Introduction

Gene expression is a complex phenomenon which is regulated by set of proteins, called transcription factors (TFs), that activate or repress various genes (Mitsuda and Ohme-Takagi 2009). In plants, TFs regulate genes involved in many important biological processes such as growth and development (Ramachandran et al. 1994), defense against pathogens and environmental stress (Singh et al. 2002), seed maturation and flower development (Jakoby et al. 2002), light-regulated mechanisms (Jiao et al. 2007), and secondary metabolism (Vom Endt et al. 2002). These TFs regulate genes by a set of highly coordinated internal/external signals, and some even interact with other TFs (Yang et al. 2012a). Secondary metabolites have long been used in pharmaceuticals, agrochemicals, fragrance ingredients, food additives, and pesticides, and are therefore of industrial importance. These metabolites do not participate directly in plant growth, development, and reproduction (Fraenkel 1959; Chae et al. 2014). However, they often play an important role in various processes including defense against pathogens, herbivores, and other interspecies defenses (Samuni-Blank et al. 2012). Also, these specialized metabolites and their associated pathways provide unique adaptive strategies for various organisms under harsh and dynamic environmental conditions (Weng and Noel 2012). Transcriptional regulation of secondary metabolite synthesis is highly controlled by a complex network of multiple TFs. Several TFs involved in the regulation of metabolic pathway genes have been studied in model plants like Arabidopsis thaliana (D’Auria and Gershenzon 2005). With advancements in developmental and molecular biology techniques for analyzing alkaloid biosynthesis, various genes have been identified to be involved in formation of metabolites like tropane (Shoji et al. 2000; Herbert 2003), benzylisoquinoline (Herbert 2003; Ziegler and Facchini 2008), and terpenoid indole alkaloids (TIAs) (Eichinger 1999; Ruppert et al. 2005). Since various studies have reported genes related to metabolite synthesis, further attempt should also be made to identify key regulators associated with their complex pathways (De Luca and St Pierre 2000; Oudin et al. 2007). Since gene expression is highly regulated by specific TFs, various biological functions get influenced by any modification in their activity, which in turn dynamically alters transcriptome profile leading to metabolic and/or phenotypic changes (Mitsuda and Ohme-Takagi 2009). Therefore, to elucidate molecular mechanisms associated to plant secondary metabolism, a prerequisite is to identify candidate regulatory TFs through a productive and precise searching procedure (Haynes et al. 2013).

Rauvolfia serpentina is an important medicinal plant of family Apocynaceae, commonly known as “Indian snakeroot,” endemic to Indian subcontinent and South-East Asian countries. It is also found in the Himalayan mountain ranges distributed over the foothills up to elevations of 1300–1400 m (Dey and De 2011). The major phytochemical constituents of this plant are indole alkaloids (Pathania et al. 2013). Its roots are also known to produce therapeutically important alkaloids which are used in the treatment of various diseases such as ventricular arrhythmias, hypertension, and diabetes (Vakil 1955; Locket 1955; Dey and De 2011; Azmi and Qureshi 2012). Reserpine is the principle component of R. serpentina used to treat hypertension (Hutcheon 1971; Nammi et al. 2005), tachycardia (Jerie 2007), and allergy (Lelek and Furedi Szabo 1961). Other compounds such as ajmaline (Bazika 1969; Köppel et al. 1989), serpentine (Beljanski and Beljanski 1982), rescinnamine (Hutcheon 1971; Nammi et al. 2005), and yohimbine (Singh et al. 2004) have also been reported in treatment of various diseases. Despite their broad pharmacological properties, limited information is available about regulatory factors involved in the synthesis of secondary metabolites (Chou and Kutchan 1998; Wink 2010) which further affect the yield and spectrum of such pharmacologically important compounds (De Luca and St Pierre 2000; Vom Endt et al. 2002). Catharanthus roseus has been considered as model plant to understand TIA pathway (Facchini and De Luca 2008; Pani and Mahapatra 2013), and TF families such as WRKY (Suttipanta et al 2011; Schluttenhofer et al 2014) and AP2EREBP have been reported to regulate metabolite synthesis (van der Fits and Memelink 2000; Li et al. 2013). Also, bHLH and MYB-like TFs are reported to be highly expressed under jasmonate response but needs to be confirmed for their regulatory role in TIA pathway (Memelink and Gantet 2007). Therefore, identification of transcriptional regulators and their target genes, involved in the synthesis of such metabolites, is crucial to understand associated metabolic pathways.

Advances in molecular and bioinformatics techniques have already made the availability of high-throughput biological data for various model plants. Fortunately, the large-scale transcriptomics data for various medicinally important plants, including R. serpentina, is also available at Medicinal Plant Genomics resource (MPGR, http://medicinalplantgenomics.msu.edu/) (Gongora-Castillo et al. 2012). Availability of this high-throughput data allowed us to conceive a hypothesis for complex regulatory insights in R. serpentina, specifically associated to secondary metabolism. Various traditional and recently developed methods are available to perform functional annotation of TFs (Mitsuda and Ohme-Takagi 2009), but as far as our knowledge, no reports are available for omics- and network-based analysis in R. serpentina. Integration of graph theoretical based methods and “-omics” data facilitates systems-level analysis of underlying molecular mechanisms and unravel various complex biological processes (Aoki et al. 2007; Mao et al. 2009; Naika et al. 2013). Moreover, network analyses using gene co-expression data has been proposed to explain biological systems since it models and decipher interactions of real biological networks (Aoki et al. 2007; Ma et al. 2013). Such analyses assist to identify functionally important regulators from a large set of genes with unknown function, and may provide information of associated pathways (Lee and Tzou 2009; Srivastava et al. 2010). Instead of performing analysis for tens of thousands of gene comparisons, a module-based analysis limits the analysis on only few orders of magnitudes for comparisons. The module represents a set of genes with very similar expression patterns that corresponds to specific pathway, therefore focusing on this amounts to a biologically meaningful data reduction scheme. Utility of such analyses has uncovered gene regulatory mechanisms based on their expression profiles (Hartwell et al. 1999; Ihmels et al. 2004). Hence, advances in network biology have made it possible to fill the gap of understanding association of individual genes by systematically investigating their association.

In this study, we have aimed to establish an integrative genome- and systems-wide framework to identify TFs, as well as to elucidate their complex regulatory behavior, associated with synthesis of secondary metabolites in R. serpentina. Gene co-expression network analysis was implemented to identify functionality of TFs, which were not annotated through simple annotation methods. Modules were obtained from co-expression network and enrichment analysis was performed to identify highly significant modules. Genes present in each of significant module was enriched with similar ontological terms and tissue-specific expression emphasizing their robustness. These modules were associated with hormone- or elicitor-mediated synthesis of indole alkaloids, epigenetic regulations, reproductive system, and upstream of TIA pathway, respectively. Interestingly, one module was found to regulate synthesis of specialized indole alkaloids known to be present in roots under hormone- or elicitor-mediated response, and do not have any shared metabolite with the remaining three modules. The identification of such regulatory factors may aid to improve quantity and diversity of secondary metabolites, and methodology presented herein can possibly assist their experimental studies. This graph theoretical approach may provide a better reference for other non-model plants to improve the established strategies for TF identification.

Materials and Methods

Toward our goal of identification of TFs and associated modules in secondary metabolism, the following procedures were implemented: (1) annotation of R. serpentina transcriptome, (2) calculation of Pearson correlation coefficient, (3) transcription factor binding site analysis, (4) construction of weighted co-expression network, (5) enrichment analysis and module detection, (6) reference-based enrichment of significant modules, and (7) integration of metabolomics data.

Annotation of R. serpentina Transcriptome

Transcriptomic sequences and expression profile data of various R. serpentina tissues were retrieved from the Medicinal Plant Genomics Resource database (MPGR, http://medicinalplantgenomics.msu.edu/) (Gongora-Castillo et al. 2012). All transcripts were annotated by performing BLASTX (Altschul et al. 1997) search against the reference Arabidopsis proteome (TAIR10, http://Arabidopsis.org). An e value cutoff of 1e-05 was considered to identify homologous genes, and annotations of associated top hits were preserved for further analysis. In order to identify TFs, transcripts with unknown primary annotation were also searched independently against Plant Transcription Factor Database (PlnTFDB; http://plntfdb.bio.uni-potsdam.de/v3.0/) (Riaño-Pachón et al. 2007) using BLASTX with the same specifications of hit selection. To describe the functional annotation of transcripts, gene ontology (GO) (Dimmer et al 2008) analysis for all three biological domains (biological processes (BP), molecular functions (MF), and cellular components (CC)) was performed. Annot8r (Schmid and Blaxter 2008) was used to obtain GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway annotation for assembled transcripts based on similarity search against annotated subsets of the UniProt database (The UniProt Consortium 2007). This annotation was performed with an e value threshold of 10-01, and top hits were retained based on highest significant e value as well as bit score.

Calculation of Pearson Correlation Coefficient

In order to construct TF-gene regulatory network, Pearson correlation coefficient (PCC) (Eisen et al. 1998) between identified TFs and all transcripts were calculated from expression data in the form of FPKM (fragments per kilobase of transcript per million fragments mapped), available for different tissues (mature leaf, young leaves, upper stem, young roots, mature roots, red stem, flower, and woody stem), using customized in-house written PERL script. PERL code has been provided at GitHub (https://github.com/ShivalikaP/Perl-script-to-calculate-Pearson-correlation-coefficient). PCC was computed based on the following equation:

$$ \mathrm{P}\mathrm{C}\mathrm{C}=\frac{{\displaystyle \sum \left(x-\overline{x}\right)\left(y-\overline{y}\right)}}{\sqrt{{\left(x-\overline{x}\right)}^2}{\sqrt{\left(y-\overline{y}\right)}}^2} $$

Where \( x \) and \( y \) denotes the corresponding expression profiles of TF and transcripts, respectively, and \( \overline{x} \) and \( \overline{y} \) represents their mean. Mean of expression profile depends upon sample size which is different for both TFs and transcripts.

Network density (ND) helps in determining biologically important PCC threshold (Aoki et al. 2007). ND was measured as follows:

$$ \mathrm{N}\mathrm{D}=\frac{2E}{K\left(K-1\right)} $$

Where \( E \) and \( \frac{K\left(K-1\right)}{2} \) specify the number of actual links and possible links of non-singleton nodes, respectively.

The “t test” was implemented to select significant TF-gene pairs with p value ≤0.05 (McCluskey and Lalkhen 2007), and further subjected to multiple testing correction using false discovery rate (FDR) (Benjamini and Hochberg 1995) method through “stats” library of R statistical package (http://www.r-project.org/). Shortlisted pairs were compared against TF-gene connectivity data (Misra and Sriram 2013), downloaded from the Arabidopsis Gene Regulatory Information Server (AGRIS; http://Arabidopsis.med.ohio-state.edu/) (Davuluri et al. 2003), to verify their regulatory relationship based on prior knowledge available for corresponding orthologs in A. thaliana.

Transcription Factor Binding Site Analysis

Transcription factor binding sites (TFBS) were identified using pPromotif tool (Jha and Shankar 2014) for those TF-gene pairs which did not have any match to already verified TF-gene connectivity data. In order to identify TFBS, 5 kb upstream region of orthologous Arabidopsis genes of corresponding transcripts were retrieved from BioMart of Ensembl Plants database (http://plants.ensembl.org/biomart/martview) (Kinsella et al. 2011). These upstream regions were further analyzed to identify TFBS based on a position weight matrix generated using the Gibbs Motif Sampler. Furthermore, pairs were selected based on comparison of TF families assigned to TFs against all possible binding sites obtained. These selected pairs were further complemented by manual literature search. Accuracy of identified TFBS was statistically validated using Z-score for orthologous Arabidopsis genes. For calculating Z-score, pPromotif was implemented on 5 kb upstream sequences of Arabidopsis (as null model/control), and mean as well as standard deviation was calculated from score obtained from TFBS hits of 13 TF families. Z-score (Sundar et al. 2008) was computed using the following equation:

$$ Z-\mathrm{score}=\frac{x-\mathrm{Mean}}{\mathrm{SD}} $$

Where x is the pPromotif score for each TF hit obtained for orthologous Arabidopsis genes. Mean and SD are the average and standard deviation of pPromotif scores, respectively, for all control genes of 13 TF families.

A two-tailed p value statistics, corresponding to Z-score, was further calculated using “stats” library of R package.

Construction of Weighted Co-expression Network

Weighted co-expression network was constructed from TF-gene pairs obtained during successive filtering steps. An assessment-based co-expression network was attained since edges connecting nodes at significant PCC threshold and p values were kept. Cytoscape v 3.1.0 (Shannon et al 2003) program was used for network construction and visualization purpose. TFs obtained in resultant co-expression network were classified into different families based on BLAST search against PlnTFDB. PCC values were used as edge weights to construct network, where weights represent the strength of connections or co-expression. Random network with same number of nodes and edges, as of weighted co-expression network, was generated using “igraph” (Csardi and Nepusz 2006) library in R package. Network topology parameters of both networks were determined to compare their topological properties.

Enrichment Analysis and Module Detection

Enrichment analysis of all transcripts corresponding to each TF family was carried out with singular enrichment analysis (SEA) of agriGO web-based tool (http://bioinfo.cau.edu.cn/agriGO/index.php) (Du et al. 2010), and significantly enriched GO terms were determined in all comparative conditions by comparing it against background reference. Hypergeometric test (Shimizu et al. 2014) with Bonferroni correction (Simes 1986) was applied for selecting these statistically significant terms. In search of comparatively smaller and functionally important gene modules, Markov cluster (MCL) (Van Dongen 2008) algorithm, a systematic graph clustering method, was used to fragmentize the network. This algorithm is based on the “random walks” which states that starting with a node randomly travel to connected nodes will make to stay within a cluster than travel between, i.e., flow is easier within dense regions than across sparse boundaries. MCL algorithm, with default inflation value of 2.5, was implemented using clusterMaker2 (Morris et al. 2011) plugin of Cytoscape. Enrichment analysis of all modules was also performed independently. Expression data was used to generate heat maps using “gplots” library of R package. “Vennerable” library from same package was used to obtain venn diagram of common target genes among TFs in each significant module.

Reference-Based Enrichment of Significant Modules

In order to perform reference-based enrichment analysis, clueGO (Bindea et al. 2009) plugin of Cytoscape was used against model plant A. thaliana. Enrichment analysis was performed using a two-sided (enrichment/depletion) tests based on hypergeometric distribution with Bonferroni (Simes 1986) p value correction method. Besides functional enrichment, it also creates an annotation network of functionally grouped non-redundant biological pathways associated with enriched GO terms for visualization. Kappa statistics (Sim and Wright 2005) is used by clueGO to link similar terms in a network that provides a measure of accordance from a range between 0 and 1, where 0 and 1 indicates no and almost perfect accordance, respectively. In this analysis, a kappaScore threshold of 0.4 was considered.

Integration of Metabolomics Data

To assess quality of significant modules and to determine the involvement of TFs in regulation of secondary metabolism, metabolome data of R. serpentina was included. Correlation data of metabolome, available at the Plant and Microbial Metabolomics Resource (PMR; http://metnetdb.org/PMR/), of corresponding transcriptome was obtained using PMR metabolomic-transcriptomic co-analysis tool at selected PCC threshold and significant p value (≤0.05). TFs present in significant modules were searched against correlation data to determine module-specific regulation of metabolite synthesis. “Vennerable” library was used to obtain venn diagram of shared metabolites in different modules. Significance of modules associated with synthesis of specialized metabolites was also statistically verified using “Mann-Whitney test” from “stats” library of R package.

Results

An integrative approach was established to identify TFs and their role in regulation of valuable metabolites and metabolic traits (Fig. 1) in R. serpentina. Initially, TFs from the transcripts of unknown function were identified which were further analyzed to determine their complex regulatory mechanism using graph theoretical approach.

Fig. 1
figure 1

Strategy implemented for identification of transcription factors

Full size image

Identification of Transcription Factors

A. thaliana has been widely used as a reference model for annotation of various plants including C. roseus, a member of Apocynaceae family (Schluttenhofer et al. 2014). Similarly, annotation of complete R. serpentina transcripts was also carried out against Arabidopsis proteome that returned 80,829 significant hits. A total of 12,670 transcript entries without primary annotation were retrieved, and 342 hits were considered as putative TFs while annotating against PlnTFDB.

GO- and KEGG-based functional analysis was performed to functionally annotate complete transcriptome using annot8r, and output has been provided as separate spreadsheets at GitHub (https://github.com/ShivalikaP/R_serpentina_transcriptome_annotation). Annot8r has been widely used to functionally annotate transcriptomes of various plants including Citrus sinensis (Yang et al. 2013) and Euphorbia fischeriana (Barrero et al. 2011). This analysis returned functional annotation for ∼71 and ∼43 % transcripts covering a broad range of GO and KEGG categories, respectively, of R. serpentina transcriptome. The most abundant BP categories were metabolic process (∼26 %), response to stimulus (∼14 %), and regulation of biological process (∼13 %) revealing that the majority of transcripts were participating in metabolic activities under response to stimulus (Fig. S1a). Most abundant GO categories such as binding (∼44 %), transferase activity (∼15 %), and catalytic activity (∼13 %) terms were determined under MF category (Fig. S1b). Similarly, major GO categories identified under CC categories included intracellular (∼68 %) and membrane (∼27 %) (Fig. S2a) terms. A total of 260 KEGG pathways were retrieved for whole trancriptome sequences, and the top 20 of them were associated with plant hormone signal transduction (∼5 %), plant-pathogen interaction (∼4 %), ribosome (∼3 %), and spliceosome (∼3 %) (Fig. S2b).

Threshold Selection Based on Network Topology

Since simple annotation was unable to annotate biological processes associated with 190 TFs (Table S1), these TFs were annotated based on their interacting transcripts. In order to determine congruity of putative TFs and associated transcripts, expression profiles among eight different tissues were used to compute PCC. This number of samples is sufficient to carry out co-expression analysis as evident from various studies to identify new candidate genes (Gongora-Castillo et al. 2012; Paul et al. 2014; Schluttenhofer et al. 2014). PCC is used to construct a correlation matrix of gene expression from which weighted co-expression network is generated (Eisen et al. 1998). A total of 4,504,588 TF-gene pairs were obtained for unique 342 TFs and 19,333 transcripts, and pairs with positive correlation were retained to analyze biological processes up-regulated by these TFs. ND demonstrates how sparse or dense a graph is according to the number of connections per node set, and has been used for choosing a statistically relevant threshold since at this cutoff, biologically significant modules are expected (Aoki et al. 2007; Mao et al. 2009). To characterize the PCC threshold, change in number of nodes, edges, and network density, as a function of various PCC cutoff values, were examined. As the cutoff value increased, node and edge number, as well as the actual number of edges and all possible edges were decreased. However, as the cutoff reached at a relatively high value, decreasing rate of edges became slower than that of all possible edges (Fig. S3a) and nodes (Fig. S3b), which leads to an increase in ND. ND was at its minimum at 0.84 cutoff (Fig. 2a), and PCC corresponding to this minimal ND was considered as threshold. TF-gene pairs at and above this threshold were analyzed to obtain biologically significant modules. Conversely, threshold was adjusted at 0.80, instead of 0.84, in order to retain a relatively large number of connections and prevent the loss of significant information in network construction. Although, PCC can be used to measure the congruity, yet it is difficult to determine its statistical significance more than expected by chance. All TF-gene pairs (151,504) at and above the threshold of 0.80 were found to be statistically significant (p value ≤0.05) (Table S2), and the number of these pairs were further reduced to 82,635 based on availability of Arabidopsis orthologs for corresponding transcript of R. serpentina in TF-gene connectivity data from AGRIS (Davuluri et al 2003). Regulatory information of these TF-gene pairs was further complemented by comparing them against this TF-gene connectivity data, and 7066 connections among all these pairs were retained, whereas the remaining 75,569 were discarded due to unavailability of the TF-gene interactions. This selection criterion helps to strengthen TF-gene pairs regularity information based on prior validated information that ensure minimum number of false positives.

Fig. 2
figure 2

Topological analysis of co-expression network

Full size image

Identification of Transcription Factor Binding Sites

Since PCC of discarded TF-gene pairs were found to be statistically significant, transcripts were analyzed for possible TFBS to match coinciding TF in each pair. Promoter sequences are often conserved between diverse plant species (Fauteux and Strömvik 2009); (Koch et al. 2001) that allow the use of A. thaliana as reference to perform TFBS analysis. Upstream sequences of orthologous Arabidopsis genes, corresponding to 9676 transcripts, were processed to identify TFBS. A total of 13,439 TF-gene pairs were selected by comparing them against pPromotif output, and the rest were discarded in further studies. In order to predict accuracy of TFBS prediction, hits obtained for orthologous Arabidopsis genes were compared against TFs from AGRIS (control) to compute Z-score. Z-score explains the direction as well as measure of deviation of number of TFBS occurrences from background distribution’s mean. Null hypothesis states that there was no significant difference in scores between orthologous Arabidopsis genes and control. It was found that for ABI3VP1, 10,829 out of 10,829 (100 %) number of sites had insignificant p value (>0.05). Similarly, the percentage of hits obtained for other TF families such as AP2EREBP, ARF, bHLH, bZIP, CCAAT, C2C2-GATA, HB, MYB, MYB-related, MADS, SBP, and WRKY were 93.59 % (9385/10,028), 79.24 % (8067/10,180), 100 % (10,758/10,758), 96.11 % (10,408/10,829), 99.88 % (10,816/10,829), 99.90 % (10,819/10,829), 94.99 % (6632/6982), 92.85 % (10,055/10,829), 98.29 % (10,644/10,829), 96.63 % (9651/9987), 100 % (1087/1087), and 94.78 % (10,132/10,690), respectively. Since p value was found to be insignificant, null hypothesis was accepted which further signifies the accuracy of TFBS prediction. These selected pairs were further augmented by critical manual literature search for TFs regulating either corresponding orthologous Arabidopsis genes or orthologs of orthologous Arabidopsis genes in some other plant species. The search was carried out for 290 TF-gene pairs randomly selected at different cutoffs in order to cover a wide range of PCC (0.8–1.0). Interestingly, ∼94 % of TF-gene pairs were observed to be supported in different literature studies (Table S3) which further enhanced the confidence of output obtained. Importantly, this analysis enhances number of connections by identifying TF-gene pair regulatory relationship left in earlier step which was supported by expression data.

Construction of Weighted Co-expression Network

Weighted co-expression network was constructed for deciphering systematic involvement of identified TFs in regulation of various biological processes. Such co-expression networks often utilizes measured PCC as weights, and has been widely used to determine biologically important modules in a variety of organisms (Aoki et al. 2007; Mao et al. 2009; Ruan et al. 2010; Gongora-Castillo et al. 2012; Fukushima et al. 2012). Co-expression network constituted of nodes and edges representing genes and connections between them, respectively. The network had 7304 nodes (including unique 69 TFs and 7235 target genes) with 20,505 edges. All identified 69 TF genes were classified into 13 TF families (Table S4, Fig. 3) based on the BLAST search against PlnTFDB. PCC values obtained in former step were used as edge attribute where these numeric attributes quantifies strength of connections among nodes, which again enhanced the method of module detection. Real networks are often compared with random networks to decipher scale-free nature of networks and to distinguish their topological properties (Alon 2006; Raval and Ray 2013). A weighted random network was constructed with same number of nodes, as of actual co-expression network, by creating random connection among edges for 2000 iterations. Compared to bell-shaped distribution of random network (Fig. 2b), the degree distribution (DD) of co-expression network was highly skewed indicating that majority of nodes had a degree, close to average degree in random graphs and thereby satisfies properties of biological networks (Albert 2005). DD of co-expression network also satisfied the power law (Fig. S4) since there was no peak around average degree. Graph rather sloped steeply downwards with increase in degree demonstrating the presence of a large number of nodes with very few connections dominated by some extremely well-connected ones. This revealed that co-expression network had scale-free behavior and robust against random perturbations since such networks are most likely to hit a node with only few neighbors, and therefore disrupt only a small portion (Emmert-streib and Dehmer 2008). Negative value of assortativity (−0.68) also augmented the scale-free nature of co-expression network as compared to random network (0.0014) (Johnson et al 2010).

Fig. 3
figure 3

Co-expression network constituted of transcription factors and associated genes

Full size image

Functional Annotation and Enrichment (Over-Representation) Analysis

GO terms of target transcripts, corresponding to each putative TF family, were used as background to perform enrichment analysis through agriGO. Analysis of each TF family revealed that several transcripts were observed to be significantly over-represented (p value ≤0.05) to participate in more than one biological process. TF families such as ABI3VP1, bHLH, HB, MYB, MYB-related, and WRKY (Figs. S5–S10) were significantly enriched with terms related to plant reproductive system. Major significant terms obtained among these TFs included floral whorl development, stamen development, floral organ development, pollination, pollen tube growth, and androecium development. TF families (ABI3VP1, AP2-EREBP, MYB-related, and WRKY) (Figs. S5, S11, S9, and S10) were enriched for terms such as response to hormone stimulus, response to endogenous stimulus, response to organic substance, and response to defense response presented their direct involvement in response to environmental conditions. HB (Fig. S7) and bZIP (Fig. S12) were observed to be significantly enriched with terms related to cell cycle such as cell cycle process and microtubule-based process, while cell division AP2-EREBP (Fig. S11) and bHLH (Fig. S6) were significantly over-represented with terms photosynthesis and photosynthesis, light reaction. Interestingly, some TF families were perceived to be enriched with terms associated with specific process such as ABI3VP1 (Fig. S5) (polarity specification of adaxial/abaxial axis and specification of axis polarity) to axis or pattern formation, AP2-EREBP (Fig. S11) (pentose-phosphate shunt, NADPH regeneration, and NADP metabolic process) to primary metabolic pathway as well as upstream of TIA pathway (Dubey et al. 2003), HB (Fig. S7) (chloroplast RNA processing, DNA alkylation, and DNA methylation) to post-translational modifications via ubiquitin-dependent proteolytic pathway, and MADS (Fig. S13) (multicellular organismal development) to organ development. TF families ARF, CCAAT, C2C2-GATA, and SBP were not found to be enriched with any significant term.

Detection of Functional Modules

Gene-enrichment analysis demonstrated that multiple biological pathways were under regulation of one TF family that did not specify pathway regulated by each TF of given family. Therefore, module-based analysis was performed to identify the TFs that may function in an integrative manner to regulate specific pathway. The identification of modules is a better choice to understand the coordinated behavior of TFs of different families since each module occupies certain set of genes shared by TFs present in it. Top-down approach (or non-targeted approach) (Aoki et al. 2007) can be applied to fragmentize network into modular structures through graph clustering algorithms based on the topology of connections (Mao et al. 2009). MCL algorithm has successfully been implemented for module detection in many biological networks (Mao et al. 2009). It sub-divided the large network, to simplify the analysis of weighted co-expression network, into 42 modules (Fig. S14) ranging from 1251 (largest) to 4 (smallest) nodes at default inflation value. These modules were referred to by their number henceforth, as a standard practice. The first five modules (1–5) were largest in size with 1,251, 995, 762, 750, and 417 nodes including 9, 5, 7, 5, and 4 regulatory TF genes, respectively. Modules from 6–10, 11–16, 17–28, 29–37, and 38–42 fragmented in ranges from 201 to 290, 101 to 150, 51 to 100, 20 to 45, and 4 to 15 nodes, respectively, whereas modules 6–42, except 7 and 18 (2 TFs), had only 1 TF.

Enrichment Analysis of Modules

Enrichment analysis was performed for all 42 modules, and only few (1–4, 6, 8–11, 13–15, 20, 21) were characterized with at least one significant enriched GO term. The first four modules (Fig. 4) were considered as significant modules since each of them presented maximum number of enriched terms with similar biological processes, and implicated in downstream analysis. A number of earlier reports have presented considering only significant modules (Childs et al. 2011; Chou et al. 2014). Transcript identifiers of all TF families in significant modules are provided in Table 1. Most transcripts present in each module were enriched with similar ontological terms indicating the robustness of modules. GO and KEGG pathways annotation for all transcripts present in these significant modules (1–4) is provided in Tables S5–S8. Details of significant modules are provided as follows:

Fig. 4
figure 4

Four significant modules and their associated functions

Full size image
Table 1 TFs with their transcript identifier as well as TF family present in the four significant modules (1–4)
Full size table
  1. Module 1

    Module 1 was associated with defense response and indole alkaloid synthesis (Fig. 5) as majority of significant terms included response to stimulus (GO:0050896; 1.02e-08), response to endogenous stimulus (GO:0009719; 7.45e-07), defense response (GO:0006952; 2.52e-06), abscisic acid-mediated signaling pathway (GO:0009738; 0.00306), indole derivative biosynthetic process (GO:0042435; 0.0208), indole derivative metabolic process (GO:0042434; 0.0239), and response to jasmonic acid stimulus (GO:0009753; 0.0427). Intentness of abscisic acid and jasmonate-mediated signaling response for major indole alkaloids synthesis has already been reported in R. verticillata (Fernandez and de Luca 1994) and C. roseus (Schluttenhofer et al. 2014), closely related plants to R. serpentina, respectively. It has already been reported that diverse metabolic pathways are induced under various stress responses (Naika et al. 2013). Therefore, all these major ontological terms may reveal possible role of module 1 in defense response (Wei et al. 2014; Zarate et al. 2007) and indole alkaloid synthesis (Fernandez and de Luca 1994; Schluttenhofer et al. 2014). Also, TF families (ABI3VP1, AP2-EREBP, MYB-related, SBP, and WRKY) (Fig. 6a) and their targets (Fig. S15a) had comparatively higher expression in roots further complementing earlier reports of synthesis of major alkaloids in roots. Most significant pathways, like plant-pathogen interaction and plant hormone signal transduction, were also found to be associated with defense response and response to stimulus (Fig. S16a).

    Fig. 5
    figure 5

    Gene ontology enrichment analysis of module 1

    Full size image
    Fig. 6
    figure 6

    Heat maps of expression profiles of significant modules

    Full size image
  2. Module 2

    Module 2 was observed to be involved in epigenetic regulations as majority of enriched terms (Fig. S17) were related to chloroplast RNA processing (GO:0031425; 5.85e-07), cell cycle process (GO:0022402; 0.000235), embryonic development ending in seed dormancy (GO:0009793; 0.00182), fruit development (GO:0010154; 0.00237), fruit development (GO:0010154; 0.00237), seed development (GO:0048316; 0.00158), and post-embryonic development (GO:0009791; 0.0247). Most significant chloroplast RNA processing term present an important regulatory step of post-translational modification via ubiquitin-dependent proteolytic pathway, and involves in stabilization of primary chloroplast transcripts (Schuster and Gruissem 1991). This ontology term also emphasized on localization of upstream of TIA pathway in chloroplast (Ershov et al. 2002; Flügge and Gao 2005; Herrmann and Weaver 1999; Schmid and Amrhein 1995). TIA pathway (Fernandez and de Luca 1994; Shukla et al. 2006) and jasmonic acid-mediated activation of ORCA (AP2-like) TF (Vom Endt et al. 2002) have already been reported to be regulated by post-translational modifications under ubiquitin-dependent proteolysis. Also, TF families (ABI3VP1, bHLH, HB, and AP2-EREBP) presented were found to be highly expressed in young leaves (Fig. 6b), and may necessitate chloroplast localized synthesis of precursors under epigenetic regulation. Interestingly, enrichment analysis of HB TF family was also enriched with few similar enriched terms signified its involvement in epigenetic regulations. Moreover, HB TF present in flowering wageningen (FWA) gene has already been reported to be involved in RNA-directed DNA methylation in A. thaliana (Matzke et al. 2007). This highlighted the involvement of HB TF family in epigenetic regulations and may contribute to growth, development, and consequently plant evolution (Grafi and Ohad 2013). Significance of module was also augmented by the involvement of cell cycle process (Perianez-Rodriguez et al. 2014), chromatin modifications/remodeling in seed development (Baroux et al. 2007), embryonic development ending in seed dormancy (Ríos et al. 2014), and post-embryonic development (de Vega-Bartol et al. 2013; Perianez-Rodriguez et al. 2014). Similarly, most significant pathways for this module comprised of processes such as cell cycle, spliceosome, mismatch repair, and RNA degradation complementing GO terms assigned, and its involvement in epigenetic regulations (Fig. S16b).

  3. Module 3

    Module 3 was associated with reproductive system since significantly enriched terms (Fig. S18) were found to be related to reproductive process (GO:0022414; 2.18e-06), reproduction (GO:0000003; 4.3e-06), pollen tube growth (GO:0009860; 2.4e-05), stamen development (GO:0048443; 3.73e-05), androecium development (GO:0048466; 3.73e-05), and pollination (GO:0009856; 4.18e-05). TF families ABI3VP1, bHLH, MYB, MYB-related, SBP, and WRKY were found to be expressed mainly in flower (Fig. 6c) which further complement the enrichment results and highlights robustness of module. Pathway annotation of module was significantly over-represented with terms like flavonoid biosynthesis and phenylalanine metabolism (Fig. S16c) that may be associated to synthesis of anthocyanins and thereby represent its involvement in pollination (Mol et al. 1998).

  4. Module 4

    Module 4 was related to primary metabolic pathway, and upstream of TIA pathway as significant GO terms (Fig. S19) presented were plastid localization (GO:0051644; 1.45e-08), chloroplast relocation (GO:0009902; 1.45e-08), photosynthesis, light reaction (GO:0019684; 3.72e-07), pentose-phosphate shunt (GO:0006098; 1.23e-06), NADPH regeneration (GO:0006740; 1.28e-06), NADPH metabolic process (GO:0006739; 1.76e-06), oxidoreduction coenzyme metabolic process (GO:0006733; 3.04e-05), and generation of precursor metabolites and energy (GO:0006091; 0.000863). Pentose-phosphate cycle is known to be a direct precursor for isoprenoid synthesis (Ershov et al. 2002; Flügge and Gao 2005) and shikimate pathway that are involved in the release of various aromatic amino acids localized inside the chloroplast (Herrmann and Weaver 1999). Similarly, isoprenoid biosynthesis utilizes NADPH which is a byproduct of methylerythritol phosphate (MEP) pathway localized in chloroplast/plastid (Dubey et al. 2003; Seemann et al. 2006). Since upstream of TIA pathway utilizes NADPH, the presence of significantly enriched terms together like NADPH regeneration, NADPH metabolic process, plastid localization, and chloroplast relocation intimated the involvement of this module in synthesis of precursor metabolites through MEP pathway (El-Sayed and Verpoorte 2007). Enrichment analysis of AP2-EREBP TF family was also enriched with similar terms like pentose-phosphate shunt (GO:0006098; 0.0179), NADPH regeneration (GO:0006740; 0.0186), NADP metabolic process (GO:0006739; 0.0251), photosynthesis (GO:0015979), and photosynthesis, light reaction (GO:0019684) (Fig. S11). Moreover, AP2EREBP TF family has also been reported to regulate primary and secondary metabolism in C. roseus (Van der Fits and Memelink 2000) that may possibly regulate identical processes in module 2. Other co-expressed TF families (bHLH, WRKY, and SBP) may correspond to unknown TFs, as revealed in the study by Oudin et al. (2007), regulating upstream of TIA pathway. Tissue-specific expression of these TF families in mature leaves also (Fig. 6d) emphasized chloroplast localized regulation of precursor synthesis. Similarly, most significant pathways such as photosynthesis and glycolysis/gluconeogenesis (Fig. S16d) also augmented its involvement in regulating primary metabolic pathways as well as synthesis of precursors in upstream of TIA pathway (Aniszewski 2007; Glenn et al. 2013). It has also been stated earlier that the initial committed steps in secondary metabolite synthesis are recruited from primary metabolic pathways (Aharoni and Galili 2011; Chu et al. 2011). Interestingly, these results revealed that this module may act as connecting link between primary and secondary metabolic pathways, and TFs presented may regulate connecting steps between them.

Putative TFs have been found to work in a highly coordinated manner to perform important biological functions since 38 % of them were highly clustered only in the first four modules. A total of 39, 123, 50, and 2 common transcripts (Fig. 7), among all TF families, were obtained for significant modules (1–4), respectively, and were associated with representative pathways. This analysis represented that TFs do not perform their functions independently but in a highly coordinated manner as composite/co-regulatory modules (CRMs) where more than one TF families work together to facilitate regulation of gene expression (Jha et al. 2011; Matys et al. 2006). This further indicated the interdependence and complex coordinated behavior among all TF families to regulate specific biological pathways.

Fig. 7
figure 7

Venn diagram of transcripts shared by transcription factors of each significant module

Full size image

Enrichment Analysis Against Reference Model

Networks obtained from reference-based enrichment consisted of significant GO terms as nodes and shared genes among these terms as edges. In all significant modules, various biological pathways were distinctly clustered by iterative merging of initially defined groups based on kappa score threshold. A total of 605 transcripts were enriched in module 1 with majority of significant terms being hormone-mediated signaling pathway, defense response, and secondary metabolic process (Fig. S20). All these terms were similar to agriGO enrichment which further highlights its involvement in indole alkaloids synthesis (Chang et al. 2013; Schluttenhofer et al. 2014). Similarly, a total of 472, 242, and 389 transcripts in module 2, 3, and 4, respectively, were over-represented with significant GO terms. In module 2 (Fig. S21), significant terms like chromatin organization and RNA processing represented its participation in epigenetic regulations that may be associated with TIA pathway (Fernandez and de Luca 1994; Shukla et al. 2006; Siberil 2002; Vom Endt et al. 2002). In module 3 (Fig. S22), pathways represented by significant GO terms such as external encapsulating structure organization, stamen development, and pollen exine formation were also related to reproductive system. Similarly, module 4 was represented with significant terms such as photosynthesis, thylakoid membrane organization, and plastid organization (Fig. S23). Conclusively, this reference-based enrichment analysis also complements agriGO enrichments, and thereby highlighting robustness and quality of significant modules.

Module-Specific Regulation of Secondary Metabolites

Integration of transcriptome and metabolome data has been successfully implemented to determine TFs involved in regulation of specific metabolic pathways in the earlier reports (Hirai et al. 2005; Yonekura-Sakakibara et al. 2008). TFs present in significant modules were compared against correlation data (40 metabolites), and significant modules (1–4) were found to regulate 43.59, 17.94, 17.94, and 7.69 %, respectively, of all secondary metabolites (39) except tetraphyllicine. Module 1 (Table 2) was observed to regulate major indole alkaloids known to be synthesized in roots, and did not even share any metabolite with the rest of the significant modules (Fig. 8). Interestingly, major root indole alkaloids such as ajmaline, ajmalicine, reserpine, sarpagine, and serpentine (O’Connor and Maresh 2006) were present only in module 1. Therefore, transcripts in module 1 may indicate a complex interaction among them to regulate synthesis of these important indole alkaloids. Similarly, our findings were also supported by an earlier study that presented hormonal- or elicitor-mediated signaling pathways (e.g., WRKY TF family) to regulate major indole alkaloids, such as ajmalicine and serpentine, in C. roseus (Schluttenhofer et al. 2014). All these important findings, along with specifically high expression of module 1 transcripts in roots (Fig. S15a), determined the quality of identified modules as well as the protocol implemented for identification of key regulators. Modules 2–4 (Fig. 8, Table S9) shared few common as well as similar isomers of precursor metabolites (tryptamine, vellosimine, strictosidine, kaempferol hexoside dirhamnoside isomer 2, and kaempferol hexoside dirhamnoside isomer 1) that may reveal their intra- and inter-cellular trafficking of biosynthetic intermediates throughout the course of associated biosynthetic pathways (Kutchan 2005; O’Connor and Maresh 2006). Based on functional annotation and enrichment analysis, these modules were represented with significant representative processes like epigenetic regulations, reproductive system, and precursor’s biosynthesis with majority of expression in young roots, young leaves, flower, and mature leaves, respectively (Fig. S15b–d). Modules 2 and 4 were found to be expressed in leaves (young and mature, respectively), and may be grouped according to their representative functions viz., epigenetic regulations and photosynthesis associated synthesis of precursors. Module 3 was found to be regulating metabolites, mainly flavonoids and its derivatives, involved in pollination which also complemented enrichment results (Mol et al. 1998; Borevitz 2000). Importantly, integration of metabolomics and transcriptomics data had strengthened the reliability of significant modules due to module-specific regulation of various secondary metabolites (Fig. 8, Tables 2 and S9).

Table 2 Secondary metabolites found to be highly correlated with TFs of module 1 at the PCC threshold (≥0.80), obtained from metabolomics data
Full size table
Fig. 8
figure 8

Venn diagram of shared metabolites among significant modules

Full size image

Significance of Gene Clustering in Module 1

Module 1 was observed to be involved in the synthesis of ajmaline/sarpagine-type alkaloids (O’Connor and Maresh 2006) of the last few steps of TIA pathway, under response of hormone- or elicitor-mediated signals (Leménager et al. 2005). Genes regulating synthesis of specialized molecules, under various environmental stresses, have been reported to physically clustered (Ehrlich and Raven 1964; Nützmann and Osbourn 2014) and represents biased clustering against non-specialized metabolites (Chae et al. 2014). In order to statistically verify biased nature of modules related to synthesis of specialized metabolites, Mann-Whitney test was implemented on co-expression values of all modules which had at least one specific alkaloid expressed only in roots (18) against remaining modules (24). Null hypothesis states that there was no significant preference of metabolic pathways genes in modules of specialized metabolites as compared to remaining modules. Mean co-expression value of specialized metabolites containing modules (0.88) was significantly higher than the remaining modules (0.87), and it was found that modules of specialized metabolites had significant preference for their synthesis over non-specialized metabolites (p value = 4.651e-16). It has been concluded that module 1 is statistically found to be most significant, and mainly under the control of diverse environmental conditions associated with specialized metabolites synthesis. Similarly, TFs present in this module may contribute into the diversity of major indole alkaloids in R. serpentina in response to various internal/external stimuli.

Discussion

R. serpentina has been widely used in Ayurvedic medicinal preparations and tribal ethnomedical systems for thousands of years (Dey and De 2011), and roots of this plant are well known to produce medicinally important alkaloids used in treatment of various diseases (Azmi and Qureshi 2012; Dey and De 2011; Locket 1955; Vakil 1955). In a previous study, we have been able to computationally identify inhibitors against aldose reductase from a repository of natural compounds compiled from the same plant (Pathania et al. 2013). Despite its medicinal importance, few studies have reported regulation of secondary metabolite synthesis in R. serpentina (W. M. Chou and Kutchan 1998; Wink 2010). The limited information of regulators that control the production of these molecules inhibit biotechnological application which may affect yield and spectrum of these pharmacologically important compounds (De Luca and St Pierre 2000; Vom Endt et al. 2002). Therefore, identification of transcriptional regulators as well as their target genes is crucial to understand metabolic pathways that lead to synthesis of such molecules. In recent years, gene co-expression networks have been used to unravel important biological processes and their regulatory mechanisms. There has been a growing trend in using large-scale protein-protein interaction (PPI) networks in plant systems (Yang et al. 2012b); however, the datasets cover only a limited range of PPI data. Furthermore, PPI networks are not static but dynamic entity and expression of protein is intrinsically controlled by different regulatory mechanisms (Liang and Li 2007). With an exponential growth of gene expression data, most of the current research has been shifted toward systems biology that largely include co-expression networks of transcriptome data (Yonekura-Sakakibara et al. 2008; Zhang et al. 2012).

Identification and annotation of TFs at genome scale are the first step toward understanding the mechanism of gene expression and its regulation. Though various genome-wide experimental studies have reported candidate TFs hitherto (Hu et al. 2014; Wang et al. 2010; Zhu et al. 2012), not much systems-level work has been done to assimilate the data. With increase in high-throughput proteomic/genomic studies and computational facilities, identification and annotation of TFs at genome scale are now being used to understand the regulation of biological mechanism (He et al. 2010; Pani and Mahapatra 2013). Although TFs are initially identified by computational approaches, they have also been validated by experimental approaches (Van Dijk et al. 2010) presenting the strength of computational protocols. In this study, an integrative approach has been implemented that involves computational identification of TFs, and their regulatory mechanisms in tissue-specific synthesis of secondary metabolite in R. serpentina. Co-expression network was constructed, with relatively stringent PCC threshold, from statistically significant TF-gene pairs. It has been augmented that genes of interest are co-expressed if their PCC is more than 0.80 and is generally preferred (Chien et al. 2014). However, choosing a strict threshold may lead to loss of significant gene pairs. Therefore, ND offers a reliable measure for choosing calibrated thresholds since at this cutoff, biologically significant modules are expected to be found (Aoki et al. 2007; Mao et al. 2009). Moreover, our approach includes identification of possible TFBS that enhanced the quality and enlarged the dataset used for network analysis. These stringent criteria were used to reduce weak connections followed by false positives among TF-gene pairs by comparing against TF-gene connectivity data and TFBS.

Gene co-expression network analysis has been successfully applied for identifying functional clusters of highly correlated genes in a variety of organisms (Aoki et al. 2007; Mao et al. 2009; Ruan et al. 2010; Gongora-Castillo et al. 2012; Fukushima et al. 2012). Gene-enrichment analysis demonstrated that various biological pathways were under regulation of multiple TF families. To determine functions of TFs, those were not annotated by simple annotation (Table S1), a module-based analysis was performed to annotate them based on interacting transcripts. Identification and characterization of modules is a better choice to understand the coordinated behavior of TFs since each module may occupy certain genes representing specific pathways with similar GO terms. Four statistically and biologically significant modules (1–4) were obtained with similar enrichments that were broadly categorized into functional classes viz., defense response, epigenetic regulations, reproductive system, and upstream of TIA pathway, respectively. Moreover, these modules have shown tissue-specific expression (Fig. S15) that further signifies tissue-specific synthesis and accumulation of metabolites. Functional annotation has identified AP2-EREBP TF family to be involved in regulation of primary metabolic pathway and precursors of TIA pathway, similar to ORCA3, which has been previously reported to regulate these pathways in C. roseus (Van der Fits and Memelink 2000). Additionally, WRKY TF family was also observed to regulate ajmalicine and serpentine synthesis that has been previously reported in C. roseus (Schluttenhofer et al. 2014). Few putative TFs (e.g., SBP) identified, which were not functionally annotated through simple annotation and TF family enrichment analysis, have been inferred based on high degree of transcriptional connectivity using module-based analysis. Therefore, graph theoretical approach implemented presented utility to functionally annotate TFs based on their interacting partners.

TFs were found to work in a highly coordinated manner to perform important biological functions since 38 % of them were highly clustered only in the first four modules. Shared genes among significant modules were found to be associated with representative pathways also highlight coordinated and complex behavior of TF families in each module. This analysis shows that TFs do not perform their functions independently but act as composite regulatory modules (CRMs) where more than one TF family work together to facilitate regulation of gene expression (Matys et al. 2006; Jha et al. 2011). TFs present in modules 1 and 4 were found to be associated with regulation of major indole alkaloids and primary metabolic pathway as well as upstream of TIA pathway with high expression in roots and mature leaves, respectively (Figs. 6 and S15). Similar to known regulatory roles of WRKY and AP2-EREBP TF families in C. roseus, this study has also been proposed to regulate the synthesis of indole alkaloids in R. serpentina.

Various studies have reported the integration of gene expression data with metabolomics data that contributes to identifying significant biological processes and molecular mechanisms more precisely (Hirai et al 2005; Yonekura-Sakakibara et al 2008; Meyer et al 2014). In our analysis, inclusion of metabolomics data has also strengthened the reliability of significant modules because of module-specific regulation of various secondary metabolites (Fig. 8). Statistical analysis inferred that transcripts present in module 1 may be close enough on chromosome to form small physical clusters to regulate synthesis of specialized metabolites as supported from previous studies (Frey et al. 2003; Nützmann and Osbourn 2014). Formation of such functional gene set has already proven to be involved in mainly defense response and expedite parallel activation of genes from physically linked genes (Chu et al. 2011) through genomic rearrangements that accelerate the expression of genes in specific pathway. Similarly, TFs present in module 1 may contribute into the diversity of major indole alkaloids in R. serpentina in response to various internal/external stimuli. Significant modules obtained were biologically conceivable with similar enrichments for transcripts in a particular functional classes and their tissue-specific expression may correlate to synthesis as well as accumulation of secondary metabolites. While limited gene expression data was available for this analysis, further experimental studies would be required to ascertain specific role(s) of putative TFs in R. serpentina.

Conclusion

The study presented herein provides an integrative approach constituting of genome- and system-wide analysis to identify regulatory TFs and associated modules involved in R. serpentina secondary metabolism. Integration of metabolomics data provided an advantage to further filter out the putative TFs involved in module-specific regulation of secondary metabolites. Of all four significant modules, module 1 was the most significant since it was found to be statistically associated with synthesis of specialized indole alkaloids, and transcripts contributing to module may be physically clustered on the genome to involve in their synthesis. TF families in modules 1 and 4 may coordinately regulate synthesis of specialized metabolites under response to stimulus and precursors of TIA pathway, respectively. Moreover, module 4 also acts as connecting link between primary and secondary metabolic pathways. Similar to known regulatory roles of WRKY and AP2-EREBP TF families in C. roseus, the present study also proposed them to regulate the synthesis of major indole alkaloids in R. serpentina. As far as our knowledge, we present the first in silico study to identify TFs and their regulatory role in R. serpentina secondary metabolism. Conclusively, the methodology implemented could help to better understand TF-mediated regulation of genes related to TIAs synthesis. Furthermore, this study may assist to identify and explore functions of transcriptional regulators which in turn may aid to improve quantity and diversity of secondary metabolites, and also provide a reference to other non-model plant systems.