading has been described (Vaistij et al., 2002; Daxinger et al., 2009). Spreading of methylation in transgenic plants does not always occur and there are still discrepancies in the genetic factors that may govern the methylation spreading (Henderson & Jacobsen, 2008; Daxinger et al., 2009). It has been suggested that locus-specific effects may account for the mechanism that underlies RdDM, which is initiated using hairpin constructs (Daxinger et al., 2009). Perhaps the chromatin condensation state could be responsible for the transcriptional co-silencing observed in the AMY2 gene cluster. The latter is in accordance with the association of the expression of the avenacin cluster in oats with chromatin decondensation (Wegel et al., 2009). Alternatively, there may be other cis elements present in the cluster genes that render the cluster more susceptible to methylation. Further investigation of the mechanisms governing co-ordinate gene regulation in plant genomes is likely to shed more light on this phenomenon.

Function of cluster genes

Expression of the L. japonicus AMY2 gene in N. benthamiana revealed that in addition to β-amyrin, AMY2 also produced dihydrolupeol. It is possible that nonspecific endogenous enzymatic activity in N. benthamiana results in the formation of the dihydrolupeol as compared to the formation of lupeol in yeast cells by AMY2. Alternatively, one possibility for the formation of this triterpene backbone is that AMY2 delivers a reducing hydride to the lupenyl cation to produce dihydrolupeol. The presence of the lupenyl cation as a biosynthetic intermediate could explain the observed formation of both dihydrolupeol and β-amyrin by AMY2. Further characterization of the enzyme kinetics and action, as well as the production of a dihydrolupeol standard, which proved challenging as yet, will allow the confirmation of the molecule identity. Our attempts to detect dihydrolupeol in MeJA-induced L. japonicus roots were not successful; this could be due to conversion to downstream products.

Not many members of the CYP71D subfamily have been functionally characterized but the involvement of certain members in the biosynthesis of terpenoids has been reported before (Lupien et al., 1999; Takahashi et al., 2005). LjCYP71D353 represents a novel cytochrome P450 enzyme acting on an unusual triterpene skeleton – dihydrolupeol – giving rise via successive reactions to 20-hydroxy-lupeol and 20-hydroxybetulinic acid. Whether these compounds represent the final or intermediate products of the AMY2 biosynthetic pathway in L. japonicus warrants investigation. Successive hydroxylation/oxidation reactions catalyzed by plant cytochrome P450 enzymes towards other substrates have been documented (Bak et al., 1997; Helliwell et al., 2001). Recently, CYP72A154 from liquorice was shown to catalyze three sequential oxidation steps at C-30 of 11-oxo-β-amyrin to produce glycyrrhetinic acid in yeast (Seki et al., 2011). LjCYP71D353 catalyzes oxidation reactions at two different positions of the triterpene skeleton (hydroxylation at C-20 and acid formation at C-28). A range of multifunctional cytochrome P450 enzymes catalyzing different oxidation reaction at different positions have also recently been reported from bacteria (Anzai et al., 2008; Carlson et al., 2010; Kudo et al., 2010) and fungi (Tokai et al., 2007); in the avenacin gene cluster of oat, CYP51H10 catalyses epoxidation of β-amyrin at C-12/C-13 and also hydroxylation at C-16 (K. Geisler & A. E. Osbourn, unpublished results). This enzyme is an addition to the arsenal of biosynthetic enzymes needed for the production of novel triterpenes by synthetic biology approaches.

The cytochrome P450 genes LjCYP88D4/5 within the AMY2 cluster belong to the legume-specific CYP88D subfamily, members of which have previously been characterized and assigned roles in triterpene biosynthesis. Specifically, GuCYP88D6 (Seki et al., 2008) exhibiting 72% similarity with LjCYP88D5 is a β-amyrin 11-oxidase. Unexpectedly, we did not detect any activity towards β-amyrin when both enzymes are expressed in N. benthamiana with AMY2. This could be attributed to the different heterologous system employed. Alternatively LjCYP88D5 may have a different function than GuCYP88D6 in Glycyrrhiza uralensis.

The fourth gene identified in the AMY2 gene cluster has extensive similarity to a SDR-like protein that has been identified and characterized in A. thaliana, namely HCF173 (Schult et al., 2007). HCF173 has been predicted to have lost its dehydrogenase activity and is reported as an RNA binding protein, with a regulatory role in the translational activity of the mRNA (pbsA) that directly interacts with it as part of a high molecular weight complex. The high similarity between LjSDRt and HCF173 suggests that LjSDRt may also represent a regulatory protein, having lost a metabolic function. However, no other homologs to HCF173 in other plants have been characterized as yet to allow for a justified functional prediction for LjSDRt.

In our functional analysis using gene silencing, we were able to detect a root growth defect in the plants silenced for AMY2. Our investigation for a phenotype was not exhaustive and we have only looked macroscopically for defects in plant growth and nodulation. The recent generation of the LORE1 reverse genetic resource for L. japonicus (Urbanski et al., 2012), in which mutant lines for the cluster genes have already been identified, offers a new opportunity to further characterize the role of cluster genes in plant development and physiology, as well as to validate the functional relation amongst them.

Most of the plant metabolic gene clusters discovered so far are for synthesis of compounds that are implicated in innate and induced disease resistance, insect resistance, abiotic stress tolerance and/or allelopathy (Chu et al., 2011; Kliebenstein & Osbourn, 2012). A number of apparently opposed activities have been reported for the role of triterpenes in plant developmental processes. For example, in the avenacin cluster the accumulation of late triterpene pathway intermediates can result in deleterious effects on plant growth (Mylona et al., 2008); elevated accumulation of the triterpene thalianol in A. thaliana results in enhanced root length (Field & Osbourn, 2008; Mylona et al., 2008; Field et al., 2011); and simple triterpenes, such as lupeol, act as negative regulators of nodule formation, and thus cell proliferation (Delis et al., 2011). These observations open up further questions about whether growth inhibition/promotion mediated by triterpenes occurs via different pathways or through antagonistic effects on a common pathway (Osbourn et al., 2011). The inducibility of the AMY2 gene cluster provides the opportunity to further pursue and define such a regulatory role, as yet unidentified, in plant growth and development processes.

Acknowledgements

We would like to thank George Lomonossoff for the pM81-FSC1 vector, Thomas Ott for the pUBI-GWS-GFP vector and P. Waterhouse for the pHANNIBAL vector. This work has been supported by the ESF/NSRF programme Heracleitus II (to A.K. and K.K.P.), the Postgraduate Programmes ‘Biotechnology: Quality of Nutrition and the Environment’ and ‘Applications of Molecular Biology-Genetics, Diagnostic Biomarkers’ (DBB) (to K.K.P.), a joint Engineering and Physical Sciences Research Council/National Science Foundation award to A.E.O. as part of the Syntegron consortium (EP/H019154/1) NSF/EPRSC, the BBSRC Institute Strategic Programme Grant ‘Understanding and Exploiting Plant and Microbial Secondary Metabolism’ (BB/J004596/1 to A.E.O.) and the John Innes Foundation (A.E.O.). K.G. was supported by a Danish Research Agency International Studentship.

Ancillary