Steroidal glycoalkaloids (SGAs) are cholesterol-derived specialized metabolites produced in species of the Solanaceae. Here, we report that a group of jasmonate-responsive transcription factors of the ETHYLENE RESPONSE FACTOR (ERF) family (JREs) are close homologs of alkaloid regulators in Cathranthus roseus and tobacco, and regulate production of SGAs in tomato. In transgenic tomato, overexpression and dominant suppression of JRE genes caused drastic changes in SGA accumulation and in the expression of genes for metabolic enzymes involved in the multistep pathway leading to SGA biosynthesis, including the upstream mevalonate pathway. Transactivation and DNA–protein binding assays demonstrate that JRE4 activates the transcription of SGA biosynthetic genes by binding to GCC box-like elements in their promoters. These JRE-binding elements occur at significantly higher frequencies in proximal promoter regions of the genes regulated by JRE genes, supporting the conclusion that JREs mediate transcriptional co-ordination of a series of metabolic genes involved in SGA biosynthesis.
Plants have evolved multitudes of defense and adaptation mechanisms to survive in fluctuating environments. To ward off threats imposed by pathogens and pests, plants produce and accumulate toxic substances, including a diverse array of alkaloids, terpenoids and other metabolites with bioactive properties ( Bednarek and Osbourn 2009 ). Although these plant-derived chemicals, or phytochemicals, are widely exploited by humans as valuable compounds, they are often unwanted in food crops because of adverse impacts on human health ( Betz 1999 ). Accordingly, elimination or reduction of such phytochemicals is a critical aim in plant breeding. Complex metabolic pathways, encompassing precursor-supplying primary pathways and downstream specialized pathways, produce defense compounds. Tight regulation, often involving transcriptional co-ordination of structural genes, is required to ensure restrained implementation of these costly chemical defenses, which often impose burdens on growth and development in plants ( Baldwin 1998 ).
Jasmonates (JAs) play central signaling roles in a wide range of plant resistance and developmental responses ( Wasternack and Hause 2013 ), including the elicitation of defense chemical pathways. Indeed, methyl jasmonate (MeJA) has been widely used as an elicitor to induce the production of secondary metabolites in plant culture ( Yukimune et al. 1996 ). Perception of JA signals and the subsequent signal transduction pathway steps, mediated by CORONATINE INSENSITIVE 1, JASMONATE ZIM-DOMAIN proteins (JAZs) and JAZ-interacting transcription factors, have been elucidated through molecular and genetic studies and are conserved among plant species ( Wasternack and Hause 2013 ). However, less is known about the molecular mechanisms linking JA signaling with downstream defense-related metabolism ( De Geyter et al. 2012 ). A group of JA-inducible ETHYLENE RESPONSE FACTOR (ERF) transcription factors, categorized into clade 2 of group IXa ( Nakano et al. 2006 , Shoji et al. 2010 , Shoji et al. 2013 ), have been found to play regulatory roles in JA-induced alkaloid biosynthesis in distinct species. For instance, OCTADECANOID-DERIVATIVE RESPONSIVE CATHARANTHUS 3 (ORCA3) controls the JA-dependent production of monoterpenoid indole alkaloids, including clinically important compounds, in the medicinal species Catharanthus roseus ( van der Fits and Memelink 2000 ). Similarly, the JA-induced formation of ornithine-derived nicotine is regulated by ERF189 in tobacco ( Shoji et al. 2010 ). Both ORCA3 and ERF189 mediate the co-ordinated transcription of a series of metabolic and transport genes ( Shoji et al. 2009 ) involved in the alkaloid pathways by recognizing specific GCC box-like elements found in promoters of the target genes ( van der Fits and Memelink 2000 , Shoji et al. 2010 , Shoji and Hashomoto 2011a ). The JA-inducible expression of ORCA3 and ERF189 is regulated by a homologs of the basic helix–loop–helix (bHLH) family MYC2 transcription factor ( Zhang et al. 2011 , Shoji and Hashimoto 2011b ). MYC2 is a JAZ-interacting factor, involved in the regulation of a wide range of downstream JA responses ( Kazan and Manners 2013 ).
Steroidal glycoalkaloids (SGAs), which occur in certain species of Solanaceae and Lilaceae families, are nitrogen-containing compounds with a glycosylated steroidal backbone derived from cholesterol. Based on their cytotoxic properties, SGAs have been proposed to function in plant defense against biotic threats ( Friedman 2002 , Friedman 2006 ). SGAs are present as toxic and anti-nutritional compounds in inedible parts of solanaceous vegetables, such as immature green fruits of tomato and sprouts and green peels of potato tubers ( Friedman 2002 , Friedman 2006 , Iijima et al. 2013 ). In tomato, α-tomatine, derived from its aglycone tomatidine through elaborate glycosylation steps, is the predominant SGA in green organs, while its less toxic derivatives are found in red ripe fruits ( Iijima et al. 2009 ). After early molecular studies on glycosyltransferases involved in the glycosylation ( Moehs et al. 1997 , Itkin et al. 2011 ), a series of GLYCOALKALOID METABOLISM ( GAME ) genes were identified as responsible for enzymatic steps from cholesterol to SGAs in tomato and potato ( Itkin et al. 2013 ). STEROL SIDE CHAIN REDUCTASE 2 (SSR2) plays a key role in diverting the metabolic flow to cholesterol formation from cycloartenol ( Sawai et al 2014 ). Similar to other metabolic genes that form clusters in plant genomes ( Boycheva et al. 2014 ), GAME genes are clustered in both the tomato and potato genomes ( Itkin et al. 2013 ).
Cárdenas et al. (2016) recently identified five tomato ERF genes phylogenetically related to alkaloid-regulating ORCA3 and ERF189 , and reported that one of them regulates the metabolic genes involved in the biosynthesis of SGAs in tomato and potato ( Cárdenas et al. 2016 ). Here, we report characterization of these genes, which we term Jasmonate-responsive ERF ( JRE ) genes, as well as of a sixth related JRE in tomato. We performed transcript profiling and metabolite analyses of transgenic tomato lines with altered JRE function. We demonstrated that JRE genes play central roles in the transcriptional regulation of pathways leading to SGA formation, from the isoprenoid-supplying mevalonate pathway to GAME-mediated steps after cholesterol. Transactivation and DNA–protein binding studies indicated that JRE4 positively controls the biosynthesis genes by recognizing GCC box-like elements in their promoters. Significant enrichment of putative JRE-binding elements in proximal 5′-flanking regions of JRE -regulated genes, including those involved in SGA biosynthesis, further supports the conclusion of JRE- mediated transcriptional co-ordination of these genes.
Six JRE genes and their expression patterns in tomato
In tomato, there is a gene cluster (spanning about 100 kb) with five clade 2 ERF genes of group IXa on chromosome I ( Cárdenas et al. 2016 ); we found that one additional ERF gene of this clade resides as a singleton on chromosome V ( Fig. 1a ). Since all six ERF genes were inducible by jasmonate (see below), we named them Jasmonate-responsive ERF ( JRE ) 1 (Solyc01g090300), JRE2 (Solyc01g090310), JRE3 (Solyc01g090320), JRE4 (Solyc01g090340) JRE5 (Solyc01g090370) and JRE6 (Solyc05g050790). Note that JRE1 , JRE2 , JRE3 , JRE4 and JRE5 , respectively, correspond to GAME9-like1 , GAME9-like2 , GAME9-like3 , GAME9 and GAME9-like4 reported in a very recent study ( Cárdenas et al. 2016 ).
Based on alignment of amino acid sequences of the conserved DNA-binding domain, a phylogenetic tree including JREs from tomato and related ERF proteins from Arabidopsis, Catharanthus roseus and tobacco was generated to examine the evolutionary relationships among the members ( Fig. 1b ). As defined previously ( Shoji et al. 2013 ), clade 2 is divided into four subgroups. JRE3 and JRE4 were placed in clade 2–2b, and the remaining four JREs were in clade 2–3 ( Fig. 1b ).
We examined the expression patterns of JRE genes along with SGA biosynthetic 3-hydroxy-3-methylglutaryl CoA reductase ( HMGR1 ), sterol 4α-methyl oxidase 2 ( SMO2 ) and GAME1 genes in tomato tissues using quantitative reverse transcription–PCR (qRT–PCR) ( Fig. 2 and numerical values in Supplementary Table S1 ). Both JRE genes and SGA biosynthesis genes were expressed in organs from 7-week-old plants, and no apparent organ specificity was observed ( Fig. 2a ). JRE4 was the most highly expressed JRE in nearly all examined tissues, though in roots and fruits at certain stages the predominance of JRE4 was diminished, mainly due to decreased JRE4 expression in the tissues. The levels of toxic SGAs, such as α-tomatine, as well as their production drastically decrease during fruit ripening in parallel with increased catabolism of toxic SGAs to less toxic forms ( Iijima et al. 2009 ). In accordance with such changes, expression of SGA biosynthesis genes and some members of JRE genes, JRE1 , JRE2 and JRE4 , progressively decreased as fruits matured ( Fig. 2b ). Except for HMGR1 , expression of which clearly declined at the breaker stage, decreases in the SGA biosynthesis and JRE gene expression were most evident during green fruit stages ( Fig. 2b ), when fruits were rapidly enlarging, rather than during later color-changing stages. Expression levels of JRE genes in cultured hairy roots were comparable with or somewhat higher than those in tissues from greenhouse-grown plants, suggesting the usefulness of the cultured material in studies on JRE genes and SGA biosynthesis. MeJA co-ordinately induced SGA biosynthesis genes in tomato hairy roots ( Fig. 2c ). All JRE genes were induced by MeJA treatment in tomato hairy roots, but the induction kinetics in terms of magnitude and timing varied among the members ( Fig. 2c ). JRE4 and JRE6 were gradually induced during the 24 h duration. Acute and strong induction within 30 min followed by a sharp decline was characteristic of JRE1 , JRE2 and JRE3 . Induction of JRE5 peaked at 4 h after the start of the treatment.
Generation of transgenic tomato lines with altered function of JRE genes
To study the effect of altered JRE function on gene expression and metabolism in tomato, stable transgenic tomato lines of plants and hairy roots were generated by Agrobacterium -mediated transformation. Two JRE4 overexpression ( JRE4-OX ) lines of plants (lines OX1 and OX12) were established by introducing JRE4 cDNA under the control of the Cauliflower mosaic virus (CaMV) 35S promoter. JRE4 was chosen as a target gene for overexpression because it is the main JRE expressed in tomato tissues ( Fig. 2 ). In leaves of the JRE4-OX lines, high expression of JRE4 transcript was confirmed; this expression did not significantly change with MeJA treatment ( Supplementary Fig. S1a ). In contrast, MeJA induced the endogenous JRE4 gene in wild-type controls ( Supplementary Fig. S1a ). No visible abnormalities were observed in the JRE4-OX plants ( Supplementary Fig. S1b ).
We chose a dominant suppression strategy to compromise JRE function, since the closely related JRE genes could have overlapping functions. For this purpose, JRE3, JRE4 and JRE5 were fused at their C-termini with the ERF-associated amphiphilic repression (EAR) motif, which dominantly suppresses expression of genes targeted by transcription factors ( Hiratsu et al. 2003 ). The JRE-EAR genes were overexpressed using the CaMV 35S promoter in transgenic tomato hairy roots; two independent lines (lines #1 and #2) for each construct were selected and analyzed. Overexpression of the introduced JRE genes was confirmed in JRE-EAR lines using qRT–PCR ( Supplementary Fig. S2 ). Consistent with different expression levels of the endogenous JRE genes ( Fig. 2 ), the degrees of overexpression relative to vector controls varied among the lines with different fusion constructs. There was significant reduction of off-target JRE gene expression (e.g. JRE1 and JRE2 in all JRE-EAR lines, and JRE6 in JRE3-EAR and JRE5-EAR lines) ( Supplementary Fig. S2 ), suggesting down-regulation of the JRE genes by the JRE-EAR fusions. There were slight variations of growth and morphology among the lines (not shown), but they were within the range normally observed for hairy root cultures.
Transcript profiling in overexpression and dominant suppression lines of JRE genes
To clarify the regulatory function of JRE transcription factors in tomato, we investigated the effects of altered JRE function on the transcriptome to reveal genes targeted by JRE genes. Comparative transcript profiling was carried out using the overexpression and dominant suppression transgenic lines with a custom tomato oligoarray representing >40,000 transcripts ( Ruiu et al. 2015 ). Total RNA from each line was labeled and hybridized to the array. As control samples for comparison, wild-type plants for overexpression and empty vector control lines for suppression were included. To examine the profiles in distinct types of tissues as well as after JA elicitation, leaves of JRE4-OX lines and hairy roots of suppression lines treated with MeJA were used for analysis.
Array oligos representing genes up-regulated in JRE4-OX lines and down-regulated in JRE-EAR lines are listed in Supplementary Tables S2 and S3 , respectively; the list of the genes represented is given in Supplementary Table S4 . A large number of metabolic genes involved in SGA biosynthesis were among the JRE -regulated genes; 24 genes regulated in either JRE4-OX or JRE-EAR lines ( Fig. 3 ; Supplementary Table S5 ). All genes, except two (Solyc05g055760 for IDI and Solyc02g069490 for SSR2 ), whether they satisfied both listing criteria ( Rox >5.0, Q <0.85) or not, showed trends of signal increases in JRE4-OX lines as well as decreases in JRE-EAR lines. The JRE -regulated metabolic genes were involved in nearly all branches of the pathway leading to SGAs from the upstream mevalonate pathway to cholesterol biosynthesis and downstream aglycone formation and glycosylation ( Fig. 3 ). The extent of regulation generally seemed greater for later steps, especially those after cholesterol ( Fig. 3 ); GAME genes were especially responsive to alterations of JRE function ( Supplementary Tables S2, S3 ). All of the biosynthesis genes were similarly suppressed by JRE3-EAR , JRE4-EAR and JRE5-EAR ; this tendency was corroborated by the unbiased distribution of probes for SGA biosynthesis genes ( Supplementary Fig. S3 ).
To validate the results of microarray analyses, transcript levels of SGA biosynthesis genes were analyzed in JRE4-OX and JRE4-EAR lines by qRT–PCR. Increased expression of the genes, except acetyl-CoA C-acetyltransferase ( ACAA ), was observed in leaves of JRE4-OX lines relative to wild-type controls in both MeJA-treated and mock-treated conditions ( Fig. 4a ). Consistent with the microarray analysis results, the degree of up-regulation in JRE4-OX lines was greater for genes involved in later parts of the pathway ( Fig. 4a ). We also found clear up-regulation of SMO2 and GAME1 but not HMGR1 of the mevalonate pathway in roots from JRE4-OX plants ( Supplementary Fig. S4 ). In transgenic hairy roots expressing JRE4-EAR , all of the examined SGA biosynthesis genes were suppressed to 28–70% relative to the controls ( Fig. 4b ).
Metabolic impact of altered JRE4 function
To clarify how altered JRE4 function affects SGA-related metabolism in tomato, we examined metabolite levels in leaves from JRE4-OX plants exposed to MeJA vapor ( Fig. 5a ) and hairy roots of a JRE4-EAR line treated with MeJA ( Fig. 5b ). SGAs, including the predominant α-tomatine, were extracted and measured using liquid chromatography–quadrupole time-of-flight–mass spectrometry (LC-QTOF-MS), whereas more hydrophilic substances, including pathway intermediates and other sterols and triperpenoids, were analyzed using gas chromatography–mass spectrometry (GC-MS). In the JRE4-OX line (line OX1), the level of α-tomatine increased 2.1-fold relative to wild-type controls, while in the JRE4-EAR line (line #1), clear reductions of α-tomatine to 47% the level of controls were observed. Similar changes were observed for other SGAs ( Supplementary Table S6 ; Supplementary Fig. S5 ). Cholesterol, a sterol precursor of SGAs, increased 1.5-fold in the JRE4-EAR line. The phytosterols campesterol and stigmasterol did not show any significant changes except for a 32% decrease of stigmasterol in the JRE4-OX line ( Fig. 5 ). Cycloartenol and lanosterol, the first tetracyclic triperpenoid intermediates, were markedly decreased in the JRE4-OX line, to 9% and 15% of levels in the control, respectively, while a 3.3-fold increase of cycloartenol and a 2.9-fold increase of lanosterol were observed in the JRE4-EAR line. Triterpenoids, namely α-amyrin, β-amyrin and lupeol, were increased 1.3- to 1.5-fold in the JRE4-OX line and decreased to 14–80% of the control levels in the JRE4-EAR line. Squalene decreased to 48% of the level in the control in the JRE4-OX line and did not significantly change in the JRE4-EAR line.
JRE4 activates transcription of SGA biosynthesis genes by binding to GCC box-like promoter elements
To examine whether JRE4 activates the transcription of SGA biosynthesis genes in vivo through their 5′ -flanking regions, transient transactivation assays were performed in tomato fruits using Agrobacterium -mediated infection, or agroinjection, for gene delivery. For transient expression, the β-glucuronidase ( GUS ) reporter was placed downstream of the 5′-flanking region of sterol reductase ( DWF5 ) (−1,500 to − 1; counted from the first ATG) or GAME4 (−1,500 to − 1) ( Fig. 6a ). The individual reporter constructs and a CaMV 35S promoter-driven JRE4 effector were co-delivered into tomato fruits by agroinjection along with a CaMV 35S promoter-driven green fluorescent protein ( GFP ) reference construct. The expression levels of the GUS reporter genes were analyzed by qRT–PCR and normalized to that of the reference GFP gene ( Fig. 6b ). DWF5 and GAME4 promoter-driven GUS reporter genes were up-regulated 3.8- and 8.2-fold by JRE4 overexpression, respectively, indicating that the JRE4 -mediated gene activation was dependent on regions included in the reporter constructs. The 1,500 bp region of DWF5 could be trimmed to a relatively short span (−285 to − 1) without losing reporter responsiveness (10.6-fold) ( Fig. 6b ) and basal activity, indicating the functional importance of this proximal region.
JRE4 recognizes two structurally related elements, the GCC box-like P box and the GCC box ( Shoji et al. 2013 ; Supplementary Table S7 ). Accordingly, we used in silico analysis to identify P box and GCC box elements within the 1,500 bp 5′-flanking regions of DWF5 and GAME4 . Using a cut-off score of 7.0, two P boxes (named D1 and D2) in DWF5 and one P box (named G1) encompassing a GCC box in GAME4 were predicted, while no GCC box, other than the one included in G1, was found. Both D1 and D2 were present within the short functional region (−285 to − 1) of DWF5 .
To clarify the role of the predicted elements in the JRE4 -dependent reporter activation, nucleotides within the elements were substituted ( Fig. 6a ) and the resultant mutant reporters were subjected to transient transactivation assays ( Fig. 6b ). For DWF5 , mutations in D1 abolished the JRE4 -dependent activation of the reporter driven by the short region (−285 to − 11). Mutations in D2, which resides in the 5′-untranslated region rather than the promoter, did not have major influences on the induction, indicating the requirement for a functional D1 element but not for D2. GAME4 promoter activation did not change much with mutations only in G1, suggesting the presence of additional elements indispensable for the promoter induction. Based on such an assumption, we lowered the cut-off value in the analysis, and found the G2 element (−1,111 to − 1,102) with a score of 6.1 as a predicted P box ( Fig. 6a ). Mutations in G2 alone did not change activation, similar to those in G1 ( Fig. 6b ). However, when the GAME4 promoter construct carried mutations at both G1 and G2, it was no longer activated by the JRE4 effector, pointing to the involvement of functionally redundant G1 and G2 in the GAME4 activation.
To validate in vitro binding of the elements to JRE4 proteins, electrophoretic mobility shift assay (EMSA) was carried out with oligonucleotide probes based on sequences of the elements ( Supplementary Table S10 ). When incubated with recombinant fusion proteins of truncated JRE4 (corresponding to amino acids 40–219) tagged with N-terminal thioredoxin and other short tags, DNA–protein complexes were detected as intense shifted triplet bands for D2 and G1 probes ( Fig. 6c ). Similar patterns of shifted bands were observed for D1 and G2, albeit with lower intensity than those for the other two probes. Interestingly, the intensities of the shifted bands corresponded to the in silico prediction scores for the elements. The shifted bands were completely abolished when mutated probes ( Fig. 6a ) were used, confirming the specificity of binding of the wild-type probes.
Putative JRE-binding elements found in proximal promoter regions of JRE -regulated genes
The JRE4-binding elements found in proximal promoter regions ( Fig. 6 ) prompted us to examine whether genes with such JRE-binding elements in promoter regions were enriched among JRE -regulated genes. In addition to the P and GCC boxes targeted by clade 2–2b JRE4, a related CS1 box, recognized by clade 2–3 JREs ( Fig. 7a ), was included in the analysis. All of the examined boxes were represented as weighted matrices ( Supplementary Table S7 ). We computationally searched for P, GCC and CS1 boxes with a cut-off score of 7.0 in promoter regions (up to − 800; counted from the first ATG) of JRE -regulated genes (group R) (180 genes in Supplementary Table S4 ) and of JRE -regulated SGA biosynthesis genes (group SR) (24 genes in Fig. 3 ), a subset of group R genes. The corresponding regions of all protein-coding genes annotated in a tomato reference genome (group W) (34,725 genes) were analyzed as controls. The genes with the JRE-binding elements in each group were counted. The values even for group W were slightly variable among examined regions, possibly reflecting biased GC contents or other genomic features in promoter regions. We detected enrichment of genes with a P box in the −200 to − 100 region for both R (2.1-fold) and SR groups (5.8-fold) and of genes with a CS1 box in the −300 to − 1 region for group R (2.2- to 2.7-fold) and in the −300 to − 200 region for group SR (5.7-fold).
To examine whether sequences related to P, GCC and CS1 boxes could be retrieved in non-targeted analysis, the 5′-flanking sequences (−1,500 to − 1) of group R and SR genes were subjected to MEME (Multiple EM for Motif Elicitation) analysis ( Bailey et al. 2006 ), retrieving the sequences shared among the queries. Multiple sequences with similarity to the JRE-binding boxes were retrieved ( Supplementary Table S8 ); match scores representing the similarities were calculated using position-specific probability matrices for P, GCC and CS1 boxes based on sequences retrieved from the 5′-flanking regions (−300 to − 1) of group R genes ( Supplementary Table S7 ).
Impacts of altered JRE function on SGA biosynthesis
Here, we used transgenic approaches, involving overexpression and dominant suppression, to elucidate the regulatory functions of tomato JRE transcription factors, which are closely related to alkaloid-regulating ORCA3 from C. roseus ( van der Fits and Memelink 2000 ) and ERF189 from tobacco ( Shoji et al. 2010 ). Based on microarray analyses, we identified a large number of JRE -regulated metabolic genes involved in SGA biosynthesis, including all clustered GAME genes except GAME2 , constituting a core pathway downstream of cholesterol ( Itkin et al. 2011 , Itkin et al. 2013 ), genes for cholesterol biosynthesis, including SSR2 ( Sawai et al. 2014 ) and flux-controlling HMGR ( Narita and Gruissem 1989 ), and others involved in the mevalonate pathway ( Fig. 2 ). Our results demonstrate the existence of JRE -mediated transcriptional regulation of the entire SGA pathway in tomato. This regulation of the upstream pathways, as far up as the isoprenoid-producing branch, may be required to meet the metabolic demands for downstream SGA production without disturbing homeostasis of other metabolites derived from the highly branched terpenoid pathways. The relatively small changes in the accumulation of the essential phytosterols ( Boutté and Grebe 2009 ) campesterol and stigmasterol in the transgenic lines support this conclusion ( Fig. 4 ). Although the overall pathway to SGA biosynthesis was generally co-ordinated, expression levels of genes involved in later steps, such as GAME genes, changed much more in the transgenic lines than did the upstream genes ( Figs. 3, 4a ; Supplementary Fig. S4 ; Supplementary Tables S2, S3, S5 ). Such differential regulation between distinct parts of the pathway was corroborated with the changes in the levels of metabolites; changes in SGAs and upstream intermediates, cholesterol, cycloartenol, lanosterol and squalene, showed opposite trends in both the overexpression and the suppression lines ( Fig. 5 ). This difference in accumulation presumably reflects imbalances between the early and late parts of the pathway. To understand the metabolic changes of triterpenoids, α-amyrin, β-amyrin and lupeol, which showed a trend opposite to that of cycloartenol and lanosterol ( Fig. 5 ), it might be useful to examine the regulation of oxidosqualene cyclase genes, none of which was identified as a regulated gene in our microarray analysis. Of course, to better understand the metabolic impacts of JRE4 overexpression, we need to examine not only the expression of the metabolic genes at the transcript level but also activities of the involved enzymes and metabolic flux reflecting them.
Our microarray-based approach to screening genes involved in a JRE -controlled regulon not only points to known structural genes ( Fig. 3 ) but also helps mine novel metabolic and transport genes required for complex SGA-related metabolism. In this regard, it is intriguing that many uncharacterized genes annotated as encoding glucosyltransferases, Cyt P450 enzymes and peptide transporters were included in the list of JRE -regulated genes ( Supplementary Table S4 ). Similar transgenic approaches combined with transcript profiling in other species producing SGAs are considered useful to elucidate the molecular bases of chemically diverse metabolites of this group ( Friedman 2002 , Friedman 2006 , Iijima et al 2013 ). It should be noted that Cárdenas et al. (2016) recently reported that potato GAME9, an ortholog of tomato GAME9/JRE4, regulates the biosynthetic genes of the SGAs chaconine and solanine ( Cárdenas et al. 2016 ). Our identification of a series of JRE -regulated metabolic genes in tomato provides new insights on cholesterol formation, which has been relatively unexplored in plants compared with other organisms ( Sawai et al. 2014 ).
To understand this regulation in the context of a greater metabolic network, we will need to address the co-ordination of the JRE -mediated transcriptional regulation with other mechanisms operating at the transcriptional and post-transcriptional levels ( Pollier et al. 2013 , Mertens et al. 2015 , Van Moerkercke et al. 2015 ). As discussed below, it is still an open question whether all parts of the pathway are similarly subjected to the transcriptional regulation by JRE genes. In addition to structural genes, molecular factors important for the regulation of the pathway could be included in the JRE -regulated genes. Notably, among the genes identified as regulated by JRE genes ( Supplementary Table S4 ) was a gene encoding a RING-finger E3 ubiquitin ligase, a homolog of MAKIBISHI1 from Medicago truncatula that controls HMGR enzyme activity ( Pollier et al. 2013 ). Studies of tomato JRE genes and GAME9 ( Cárdenas 2016 ; this study), along with recent studies on a group of related bHLH transcription factors ( Van Moerkercke et al. 2015 , Mertens et al. 2015 ), open a new chapter in the study of the regulation of terpenoid pathways, which had remained unexplored until recently.
Transcriptional regulation of SGA biosynthesis genes
JRE4 directly activates the transcription of DWF5 and GAME4 genes by recognizing the GCC-like elements in their promoter regions ( Fig. 6 ). Two of the three functional elements found in DWF5 and GAME4 are present in similarly situated proximal regions (−248 to − 219 relative to the ATG) ( Fig. 6a ). This finding is consistent with previous reports on ORCA3- and ERF189-recognizing elements present in similar regions of the targeted genes ( Van der Fits and Memelink 2000 , Shoji et al. 2010 , Shoji and Hashimoto 2011a ) and also with the computational predictions of the elements in the JRE -regulated genes ( Fig. 7 ), suggesting a common mechanistic feature among transcription factors of this group. Our results for GAME4 , which is present in a two-gene metabolic cluster ( Fig. 6 ), point to a role for promoter-binding transcription factors in the regulation of such clusters, possibly in addition to their proposed regulation at the chromosomal level ( Wegel et al. 2009 ).
It remains to be addressed whether all JRE -regulated SGA biosynthesis genes, including clustered GAME genes other than GAME4 , are regulated directly by JRE genes in a similar manner to that demonstrated for DWF5 and GAME4. Considering the large number of regulated steps and the differential regulation we observed, we cannot exclude the possibility of the involvement of additional mechanisms, such as indirect regulation through other transcription factors or metabolite-mediated feedback regulation. As demonstrated for regulation of nicotine biosynthesis genes by ERF189 and the JAZ-interacting bHLH transcription factor MYC2 ( Shoji and Hashimoto 2011b ), it is also plausible that JREs regulate the downstream genes in co-operation with other transcription factors. In this regard, it is interesting that both MYC2 and GAME4/JRE4 were required for transactivation of promoters of SGA biosynthesis genes in tobacco cells ( Cárdenas et al. 2016 ). Nevertheless, the significant enrichment among the JRE -regulated genes of those bearing a P or CS1 box in the proximal promoter ( Fig. 7 ) and the complementary results of the MEME analysis ( Supplementary Table S8 ) suggest that JRE4, and possibly other JREs, including clade 2–3 members that can recognize the CS1 box, participate in the transcriptional regulation of many, but not necessarily all, of the genes by binding to the predicted elements. The frequent occurrence of the JRE-binding elements in the regulated genes supports the notion that genes acting downstream are recruited into regulons under the control of transcription factors through acquisition of functional cis -elements in the appropriate promoter regions ( Shoji and Hashimoto 2011a , Moghe and Last 2015 ).
The results of our promoter-related analyses ( Figs. 6, 7 ; Supplementary Table S8 ) point to the possible importance of GCC-like P and CS1 boxes rather than the canonical GCC box for JRE -mediated regulation, although all JREs presumably have substantial abilities to bind to GCC boxes in vitro ( Fig. 7a , Shoji et al. 2013 ). As proposed for tobacco ERF189, which exclusively targets the P box ( Shoji et al. 2013 ), such a preference for the GCC-like box of JREs may allow the JRE -controlled regulon to be free from influence by the large number of GCC box-recognizing ERFs.
Regulatory function of multiple JRE genes
Gene clustering is common to JRE genes ( Fig. 1a ) and related ERF genes ( Shoji et al. 2013 ); in tobacco, ERF189 is clustered with related genes on a nicotine-controlling NIC2 locus, and the NIC2 -locus cluster was found to be deleted in a low-nicotine mutant ( Shoji et al. 2010 ), while ORCA3 was found to reside on the same genomic contig with a similar gene in C. roseus ( Kellner et al. 2015 ). When overexpressed in a low-nicotine mutant, ERF189 recovered nicotine accumulation to the wild-type levels, and thus ERF189 has been considered to work most effectively as a regulator of nicotine biosynthesis among the clustered ERF genes ( Shoji et al. 2010 ). Similar to the NIC2 -locus cluster in tobacco, the tomato JRE cluster includes members of different clades, namely clades 2–2b and 2–3 ( Fig. 1 ). Dominant suppression of clade 2–2b JRE3 , 2–2b JRE4 or 2–3 JRE5 similarly repressed the expression of SGA biosynthesis genes ( Fig. 3 ; Supplementary Fig. S3 ), suggesting overlapping functions of the three JRE genes. Of course, the effects of ectopic expression of the dominant repressive forms of these proteins should be interpreted carefully. The involvement of clade 2–3 JRE genes in SGA regulation is also supported by the frequent occurrences of CS1 boxes in promoters of the regulated genes. In addition to overexpression ( Fig. 3 ; Supplementary Fig. S4 ; Supplementary Table S2 ) and promoter binding ( Fig. 6 ) analyses as performed herein for JRE4 , knock-out or knock-down experiments for individual members would be required to confirm the in planta contributions of each JRE gene to SGA regulation.
The gene expression patterns strongly support a role for JRE4 in SGA regulation ( Fig. 2 ). JRE4 is expressed the most highly among JRE genes at the transcript level and, as pointed out in Itkin et al. (2013) , its expression is clearly co-ordinated with SGA biosynthesis genes in various samples ( Fig. 2 ). In fruits, progressive decreases of expression were evident for JRE4 , SMO2 and GAME1 during the green fruit stages ( Fig. 2b ), indicating that SGA formation mainly declines during the green stages rather than later color-changing stages ( Iijima et al. 2009 ). In hairy roots, all JRE genes were clearly induced by MeJA treatment, but their induction patterns were variable between the members ( Fig. 2c ). Again, the gradual induction of JRE4 paralleled those of SGA biosynthesis genes during the 24 h duration of treatment. According to the co-expression analysis and other evidence, we can infer that one or a few select members, such as tomato JRE4 , tobacco ERF189 and possibly C. roseus ORCA3 , play a predominant role in regulation of targeted metabolic pathways in each species. We need to understand further the functional redundancy and divergence among the multiple ERF members to address why these ERF genes are maintained in a form of gene clusters during plant evolution.
SGAs as defense chemicals in tomato
Plants usually adapt particular classes of metabolites for chemical defense. A variety of compounds, including SGAs, methyl ketones and sesquiterpenes, are considered to mediate the herbivore resistance in Solanum species ( Antonious et al. 2014 ). SGAs are a group of bioactive compounds with abilities to bind to cholesterol, disrupt cellular membranes and inhibit cholinesterases ( Friedman 2015 ). Based on their toxic and pharmacological properties, SGAs have been proposed to be involved in plant host resistance against a wide range of biotic agents, such as bacteria, fungi, viruses, insects and animals ( Friedman 2002 , Friedman 2006 ). A series of genes encoding biosynthetic enzymes and transcription factors involved in SGA biosynthesis were induced by MeJA treatment in tomato tissues ( Figs. 2c, 4a ; Supplementary Figs. S1, S4 ). The JA-induced expression of the genes and involvement of tomato homologs of ORCA3 and ERF189 in their induction underline the committed roles of SGAs and JA signaling in induced chemical defense against herbivores in tomato, as in the case of nicotine in tobacco ( Baldwin 1998 , Shoji and Hashimoto 2013 ). Although induced significantly after elicitation, SGAs and nicotine are substantially produced even at basal levels and the amounts of SGA and nicotine accumulation in the tissues seem to be in a similar range (in the order of mg g −1 DW), indicating the similarities of these two alkaloid groups with analogous regulatory mechanisms. Drastic declines of expression of SGA biosynthetic and regulatory JRE genes during early fruit development in tomato ( Fig. 2c ) are an example of developmental regulation of defense chemical pathways, which may operate more generally, ensuring the removal of toxic substances from seed-bearing mature fruits to allow seed dispersal assisted by fruit-eating herbivores.
One of the main goals in tomato and potato breeding is the removal of toxic and anti-nutritional SGAs, which are not critically required for plant protection during normal cultivation and occasionally cause poisoning. The removal of SGAs becomes critical when considering the introduction of desirable genetic traits into cultivated species from wild counterparts that usually produce SGAs at higher levels ( Iijima et al. 2013 ). Our identification of the transcriptional regulators of the SGA pathway provides a promising molecular tool applicable for the generation of the crops with low SGA contents.
Materials and Methods
Plant growth, transformation and treatment
Sterilized seeds of tomato, Solanum lycopersicum cv. Micro-Tom, were germinated and grown to seedlings on half-strength Gamborg B5 medium solidified with 0.6% (w/v) agar and supplemented with 2% (w/v) sucrose. Two-week-old seedlings were transferred onto soil in pots and grown to maturity in the greenhouse.
The coding region of JRE4 was cloned into the Bam HI and Sac I sites on pBI121 to generate a binary vector p35S::JRE4 for overexpression. The p35S::JRE4 vector was introduced into Agrobacterium tumefaciens strain GV2260 by electroporation. Agrobacterium -mediated transformation to generate transgenic tomato plants was done according to Sun et al. (2006) . Shoots were selected on solidified Murashige and Skoog medium containing 100 mg l −1 kanamycin. Diploid individuals were screened in the T 0 generation. Transgenic plants of the T 3 generation were analyzed.
For gene expression analyses, leaves and roots from 4-week-old plantlets grown in a greenhouse were submerged in B5 medium with 100 μM MeJA and incubated for 24 h in the dark. Four individual 7-week-old plants were placed in air-tight plastic containers (approximately 81 liters in volume), with a cotton sheet soaked with 5 ml of 100 μM MeJA. The cotton sheet was replaced with a newly soaked one every day during the duration of treatment. Leaves detached from plants exposed to MeJA vapor for 4 d were used for metabolite analysis.
To generate the dominant suppression vectors, coding sequences of JRE3 , JRE4 and JRE5 were amplified by PCR with primers including the restriction sites and a sequence for the EAR motif (5′-CGGCCGCTTGATTTGGATCTTGAACTCAGACTTGGATTTGCTTA-3′; encoding LDLDLELRLGFA; Hiratsu et al., 2003 ) and inserted into the Xba I and Sac I sites of pBI121. To generate transgenic hairy roots, tomato hypocotyls from 7-day-old seedlings were infected with Agrobacterium rhizogenes strain ATCC15834 harboring a binary vector by briefly touching one end of a hypocotyl segment (1.5 cm in length) to a bacterial colony and then standing the segments on the same agar medium with the contacted end up. Hairy roots emerging from infected sites were excised and subcultured twice every week on solidified B5 medium containing 300 mg l −1 cefotaxime for disinfection and 50 mg l −1 kanamycin for drug resistance selection. The selected lines were maintained by subculturing every week in 125 ml glass flasks filled with 25 ml of liquid B5 medium supplemented with 2% (w/v) sucrose with shaking at 100 r.p.m. in the dark. MeJA was directly added to 4-day-old cultures to a final concentration of 100 μM.
cDNA microarray analysis
Total RNA was isolated from leaves treated with MeJA for 24 h and hairy roots treated with MeJA for 24 h using an RNeasy kit (Qiagen). RNA integrity was checked with an Agilent 2100 Bioanalyzer (Agilent). Total RNA (500 ng) was used to generate Cyanine 3-labeled cRNA probes using a Quick Amp Labeling Kit, One-Color (Agilent). A tomato custom oligoarray with 60-mer probes of >40,000 sequences, designed using transcript sequences of the Tomato Gene Index Version 11 (LeGI v.11; http://www.danafarber.org/ ), was hybridized with the labeled samples and scanned, and data were captured and processed as described ( Ruiu et al. 2015 ). Due to poor labeling, hybridization was cancelled for the sample from JRE4-EAR line #2.
Probes with low signal intensity (averages for the two controls <0.2) and variable intensities between lines (differences between the two controls >2.5-fold) were excluded from the analysis. Values relative to the controls were obtained by pairwise comparisons and averaged for each construct; values were defined as Rox , Rj3 , Rj4 and Rj5 derived from data for JRE4-OX , JRE3-EAR , JRE4-EAR and JRE5-EAR lines, respectively. For the overexpression experiment, probes with Rox >5 are listed in Supplementary Table S2 . For the suppression experiment, Q was defined as Q2 = Rj32 + Rj42 + Rj52 , and probes with Q < 0.85 are listed in Supplementary Table S3 .
Total RNA was isolated from plant samples that had been ground in liquid nitrogen using an RNeasy kit (Qiagen) and then converted to first-strand cDNA using ReverTra Ace qPCR RT Master Mix (Toyobo) with oligo(dT) primer. The cDNA templates were amplified using a LightCycler 96 (Roche) with SYBR Premix Ex Taq (Takara) according to Shoji et al. (2010) . The primer sequences are given in Supplementary Table S9 . EF1α (Solyc05g005060) was used as a reference gene. Each assay was repeated at least three times. Based on amplifications from equal molar quantities of cloned amplicons, amplifications from different primer pairs were normalized.
For measurement of SGAs, freeze-dried samples (2 mg) were extracted with 250 μl (for hairy root samples) or 500 μl (for leaf samples) of 80% (v/v) methanol containing 2.5 μM lidocaine and 2.5 μM 10-camphor sulfonic acid using a mixer mill with zirconia beads for 7 min at 18 Hz and 4 °C. After centrifugation for 10 min at 12,000× g , the supernatant was filtered using an HLB μElution plate (Waters). The extracts (1 μl) were analyzed using LC-QTOF-MS (LC, Waters Acquity UPLC system; MS, Waters Xevo G2 Q-Tof). Positive ion mode was used and analytical conditions were as follows: LC column, acquity bridged ethyl hybrid (BEH) C18 (1.7 μm, 2.1 mm×100 mm, Waters); solvent system, solvent A (water including 0.1% formic acid) and solvent B (acetonitrile including 0.1% formic acid); gradient program, 99.5% A/0.5% B at 0 min, 99.5% A/0.5% B at 0.1 min, 20% A/80% B at 10 min, 0.5% A/99.5% B at 10.1 min, 0.5% A/99.5% B at 12.0 min, 99.5% A/0.5% B at 12.1 min and 99.5% A/0.5% B at 15.0 min; flow rate, 0.3 ml min −1 at 0 min, 0.3 ml min −1 at 10 min, 0.4 ml min −1 at 10.1 min, 0.4 min min −1 at 14.4 min and 0.3 ml min −1 at 14.5 min; column temperature, 40°C; MS detection: capillary voltage, +3.0 keV, cone voltage, 25.0 V, source temperature, 120°C, desolvation temperature, 450°C, cone gas flow, 50 l h −1 ; desolvation gas flow, 800 l h −1 ; collision energy, 6 V; mass range, m/z 50–1,500; scan duration, 0.1 s; inter-scan delay, 0.014 s; data acquisition, centroid mode; Lockspray (leucine enkephalin): scan duration, 1.0 s; inter-scan delay, 0.1 s. MS/MS data were acquired in ramp mode using the following analytical conditions: (i) MS: mass range, m/z 50–1,500; scan duration, 0.1 s; inter-scan delay, 0.014 s; data acquisition, centroid mode; polarity, positive/negative; and (ii) MS/MS: mass range, m/z 50–1,500; scan duration, 0.02 s; inter-scan delay, 0.014 s; data acquisition, centroid mode; polarity, positive/negative collision energy, ramped from 10 to 50 V. In this mode, MS/MS spectra of the top 10 ions (>1,000 counts) in an MS scan were automatically obtained. If the ion intensity was <1,000, MS/MS data acquisition was not performed and moved to the next top 10 ions. Chemical assignment of SGAs was performed using the MS/MS spectra reported in Itkin et al. (2011) . All SGA levels were calculated based on a calibration curve of α-tomatine (Tokyo Chemical Industry Co., Ltd.), assuming the same molar responses of SGAs.
For measurement of triterpenes, the extraction of plant tissues with added standards ([25,26,26,26,27,27,27- 2 H 7 ]cholesterol (98% D, Cambridge Isotope Laboratories, Inc.) or synthesized [3,28,28,28- 2 H 4 ]β-amyrin, [28,28,28- 2 H 3 ]α-amyrin and 28,28,28- 2 H 3 ]lupeol; Ohyama et al. 2007 ) was carried out using the method previously described ( Tsukagoshi et al. 2016 ). Quantification of triterpenes except for lanosterol and squalene using GC-MS analysis was performed as described previously (for sterols, Choi et al, 2014 ; for triterpenols, Ohyama et al. 2007 ). Lanosterol amounts were calculated using the peak area ratio of the fragment ion ( m/z : 393) of trimethylsilylated lanosterol and that ( m/z : 336) of the standard (TMS derivative of the labeled cholesterol). Quantification of squalene was performed using the standard calibration curve with coefficients of determination, r2 > 0.9996. The curves were constructed using the peak area value of TIC (total ion chromatogram).
The 5′-flanking regions of DWF5 and GAME5 were amplified by PCR from tomato genomic DNA with primers attached to the restriction sites and cloned into the Hin dIII and Bam HI sites on pBI121 to generate the GUS reporter constructs. For mutant reporter constructs, before cloning the fragments into pBI121, PCR-based mutagenesis ( Hemsley et al. 1989 ) was used to substitute the nucleotides in the cloned promoter sequences; nucleotide substitutions introduced are shown in Fig. 6a . pGBW17 was used as the empty vector control for the effector plasmids. The binary vectors, p19s for P19 silencing suppressor ( Voinnet et al. 2003 ) and p35S-GFP for the GFP reference gene, were used.
Transient gene expression was performed in tomato fruits according to an agroinjection protocol ( Orzaez et al. 2006 ). Briefly, Agrobacteriun tumefaciens strain EHA105 with a binary vector was grown overnight in YEB medium supplemented with 20 μM acetosyringone and appropriate antibiotics, and recovered by centrifugation at 4,000× g for 20 min. The cells were re-suspended in infiltration buffer (10 mM MES, 10 mM MgCl 2 , 200 μM acetosyringone, pH 5.7) by adjusting the optical density at 600 nm to 0.25, and then incubated for at least 2 h at room temperature in the dark. The bacterial suspensions for reporter and effector vectors plus those for p19s and p35S-GFP vectors were combined, and the resultant solution (200–300 μl per fruit) was injected into mature green fruits (1–1.5 cm in size) using a 1 ml plastic syringe attached to a 27-gauge needle. Gene expression was analyzed by qRT–PCR using fruits harvested 3 d after the injection.
Bacterial expression and purification of recombinant fusion proteins of JRE4, called Sl1g90340 in Shoji et al. (2013) , were done as described ( Shoji et al. 2013 ). In the pET32b-based expression vector, a portion of JRE4 (corresponding to amino acid residues 40–219) was placed downstream of a sequence for N-terminal tags, a thioredoxin, an S-tag and a His-tag ( Shoji et al. 2013 ). Proteins comprising only the tag portion from empty pET32b were purified and used as the controls.
Sequences of sense and antisense oligonucleotides used are given in Supplementary Table S10 . Probe preparation, DNA–protein binding assays, gel separation and detection of the reaction products were carried out as described ( Shoji et al. 2010 ).
We used Regulatory Sequence Analysis Tools (RSAT; Turatsinze et al. 2008 , http://rsat.ulb.ac.be/rsat/ ) to search for putative JRE-binding elements in the query genomic sequences by weight matrix scoring. Weight matrices for the P, GCC and CS1 boxes ( Shoji et al. 2013 ) are given in Supplementary Table S7 . The default settings were used for all parameters. Elements with scores higher than 7.0 were retained for analysis, along with element G2, which had a score of 6.1 in GAME4 . Genes with putative JRE-binding elements in the examined regions were counted. Significant differences of the values compared with those of group W including all tomato protein-coding genes was determined by one-sided Fisher’s exact test (α = 0.05) using the Fisher test function of R (V. 3.2.2).
Based on sequences predicted to be the JRE-binding elements by RSAT in the regions (−300 to − 1) of group R genes, position-specific probability matrices for P, GCC and CS1 boxes ( Supplementary Table S7 ) were generated by MEME software (v. 4. 10. 2., Bailey et al. 2006 ) with option settings; ‘2’ for ‘-nmotif’, ‘10’ for ‘-minw’ and applying ‘-revcomp’. To retrieve the sequences commonly found in the queries, MEME was used with option settings; ‘10’ for ‘-nmotif’, ‘7’ for ‘-minw’ and applying ‘-revcomp’. Match scores representing the similarities between a pair of the retrieved sequence and the JRE-binding box (P, GCC or CS1) were calculated using position-specific probability matrices for the pairs. When a pair of sequences with lengths of L 1 and L 2 were aligned with the overlapping length of L 3 , the alignment length Le was defined as Le = L 1 + L 2 – L 3 . Differences between 1 and normalized vector distances were summed for overlapping positions and divided by Le to give scores for the alignments. The alignment scores were calculated for all possible alignments with no gap in both orientations for a sequence pair, and the highest alignment score was adapted as a match score for the pair.
Supplementary data are available at PCP online.
This work was supported by the Japan Society for the Promotion of Science [a Grant-in-Aid for Scientific Research (C) (grant No. 26440144 to T.S.)]; the Plant Transgenic Design Initiative, Gene Research Center, University of Tsukuba [a Cooperative Research Grant]; RIKEN Plant Transformation Network; the Japan Advanced Plant Science Network.
We thank Drs. Tsuyoshi Nakagawa and Yoshinori Yagi for providing pGBW17 and p35S-GFP , respectively.
The authors have no conflicts of interest to declare.
Cauliflower mosaic virus
ERF-associated amphiphilic repression
electrophoretic mobility shift assay
ethylene response factor
gas chromatography–mass spectrometry
green fluorescent protein
jasmonate ZIM-domain protein
liquid chromatography–quadrupole time-of-flight–mass spectrometry
octadecanoid-derivative responsive Catharanthus 3
quantitative reverse transcription–PCR