Summary
Plants produce several hundreds of thousands of secondary metabolites that are important for adaptation to various environmental conditions. Although different groups of secondary metabolites are synthesized through unique biosynthetic pathways, plants must orchestrate their production simultaneously.
Phenylpropanoids and glucosinolates are two classes of secondary metabolites that are synthesized through apparently independent biosynthetic pathways. Genetic evidence has revealed that the accumulation of glucosinolate intermediates limits phenylpropanoid production in a Mediator Subunit 5 (MED5) dependent manner.
To elucidate the molecular mechanism underlying this process, we analyzed the transcriptomes of a suite of glucosinolate-deficient mutants using RNAseq and identified mis-regulated genes that are rescued by the disruption of MED5. The expression of a group of Kelch Domain F-Box genes (KFBs) that function in PAL degradation is affected in glucosinolate biosynthesis mutants and the disruption of these KFBs restores phenylpropanoid deficiency, dwarfism and sterility in the mutants.
Our study suggests that glucosinolate/phenylpropanoid metabolic crosstalk involves the transcriptional regulation of KFB genes that initiate the degradation of the enzyme phenylalanine ammonia-lyase, which catalyzes the first step of the phenylpropanoid biosynthesis pathway. Nevertheless, KFB mutant plants remain partially sensitive to glucosinolate pathway mutations, suggesting that other mechanisms that link the two pathways also exist.
Introduction
Phenylpropanoids are a group of metabolites made primarily from the amino acid phenylalanine and include a variety of hydroxycinnamoyl derivatives, anthocyanins and flavonols, as well as other low molecular weight conjugates. Another product of phenylpropanoid metabolism is lignin, a major component of the cell wall that provides rigidity and cell wall hydrophobicity to the vasculature. Plants belonging to Brassicales produce glucosinolates, which function as defense molecules by deterring herbivores and pathogens. Over 100 different types of glucosinolates have been found in plants (Halkier, 2016; Agerbirk & Olsen, 2012). Glucosinolates that are derived from tyrosine and phenylalanine are called aromatic glucosinolates whereas glucosinolates derived from tryptophan or from aliphatic amino acids are called indole or aliphatic glucosinolates, respectively (Halkier & Gershenzon, 2006). It has been shown that both the phenylpropanoid and glucosinolate biosynthetic pathways can be regulated at various levels including the transcriptional, and post-translational levels, as well as at the metabolic level through feed-forward and feed-back enzyme inhibition (Yin et al., 2012; Nintemann et al., 2017). In addition to the regulation of their own pathways, there is crosstalk between the glucosinolate and phenylpropanoid pathways (Hemm et al., 2003; Kim et al., 2015). Studies with two Arabidopsis mutants, reduced epidermal fluorescence2 (ref2) and ref5, demonstrated that the accumulation of glucosinolate biosynthetic intermediates can limit the production of phenylpropanoids (Hemm et al., 2003; Kim et al., 2015). The ref5 and ref2 mutants were originally isolated from a screen that was aimed at the isolation of phenylpropanoid-deficient mutants (Ruegger & Chapple, 2001). Wild-type Arabidopsis plants fluoresce blue upon ultra violet (UV) light exposure due to the accumulation of sinapoylmalate, a UV-protective phenylpropanoid in the leaf epidermis (Ruegger & Chapple, 2001; Landry et al., 1995). Plants having defects in sinapoylmalate accumulation fluoresce red because the absence of sinapoylmalate enhances and reveals UV-induced chlorophyll fluorescence in the mesophyll cells. Using this characteristic color difference, a series of ref mutants were isolated, and their characterization led to the identification of several key enzymes and regulators of phenylpropanoid biosynthesis (Meyer et al., 1996; Ruegger & Chapple, 2001; Franke et al., 2002; Stout et al., 2008; Schilmiller et al., 2009; Bonawitz et al., 2012). Unexpectedly, genetic mapping studies revealed that ref2 and ref5 mutants have mutations in CYP83A1 (REF2) (Hemm et al., 2003), and CYP83B1 (REF5) (Kim et al., 2015), which are involved in aliphatic and indole glucosinolate biosynthesis, respectively. Consistent with the biochemical functions of CYP83A1 and CYP83B1 in the synthesis of aliphatic and indolic glucosinolates, ref2 and ref5 mutants also show glucosinolate deficiency phenotypes, but the absence of these compounds does not lead to alterations in phenylpropanoid metabolism. Instead, biochemical and genetic analyses revealed that the accumulation of aldoximes or their derivatives negatively impacts the phenylpropanoid pathway (Kim et al., 2015).
When the production of indole-3-acetaldoxime (IAOx), the substrate of REF5/CYP83B1, was blocked by knocking out the two enzymes that are responsible for converting tryptophan to IAOx, CYP79B2 and CYP79B3, indole glucosinolates were eliminated and the phenylpropanoid content of ref5 was increased (Kim et al., 2015). These data suggest that IAOx, or a derivative thereof, is responsible for the crosstalk between the glucosinolate and phenylpropanoid pathways. In agreement with this model, increased production of IAOx induced by overexpression of CYP79B2 in wild type represses the production of phenylpropanoids (Kim et al., 2015). Importantly, cyp79b2 cyp79b3 double mutants, which lack IAOx, have elevated levels of soluble phenylpropanoids, indicating that this cross-pathway interaction occurs in wild-type plants.
A ref5 suppressor screen revealed that MED5a, a subunit of the transcriptional co-regulator Mediator complex, is involved in the crosstalk between glucosinolate and phenylpropanoid metabolism. Arabidopsis has two MED5 paralogs, MED5a and MED5b (REF4)(Bonawitz et al., 2012, 2014). med5a/b double mutants contain substantially increased levels of phenylpropanoids and exhibit increased expression of several phenylpropanoid biosynthesis genes whereas med5a and med5b single mutants exhibit either no obvious phenotype or a more modest phenotype compared to med5a/b (Bonawitz et al., 2012, 2014). Although the accumulation of aldoximes or their derivatives represses the early step(s) of phenylpropanoid pathway through MED5, it is still not clear how aldoxime metabolism affects the phenylpropanoid pathway.
To understand the molecular mechanisms underlying the crosstalk, we conducted transcriptome profiling using a suite of glucosinolate-deficient mutants. Here we report that RNAseq analysis identified 71 candidate genes that are mis-regulated in ref5 and ref2 in a MED5-dependent manner. We show that the expression of a set of F-box genes functioning in PAL ubiquitination is increased in both ref5 and ref2. Consistent with the known function of these F-box proteins, PAL activity and the amount of PAL protein are reduced in ref5 and ref2 and disruption of these F-box genes restores the phenylpropanoid deficiency in the mutants.
Materials and Methods
Plant Material and Growth Conditions
Arabidopsis thaliana (ecotype Col-0) plants were cultivated at a light intensity of 100 μE m−2 sec−1 at 22°C under a photoperiod of 16 hr light/8 hr dark. Seeds were held for 2 days at 4°C before transferring to the growth chamber. The ref5-1 allele (ref5) has a missense mutation in CYP83B1 (Kim et al., 2015), the ref2-1 allele (ref2) has a nonsense mutation in CYP83A1 (Hemm et al., 2003) and the med5 mutant is med5a med5b double mutant which are T-DNA insertion lines (Bonawitz et al., 2012; Kim et al., 2015). The kfb1-1, kfb20-1 and kfb50-1 alleles were previously reported (Zhang et al., 2013). The kfb39 mutant and the myb4 mutant were generated using CRISPR/Cas9 system (see below for detail information) (Panda et al., unpublished). The homozygosity of all mutants used in this study was confirmed by following their genotyping methods as were previously reported (Panda C Wager A Chen H-Y Xu Li; Hemm et al., 2003; Bonawitz et al., 2012; Zhang et al., 2013; Kim et al., 2015). The homozygosity of kfb39 mutant was confirmed by the presence or absence of DdeI site in the CRISPR/Cas9 target sequences (see below for detail information).
Metabolite analysis
For soluble metabolite analyses, leaves from three-week old plants were extracted with 50% methanol (v/v) at 65°C for 90 minutes at a tissue concentration of 100 mg mL −1. Samples were centrifuged at 16000xg for minutes. 10 μL of extraction was loaded on a Shim-pack XR-ODS column (Shimadzu, Japan) and separated at an increasing concentration of acetonitrile from 2% to 25% for 18 min in 0.1% formic acid at a flow rate of 0.8 mL min−1. Compounds were identified based on their retention times and UV spectra. Sinapoylmalate was quantified using sinapic acid as a standard.
High-throughput mRNA sequencing
For RNAseq analysis, we collected three biological replicates of whole rosette leaves from 18-day-old plants. Total RNA was prepared using the Qiagen RNeasy Plant Mini Kit (Qiagen Sciences) following the manufacturere's instructions. RNA samples were treated with RNase-free DNase Set (Qiagen Sciences) according to the manufacturer's instructions. Total RNA was submitted to the Purdue Genomics Core Facility (Purdue University) for library construction and sequencing. cDNA libraries for sequencing were prepared using a TruSeq Stranded mRNA Library Prep Kit (Illumina) according to the manufacturer’s instructions but 8 cycles of PCR amplification were performed rather than the recommended 15. All 18 libraries were then titrated using a KAPA quantification kit (Kapa Biosystems) and added to a single library pool. That pool was denatured and clustered on two lanes of Illumina High Output Chemistry flow cells at 16.5 pM. Paired-end, 100-bp sequencing was performed by an Illumina HiSeq 2500 in rapid mode. To pre-process filtered Illumina reads for mapping, sequences were first assessed for quality using FastQC (v.0.11.2; http://www.bioinformatics.babraham.ac.uk), then trimmed using the Trimmomatic toolkit (v.0.32; http://www.usadellab.org/cms/?page=trimmomatic) to remove bases with a Phred33 score less than 20 and a minimum length of 30. Quality-trimmed reads were then mapped to the bowtie2- indexed Arabidopsis genome (Bowtie ver 2.2.5.0 and TAIR10) using Tophat (v.2.0.13; http://tophat.cbcb.umd.edu/) with default parameters (Trapnell et al., 2009). A counts matrix consisting of the raw read count was generated using HTSeq (v.0.6.1; http://www-huber.embl.de/users/anders/HTSeq/) (Anders et al., 2015). To eliminate fold changes of infinity and divide-by-zero errors, the counts matrix was modified such that genes with 0 counts across all samples were removed, and remaining 0 count values were changed to 1. Differential expression analysis was performed using the statistical program R (v.3.2.0; http://www.r-project.org/) in conjunction with three analytical methods available from Bioconductor (http://www.bioconductor.org): DESeq2 (v.1.8.1), edgeR (v.3.10.2) and the Cufflinks (v.2.2.1) suite of programs (http://cufflinks.cbcb.umd.edu/). All three statistical methods gave similar overall conclusions. We selected the most conservative results (edgeR; false discovery rate, 0.05) for further investigation and reporting (Robinson et al., 2010). Venn diagrams were adapted from those created with the online tool Venny (v.2.0.2; http://bioinfogp.cnb.csic.es/tools/venny/) and gene ontology term analysis was performed using the online tool DAVID (v.6.7; http://david.ncifcrf.gov)(Huang et al., 2009b; Huang et al., 2009a). The RNAseq data used in this study have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al., 2002) under GEO Series accession number GSE99581.
Generation of kfb39 knock-out mutant using CRISPR/Cas9 system
To generate kfb39 mutants, we used the psgRNA-Cas9-KFB39 vector which was previously reported to generate Arabidopsis mutants (Feng et al., 2014). The psgRNA-CAS9 vector was cut with BbsI and then a KFB39 target sequence was inserted into a cut psgRNA-CAS9 vector using the primer pair CC4841 (5’- GATTGCA TCG CGT TGC GTT TCC AG-3’) and CC4842 (5’- AAACCT GGA AAC GCA ACG CGA TGC-3’). The resulting KFB39 target sequence together with guide RNA and CAS9 was subsequently cloned into pCAMBIA1300 vector after cutting with HindIII and KpnI. The pCAMBIA1300 vector carrying KFB39-CRISPR construct was introduced into ref5 and ref2 plants via Agrobacterium tumefaciens (strain C58 pGV3850) mediated transformation. To select hygromycin resistant transformants, T1 seeds were planted on MS media containing hygromycin after surface sterilization. Hygromycin-resistant T1 transgenic Arabidopsis plants were transferred to soil and genomic DNA was extracted from leaves to identify mutations in KFB39. The target regions were amplified by PCR with the primer pair CC4964 (5’- TGC TCA ATC AAA CAC ACT CTC T-3’) and CC4928 (5’- GTT TCC CAT CGA CGT TCA CT -3’) and the PCR products were sequenced to identify mutations. We found homozygous mutants carrying ‘T’ insertion at the 3 bases upstream of PAM site which causes frame shift resulting in truncated KFB39 protein with 268 amino acids instead of 410 amino acids. Homozygosity of kfb39 mutation in ref5 kfb39 and ref2 kfb39 was confirmed with DdeI digestion because the ‘T’ insertion generates a DdeI recognition site.
PAL activity and PAL protein quantification
For crude protein extraction, 3rd and 4th rosette leaves from 3-week old seedlings were harvested and ground in liquid nitrogen. 500 μL of extraction buffer containing 100 mM Tris-HCl pH 8.3, 2 mM dithiothreitol, and 10% (v/v) glycerol was added and the mixture was stirred at 4°C for 1 hour. After centrifugation at 10,000xg for 10 min at 4°C, the supernatant was collected, desalted into extraction buffer on a column filled with Sephadex G-50 (GE Healthcare Life Science, Pittsburgh, USA) and assayed for PAL activity. Total protein content was measured using Quick Start Bradford protein assay kit (Bio-Rad, Hercules, CA USA) using bovine serum albumin as a standard. To measure the activity of PAL, 100 μL aliquots of the protein extracts were incubated with 4 mM Phe in reaction buffer (100 mM Tris-HCl pH 8.3, 2 mM DTT, 10% (v/v) glycerol) in a total volume of 280 μL. Reactions were incubated at 37°C for 60 min and then stopped by addition of 20 μL of glacial acetic acid. Reaction products were extracted with 400 μL ethyl acetate, 300 μL of which was removed, dried in a speed vac and redissolved in 50 μL of 50% methanol. 10 μL of the final extract was analyzed by HPLC to quantify the reaction product, cinnamic acid. The PAL activity and protein quantification shown in Figure 5a were performed with whole rosette leaves from three-week old plants by following previous reports (Zhang et al., 2015).
Results
Disruption of MED5 rescues the phenylpropanoid deficiency in ref5 and ref2 mutants
As shown in Fig. 1a and b, ref5 and ref2 mutants appear reddish under UV, whereas wild-type plants fluoresce blue due to the accumulation of sinapoylmalate. Consistent with their ref phenotype, both ref5 and ref2 mutants contain reduced levels of sinapoylmalate compared to wild type (Fig. 1b). ref5 med5 (ref5 med5a/b) restores the phenylpropanoid accumulation in ref5 as was previously reported (Fig. 1a, b) (Kim et al., 2015). Because CYP83A1/REF2 and CYP83B1/REF5 functions similarly in the glucosinolate biosynthetic pathway (Hemm et al., 2003; Kim et al., 2015), we speculated that the disruption of MED5 would rescue the phenylpropanoid deficiency of the ref2 mutant. To test this hypothesis, ref2 med5 (ref2 med5a/b) plants were generated and their biochemical phenotype was assessed. The ref2 med5 mutant shows bluish fluorescence under UV and the level of sinapoylmalate in ref2 med5 is increased compared to the ref2 single mutant, indicating an involvement of MED5 in the crosstalk in ref2 (Fig. 1a, b). It is important to note, however, that the sinapoylmalate content of ref5 med5 and ref2 med5 is less than that of med5, suggesting that there may be MED5-dependent and -independent mechanisms involved in glucosinolate/phenylpropanoid crosstalk. Alternatively, given the repressive effect of MED5 on the phenylpropanoid pathway, we continued to consider the possibility that the biochemical phenotype of the med5 mutations in some way bypasses the phenylpropanoid deficiency of ref5 and ref2 mutants and that MED5 is not actually involved in this interaction. Nevertheless, because the disruption of MED5 restores the phenylpropanoid deficiency in both ref5 and ref2 mutants, we hypothesized that there is a common mechanistic basis for how glucosinolate metabolites affect phenylpropanoid production that involves MED5 (Fig. 1c).
Transcriptome profiles identify mis-regulated genes in glucosinolate-deficient mutants
Given that MED5 is a subunit of a transcriptional co-regulator, it seemed likely that it functions in the crosstalk by influencing the expression of genes that play roles in phenylpropanoid metabolism. To explore this possibility, we performed RNAseq analysis using total RNA prepared from whole rosette leaves of wild type (Col-0), ref2, ref5, med5, ref2 med5, and ref5 med5 to identify genes that are mis-regulated in ref5 and ref2 but not in ref5 med5 and ref2 med5. RNAseq analysis identified that 1074 genes were up-regulated in ref5 compared to wild type and 2423 genes were up-regulated in ref2 compared to wild type. Among these, 402 genes were increased in both ref5 and ref2 mutants compared to wild type (Fig. 2a), 53 of which show restored expression in the absence of MED5 (Fig. 2a, 2c, Table 1). Similarly, we found that 18 genes are down-regulated in both ref5 and ref2 compared to wild type in a MED5-dependent manner (Fig. 2b, 2d and Table 2). Thus, this analysis identified a total of 71 candidate genes that may be involved in the MED5-dependent crosstalk between glucosinolate and phenylpropanoid biosynthesis (Table 1, 2).
Altered expression of MYB4, BRT1 and HCT does not play a major role in glucosinolate/phenylpropanoid crosstalk
Among the 71 candidate genes, two phenylpropanoid biosynthetic genes, hydroxycinnamoyl-CoA shikimate/quinate hydroxycinnamoyl transferase (HCT) and Bright Trichomes 1 (BRT1) and one phenylpropanoid-related transcription factor, MYB4, were identified (Table 1, 2, Fig. 3a-c). The expression of HCT was decreased whereas the expression of BRT1 was increased in ref5 and ref2 (Fig. 3a-b). BRT1 (At3g21560) encodes UGT84A2, a glucosyltransferase previously shown to have activity toward sinapic acid as a substrate (Sinlapadech et al., 2007). Like the ref5 and ref2 mutants, the brt1 mutant accumulates less sinapoylmalate than wild type. Since the expression of BRT1 was increased in both ref5 and ref2, it seemed unlikely that the enhanced expression of BRT1 is directly involved in phenylpropanoid repression in ref5 and ref2 (Fig. 2c, d, Fig. 3a).
HCT catalyzes the synthesis of p-coumaroyl shikimate in the lignin biosynthetic pathway (Hoffmann et al., 2004; Li et al., 2010). The expression of HCT was reduced by about 15% and 30% in ref5 and ref2 mutants, respectively (Fig. 3b). We also excluded this gene from further consideration because previous studies with HCT RNAi lines showed that obvious metabolic phenotypes were not observed until HCT activity was reduced to less than 2% of the wild-type value (Hoffmann et al., 2004; Li et al., 2010). In addition, the repression of HCT results in the accumulation of flavonoids but ref5 and ref2 mutants accumulate less flavonoids than wild type (Fig. S1) (Hoffmann et al., 2004; Li et al., 2010; Kim et al., 2015).
MYB4 is a transcriptional factor the function of which includes the repression of C4H expression(Jin et al., 2000). C4H expression was reduced about 20% in ref2 but it was not altered in ref5 compared to wild type (Fig. 3d). To further evaluate the possible impact of altered MYB4 expression, we generated ref5 myb4 and ref2 myb4 double mutants. In agreement with the function of MYB4, myb4 mutants produced more sinapoylmalate than wild type (Fig. 3e) (Jin et al., 2000; Panda et al., unpublished) but the level of sinapoylmalate in ref5 myb4 and ref2 myb4 mutants was not different from that in ref5 and ref2 mutants (Fig. 3e). Taken together, we concluded that the altered expression of MYB4 is not related to phenylpropanoid repression in ref5 and ref2.
The expression of KFBs functioning in PAL degradation is affected in ref5 and ref2
Besides HCT, BRT1 and MYB4, two additional genes, At3g59930 and At2g44130, caught our attention due to their known functions in phenylpropanoid metabolism. At3g59930 and At2g44130 encode the kelch-domain containing F-box (KFB) proteins KFB39 and KFB50, respectively, both of which are involved in the ubiquitination and proteasome-mediated degradation of PAL (Zhang et al., 2015; Zhang et al., 2013). As shown in Fig. 4, the expression of KFB39 and KFB50 in both ref5 and ref2 mutants was increased two- to four-fold compared to wild type and restored in ref5 med5 and ref2 med5. It is notable that the expression of KFB39 and KFB50 was significantly reduced in med5 compared to wild type (Fig. 4a, b) indicating that MED5 is required for the activation of KFB39 and KFB50 expression not only in ref5 and ref2 mutants but also in wild type. Two other KFBs, KFB1 and KFB20, function redundantly with KFB39 and KFB50 in PAL ubiquitination (Zhang et al., 2013). We found that the expression of KFB1 and KFB20 were generally increased in ref5 and ref2 mutants compared to wild type but not returned to normal in ref5 med5 and ref2 med5 (Fig. 4c, d). These data suggest that there may be MED5-dependent and MED5-independent mechanisms for the regulation of KFB expression.
In addition to the role of these four KFBs in phenylpropanoid metabolism, KFBCHS is involved in the ubiquitination of chalcone synthase (CHS), an enzyme required for flavonoid production (Zhang et al., 2017). We found that although ref5 and ref2 mutants showed reduced flavonoid content (Fig. S1), the expression of KFBCHS was not affected (Fig. S2). This suggests that aldoxime-induced misregulation of KFBs is restricted to specific targets in these mutants.
PAL activity and the amount of PAL protein are decreased in ref5 and ref2 mutants
Because the four KFBs we identified play a role in PAL degradation, we hypothesized that ref5 and ref2 mutants have enhanced PAL turnover, leading to the repression of PAL activity and phenylpropanoid production. Indeed, we found that PAL activity and protein levels were slightly reduced in ref5 and ref2 mutants and enhanced by elimination of MED5 (Fig 5a, b) but PAL protein and enzyme levels were lower in ref2 med5 and ref5 med5 than in med5 alone, again indicating that MED5-independent mechanisms may also play a role in the crosstalk. Although PAL expression is increased in the med5 mutant (Bonawitz et al., 2014) (Fig 5c-f), it was affected only modestly in the ref2 and ref5 mutants suggesting that reduced PAL activity and PAL protein level in ref2 and ref5 (Fig 5a-b) is largely due to KFB-dependent PAL degradation. To evaluate this hypothesis, we generated multiple mutants combining ref2 and ref5 with KFB knockout lines. Elimination of all four KFB genes led to a substantial increase in PAL activity that was unaffected by the introduction of the ref2 or ref5 mutations into the same genetic background (Fig 6) indicating that the reduction of PAL activity in these mutants is due to KFB-mediated protein turnover. In contrast, PAL activity in ref5 kfb1/20/50 or ref2 kfb1/20/50 quadruple mutants was significantly lower than that in kfb1/20/50 triple mutant suggesting that the accumulation of aldoximes in ref2 and ref5 affects PAL activity through KFB39 (Fig 6).
KFB-mediated PAL degradation is one of multiple mechanisms involved in the repression of phenylpropanoid production in ref5 and ref2
Given that disruption of KFBs restored PAL activity to ref2 and ref5, we tested whether this restoration would rescue the reduced phenylpropanoid production in ref5 and ref2 (Fig. 7). Because the expression of only KFB39 and KFB50 is increased in ref5 and ref2 in a MED5-dependent manner (Fig.4), we first tested the impact of these two genes on phenylpropanoid production in ref5. The knockout of either KFB39 or KFB50 alone did not affect sinapoylmalate accumulation in ref5 mutant (Fig. 7a). When both KFB39 and KFB50 were eliminated, sinapoylmalate content was slightly increased compared to the ref5 mutant but its level was still less than wild type (Fig. 7a). When all four of the KFBs were removed, sinapoylmalate content in the ref5 and ref2 mutant backgrounds was restored to the wild-type level (Fig. 7b) but was not as high as that in kfb1/20/39/50 quadruple mutant (Fig. 7b). Considering that there is no statistical significance among PAL activities of these mutants (Fig. 6a, b), in addition to PAL, there must be other step(s) of the phenylpropanoid pathway that are affected in ref5 and ref2 mutants in addition to PAL.
Discussion
Glucosinolates and phenylpropanoids are secondary metabolites that are made through unique biosynthetic pathways. The first step of glucosinolate core structure formation is the production of aldoximes, which are subsequently used as substrates by CYP83A1/REF2 or CYP83B1/REF5 in Arabidopsis. Previous studies with two Arabidopsis mutants have demonstrated an unexpected link between glucosinolate and phenylpropanoid production (Hemm et al., 2003; Kim et al., 2015). It appears that the accumulation of aldoximes or their derivatives limits phenylpropanoid production (Hemm et al., 2003; Kim et al., 2015). Because aldoximes are precursors of many other biomolecules besides glucosinolates, including cyanogenic glucosides and nitriles, and can be found in diverse plant species such as maize, poplar, yew and coca (Irmisch et al., 2013, 2015; Luck et al., 2017; Sørensen et al., 2018; Bhalla et al., 2018; Dhandapani et al., 2019), aldoxime-mediated suppression of phenylpropanoid metabolism may be relevant to other species and metabolic pathways beyond Arabidopsis.
Previous work showed that glucosinolate/phenylpropanoid crosstalk requires MED5 (Kim et al., 2015) but its underlying mechanism was not identified. In this study, we demonstrate that the crosstalk involves transcriptional regulation of four KFB genes functioning in PAL ubiquitination and that MED5 is required for the upregulation of two of them. Interestingly, MED5 is only required for the increased expression of KFB39 and KFB50 in the context of the ref5 and ref2 mutations, but not for the expression of KFB1 and KFB20 (Fig 4). Apparently, there exist distinct regulatory mechanisms between the two sub families of KFB proteins even though they function redundantly in PAL degradation. In contrast, the expression of KFBCHS, which functions in flavonoid biosynthesis, (Zhang et al., 2017) was not affected in the high aldoxime containing mutants, although both showed reduced flavonoid contents indicating that aldoxime metabolism affects a specific group of KFBs and not others (Kim et al., 2015) (Fig. S1, S2). Instead, KFB-mediated PAL degradation and the ensuing reduction in PAL activity is a more likely explanation for the reduced flavonoid content of ref5 and ref2.
We previously suspected that the influence of aldoxime metabolites on phenylpropanoid metabolism might be just a pharmacological effect rather than a biologically relevant regulatory mechanism; however, the data presented here suggest that there may be a more sophisticated function for this crosstalk. First, at least one aspect of the crosstalk requires a specific subunit of the transcriptional regulatory complex Mediator (Kim et al., 2015) which enhances the expression of KFB39 and KFB50 and thereby reduces PAL activity (Fig. 4a, b). Second, this regulatory circuit affects a certain group of KFB genes but not others, indicating some level of specificity to the process (Fig. 4, Fig. S2). Finally, a certain level of aldoxime-sensitivity is maintained in KFB-deficient lines, indicating that disruptions in glucosinolate metabolism impinge on phenylpropanoid metabolism in multiple ways. Further study will be required to elucidate how plants sense aldoximes or their derivatives and transduce signals to regulate gene expression and enzyme activity.
The phenylpropanoid pathway is subject to multiple layers of regulation at various steps. As PAL is generally considered to be the first rate-limiting enzyme in the phenylpropanoid pathway, the transcriptional and post-translational modification regulation of its activity can modulate the production of a variety of phenolic compounds. For example, developmental cues and environmental stresses affect PAL expression and its activity is highly correlated with PAL transcript abundance (Olsen et al., 2008; Huang et al., 2010; Zhang et al., 2013). Further, ubiquitination or phosphorylation of PAL can affect its stability or specific subcellular localization (Zhang et al., 2013, 2015; Allwood et al., 1999; Cheng et al., 2001). Finally, PAL is subject to feedback regulation through biosynthetic intermediates (Blount et al., 2000; Bolwell et al., 1986). Cinnamic acid, a product of PAL can influence its expression and activity (Mavandad et al., 1990; Blount et al., 2000) and various flavonoids including naringenin, a precursor of flavonoids, can inhibit its activity (Sato & Sankawa, 1983; Yin et al., 2012). Our data now suggest that an intermediate in glucosinolate biosynthesis or a derivative thereof, made through an independent pathway, influences PAL activity by regulating its protein stability.
The fact that KFBs are at least partially responsible for the phenylpropanoid deficiency in ref5 and ref2, and that REF5 and REF2 act in parallel glucosinolate pathways suggests that there may be a common metabolite or class of metabolite that triggers the crosstalk in these mutants. Although both CYP83A1/REF2 and CYP83B1/REF5 have activity toward various aldoximes, they show distinctive substrate specificity (Naur et al., 2003). CYP83A1/REF2 mainly metabolizes the aliphatic oximes derived from chain-elongated homologs of methionine. In contrast, CYP83B1/REF5 has a higher affinity for indole-3-acetaldoxime (IAOx), with a 50-fold difference in Km between CYP83A1/REF2 and CYP83B1/REF5 for IAOx (Naur et al., 2003). Consistently, plants having defects in CYP83B1/REF5 exhibit characteristic phenotypes that are associated with increased level of IAOx-derived auxin (Boerjan et al., 1995; Wagner et al., 1997; Barlier et al., 2000; Bak & Feyereisen, 2001; Kim et al., 2015), whereas the ref2 mutant does not (Hemm et al., 2003). Thus, it seems improbable that the level of IAOx or its derivatives is increased in ref2. Interestingly, Arabidopsis transgenic plants overexpressing sorghum CYP79A1, or both CYP79A1 and CYP71E1 together showed reduced production of phenylpropanoids such as sinapoylglucose, sinapoylmalate and flavonoids (Kristensen et al., 2005). Since sorghum CYP79A1 has specific activity toward tyrosine to produce p-hydroxyphenylacetaldoxime (Kristensen et al., 2005), the accumulation of p-hydroxyphenylacetaldoxime or its derivatives may also affect phenylpropanoid production. Given that the altered level of various types of aldoxime results in similar impact on phenylpropanoid production, it is possible that a common structure of the oximes, rather than the precise identity of the amino acid R-group from which the compound was derived may be the factor important for the crosstalk.
A recent study identified scaffolding proteins that structure the monolignol biosynthetic P450 enzyme complex (monolignol metabolon) in the endoplasmic reticulum membrane (Gou et al., 2018). This complex appears to function in regulating monolignol biosynthesis but not flavonoid synthesis (Gou et al., 2018). It has been proposed that metabolic complexes of dhurrin and glucosinolate biosynthetic enzymes play a role in optimizing the production of these phytochemicals with the CYP79 aldoxime-producing enzymes and the CYP71E1 or CYP83 aldoxime-consuming enzymes forming similar complexes on the endoplasmic reticulum membrane (Møller, 2010). Considering that aldoximes or their derivatives can influence phenylpropanoid production by affecting PAL, the evolution of these complexes may have been important not only to increase local concentration of metabolites but also to reduce leakage of biologically active metabolites that can affect other pathways.
In summary, our transcriptome analyses and genetic study with a set of glucosinolate deficient mutants reveal that the accumulation of aldoximes or their derivatives affects expression of a specific group of F-box genes, thereby altering the stability of PAL as well as act to negatively affect at least one other step of phenylpropanoid metabolism. Considering that aldoximes are precursors of important biomolecules and the oxime-producing CYP79s are widespread in higher plants (Sørensen et al., 2018), this crosstalk may be relevant to not only glucosinolates but also other metabolic pathways.
Author contributions
J.K., and C.C. conceived and designed the experiments; J.K, and X.Z performed the experiments; J.K, X.Z, P.P., C-J.L, and C.C., analyzed and interpreted the data; J.K. and C.C. wrote the manuscript;
Data availability
The authors declare that all data supporting the findings of this study are included in the manuscript and Supplementary Information files are available from the corresponding authors upon request. RNAseq data have been deposited in NCBI’s Gene Expression Omnibus60 under GEO Series accession number GSE99581.
ACKNOWLEDGMENTS
The authors thank to Dr. Jian-Kang Zhu for generous gift with pCrisper-CAS9 vector and Dr. Xu (Sirius) Li for myb4 mutant seeds. This work was supported as part of the Center for Direct Catalytic Conversion of Biomass to Biofuels, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences under Award # DE-SC0000997.